diff --git a/sql/upgrades/156.sql b/sql/upgrades/156.sql
new file mode 100644
index 00000000..5abcd218
--- /dev/null
+++ b/sql/upgrades/156.sql
@@ -0,0 +1,25 @@
+-- SWH DB schema upgrade
+-- from_version: 155
+-- to_version: 156
+-- description: Make swh_release_add properly idempotent
+
+-- latest schema version
+insert into dbversion(version, release, description)
+      values(156, now(), 'Work In Progress');
+
+-- Create entries in release from tmp_release
+create or replace function swh_release_add()
+    returns void
+    language plpgsql
+as $$
+begin
+    perform swh_person_add_from_release();
+
+    insert into release (id, target, target_type, date, date_offset, date_neg_utc_offset, name, comment, author, synthetic)
+      select distinct t.id, t.target, t.target_type, t.date, t.date_offset, t.date_neg_utc_offset, t.name, t.comment, a.id, t.synthetic
+        from tmp_release t
+        left join person a on a.fullname = t.author_fullname
+        where not exists (select 1 from release where t.id = release.id);
+    return;
+end
+$$;
diff --git a/swh/storage/cassandra/storage.py b/swh/storage/cassandra/storage.py
index c2e91f1e..cbd0b4b2 100644
--- a/swh/storage/cassandra/storage.py
+++ b/swh/storage/cassandra/storage.py
@@ -1,1165 +1,1169 @@
 # Copyright (C) 2019-2020  The Software Heritage developers
 # See the AUTHORS file at the top-level directory of this distribution
 # License: GNU General Public License version 3, or any later version
 # See top-level LICENSE file for more information
 
 import datetime
 import itertools
 import json
 import random
 import re
 from typing import Any, Dict, List, Iterable, Optional
 
 import attr
 from deprecated import deprecated
 
 from swh.core.api.serializers import msgpack_loads, msgpack_dumps
 from swh.model.model import (
     Revision,
     Release,
     Directory,
     DirectoryEntry,
     Content,
     SkippedContent,
     OriginVisit,
     OriginVisitStatus,
     Snapshot,
     Origin,
 )
 from swh.model.hashutil import DEFAULT_ALGORITHMS
 from swh.storage.objstorage import ObjStorage
 from swh.storage.writer import JournalWriter
 from swh.storage.utils import now
 
 from ..exc import StorageArgumentException, HashCollision
 from .common import TOKEN_BEGIN, TOKEN_END
 from .converters import (
     revision_to_db,
     revision_from_db,
     release_to_db,
     release_from_db,
     row_to_visit_status,
 )
 from .cql import CqlRunner
 from .schema import HASH_ALGORITHMS
 
 
 # Max block size of contents to return
 BULK_BLOCK_CONTENT_LEN_MAX = 10000
 
 
 class CassandraStorage:
     def __init__(self, hosts, keyspace, objstorage, port=9042, journal_writer=None):
         self._cql_runner = CqlRunner(hosts, keyspace, port)
         self.journal_writer = JournalWriter(journal_writer)
         self.objstorage = ObjStorage(objstorage)
 
     def check_config(self, *, check_write):
         self._cql_runner.check_read()
 
         return True
 
     def _content_get_from_hash(self, algo, hash_) -> Iterable:
         """From the name of a hash algorithm and a value of that hash,
         looks up the "hash -> token" secondary table (content_by_{algo})
         to get tokens.
         Then, looks up the main table (content) to get all contents with
         that token, and filters out contents whose hash doesn't match."""
         found_tokens = self._cql_runner.content_get_tokens_from_single_hash(algo, hash_)
 
         for token in found_tokens:
             # Query the main table ('content').
             res = self._cql_runner.content_get_from_token(token)
 
             for row in res:
                 # re-check the the hash (in case of murmur3 collision)
                 if getattr(row, algo) == hash_:
                     yield row
 
     def _content_add(self, contents: List[Content], with_data: bool) -> Dict:
         # Filter-out content already in the database.
         contents = [
             c for c in contents if not self._cql_runner.content_get_from_pk(c.to_dict())
         ]
 
         self.journal_writer.content_add(contents)
 
         if with_data:
             # First insert to the objstorage, if the endpoint is
             # `content_add` (as opposed to `content_add_metadata`).
             # TODO: this should probably be done in concurrently to inserting
             # in index tables (but still before the main table; so an entry is
             # only added to the main table after everything else was
             # successfully inserted.
             summary = self.objstorage.content_add(
                 c for c in contents if c.status != "absent"
             )
             content_add_bytes = summary["content:add:bytes"]
 
         content_add = 0
         for content in contents:
             content_add += 1
 
             # Check for sha1 or sha1_git collisions. This test is not atomic
             # with the insertion, so it won't detect a collision if both
             # contents are inserted at the same time, but it's good enough.
             #
             # The proper way to do it would probably be a BATCH, but this
             # would be inefficient because of the number of partitions we
             # need to affect (len(HASH_ALGORITHMS)+1, which is currently 5)
             for algo in {"sha1", "sha1_git"}:
                 collisions = []
                 # Get tokens of 'content' rows with the same value for
                 # sha1/sha1_git
                 rows = self._content_get_from_hash(algo, content.get_hash(algo))
                 for row in rows:
                     if getattr(row, algo) != content.get_hash(algo):
                         # collision of token(partition key), ignore this
                         # row
                         continue
 
                     for algo in HASH_ALGORITHMS:
                         if getattr(row, algo) != content.get_hash(algo):
                             # This hash didn't match; discard the row.
                             collisions.append(
                                 {algo: getattr(row, algo) for algo in HASH_ALGORITHMS}
                             )
 
                 if collisions:
                     collisions.append(content.hashes())
                     raise HashCollision(algo, content.get_hash(algo), collisions)
 
             (token, insertion_finalizer) = self._cql_runner.content_add_prepare(content)
 
             # Then add to index tables
             for algo in HASH_ALGORITHMS:
                 self._cql_runner.content_index_add_one(algo, content, token)
 
             # Then to the main table
             insertion_finalizer()
 
         summary = {
             "content:add": content_add,
         }
 
         if with_data:
             summary["content:add:bytes"] = content_add_bytes
 
         return summary
 
     def content_add(self, content: Iterable[Content]) -> Dict:
         contents = [attr.evolve(c, ctime=now()) for c in content]
         return self._content_add(list(contents), with_data=True)
 
     def content_update(self, content, keys=[]):
         raise NotImplementedError(
             "content_update is not supported by the Cassandra backend"
         )
 
     def content_add_metadata(self, content: Iterable[Content]) -> Dict:
         return self._content_add(list(content), with_data=False)
 
     def content_get(self, content):
         if len(content) > BULK_BLOCK_CONTENT_LEN_MAX:
             raise StorageArgumentException(
                 "Sending at most %s contents." % BULK_BLOCK_CONTENT_LEN_MAX
             )
         yield from self.objstorage.content_get(content)
 
     def content_get_partition(
         self,
         partition_id: int,
         nb_partitions: int,
         limit: int = 1000,
         page_token: str = None,
     ):
         if limit is None:
             raise StorageArgumentException("limit should not be None")
 
         # Compute start and end of the range of tokens covered by the
         # requested partition
         partition_size = (TOKEN_END - TOKEN_BEGIN) // nb_partitions
         range_start = TOKEN_BEGIN + partition_id * partition_size
         range_end = TOKEN_BEGIN + (partition_id + 1) * partition_size
 
         # offset the range start according to the `page_token`.
         if page_token is not None:
             if not (range_start <= int(page_token) <= range_end):
                 raise StorageArgumentException("Invalid page_token.")
             range_start = int(page_token)
 
         # Get the first rows of the range
         rows = self._cql_runner.content_get_token_range(range_start, range_end, limit)
         rows = list(rows)
 
         if len(rows) == limit:
             next_page_token: Optional[str] = str(rows[-1].tok + 1)
         else:
             next_page_token = None
 
         return {
             "contents": [row._asdict() for row in rows if row.status != "absent"],
             "next_page_token": next_page_token,
         }
 
     def content_get_metadata(self, contents: List[bytes]) -> Dict[bytes, List[Dict]]:
         result: Dict[bytes, List[Dict]] = {sha1: [] for sha1 in contents}
         for sha1 in contents:
             # Get all (sha1, sha1_git, sha256, blake2s256) whose sha1
             # matches the argument, from the index table ('content_by_sha1')
             for row in self._content_get_from_hash("sha1", sha1):
                 content_metadata = row._asdict()
                 content_metadata.pop("ctime")
                 result[content_metadata["sha1"]].append(content_metadata)
         return result
 
     def content_find(self, content):
         # Find an algorithm that is common to all the requested contents.
         # It will be used to do an initial filtering efficiently.
         filter_algos = list(set(content).intersection(HASH_ALGORITHMS))
         if not filter_algos:
             raise StorageArgumentException(
                 "content keys must contain at least one of: "
                 "%s" % ", ".join(sorted(HASH_ALGORITHMS))
             )
         common_algo = filter_algos[0]
 
         results = []
         rows = self._content_get_from_hash(common_algo, content[common_algo])
         for row in rows:
             # Re-check all the hashes, in case of collisions (either of the
             # hash of the partition key, or the hashes in it)
             for algo in HASH_ALGORITHMS:
                 if content.get(algo) and getattr(row, algo) != content[algo]:
                     # This hash didn't match; discard the row.
                     break
             else:
                 # All hashes match, keep this row.
                 results.append(
                     {
                         **row._asdict(),
                         "ctime": row.ctime.replace(tzinfo=datetime.timezone.utc),
                     }
                 )
         return results
 
     def content_missing(self, content, key_hash="sha1"):
         for cont in content:
             res = self.content_find(cont)
             if not res:
                 yield cont[key_hash]
             if any(c["status"] == "missing" for c in res):
                 yield cont[key_hash]
 
     def content_missing_per_sha1(self, contents):
         return self.content_missing([{"sha1": c for c in contents}])
 
     def content_missing_per_sha1_git(self, contents):
         return self.content_missing(
             [{"sha1_git": c for c in contents}], key_hash="sha1_git"
         )
 
     def content_get_random(self):
         return self._cql_runner.content_get_random().sha1_git
 
     def _skipped_content_get_from_hash(self, algo, hash_) -> Iterable:
         """From the name of a hash algorithm and a value of that hash,
         looks up the "hash -> token" secondary table
         (skipped_content_by_{algo}) to get tokens.
         Then, looks up the main table (content) to get all contents with
         that token, and filters out contents whose hash doesn't match."""
         found_tokens = self._cql_runner.skipped_content_get_tokens_from_single_hash(
             algo, hash_
         )
 
         for token in found_tokens:
             # Query the main table ('content').
             res = self._cql_runner.skipped_content_get_from_token(token)
 
             for row in res:
                 # re-check the the hash (in case of murmur3 collision)
                 if getattr(row, algo) == hash_:
                     yield row
 
     def _skipped_content_add(self, contents: Iterable[SkippedContent]) -> Dict:
         # Filter-out content already in the database.
         contents = [
             c
             for c in contents
             if not self._cql_runner.skipped_content_get_from_pk(c.to_dict())
         ]
 
         self.journal_writer.skipped_content_add(contents)
 
         for content in contents:
             # Compute token of the row in the main table
             (token, insertion_finalizer) = self._cql_runner.skipped_content_add_prepare(
                 content
             )
 
             # Then add to index tables
             for algo in HASH_ALGORITHMS:
                 self._cql_runner.skipped_content_index_add_one(algo, content, token)
 
             # Then to the main table
             insertion_finalizer()
 
         return {"skipped_content:add": len(contents)}
 
     def skipped_content_add(self, content: Iterable[SkippedContent]) -> Dict:
         contents = [attr.evolve(c, ctime=now()) for c in content]
         return self._skipped_content_add(contents)
 
     def skipped_content_missing(self, contents):
         for content in contents:
             if not self._cql_runner.skipped_content_get_from_pk(content):
                 yield {algo: content[algo] for algo in DEFAULT_ALGORITHMS}
 
     def directory_add(self, directories: Iterable[Directory]) -> Dict:
         directories = list(directories)
 
         # Filter out directories that are already inserted.
         missing = self.directory_missing([dir_.id for dir_ in directories])
         directories = [dir_ for dir_ in directories if dir_.id in missing]
 
         self.journal_writer.directory_add(directories)
 
         for directory in directories:
             # Add directory entries to the 'directory_entry' table
             for entry in directory.entries:
                 self._cql_runner.directory_entry_add_one(
                     {**entry.to_dict(), "directory_id": directory.id}
                 )
 
             # Add the directory *after* adding all the entries, so someone
             # calling snapshot_get_branch in the meantime won't end up
             # with half the entries.
             self._cql_runner.directory_add_one(directory.id)
 
         return {"directory:add": len(missing)}
 
     def directory_missing(self, directories):
         return self._cql_runner.directory_missing(directories)
 
     def _join_dentry_to_content(self, dentry):
         keys = (
             "status",
             "sha1",
             "sha1_git",
             "sha256",
             "length",
         )
         ret = dict.fromkeys(keys)
         ret.update(dentry.to_dict())
         if ret["type"] == "file":
             content = self.content_find({"sha1_git": ret["target"]})
             if content:
                 content = content[0]
                 for key in keys:
                     ret[key] = content[key]
         return ret
 
     def _directory_ls(self, directory_id, recursive, prefix=b""):
         if self.directory_missing([directory_id]):
             return
         rows = list(self._cql_runner.directory_entry_get([directory_id]))
 
         for row in rows:
             # Build and yield the directory entry dict
             entry = row._asdict()
             del entry["directory_id"]
             entry = DirectoryEntry.from_dict(entry)
             ret = self._join_dentry_to_content(entry)
             ret["name"] = prefix + ret["name"]
             ret["dir_id"] = directory_id
             yield ret
 
             if recursive and ret["type"] == "dir":
                 yield from self._directory_ls(
                     ret["target"], True, prefix + ret["name"] + b"/"
                 )
 
     def directory_entry_get_by_path(self, directory, paths):
         return self._directory_entry_get_by_path(directory, paths, b"")
 
     def _directory_entry_get_by_path(self, directory, paths, prefix):
         if not paths:
             return
 
         contents = list(self.directory_ls(directory))
 
         if not contents:
             return
 
         def _get_entry(entries, name):
             """Finds the entry with the requested name, prepends the
             prefix (to get its full path), and returns it.
 
             If no entry has that name, returns None."""
             for entry in entries:
                 if entry["name"] == name:
                     entry = entry.copy()
                     entry["name"] = prefix + entry["name"]
                     return entry
 
         first_item = _get_entry(contents, paths[0])
 
         if len(paths) == 1:
             return first_item
 
         if not first_item or first_item["type"] != "dir":
             return
 
         return self._directory_entry_get_by_path(
             first_item["target"], paths[1:], prefix + paths[0] + b"/"
         )
 
     def directory_ls(self, directory, recursive=False):
         yield from self._directory_ls(directory, recursive)
 
     def directory_get_random(self):
         return self._cql_runner.directory_get_random().id
 
     def revision_add(self, revisions: Iterable[Revision]) -> Dict:
         revisions = list(revisions)
 
         # Filter-out revisions already in the database
         missing = self.revision_missing([rev.id for rev in revisions])
         revisions = [rev for rev in revisions if rev.id in missing]
         self.journal_writer.revision_add(revisions)
 
         for revision in revisions:
             revobject = revision_to_db(revision)
             if revobject:
                 # Add parents first
                 for (rank, parent) in enumerate(revobject["parents"]):
                     self._cql_runner.revision_parent_add_one(
                         revobject["id"], rank, parent
                     )
 
                 # Then write the main revision row.
                 # Writing this after all parents were written ensures that
                 # read endpoints don't return a partial view while writing
                 # the parents
                 self._cql_runner.revision_add_one(revobject)
 
         return {"revision:add": len(revisions)}
 
     def revision_missing(self, revisions):
         return self._cql_runner.revision_missing(revisions)
 
     def revision_get(self, revisions):
         rows = self._cql_runner.revision_get(revisions)
         revs = {}
         for row in rows:
             # TODO: use a single query to get all parents?
             # (it might have lower latency, but requires more code and more
             # bandwidth, because revision id would be part of each returned
             # row)
             parent_rows = self._cql_runner.revision_parent_get(row.id)
             # parent_rank is the clustering key, so results are already
             # sorted by rank.
             parents = tuple(row.parent_id for row in parent_rows)
             rev = revision_from_db(row, parents=parents)
             revs[rev.id] = rev.to_dict()
 
         for rev_id in revisions:
             yield revs.get(rev_id)
 
     def _get_parent_revs(self, rev_ids, seen, limit, short):
         if limit and len(seen) >= limit:
             return
         rev_ids = [id_ for id_ in rev_ids if id_ not in seen]
         if not rev_ids:
             return
         seen |= set(rev_ids)
 
         # We need this query, even if short=True, to return consistent
         # results (ie. not return only a subset of a revision's parents
         # if it is being written)
         if short:
             rows = self._cql_runner.revision_get_ids(rev_ids)
         else:
             rows = self._cql_runner.revision_get(rev_ids)
 
         for row in rows:
             # TODO: use a single query to get all parents?
             # (it might have less latency, but requires less code and more
             # bandwidth (because revision id would be part of each returned
             # row)
             parent_rows = self._cql_runner.revision_parent_get(row.id)
 
             # parent_rank is the clustering key, so results are already
             # sorted by rank.
             parents = tuple(row.parent_id for row in parent_rows)
 
             if short:
                 yield (row.id, parents)
             else:
                 rev = revision_from_db(row, parents=parents)
                 yield rev.to_dict()
             yield from self._get_parent_revs(parents, seen, limit, short)
 
     def revision_log(self, revisions, limit=None):
         seen = set()
         yield from self._get_parent_revs(revisions, seen, limit, False)
 
     def revision_shortlog(self, revisions, limit=None):
         seen = set()
         yield from self._get_parent_revs(revisions, seen, limit, True)
 
     def revision_get_random(self):
         return self._cql_runner.revision_get_random().id
 
     def release_add(self, releases: Iterable[Release]) -> Dict:
-        missing = self.release_missing([rel.id for rel in releases])
-        releases = [rel for rel in releases if rel.id in missing]
+        to_add = []
+        for rel in releases:
+            if rel not in to_add:
+                to_add.append(rel)
+        missing = set(self.release_missing([rel.id for rel in to_add]))
+        to_add = [rel for rel in to_add if rel.id in missing]
 
-        self.journal_writer.release_add(releases)
+        self.journal_writer.release_add(to_add)
 
-        for release in releases:
+        for release in to_add:
             if release:
                 self._cql_runner.release_add_one(release_to_db(release))
 
-        return {"release:add": len(missing)}
+        return {"release:add": len(to_add)}
 
     def release_missing(self, releases):
         return self._cql_runner.release_missing(releases)
 
     def release_get(self, releases):
         rows = self._cql_runner.release_get(releases)
         rels = {}
         for row in rows:
             release = release_from_db(row)
             rels[row.id] = release.to_dict()
 
         for rel_id in releases:
             yield rels.get(rel_id)
 
     def release_get_random(self):
         return self._cql_runner.release_get_random().id
 
     def snapshot_add(self, snapshots: Iterable[Snapshot]) -> Dict:
         missing = self._cql_runner.snapshot_missing([snp.id for snp in snapshots])
         snapshots = [snp for snp in snapshots if snp.id in missing]
 
         for snapshot in snapshots:
             self.journal_writer.snapshot_add([snapshot])
 
             # Add branches
             for (branch_name, branch) in snapshot.branches.items():
                 if branch is None:
                     target_type = None
                     target = None
                 else:
                     target_type = branch.target_type.value
                     target = branch.target
                 self._cql_runner.snapshot_branch_add_one(
                     {
                         "snapshot_id": snapshot.id,
                         "name": branch_name,
                         "target_type": target_type,
                         "target": target,
                     }
                 )
 
             # Add the snapshot *after* adding all the branches, so someone
             # calling snapshot_get_branch in the meantime won't end up
             # with half the branches.
             self._cql_runner.snapshot_add_one(snapshot.id)
 
         return {"snapshot:add": len(snapshots)}
 
     def snapshot_missing(self, snapshots):
         return self._cql_runner.snapshot_missing(snapshots)
 
     def snapshot_get(self, snapshot_id):
         return self.snapshot_get_branches(snapshot_id)
 
     def snapshot_get_by_origin_visit(self, origin, visit):
         try:
             visit = self.origin_visit_get_by(origin, visit)
         except IndexError:
             return None
 
         return self.snapshot_get(visit["snapshot"])
 
     def snapshot_count_branches(self, snapshot_id):
         if self._cql_runner.snapshot_missing([snapshot_id]):
             # Makes sure we don't fetch branches for a snapshot that is
             # being added.
             return None
         rows = list(self._cql_runner.snapshot_count_branches(snapshot_id))
         assert len(rows) == 1
         (nb_none, counts) = rows[0].counts
         counts = dict(counts)
         if nb_none:
             counts[None] = nb_none
         return counts
 
     def snapshot_get_branches(
         self, snapshot_id, branches_from=b"", branches_count=1000, target_types=None
     ):
         if self._cql_runner.snapshot_missing([snapshot_id]):
             # Makes sure we don't fetch branches for a snapshot that is
             # being added.
             return None
 
         branches = []
         while len(branches) < branches_count + 1:
             new_branches = list(
                 self._cql_runner.snapshot_branch_get(
                     snapshot_id, branches_from, branches_count + 1
                 )
             )
 
             if not new_branches:
                 break
 
             branches_from = new_branches[-1].name
 
             new_branches_filtered = new_branches
 
             # Filter by target_type
             if target_types:
                 new_branches_filtered = [
                     branch
                     for branch in new_branches_filtered
                     if branch.target is not None and branch.target_type in target_types
                 ]
 
             branches.extend(new_branches_filtered)
 
             if len(new_branches) < branches_count + 1:
                 break
 
         if len(branches) > branches_count:
             last_branch = branches.pop(-1).name
         else:
             last_branch = None
 
         branches = {
             branch.name: {"target": branch.target, "target_type": branch.target_type,}
             if branch.target
             else None
             for branch in branches
         }
 
         return {
             "id": snapshot_id,
             "branches": branches,
             "next_branch": last_branch,
         }
 
     def snapshot_get_random(self):
         return self._cql_runner.snapshot_get_random().id
 
     def object_find_by_sha1_git(self, ids):
         results = {id_: [] for id_ in ids}
         missing_ids = set(ids)
 
         # Mind the order, revision is the most likely one for a given ID,
         # so we check revisions first.
         queries = [
             ("revision", self._cql_runner.revision_missing),
             ("release", self._cql_runner.release_missing),
             ("content", self._cql_runner.content_missing_by_sha1_git),
             ("directory", self._cql_runner.directory_missing),
         ]
 
         for (object_type, query_fn) in queries:
             found_ids = missing_ids - set(query_fn(missing_ids))
             for sha1_git in found_ids:
                 results[sha1_git].append(
                     {"sha1_git": sha1_git, "type": object_type,}
                 )
                 missing_ids.remove(sha1_git)
 
             if not missing_ids:
                 # We found everything, skipping the next queries.
                 break
 
         return results
 
     def origin_get(self, origins):
         if isinstance(origins, dict):
             # Old API
             return_single = True
             origins = [origins]
         else:
             return_single = False
 
         if any("id" in origin for origin in origins):
             raise StorageArgumentException("Origin ids are not supported.")
 
         results = [self.origin_get_one(origin) for origin in origins]
 
         if return_single:
             assert len(results) == 1
             return results[0]
         else:
             return results
 
     def origin_get_one(self, origin: Dict[str, Any]) -> Optional[Dict[str, Any]]:
         if "id" in origin:
             raise StorageArgumentException("Origin ids are not supported.")
         if "url" not in origin:
             raise StorageArgumentException("Missing origin url")
         rows = self._cql_runner.origin_get_by_url(origin["url"])
 
         rows = list(rows)
         if rows:
             assert len(rows) == 1
             result = rows[0]._asdict()
             return {
                 "url": result["url"],
             }
         else:
             return None
 
     def origin_get_by_sha1(self, sha1s):
         results = []
         for sha1 in sha1s:
             rows = self._cql_runner.origin_get_by_sha1(sha1)
             if rows:
                 results.append({"url": rows.one().url})
             else:
                 results.append(None)
         return results
 
     def origin_list(self, page_token: Optional[str] = None, limit: int = 100) -> dict:
         # Compute what token to begin the listing from
         start_token = TOKEN_BEGIN
         if page_token:
             start_token = int(page_token)
             if not (TOKEN_BEGIN <= start_token <= TOKEN_END):
                 raise StorageArgumentException("Invalid page_token.")
 
         rows = self._cql_runner.origin_list(start_token, limit)
         rows = list(rows)
 
         if len(rows) == limit:
             next_page_token: Optional[str] = str(rows[-1].tok + 1)
         else:
             next_page_token = None
 
         return {
             "origins": [{"url": row.url} for row in rows],
             "next_page_token": next_page_token,
         }
 
     def origin_search(
         self, url_pattern, offset=0, limit=50, regexp=False, with_visit=False
     ):
         # TODO: remove this endpoint, swh-search should be used instead.
         origins = self._cql_runner.origin_iter_all()
         if regexp:
             pat = re.compile(url_pattern)
             origins = [orig for orig in origins if pat.search(orig.url)]
         else:
             origins = [orig for orig in origins if url_pattern in orig.url]
 
         if with_visit:
             origins = [orig for orig in origins if orig.next_visit_id > 1]
 
         return [{"url": orig.url,} for orig in origins[offset : offset + limit]]
 
     def origin_add(self, origins: Iterable[Origin]) -> Dict[str, int]:
         known_origins = [
             Origin.from_dict(d)
             for d in self.origin_get([origin.to_dict() for origin in origins])
             if d is not None
         ]
         to_add = [origin for origin in origins if origin not in known_origins]
         self.journal_writer.origin_add(to_add)
         for origin in to_add:
             self._cql_runner.origin_add_one(origin)
         return {"origin:add": len(to_add)}
 
     @deprecated("Use origin_add([origin]) instead")
     def origin_add_one(self, origin: Origin) -> str:
         known_origin = self.origin_get_one(origin.to_dict())
 
         if known_origin:
             origin_url = known_origin["url"]
         else:
             self.journal_writer.origin_add([origin])
 
             self._cql_runner.origin_add_one(origin)
             origin_url = origin.url
 
         return origin_url
 
     def origin_visit_add(self, visits: Iterable[OriginVisit]) -> Iterable[OriginVisit]:
         for visit in visits:
             origin = self.origin_get({"url": visit.origin})
             if not origin:  # Cannot add a visit without an origin
                 raise StorageArgumentException("Unknown origin %s", visit.origin)
 
         all_visits = []
         nb_visits = 0
         for visit in visits:
             nb_visits += 1
             if not visit.visit:
                 visit_id = self._cql_runner.origin_generate_unique_visit_id(
                     visit.origin
                 )
                 visit = attr.evolve(visit, visit=visit_id)
             self.journal_writer.origin_visit_add([visit])
             self._cql_runner.origin_visit_add_one(visit)
             assert visit.visit is not None
             all_visits.append(visit)
             self._origin_visit_status_add(
                 OriginVisitStatus(
                     origin=visit.origin,
                     visit=visit.visit,
                     date=visit.date,
                     status="created",
                     snapshot=None,
                 )
             )
 
         return all_visits
 
     def _origin_visit_status_add(self, visit_status: OriginVisitStatus) -> None:
         """Add an origin visit status"""
         self.journal_writer.origin_visit_status_add([visit_status])
         self._cql_runner.origin_visit_status_add_one(visit_status)
 
     def origin_visit_status_add(
         self, visit_statuses: Iterable[OriginVisitStatus]
     ) -> None:
         # First round to check existence (fail early if any is ko)
         for visit_status in visit_statuses:
             origin_url = self.origin_get({"url": visit_status.origin})
             if not origin_url:
                 raise StorageArgumentException(f"Unknown origin {visit_status.origin}")
 
         for visit_status in visit_statuses:
             self._origin_visit_status_add(visit_status)
 
     def _origin_visit_merge(
         self, visit: Dict[str, Any], visit_status: OriginVisitStatus,
     ) -> Dict[str, Any]:
         """Merge origin_visit and visit_status together.
 
         """
         return OriginVisit.from_dict(
             {
                 # default to the values in visit
                 **visit,
                 # override with the last update
                 **visit_status.to_dict(),
                 # visit['origin'] is the URL (via a join), while
                 # visit_status['origin'] is only an id.
                 "origin": visit["origin"],
                 # but keep the date of the creation of the origin visit
                 "date": visit["date"],
             }
         ).to_dict()
 
     def _origin_visit_apply_last_status(self, visit: Dict[str, Any]) -> Dict[str, Any]:
         """Retrieve the latest visit status information for the origin visit.
         Then merge it with the visit and return it.
 
         """
         row = self._cql_runner.origin_visit_status_get_latest(
             visit["origin"], visit["visit"]
         )
         assert row is not None
         return self._origin_visit_merge(visit, row_to_visit_status(row))
 
     def _origin_visit_get_updated(self, origin: str, visit_id: int) -> Dict[str, Any]:
         """Retrieve origin visit and latest origin visit status and merge them
         into an origin visit.
 
         """
         row_visit = self._cql_runner.origin_visit_get_one(origin, visit_id)
         assert row_visit is not None
         visit = self._format_origin_visit_row(row_visit)
         return self._origin_visit_apply_last_status(visit)
 
     @staticmethod
     def _format_origin_visit_row(visit):
         return {
             **visit._asdict(),
             "origin": visit.origin,
             "date": visit.date.replace(tzinfo=datetime.timezone.utc),
         }
 
     def origin_visit_get(
         self,
         origin: str,
         last_visit: Optional[int] = None,
         limit: Optional[int] = None,
         order: str = "asc",
     ) -> Iterable[Dict[str, Any]]:
         rows = self._cql_runner.origin_visit_get(origin, last_visit, limit, order)
 
         for row in rows:
             visit = self._format_origin_visit_row(row)
             yield self._origin_visit_apply_last_status(visit)
 
     def origin_visit_find_by_date(
         self, origin: str, visit_date: datetime.datetime
     ) -> Optional[Dict[str, Any]]:
         # Iterator over all the visits of the origin
         # This should be ok for now, as there aren't too many visits
         # per origin.
         rows = list(self._cql_runner.origin_visit_get_all(origin))
 
         def key(visit):
             dt = visit.date.replace(tzinfo=datetime.timezone.utc) - visit_date
             return (abs(dt), -visit.visit)
 
         if rows:
             row = min(rows, key=key)
             visit = self._format_origin_visit_row(row)
             return self._origin_visit_apply_last_status(visit)
         return None
 
     def origin_visit_get_by(self, origin: str, visit: int) -> Optional[Dict[str, Any]]:
         row = self._cql_runner.origin_visit_get_one(origin, visit)
         if row:
             visit_ = self._format_origin_visit_row(row)
             return self._origin_visit_apply_last_status(visit_)
         return None
 
     def origin_visit_get_latest(
         self,
         origin: str,
         type: Optional[str] = None,
         allowed_statuses: Optional[List[str]] = None,
         require_snapshot: bool = False,
     ) -> Optional[Dict[str, Any]]:
         # TODO: Do not fetch all visits
         rows = self._cql_runner.origin_visit_get_all(origin)
         latest_visit = None
         for row in rows:
             visit = self._format_origin_visit_row(row)
             updated_visit = self._origin_visit_apply_last_status(visit)
             if type is not None and updated_visit["type"] != type:
                 continue
             if allowed_statuses and updated_visit["status"] not in allowed_statuses:
                 continue
             if require_snapshot and updated_visit["snapshot"] is None:
                 continue
 
             # updated_visit is a candidate
             if latest_visit is not None:
                 if updated_visit["date"] < latest_visit["date"]:
                     continue
                 if updated_visit["visit"] < latest_visit["visit"]:
                     continue
 
             latest_visit = updated_visit
 
         return latest_visit
 
     def origin_visit_status_get_latest(
         self,
         origin_url: str,
         visit: int,
         allowed_statuses: Optional[List[str]] = None,
         require_snapshot: bool = False,
     ) -> Optional[OriginVisitStatus]:
         rows = self._cql_runner.origin_visit_status_get(
             origin_url, visit, allowed_statuses, require_snapshot
         )
         # filtering is done python side as we cannot do it server side
         if allowed_statuses:
             rows = [row for row in rows if row.status in allowed_statuses]
         if require_snapshot:
             rows = [row for row in rows if row.snapshot is not None]
         if not rows:
             return None
         return row_to_visit_status(rows[0])
 
     def origin_visit_get_random(self, type: str) -> Optional[Dict[str, Any]]:
         back_in_the_day = now() - datetime.timedelta(weeks=12)  # 3 months back
 
         # Random position to start iteration at
         start_token = random.randint(TOKEN_BEGIN, TOKEN_END)
 
         # Iterator over all visits, ordered by token(origins) then visit_id
         rows = self._cql_runner.origin_visit_iter(start_token)
         for row in rows:
             visit = self._format_origin_visit_row(row)
             visit_status = self._origin_visit_apply_last_status(visit)
             if (
                 visit_status["date"] > back_in_the_day
                 and visit_status["status"] == "full"
             ):
                 return visit_status
         else:
             return None
 
     def stat_counters(self):
         rows = self._cql_runner.stat_counters()
         keys = (
             "content",
             "directory",
             "origin",
             "origin_visit",
             "release",
             "revision",
             "skipped_content",
             "snapshot",
         )
         stats = {key: 0 for key in keys}
         stats.update({row.object_type: row.count for row in rows})
         return stats
 
     def refresh_stat_counters(self):
         pass
 
     def origin_metadata_add(
         self,
         origin_url: str,
         discovery_date: datetime.datetime,
         authority: Dict[str, Any],
         fetcher: Dict[str, Any],
         format: str,
         metadata: bytes,
     ) -> None:
         if not isinstance(origin_url, str):
             raise StorageArgumentException(
                 "origin_id must be str, not %r" % (origin_url,)
             )
         if not self._cql_runner.metadata_authority_get(**authority):
             raise StorageArgumentException(f"Unknown authority {authority}")
         if not self._cql_runner.metadata_fetcher_get(**fetcher):
             raise StorageArgumentException(f"Unknown fetcher {fetcher}")
 
         try:
             self._cql_runner.origin_metadata_add(
                 origin_url,
                 authority["type"],
                 authority["url"],
                 discovery_date,
                 fetcher["name"],
                 fetcher["version"],
                 format,
                 metadata,
             )
         except TypeError as e:
             raise StorageArgumentException(*e.args)
 
     def origin_metadata_get(
         self,
         origin_url: str,
         authority: Dict[str, str],
         after: Optional[datetime.datetime] = None,
         page_token: Optional[bytes] = None,
         limit: int = 1000,
     ) -> Dict[str, Any]:
         if not isinstance(origin_url, str):
             raise TypeError("origin_url must be str, not %r" % (origin_url,))
 
         if page_token is not None:
             (after_date, after_fetcher_name, after_fetcher_url) = msgpack_loads(
                 page_token
             )
             if after and after_date < after:
                 raise StorageArgumentException(
                     "page_token is inconsistent with the value of 'after'."
                 )
             entries = self._cql_runner.origin_metadata_get_after_date_and_fetcher(
                 origin_url,
                 authority["type"],
                 authority["url"],
                 after_date,
                 after_fetcher_name,
                 after_fetcher_url,
             )
         elif after is not None:
             entries = self._cql_runner.origin_metadata_get_after_date(
                 origin_url, authority["type"], authority["url"], after
             )
         else:
             entries = self._cql_runner.origin_metadata_get(
                 origin_url, authority["type"], authority["url"]
             )
 
         if limit:
             entries = itertools.islice(entries, 0, limit + 1)
 
         results = []
         for entry in entries:
             discovery_date = entry.discovery_date.replace(tzinfo=datetime.timezone.utc)
             results.append(
                 {
                     "origin_url": entry.origin,
                     "authority": {
                         "type": entry.authority_type,
                         "url": entry.authority_url,
                     },
                     "fetcher": {
                         "name": entry.fetcher_name,
                         "version": entry.fetcher_version,
                     },
                     "discovery_date": discovery_date,
                     "format": entry.format,
                     "metadata": entry.metadata,
                 }
             )
 
         if len(results) > limit:
             results.pop()
             assert len(results) == limit
             last_result = results[-1]
             next_page_token: Optional[bytes] = msgpack_dumps(
                 (
                     last_result["discovery_date"],
                     last_result["fetcher"]["name"],
                     last_result["fetcher"]["version"],
                 )
             )
         else:
             next_page_token = None
 
         return {
             "next_page_token": next_page_token,
             "results": results,
         }
 
     def metadata_fetcher_add(
         self, name: str, version: str, metadata: Dict[str, Any]
     ) -> None:
         self._cql_runner.metadata_fetcher_add(name, version, json.dumps(metadata))
 
     def metadata_fetcher_get(self, name: str, version: str) -> Optional[Dict[str, Any]]:
         fetcher = self._cql_runner.metadata_fetcher_get(name, version)
         if fetcher:
             return {
                 "name": fetcher.name,
                 "version": fetcher.version,
                 "metadata": json.loads(fetcher.metadata),
             }
         else:
             return None
 
     def metadata_authority_add(
         self, type: str, url: str, metadata: Dict[str, Any]
     ) -> None:
         self._cql_runner.metadata_authority_add(url, type, json.dumps(metadata))
 
     def metadata_authority_get(self, type: str, url: str) -> Optional[Dict[str, Any]]:
         authority = self._cql_runner.metadata_authority_get(type, url)
         if authority:
             return {
                 "type": authority.type,
                 "url": authority.url,
                 "metadata": json.loads(authority.metadata),
             }
         else:
             return None
 
     def clear_buffers(self, object_types: Optional[Iterable[str]] = None) -> None:
         """Do nothing
 
         """
         return None
 
     def flush(self, object_types: Optional[Iterable[str]] = None) -> Dict:
         return {}
diff --git a/swh/storage/in_memory.py b/swh/storage/in_memory.py
index e9d13677..ea1cb044 100644
--- a/swh/storage/in_memory.py
+++ b/swh/storage/in_memory.py
@@ -1,1211 +1,1212 @@
 # Copyright (C) 2015-2020  The Software Heritage developers
 # See the AUTHORS file at the top-level directory of this distribution
 # License: GNU General Public License version 3, or any later version
 # See top-level LICENSE file for more information
 
 import re
 import bisect
 import collections
 import copy
 import datetime
 import itertools
 import random
 
 from collections import defaultdict
 from datetime import timedelta
 from typing import (
     Any,
     Callable,
     Dict,
     Generic,
     Hashable,
     Iterable,
     Iterator,
     List,
     Optional,
     Tuple,
     TypeVar,
 )
 
 import attr
 
 from deprecated import deprecated
 
 from swh.core.api.serializers import msgpack_loads, msgpack_dumps
 from swh.model.model import (
     BaseContent,
     Content,
     SkippedContent,
     Directory,
     Revision,
     Release,
     Snapshot,
     OriginVisit,
     OriginVisitStatus,
     Origin,
     SHA1_SIZE,
 )
 from swh.model.hashutil import DEFAULT_ALGORITHMS, hash_to_bytes, hash_to_hex
 from swh.storage.objstorage import ObjStorage
 from swh.storage.utils import now
 
 from .exc import StorageArgumentException, HashCollision
 
 from .converters import origin_url_to_sha1
 from .utils import get_partition_bounds_bytes
 from .writer import JournalWriter
 
 # Max block size of contents to return
 BULK_BLOCK_CONTENT_LEN_MAX = 10000
 
 
 SortedListItem = TypeVar("SortedListItem")
 SortedListKey = TypeVar("SortedListKey")
 
 FetcherKey = Tuple[str, str]
 
 
 class SortedList(collections.UserList, Generic[SortedListKey, SortedListItem]):
     data: List[Tuple[SortedListKey, SortedListItem]]
 
     # https://github.com/python/mypy/issues/708
     # key: Callable[[SortedListItem], SortedListKey]
 
     def __init__(
         self,
         data: List[SortedListItem] = None,
         key: Optional[Callable[[SortedListItem], SortedListKey]] = None,
     ):
         if key is None:
 
             def key(item):
                 return item
 
         assert key is not None  # for mypy
         super().__init__(sorted((key(x), x) for x in data or []))
 
         self.key: Callable[[SortedListItem], SortedListKey] = key
 
     def add(self, item: SortedListItem):
         k = self.key(item)
         bisect.insort(self.data, (k, item))
 
     def __iter__(self) -> Iterator[SortedListItem]:
         for (k, item) in self.data:
             yield item
 
     def iter_from(self, start_key: Any) -> Iterator[SortedListItem]:
         """Returns an iterator over all the elements whose key is greater
         or equal to `start_key`.
         (This is an efficient equivalent to:
         `(x for x in L if key(x) >= start_key)`)
         """
         from_index = bisect.bisect_left(self.data, (start_key,))
         for (k, item) in itertools.islice(self.data, from_index, None):
             yield item
 
     def iter_after(self, start_key: Any) -> Iterator[SortedListItem]:
         """Same as iter_from, but using a strict inequality."""
         it = self.iter_from(start_key)
         for item in it:
             if self.key(item) > start_key:  # type: ignore
                 yield item
                 break
 
         yield from it
 
 
 class InMemoryStorage:
     def __init__(self, journal_writer=None):
 
         self.reset()
         self.journal_writer = JournalWriter(journal_writer)
 
     def reset(self):
         self._contents = {}
         self._content_indexes = defaultdict(lambda: defaultdict(set))
         self._skipped_contents = {}
         self._skipped_content_indexes = defaultdict(lambda: defaultdict(set))
         self._directories = {}
         self._revisions = {}
         self._releases = {}
         self._snapshots = {}
         self._origins = {}
         self._origins_by_id = []
         self._origins_by_sha1 = {}
         self._origin_visits = {}
         self._origin_visit_statuses: Dict[Tuple[str, int], List[OriginVisitStatus]] = {}
         self._persons = {}
 
         # {origin_url: {authority: [metadata]}}
         self._origin_metadata: Dict[
             str,
             Dict[
                 Hashable,
                 SortedList[Tuple[datetime.datetime, FetcherKey], Dict[str, Any]],
             ],
         ] = defaultdict(
             lambda: defaultdict(
                 lambda: SortedList(key=lambda x: (x["discovery_date"], x["fetcher"]))
             )
         )  # noqa
 
         self._metadata_fetchers: Dict[FetcherKey, Dict[str, Any]] = {}
         self._metadata_authorities: Dict[Hashable, Dict[str, Any]] = {}
         self._objects = defaultdict(list)
         self._sorted_sha1s = SortedList[bytes, bytes]()
 
         self.objstorage = ObjStorage({"cls": "memory", "args": {}})
 
     def check_config(self, *, check_write):
         return True
 
     def _content_add(self, contents: Iterable[Content], with_data: bool) -> Dict:
         self.journal_writer.content_add(contents)
 
         content_add = 0
         if with_data:
             summary = self.objstorage.content_add(
                 c for c in contents if c.status != "absent"
             )
             content_add_bytes = summary["content:add:bytes"]
 
         for content in contents:
             key = self._content_key(content)
             if key in self._contents:
                 continue
             for algorithm in DEFAULT_ALGORITHMS:
                 hash_ = content.get_hash(algorithm)
                 if hash_ in self._content_indexes[algorithm] and (
                     algorithm not in {"blake2s256", "sha256"}
                 ):
                     colliding_content_hashes = []
                     # Add the already stored contents
                     for content_hashes_set in self._content_indexes[algorithm][hash_]:
                         hashes = dict(content_hashes_set)
                         colliding_content_hashes.append(hashes)
                     # Add the new colliding content
                     colliding_content_hashes.append(content.hashes())
                     raise HashCollision(algorithm, hash_, colliding_content_hashes)
             for algorithm in DEFAULT_ALGORITHMS:
                 hash_ = content.get_hash(algorithm)
                 self._content_indexes[algorithm][hash_].add(key)
             self._objects[content.sha1_git].append(("content", content.sha1))
             self._contents[key] = content
             self._sorted_sha1s.add(content.sha1)
             self._contents[key] = attr.evolve(self._contents[key], data=None)
             content_add += 1
 
         summary = {
             "content:add": content_add,
         }
         if with_data:
             summary["content:add:bytes"] = content_add_bytes
 
         return summary
 
     def content_add(self, content: Iterable[Content]) -> Dict:
         content = [attr.evolve(c, ctime=now()) for c in content]
         return self._content_add(content, with_data=True)
 
     def content_update(self, content, keys=[]):
         self.journal_writer.content_update(content)
 
         for cont_update in content:
             cont_update = cont_update.copy()
             sha1 = cont_update.pop("sha1")
             for old_key in self._content_indexes["sha1"][sha1]:
                 old_cont = self._contents.pop(old_key)
 
                 for algorithm in DEFAULT_ALGORITHMS:
                     hash_ = old_cont.get_hash(algorithm)
                     self._content_indexes[algorithm][hash_].remove(old_key)
 
                 new_cont = attr.evolve(old_cont, **cont_update)
                 new_key = self._content_key(new_cont)
 
                 self._contents[new_key] = new_cont
 
                 for algorithm in DEFAULT_ALGORITHMS:
                     hash_ = new_cont.get_hash(algorithm)
                     self._content_indexes[algorithm][hash_].add(new_key)
 
     def content_add_metadata(self, content: Iterable[Content]) -> Dict:
         return self._content_add(content, with_data=False)
 
     def content_get(self, content):
         # FIXME: Make this method support slicing the `data`.
         if len(content) > BULK_BLOCK_CONTENT_LEN_MAX:
             raise StorageArgumentException(
                 "Sending at most %s contents." % BULK_BLOCK_CONTENT_LEN_MAX
             )
         yield from self.objstorage.content_get(content)
 
     def content_get_range(self, start, end, limit=1000):
         if limit is None:
             raise StorageArgumentException("limit should not be None")
         sha1s = (
             (sha1, content_key)
             for sha1 in self._sorted_sha1s.iter_from(start)
             for content_key in self._content_indexes["sha1"][sha1]
         )
         matched = []
         next_content = None
         for sha1, key in sha1s:
             if sha1 > end:
                 break
             if len(matched) >= limit:
                 next_content = sha1
                 break
             matched.append(self._contents[key].to_dict())
         return {
             "contents": matched,
             "next": next_content,
         }
 
     def content_get_partition(
         self,
         partition_id: int,
         nb_partitions: int,
         limit: int = 1000,
         page_token: str = None,
     ):
         if limit is None:
             raise StorageArgumentException("limit should not be None")
         (start, end) = get_partition_bounds_bytes(
             partition_id, nb_partitions, SHA1_SIZE
         )
         if page_token:
             start = hash_to_bytes(page_token)
         if end is None:
             end = b"\xff" * SHA1_SIZE
         result = self.content_get_range(start, end, limit)
         result2 = {
             "contents": result["contents"],
             "next_page_token": None,
         }
         if result["next"]:
             result2["next_page_token"] = hash_to_hex(result["next"])
         return result2
 
     def content_get_metadata(self, contents: List[bytes]) -> Dict[bytes, List[Dict]]:
         result: Dict = {sha1: [] for sha1 in contents}
         for sha1 in contents:
             if sha1 in self._content_indexes["sha1"]:
                 objs = self._content_indexes["sha1"][sha1]
                 # only 1 element as content_add_metadata would have raised a
                 # hash collision otherwise
                 for key in objs:
                     d = self._contents[key].to_dict()
                     del d["ctime"]
                     if "data" in d:
                         del d["data"]
                     result[sha1].append(d)
         return result
 
     def content_find(self, content):
         if not set(content).intersection(DEFAULT_ALGORITHMS):
             raise StorageArgumentException(
                 "content keys must contain at least one of: %s"
                 % ", ".join(sorted(DEFAULT_ALGORITHMS))
             )
         found = []
         for algo in DEFAULT_ALGORITHMS:
             hash = content.get(algo)
             if hash and hash in self._content_indexes[algo]:
                 found.append(self._content_indexes[algo][hash])
 
         if not found:
             return []
 
         keys = list(set.intersection(*found))
         return [self._contents[key].to_dict() for key in keys]
 
     def content_missing(self, content, key_hash="sha1"):
         for cont in content:
             for (algo, hash_) in cont.items():
                 if algo not in DEFAULT_ALGORITHMS:
                     continue
                 if hash_ not in self._content_indexes.get(algo, []):
                     yield cont[key_hash]
                     break
             else:
                 for result in self.content_find(cont):
                     if result["status"] == "missing":
                         yield cont[key_hash]
 
     def content_missing_per_sha1(self, contents):
         for content in contents:
             if content not in self._content_indexes["sha1"]:
                 yield content
 
     def content_missing_per_sha1_git(self, contents):
         for content in contents:
             if content not in self._content_indexes["sha1_git"]:
                 yield content
 
     def content_get_random(self):
         return random.choice(list(self._content_indexes["sha1_git"]))
 
     def _skipped_content_add(self, contents: List[SkippedContent]) -> Dict:
         self.journal_writer.skipped_content_add(contents)
 
         summary = {"skipped_content:add": 0}
 
         missing_contents = self.skipped_content_missing([c.hashes() for c in contents])
         missing = {self._content_key(c) for c in missing_contents}
         contents = [c for c in contents if self._content_key(c) in missing]
         for content in contents:
             key = self._content_key(content)
             for algo in DEFAULT_ALGORITHMS:
                 if content.get_hash(algo):
                     self._skipped_content_indexes[algo][content.get_hash(algo)].add(key)
             self._skipped_contents[key] = content
             summary["skipped_content:add"] += 1
 
         return summary
 
     def skipped_content_add(self, content: Iterable[SkippedContent]) -> Dict:
         content = [attr.evolve(c, ctime=now()) for c in content]
         return self._skipped_content_add(content)
 
     def skipped_content_missing(self, contents):
         for content in contents:
             matches = list(self._skipped_contents.values())
             for (algorithm, key) in self._content_key(content):
                 if algorithm == "blake2s256":
                     continue
                 # Filter out skipped contents with the same hash
                 matches = [
                     match for match in matches if match.get_hash(algorithm) == key
                 ]
             # if none of the contents match
             if not matches:
                 yield {algo: content[algo] for algo in DEFAULT_ALGORITHMS}
 
     def directory_add(self, directories: Iterable[Directory]) -> Dict:
         directories = [dir_ for dir_ in directories if dir_.id not in self._directories]
         self.journal_writer.directory_add(directories)
 
         count = 0
         for directory in directories:
             count += 1
             self._directories[directory.id] = directory
             self._objects[directory.id].append(("directory", directory.id))
 
         return {"directory:add": count}
 
     def directory_missing(self, directories):
         for id in directories:
             if id not in self._directories:
                 yield id
 
     def _join_dentry_to_content(self, dentry):
         keys = (
             "status",
             "sha1",
             "sha1_git",
             "sha256",
             "length",
         )
         ret = dict.fromkeys(keys)
         ret.update(dentry)
         if ret["type"] == "file":
             # TODO: Make it able to handle more than one content
             content = self.content_find({"sha1_git": ret["target"]})
             if content:
                 content = content[0]
                 for key in keys:
                     ret[key] = content[key]
         return ret
 
     def _directory_ls(self, directory_id, recursive, prefix=b""):
         if directory_id in self._directories:
             for entry in self._directories[directory_id].entries:
                 ret = self._join_dentry_to_content(entry.to_dict())
                 ret["name"] = prefix + ret["name"]
                 ret["dir_id"] = directory_id
                 yield ret
                 if recursive and ret["type"] == "dir":
                     yield from self._directory_ls(
                         ret["target"], True, prefix + ret["name"] + b"/"
                     )
 
     def directory_ls(self, directory, recursive=False):
         yield from self._directory_ls(directory, recursive)
 
     def directory_entry_get_by_path(self, directory, paths):
         return self._directory_entry_get_by_path(directory, paths, b"")
 
     def directory_get_random(self):
         if not self._directories:
             return None
         return random.choice(list(self._directories))
 
     def _directory_entry_get_by_path(self, directory, paths, prefix):
         if not paths:
             return
 
         contents = list(self.directory_ls(directory))
 
         if not contents:
             return
 
         def _get_entry(entries, name):
             for entry in entries:
                 if entry["name"] == name:
                     entry = entry.copy()
                     entry["name"] = prefix + entry["name"]
                     return entry
 
         first_item = _get_entry(contents, paths[0])
 
         if len(paths) == 1:
             return first_item
 
         if not first_item or first_item["type"] != "dir":
             return
 
         return self._directory_entry_get_by_path(
             first_item["target"], paths[1:], prefix + paths[0] + b"/"
         )
 
     def revision_add(self, revisions: Iterable[Revision]) -> Dict:
         revisions = [rev for rev in revisions if rev.id not in self._revisions]
         self.journal_writer.revision_add(revisions)
 
         count = 0
         for revision in revisions:
             revision = attr.evolve(
                 revision,
                 committer=self._person_add(revision.committer),
                 author=self._person_add(revision.author),
             )
             self._revisions[revision.id] = revision
             self._objects[revision.id].append(("revision", revision.id))
             count += 1
 
         return {"revision:add": count}
 
     def revision_missing(self, revisions):
         for id in revisions:
             if id not in self._revisions:
                 yield id
 
     def revision_get(self, revisions):
         for id in revisions:
             if id in self._revisions:
                 yield self._revisions.get(id).to_dict()
             else:
                 yield None
 
     def _get_parent_revs(self, rev_id, seen, limit):
         if limit and len(seen) >= limit:
             return
         if rev_id in seen or rev_id not in self._revisions:
             return
         seen.add(rev_id)
         yield self._revisions[rev_id].to_dict()
         for parent in self._revisions[rev_id].parents:
             yield from self._get_parent_revs(parent, seen, limit)
 
     def revision_log(self, revisions, limit=None):
         seen = set()
         for rev_id in revisions:
             yield from self._get_parent_revs(rev_id, seen, limit)
 
     def revision_shortlog(self, revisions, limit=None):
         yield from (
             (rev["id"], rev["parents"]) for rev in self.revision_log(revisions, limit)
         )
 
     def revision_get_random(self):
         return random.choice(list(self._revisions))
 
     def release_add(self, releases: Iterable[Release]) -> Dict:
-        releases = [rel for rel in releases if rel.id not in self._releases]
-        self.journal_writer.release_add(releases)
-
-        count = 0
+        to_add = []
         for rel in releases:
+            if rel.id not in self._releases and rel not in to_add:
+                to_add.append(rel)
+        self.journal_writer.release_add(to_add)
+
+        for rel in to_add:
             if rel.author:
                 self._person_add(rel.author)
             self._objects[rel.id].append(("release", rel.id))
             self._releases[rel.id] = rel
-            count += 1
 
-        return {"release:add": count}
+        return {"release:add": len(to_add)}
 
     def release_missing(self, releases):
         yield from (rel for rel in releases if rel not in self._releases)
 
     def release_get(self, releases):
         for rel_id in releases:
             if rel_id in self._releases:
                 yield self._releases[rel_id].to_dict()
             else:
                 yield None
 
     def release_get_random(self):
         return random.choice(list(self._releases))
 
     def snapshot_add(self, snapshots: Iterable[Snapshot]) -> Dict:
         count = 0
         snapshots = (snap for snap in snapshots if snap.id not in self._snapshots)
         for snapshot in snapshots:
             self.journal_writer.snapshot_add([snapshot])
             self._snapshots[snapshot.id] = snapshot
             self._objects[snapshot.id].append(("snapshot", snapshot.id))
             count += 1
 
         return {"snapshot:add": count}
 
     def snapshot_missing(self, snapshots):
         for id in snapshots:
             if id not in self._snapshots:
                 yield id
 
     def snapshot_get(self, snapshot_id):
         return self.snapshot_get_branches(snapshot_id)
 
     def snapshot_get_by_origin_visit(self, origin, visit):
         origin_url = self._get_origin_url(origin)
         if not origin_url:
             return
 
         if origin_url not in self._origins or visit > len(
             self._origin_visits[origin_url]
         ):
             return None
 
         visit = self._origin_visit_get_updated(origin_url, visit)
         snapshot_id = visit.snapshot
         if snapshot_id:
             return self.snapshot_get(snapshot_id)
         else:
             return None
 
     def snapshot_count_branches(self, snapshot_id):
         snapshot = self._snapshots[snapshot_id]
         return collections.Counter(
             branch.target_type.value if branch else None
             for branch in snapshot.branches.values()
         )
 
     def snapshot_get_branches(
         self, snapshot_id, branches_from=b"", branches_count=1000, target_types=None
     ):
         snapshot = self._snapshots.get(snapshot_id)
         if snapshot is None:
             return None
         sorted_branch_names = sorted(snapshot.branches)
         from_index = bisect.bisect_left(sorted_branch_names, branches_from)
         if target_types:
             next_branch = None
             branches = {}
             for branch_name in sorted_branch_names[from_index:]:
                 branch = snapshot.branches[branch_name]
                 if branch and branch.target_type.value in target_types:
                     if len(branches) < branches_count:
                         branches[branch_name] = branch
                     else:
                         next_branch = branch_name
                         break
         else:
             # As there is no 'target_types', we can do that much faster
             to_index = from_index + branches_count
             returned_branch_names = sorted_branch_names[from_index:to_index]
             branches = {
                 branch_name: snapshot.branches[branch_name]
                 for branch_name in returned_branch_names
             }
             if to_index >= len(sorted_branch_names):
                 next_branch = None
             else:
                 next_branch = sorted_branch_names[to_index]
 
         branches = {
             name: branch.to_dict() if branch else None
             for (name, branch) in branches.items()
         }
 
         return {
             "id": snapshot_id,
             "branches": branches,
             "next_branch": next_branch,
         }
 
     def snapshot_get_random(self):
         return random.choice(list(self._snapshots))
 
     def object_find_by_sha1_git(self, ids):
         ret = {}
         for id_ in ids:
             objs = self._objects.get(id_, [])
             ret[id_] = [{"sha1_git": id_, "type": obj[0],} for obj in objs]
         return ret
 
     def _convert_origin(self, t):
         if t is None:
             return None
 
         return t.to_dict()
 
     def origin_get(self, origins):
         if isinstance(origins, dict):
             # Old API
             return_single = True
             origins = [origins]
         else:
             return_single = False
 
         # Sanity check to be error-compatible with the pgsql backend
         if any("id" in origin for origin in origins) and not all(
             "id" in origin for origin in origins
         ):
             raise StorageArgumentException(
                 'Either all origins or none at all should have an "id".'
             )
         if any("url" in origin for origin in origins) and not all(
             "url" in origin for origin in origins
         ):
             raise StorageArgumentException(
                 "Either all origins or none at all should have " 'an "url" key.'
             )
 
         results = []
         for origin in origins:
             result = None
             if "url" in origin:
                 if origin["url"] in self._origins:
                     result = self._origins[origin["url"]]
             else:
                 raise StorageArgumentException("Origin must have an url.")
             results.append(self._convert_origin(result))
 
         if return_single:
             assert len(results) == 1
             return results[0]
         else:
             return results
 
     def origin_get_by_sha1(self, sha1s):
         return [self._convert_origin(self._origins_by_sha1.get(sha1)) for sha1 in sha1s]
 
     def origin_get_range(self, origin_from=1, origin_count=100):
         origin_from = max(origin_from, 1)
         if origin_from <= len(self._origins_by_id):
             max_idx = origin_from + origin_count - 1
             if max_idx > len(self._origins_by_id):
                 max_idx = len(self._origins_by_id)
             for idx in range(origin_from - 1, max_idx):
                 origin = self._convert_origin(self._origins[self._origins_by_id[idx]])
                 yield {"id": idx + 1, **origin}
 
     def origin_list(self, page_token: Optional[str] = None, limit: int = 100) -> dict:
         origin_urls = sorted(self._origins)
         if page_token:
             from_ = bisect.bisect_left(origin_urls, page_token)
         else:
             from_ = 0
 
         result = {
             "origins": [
                 {"url": origin_url} for origin_url in origin_urls[from_ : from_ + limit]
             ]
         }
 
         if from_ + limit < len(origin_urls):
             result["next_page_token"] = origin_urls[from_ + limit]
 
         return result
 
     def origin_search(
         self, url_pattern, offset=0, limit=50, regexp=False, with_visit=False
     ):
         origins = map(self._convert_origin, self._origins.values())
         if regexp:
             pat = re.compile(url_pattern)
             origins = [orig for orig in origins if pat.search(orig["url"])]
         else:
             origins = [orig for orig in origins if url_pattern in orig["url"]]
         if with_visit:
             filtered_origins = []
             for orig in origins:
                 visits = (
                     self._origin_visit_get_updated(ov.origin, ov.visit)
                     for ov in self._origin_visits[orig["url"]]
                 )
                 for ov in visits:
                     if ov.snapshot and ov.snapshot in self._snapshots:
                         filtered_origins.append(orig)
                         break
         else:
             filtered_origins = origins
 
         return filtered_origins[offset : offset + limit]
 
     def origin_count(self, url_pattern, regexp=False, with_visit=False):
         return len(
             self.origin_search(
                 url_pattern,
                 regexp=regexp,
                 with_visit=with_visit,
                 limit=len(self._origins),
             )
         )
 
     def origin_add(self, origins: Iterable[Origin]) -> Dict[str, int]:
         origins = list(origins)
         added = 0
         for origin in origins:
             if origin.url not in self._origins:
                 self.origin_add_one(origin)
                 added += 1
 
         return {"origin:add": added}
 
     @deprecated("Use origin_add([origin]) instead")
     def origin_add_one(self, origin: Origin) -> str:
         if origin.url not in self._origins:
             self.journal_writer.origin_add([origin])
             # generate an origin_id because it is needed by origin_get_range.
             # TODO: remove this when we remove origin_get_range
             origin_id = len(self._origins) + 1
             self._origins_by_id.append(origin.url)
             assert len(self._origins_by_id) == origin_id
 
             self._origins[origin.url] = origin
             self._origins_by_sha1[origin_url_to_sha1(origin.url)] = origin
             self._origin_visits[origin.url] = []
             self._objects[origin.url].append(("origin", origin.url))
 
         return origin.url
 
     def origin_visit_add(self, visits: Iterable[OriginVisit]) -> Iterable[OriginVisit]:
         for visit in visits:
             origin = self.origin_get({"url": visit.origin})
             if not origin:  # Cannot add a visit without an origin
                 raise StorageArgumentException("Unknown origin %s", visit.origin)
 
         all_visits = []
         for visit in visits:
             origin_url = visit.origin
             if origin_url in self._origins:
                 origin = self._origins[origin_url]
                 if visit.visit:
                     self.journal_writer.origin_visit_add([visit])
                     while len(self._origin_visits[origin_url]) < visit.visit:
                         self._origin_visits[origin_url].append(None)
                     self._origin_visits[origin_url][visit.visit - 1] = visit
                 else:
                     # visit ids are in the range [1, +inf[
                     visit_id = len(self._origin_visits[origin_url]) + 1
                     visit = attr.evolve(visit, visit=visit_id)
                     self.journal_writer.origin_visit_add([visit])
                     self._origin_visits[origin_url].append(visit)
                     visit_key = (origin_url, visit.visit)
                     self._objects[visit_key].append(("origin_visit", None))
                 assert visit.visit is not None
                 self._origin_visit_status_add_one(
                     OriginVisitStatus(
                         origin=visit.origin,
                         visit=visit.visit,
                         date=visit.date,
                         status="created",
                         snapshot=None,
                     )
                 )
                 all_visits.append(visit)
 
         return all_visits
 
     def _origin_visit_status_add_one(self, visit_status: OriginVisitStatus) -> None:
         """Add an origin visit status without checks. If already present, do nothing.
 
         """
         self.journal_writer.origin_visit_status_add([visit_status])
         visit_key = (visit_status.origin, visit_status.visit)
         self._origin_visit_statuses.setdefault(visit_key, [])
         visit_statuses = self._origin_visit_statuses[visit_key]
         if visit_status not in visit_statuses:
             visit_statuses.append(visit_status)
 
     def origin_visit_status_add(
         self, visit_statuses: Iterable[OriginVisitStatus],
     ) -> None:
         # First round to check existence (fail early if any is ko)
         for visit_status in visit_statuses:
             origin_url = self.origin_get({"url": visit_status.origin})
             if not origin_url:
                 raise StorageArgumentException(f"Unknown origin {visit_status.origin}")
 
         for visit_status in visit_statuses:
             self._origin_visit_status_add_one(visit_status)
 
     def _origin_visit_get_updated(self, origin: str, visit_id: int) -> OriginVisit:
         """Merge origin visit and latest origin visit status
 
         """
         assert visit_id >= 1
         visit = self._origin_visits[origin][visit_id - 1]
         assert visit is not None
         visit_key = (origin, visit_id)
 
         visit_update = max(self._origin_visit_statuses[visit_key], key=lambda v: v.date)
         return OriginVisit.from_dict(
             {
                 # default to the values in visit
                 **visit.to_dict(),
                 # override with the last update
                 **visit_update.to_dict(),
                 # but keep the date of the creation of the origin visit
                 "date": visit.date,
             }
         )
 
     def origin_visit_get(
         self,
         origin: str,
         last_visit: Optional[int] = None,
         limit: Optional[int] = None,
         order: str = "asc",
     ) -> Iterable[Dict[str, Any]]:
         order = order.lower()
         assert order in ["asc", "desc"]
         origin_url = self._get_origin_url(origin)
         if origin_url in self._origin_visits:
             visits = self._origin_visits[origin_url]
             visits = sorted(visits, key=lambda v: v.visit, reverse=(order == "desc"))
             if last_visit is not None:
                 if order == "asc":
                     visits = [v for v in visits if v.visit > last_visit]
                 else:
                     visits = [v for v in visits if v.visit < last_visit]
             if limit is not None:
                 visits = visits[:limit]
             for visit in visits:
                 if not visit:
                     continue
                 visit_id = visit.visit
 
                 visit_update = self._origin_visit_get_updated(origin_url, visit_id)
                 assert visit_update is not None
                 yield visit_update.to_dict()
 
     def origin_visit_find_by_date(
         self, origin: str, visit_date: datetime.datetime
     ) -> Optional[Dict[str, Any]]:
         origin_url = self._get_origin_url(origin)
         if origin_url in self._origin_visits:
             visits = self._origin_visits[origin_url]
             visit = min(visits, key=lambda v: (abs(v.date - visit_date), -v.visit))
             visit_update = self._origin_visit_get_updated(origin, visit.visit)
             assert visit_update is not None
             return visit_update.to_dict()
         return None
 
     def origin_visit_get_by(self, origin: str, visit: int) -> Optional[Dict[str, Any]]:
         origin_url = self._get_origin_url(origin)
         if origin_url in self._origin_visits and visit <= len(
             self._origin_visits[origin_url]
         ):
             visit_update = self._origin_visit_get_updated(origin_url, visit)
             assert visit_update is not None
             return visit_update.to_dict()
         return None
 
     def origin_visit_get_latest(
         self,
         origin: str,
         type: Optional[str] = None,
         allowed_statuses: Optional[List[str]] = None,
         require_snapshot: bool = False,
     ) -> Optional[Dict[str, Any]]:
         ori = self._origins.get(origin)
         if not ori:
             return None
         visits = self._origin_visits[ori.url]
 
         visits = [
             self._origin_visit_get_updated(visit.origin, visit.visit)
             for visit in visits
             if visit is not None
         ]
 
         if type is not None:
             visits = [visit for visit in visits if visit.type == type]
         if allowed_statuses is not None:
             visits = [visit for visit in visits if visit.status in allowed_statuses]
         if require_snapshot:
             visits = [visit for visit in visits if visit.snapshot]
 
         visit = max(visits, key=lambda v: (v.date, v.visit), default=None)
         if visit is None:
             return None
         return visit.to_dict()
 
     def origin_visit_status_get_latest(
         self,
         origin_url: str,
         visit: int,
         allowed_statuses: Optional[List[str]] = None,
         require_snapshot: bool = False,
     ) -> Optional[OriginVisitStatus]:
         ori = self._origins.get(origin_url)
         if not ori:
             return None
 
         visit_key = (origin_url, visit)
         visits = self._origin_visit_statuses.get(visit_key)
         if not visits:
             return None
 
         if allowed_statuses is not None:
             visits = [visit for visit in visits if visit.status in allowed_statuses]
         if require_snapshot:
             visits = [visit for visit in visits if visit.snapshot]
 
         visit_status = max(visits, key=lambda v: (v.date, v.visit), default=None)
         return visit_status
 
     def _select_random_origin_visit_by_type(self, type: str) -> str:
         while True:
             url = random.choice(list(self._origin_visits.keys()))
             random_origin_visits = self._origin_visits[url]
             if random_origin_visits[0].type == type:
                 return url
 
     def origin_visit_get_random(self, type: str) -> Optional[Dict[str, Any]]:
         url = self._select_random_origin_visit_by_type(type)
         random_origin_visits = copy.deepcopy(self._origin_visits[url])
         random_origin_visits.reverse()
         back_in_the_day = now() - timedelta(weeks=12)  # 3 months back
         # This should be enough for tests
         for visit in random_origin_visits:
             updated_visit = self._origin_visit_get_updated(url, visit.visit)
             assert updated_visit is not None
             if updated_visit.date > back_in_the_day and updated_visit.status == "full":
                 return updated_visit.to_dict()
         else:
             return None
 
     def stat_counters(self):
         keys = (
             "content",
             "directory",
             "origin",
             "origin_visit",
             "person",
             "release",
             "revision",
             "skipped_content",
             "snapshot",
         )
         stats = {key: 0 for key in keys}
         stats.update(
             collections.Counter(
                 obj_type
                 for (obj_type, obj_id) in itertools.chain(*self._objects.values())
             )
         )
         return stats
 
     def refresh_stat_counters(self):
         pass
 
     def origin_metadata_add(
         self,
         origin_url: str,
         discovery_date: datetime.datetime,
         authority: Dict[str, Any],
         fetcher: Dict[str, Any],
         format: str,
         metadata: bytes,
     ) -> None:
         if not isinstance(origin_url, str):
             raise StorageArgumentException(
                 "origin_id must be str, not %r" % (origin_url,)
             )
         if not isinstance(metadata, bytes):
             raise StorageArgumentException(
                 "metadata must be bytes, not %r" % (metadata,)
             )
         authority_key = self._metadata_authority_key(authority)
         if authority_key not in self._metadata_authorities:
             raise StorageArgumentException(f"Unknown authority {authority}")
         fetcher_key = self._metadata_fetcher_key(fetcher)
         if fetcher_key not in self._metadata_fetchers:
             raise StorageArgumentException(f"Unknown fetcher {fetcher}")
 
         origin_metadata_list = self._origin_metadata[origin_url][authority_key]
 
         origin_metadata = {
             "origin_url": origin_url,
             "discovery_date": discovery_date,
             "authority": authority_key,
             "fetcher": fetcher_key,
             "format": format,
             "metadata": metadata,
         }
 
         for existing_origin_metadata in origin_metadata_list:
             if (
                 existing_origin_metadata["fetcher"] == fetcher_key
                 and existing_origin_metadata["discovery_date"] == discovery_date
             ):
                 # Duplicate of an existing one; replace it.
                 existing_origin_metadata.update(origin_metadata)
                 break
         else:
             origin_metadata_list.add(origin_metadata)
         return None
 
     def origin_metadata_get(
         self,
         origin_url: str,
         authority: Dict[str, str],
         after: Optional[datetime.datetime] = None,
         page_token: Optional[bytes] = None,
         limit: int = 1000,
     ) -> Dict[str, Any]:
         if not isinstance(origin_url, str):
             raise TypeError("origin_url must be str, not %r" % (origin_url,))
 
         authority_key = self._metadata_authority_key(authority)
 
         if page_token is not None:
             (after_time, after_fetcher) = msgpack_loads(page_token)
             after_fetcher = tuple(after_fetcher)
             if after is not None and after > after_time:
                 raise StorageArgumentException(
                     "page_token is inconsistent with the value of 'after'."
                 )
             entries = self._origin_metadata[origin_url][authority_key].iter_after(
                 (after_time, after_fetcher)
             )
         elif after is not None:
             entries = self._origin_metadata[origin_url][authority_key].iter_from(
                 (after,)
             )
             entries = (entry for entry in entries if entry["discovery_date"] > after)
         else:
             entries = iter(self._origin_metadata[origin_url][authority_key])
 
         if limit:
             entries = itertools.islice(entries, 0, limit + 1)
 
         results = []
         for entry in entries:
             authority = self._metadata_authorities[entry["authority"]]
             fetcher = self._metadata_fetchers[entry["fetcher"]]
             if after:
                 assert entry["discovery_date"] > after
             results.append(
                 {
                     **entry,
                     "authority": {"type": authority["type"], "url": authority["url"],},
                     "fetcher": {
                         "name": fetcher["name"],
                         "version": fetcher["version"],
                     },
                 }
             )
 
         if len(results) > limit:
             results.pop()
             assert len(results) == limit
             last_result = results[-1]
             next_page_token: Optional[bytes] = msgpack_dumps(
                 (
                     last_result["discovery_date"],
                     self._metadata_fetcher_key(last_result["fetcher"]),
                 )
             )
         else:
             next_page_token = None
 
         return {
             "next_page_token": next_page_token,
             "results": results,
         }
 
     def metadata_fetcher_add(
         self, name: str, version: str, metadata: Dict[str, Any]
     ) -> None:
         fetcher = {
             "name": name,
             "version": version,
             "metadata": metadata,
         }
         key = self._metadata_fetcher_key(fetcher)
         if key not in self._metadata_fetchers:
             self._metadata_fetchers[key] = fetcher
 
     def metadata_fetcher_get(self, name: str, version: str) -> Optional[Dict[str, Any]]:
         return self._metadata_fetchers.get(
             self._metadata_fetcher_key({"name": name, "version": version})
         )
 
     def metadata_authority_add(
         self, type: str, url: str, metadata: Dict[str, Any]
     ) -> None:
         authority = {
             "type": type,
             "url": url,
             "metadata": metadata,
         }
         key = self._metadata_authority_key(authority)
         self._metadata_authorities[key] = authority
 
     def metadata_authority_get(self, type: str, url: str) -> Optional[Dict[str, Any]]:
         return self._metadata_authorities.get(
             self._metadata_authority_key({"type": type, "url": url})
         )
 
     def _get_origin_url(self, origin):
         if isinstance(origin, str):
             return origin
         else:
             raise TypeError("origin must be a string.")
 
     def _person_add(self, person):
         key = ("person", person.fullname)
         if key not in self._objects:
             self._persons[person.fullname] = person
             self._objects[key].append(key)
 
         return self._persons[person.fullname]
 
     @staticmethod
     def _content_key(content):
         """ A stable key and the algorithm for a content"""
         if isinstance(content, BaseContent):
             content = content.to_dict()
         return tuple((key, content.get(key)) for key in sorted(DEFAULT_ALGORITHMS))
 
     @staticmethod
     def _metadata_fetcher_key(fetcher: Dict) -> FetcherKey:
         return (fetcher["name"], fetcher["version"])
 
     @staticmethod
     def _metadata_authority_key(authority: Dict) -> Hashable:
         return (authority["type"], authority["url"])
 
     def diff_directories(self, from_dir, to_dir, track_renaming=False):
         raise NotImplementedError("InMemoryStorage.diff_directories")
 
     def diff_revisions(self, from_rev, to_rev, track_renaming=False):
         raise NotImplementedError("InMemoryStorage.diff_revisions")
 
     def diff_revision(self, revision, track_renaming=False):
         raise NotImplementedError("InMemoryStorage.diff_revision")
 
     def clear_buffers(self, object_types: Optional[Iterable[str]] = None) -> None:
         """Do nothing
 
         """
         return None
 
     def flush(self, object_types: Optional[Iterable[str]] = None) -> Dict:
         return {}
diff --git a/swh/storage/sql/30-swh-schema.sql b/swh/storage/sql/30-swh-schema.sql
index 8de14afd..e9e91cad 100644
--- a/swh/storage/sql/30-swh-schema.sql
+++ b/swh/storage/sql/30-swh-schema.sql
@@ -1,482 +1,482 @@
 ---
 --- SQL implementation of the Software Heritage data model
 ---
 
 -- schema versions
 create table dbversion
 (
   version     int primary key,
   release     timestamptz,
   description text
 );
 
 comment on table dbversion is 'Details of current db version';
 comment on column dbversion.version is 'SQL schema version';
 comment on column dbversion.release is 'Version deployment timestamp';
 comment on column dbversion.description is 'Release description';
 
 -- latest schema version
 insert into dbversion(version, release, description)
-      values(155, now(), 'Work In Progress');
+      values(156, now(), 'Work In Progress');
 
 -- a SHA1 checksum
 create domain sha1 as bytea check (length(value) = 20);
 
 -- a Git object ID, i.e., a Git-style salted SHA1 checksum
 create domain sha1_git as bytea check (length(value) = 20);
 
 -- a SHA256 checksum
 create domain sha256 as bytea check (length(value) = 32);
 
 -- a blake2 checksum
 create domain blake2s256 as bytea check (length(value) = 32);
 
 -- UNIX path (absolute, relative, individual path component, etc.)
 create domain unix_path as bytea;
 
 -- a set of UNIX-like access permissions, as manipulated by, e.g., chmod
 create domain file_perms as int;
 
 
 -- Checksums about actual file content. Note that the content itself is not
 -- stored in the DB, but on external (key-value) storage. A single checksum is
 -- used as key there, but the other can be used to verify that we do not inject
 -- content collisions not knowingly.
 create table content
 (
   sha1       sha1 not null,
   sha1_git   sha1_git not null,
   sha256     sha256 not null,
   blake2s256 blake2s256 not null,
   length     bigint not null,
   ctime      timestamptz not null default now(),
              -- creation time, i.e. time of (first) injection into the storage
   status     content_status not null default 'visible',
   object_id  bigserial
 );
 
 comment on table content is 'Checksums of file content which is actually stored externally';
 comment on column content.sha1 is 'Content sha1 hash';
 comment on column content.sha1_git is 'Git object sha1 hash';
 comment on column content.sha256 is 'Content Sha256 hash';
 comment on column content.blake2s256 is 'Content blake2s hash';
 comment on column content.length is 'Content length';
 comment on column content.ctime is 'First seen time';
 comment on column content.status is 'Content status (absent, visible, hidden)';
 comment on column content.object_id is 'Content identifier';
 
 
 -- An origin is a place, identified by an URL, where software source code
 -- artifacts can be found. We support different kinds of origins, e.g., git and
 -- other VCS repositories, web pages that list tarballs URLs (e.g.,
 -- http://www.kernel.org), indirect tarball URLs (e.g.,
 -- http://www.example.org/latest.tar.gz), etc. The key feature of an origin is
 -- that it can be *fetched* from (wget, git clone, svn checkout, etc.) to
 -- retrieve all the contained software.
 create table origin
 (
   id       bigserial not null,
   url      text not null
 );
 
 comment on column origin.id is 'Artifact origin id';
 comment on column origin.url is 'URL of origin';
 
 
 -- Content blobs observed somewhere, but not ingested into the archive for
 -- whatever reason. This table is separate from the content table as we might
 -- not have the sha1 checksum of skipped contents (for instance when we inject
 -- git repositories, objects that are too big will be skipped here, and we will
 -- only know their sha1_git). 'reason' contains the reason the content was
 -- skipped. origin is a nullable column allowing to find out which origin
 -- contains that skipped content.
 create table skipped_content
 (
   sha1       sha1,
   sha1_git   sha1_git,
   sha256     sha256,
   blake2s256 blake2s256,
   length     bigint not null,
   ctime      timestamptz not null default now(),
   status     content_status not null default 'absent',
   reason     text not null,
   origin     bigint,
   object_id  bigserial
 );
 
 comment on table skipped_content is 'Content blobs observed, but not ingested in the archive';
 comment on column skipped_content.sha1 is 'Skipped content sha1 hash';
 comment on column skipped_content.sha1_git is 'Git object sha1 hash';
 comment on column skipped_content.sha256 is 'Skipped content sha256 hash';
 comment on column skipped_content.blake2s256 is 'Skipped content blake2s hash';
 comment on column skipped_content.length is 'Skipped content length';
 comment on column skipped_content.ctime is 'First seen time';
 comment on column skipped_content.status is 'Skipped content status (absent, visible, hidden)';
 comment on column skipped_content.reason is 'Reason for skipping';
 comment on column skipped_content.origin is 'Origin table identifier';
 comment on column skipped_content.object_id is 'Skipped content identifier';
 
 
 -- A file-system directory.  A directory is a list of directory entries (see
 -- tables: directory_entry_{dir,file}).
 --
 -- To list the contents of a directory:
 -- 1. list the contained directory_entry_dir using array dir_entries
 -- 2. list the contained directory_entry_file using array file_entries
 -- 3. list the contained directory_entry_rev using array rev_entries
 -- 4. UNION
 --
 -- Synonyms/mappings:
 -- * git: tree
 create table directory
 (
   id            sha1_git not null,
   dir_entries   bigint[],  -- sub-directories, reference directory_entry_dir
   file_entries  bigint[],  -- contained files, reference directory_entry_file
   rev_entries   bigint[],  -- mounted revisions, reference directory_entry_rev
   object_id     bigserial  -- short object identifier
 );
 
 comment on table directory is 'Contents of a directory, synonymous to tree (git)';
 comment on column directory.id is 'Git object sha1 hash';
 comment on column directory.dir_entries is 'Sub-directories, reference directory_entry_dir';
 comment on column directory.file_entries is 'Contained files, reference directory_entry_file';
 comment on column directory.rev_entries is 'Mounted revisions, reference directory_entry_rev';
 comment on column directory.object_id is 'Short object identifier';
 
 
 -- A directory entry pointing to a (sub-)directory.
 create table directory_entry_dir
 (
   id      bigserial,
   target  sha1_git not null,   -- id of target directory
   name    unix_path not null,  -- path name, relative to containing dir
   perms   file_perms not null  -- unix-like permissions
 );
 
 comment on table directory_entry_dir is 'Directory entry for directory';
 comment on column directory_entry_dir.id is 'Directory identifier';
 comment on column directory_entry_dir.target is 'Target directory identifier';
 comment on column directory_entry_dir.name is 'Path name, relative to containing directory';
 comment on column directory_entry_dir.perms is 'Unix-like permissions';
 
 
 -- A directory entry pointing to a file content.
 create table directory_entry_file
 (
   id      bigserial,
   target  sha1_git not null,   -- id of target file
   name    unix_path not null,  -- path name, relative to containing dir
   perms   file_perms not null  -- unix-like permissions
 );
 
 comment on table directory_entry_file is 'Directory entry for file';
 comment on column directory_entry_file.id is 'File identifier';
 comment on column directory_entry_file.target is 'Target file identifier';
 comment on column directory_entry_file.name is 'Path name, relative to containing directory';
 comment on column directory_entry_file.perms is 'Unix-like permissions';
 
 
 -- A directory entry pointing to a revision.
 create table directory_entry_rev
 (
   id      bigserial,
   target  sha1_git not null,   -- id of target revision
   name    unix_path not null,  -- path name, relative to containing dir
   perms   file_perms not null  -- unix-like permissions
 );
 
 comment on table directory_entry_rev is 'Directory entry for revision';
 comment on column directory_entry_dir.id is 'Revision identifier';
 comment on column directory_entry_dir.target is 'Target revision in identifier';
 comment on column directory_entry_dir.name is 'Path name, relative to containing directory';
 comment on column directory_entry_dir.perms is 'Unix-like permissions';
 
 
 -- A person referenced by some source code artifacts, e.g., a VCS revision or
 -- release metadata.
 create table person
 (
   id        bigserial,
   name      bytea,          -- advisory: not null if we managed to parse a name
   email     bytea,          -- advisory: not null if we managed to parse an email
   fullname  bytea not null  -- freeform specification; what is actually used in the checksums
                             --     will usually be of the form 'name <email>'
 );
 
 comment on table person is 'Person referenced in code artifact release metadata';
 comment on column person.id is 'Person identifier';
 comment on column person.name is 'Name';
 comment on column person.email is 'Email';
 comment on column person.fullname is 'Full name (raw name)';
 
 
 -- The state of a source code tree at a specific point in time.
 --
 -- Synonyms/mappings:
 -- * git / subversion / etc: commit
 -- * tarball: a specific tarball
 --
 -- Revisions are organized as DAGs. Each revision points to 0, 1, or more (in
 -- case of merges) parent revisions. Each revision points to a directory, i.e.,
 -- a file-system tree containing files and directories.
 create table revision
 (
   id                    sha1_git not null,
   date                  timestamptz,
   date_offset           smallint,
   committer_date        timestamptz,
   committer_date_offset smallint,
   type                  revision_type not null,
   directory             sha1_git,  -- source code 'root' directory
   message               bytea,
   author                bigint,
   committer             bigint,
   synthetic             boolean not null default false,  -- true iff revision has been created by Software Heritage
   metadata              jsonb,  -- extra metadata (tarball checksums, extra commit information, etc...)
   object_id             bigserial,
   date_neg_utc_offset   boolean,
   committer_date_neg_utc_offset boolean
 );
 
 comment on table revision is 'A revision represents the state of a source code tree at a specific point in time';
 comment on column revision.id is 'Git-style SHA1 commit identifier';
 comment on column revision.date is 'Author timestamp as UNIX epoch';
 comment on column revision.date_offset is 'Author timestamp timezone, as minute offsets from UTC';
 comment on column revision.date_neg_utc_offset is 'True indicates a -0 UTC offset on author timestamp';
 comment on column revision.committer_date is 'Committer timestamp as UNIX epoch';
 comment on column revision.committer_date_offset is 'Committer timestamp timezone, as minute offsets from UTC';
 comment on column revision.committer_date_neg_utc_offset is 'True indicates a -0 UTC offset on committer timestamp';
 comment on column revision.type is 'Type of revision';
 comment on column revision.directory is 'Directory identifier';
 comment on column revision.message is 'Commit message';
 comment on column revision.author is 'Author identity';
 comment on column revision.committer is 'Committer identity';
 comment on column revision.synthetic is 'True iff revision has been synthesized by Software Heritage';
 comment on column revision.metadata is 'Extra revision metadata';
 comment on column revision.object_id is 'Non-intrinsic, sequential object identifier';
 
 
 -- either this table or the sha1_git[] column on the revision table
 create table revision_history
 (
   id           sha1_git not null,
   parent_id    sha1_git not null,
   parent_rank  int not null default 0
     -- parent position in merge commits, 0-based
 );
 
 comment on table revision_history is 'Sequence of revision history with parent and position in history';
 comment on column revision_history.id is 'Revision history git object sha1 checksum';
 comment on column revision_history.parent_id is 'Parent revision git object identifier';
 comment on column revision_history.parent_rank is 'Parent position in merge commits, 0-based';
 
 
 -- Crawling history of software origins visited by Software Heritage. Each
 -- visit is a 3-way mapping between a software origin, a timestamp, and a
 -- snapshot object capturing the full-state of the origin at visit time.
 create table origin_visit
 (
   origin       bigint not null,
   visit        bigint not null,
   date         timestamptz not null,
   type         text not null
 );
 
 comment on column origin_visit.origin is 'Visited origin';
 comment on column origin_visit.visit is 'Sequential visit number for the origin';
 comment on column origin_visit.date is 'Visit timestamp';
 comment on column origin_visit.type is 'Type of loader that did the visit (hg, git, ...)';
 
 
 -- Crawling history of software origin visits by Software Heritage. Each
 -- visit see its history change through new origin visit status updates
 create table origin_visit_status
 (
   origin   bigint not null,
   visit    bigint not null,
   date     timestamptz not null,
   status   origin_visit_state not null,
   metadata jsonb,
   snapshot sha1_git
 );
 
 comment on column origin_visit_status.origin is 'Origin concerned by the visit update';
 comment on column origin_visit_status.visit is 'Visit concerned by the visit update';
 comment on column origin_visit_status.date is 'Visit update timestamp';
 comment on column origin_visit_status.status is 'Visit status (ongoing, failed, full)';
 comment on column origin_visit_status.metadata is 'Optional origin visit metadata';
 comment on column origin_visit_status.snapshot is 'Optional, possibly partial, snapshot of the origin visit. It can be partial.';
 
 
 -- A snapshot represents the entire state of a software origin as crawled by
 -- Software Heritage. This table is a simple mapping between (public) intrinsic
 -- snapshot identifiers and (private) numeric sequential identifiers.
 create table snapshot
 (
   object_id  bigserial not null,  -- PK internal object identifier
   id         sha1_git not null    -- snapshot intrinsic identifier
 );
 
 comment on table snapshot is 'State of a software origin as crawled by Software Heritage';
 comment on column snapshot.object_id is 'Internal object identifier';
 comment on column snapshot.id is 'Intrinsic snapshot identifier';
 
 
 -- Each snapshot associate "branch" names to other objects in the Software
 -- Heritage Merkle DAG. This table describes branches as mappings between names
 -- and target typed objects.
 create table snapshot_branch
 (
   object_id    bigserial not null,  -- PK internal object identifier
   name         bytea not null,      -- branch name, e.g., "master" or "feature/drag-n-drop"
   target       bytea,               -- target object identifier, e.g., a revision identifier
   target_type  snapshot_target      -- target object type, e.g., "revision"
 );
 
 comment on table snapshot_branch is 'Associates branches with objects in Heritage Merkle DAG';
 comment on column snapshot_branch.object_id is 'Internal object identifier';
 comment on column snapshot_branch.name is 'Branch name';
 comment on column snapshot_branch.target is 'Target object identifier';
 comment on column snapshot_branch.target_type is 'Target object type';
 
 
 -- Mapping between snapshots and their branches.
 create table snapshot_branches
 (
   snapshot_id  bigint not null,  -- snapshot identifier, ref. snapshot.object_id
   branch_id    bigint not null   -- branch identifier, ref. snapshot_branch.object_id
 );
 
 comment on table snapshot_branches is 'Mapping between snapshot and their branches';
 comment on column snapshot_branches.snapshot_id is 'Snapshot identifier';
 comment on column snapshot_branches.branch_id is 'Branch identifier';
 
 
 -- A "memorable" point in time in the development history of a software
 -- project.
 --
 -- Synonyms/mappings:
 -- * git: tag (of the annotated kind, otherwise they are just references)
 -- * tarball: the release version number
 create table release
 (
   id          sha1_git not null,
   target      sha1_git,
   date        timestamptz,
   date_offset smallint,
   name        bytea,
   comment     bytea,
   author      bigint,
   synthetic   boolean not null default false,  -- true iff release has been created by Software Heritage
   object_id   bigserial,
   target_type object_type not null,
   date_neg_utc_offset  boolean
 );
 
 comment on table release is 'Details of a software release, synonymous with
  a tag (git) or version number (tarball)';
 comment on column release.id is 'Release git identifier';
 comment on column release.target is 'Target git identifier';
 comment on column release.date is 'Release timestamp';
 comment on column release.date_offset is 'Timestamp offset from UTC';
 comment on column release.name is 'Name';
 comment on column release.comment is 'Comment';
 comment on column release.author is 'Author';
 comment on column release.synthetic is 'Indicates if created by Software Heritage';
 comment on column release.object_id is 'Object identifier';
 comment on column release.target_type is 'Object type (''content'', ''directory'', ''revision'',
  ''release'', ''snapshot'')';
 comment on column release.date_neg_utc_offset is 'True indicates -0 UTC offset for release timestamp';
 
 -- Tools
 create table metadata_fetcher
 (
   id            serial  not null,
   name          text    not null,
   version       text    not null,
   metadata      jsonb   not null
 );
 
 comment on table metadata_fetcher is 'Tools used to retrieve metadata';
 comment on column metadata_fetcher.id is 'Internal identifier of the fetcher';
 comment on column metadata_fetcher.name is 'Fetcher name';
 comment on column metadata_fetcher.version is 'Fetcher version';
 comment on column metadata_fetcher.metadata is 'Extra information about the fetcher';
 
 
 create table metadata_authority
 (
   id            serial  not null,
   type          text    not null,
   url           text    not null,
   metadata      jsonb   not null
 );
 
 comment on table metadata_authority is 'Metadata authority information';
 comment on column metadata_authority.id is 'Internal identifier of the authority';
 comment on column metadata_authority.type is 'Type of authority (deposit/forge/registry)';
 comment on column metadata_authority.url is 'Authority''s uri';
 comment on column metadata_authority.metadata is 'Other metadata about authority';
 
 
 -- Discovery of metadata during a listing, loading, deposit or external_catalog of an origin
 -- also provides a translation to a defined json schema using a translation tool (tool_id)
 create table origin_metadata
 (
   id             bigserial     not null,  -- PK internal object identifier
   origin_id      bigint        not null,  -- references origin(id)
   discovery_date timestamptz   not null,  -- when it was extracted
   authority_id   bigint        not null,
   fetcher_id     bigint        not null,
   format         text          not null default 'sword-v2-atom-codemeta-v2-in-json',
   metadata       bytea         not null
 );
 
 comment on table origin_metadata is 'keeps all metadata found concerning an origin';
 comment on column origin_metadata.id is 'the origin_metadata object''s id';
 comment on column origin_metadata.origin_id is 'the origin id for which the metadata was found';
 comment on column origin_metadata.discovery_date is 'the date of retrieval';
 comment on column origin_metadata.authority_id is 'the metadata provider: github, openhub, deposit, etc.';
 comment on column origin_metadata.fetcher_id is 'the tool used for extracting metadata: loaders, crawlers, etc.';
 comment on column origin_metadata.format is 'name of the format of metadata, used by readers to interpret it.';
 comment on column origin_metadata.metadata is 'original metadata in opaque format';
 
 
 -- Keep a cache of object counts
 create table object_counts
 (
   object_type text,             -- table for which we're counting objects (PK)
   value bigint,                 -- count of objects in the table
   last_update timestamptz,      -- last update for the object count in this table
   single_update boolean         -- whether we update this table standalone (true) or through bucketed counts (false)
 );
 
 comment on table object_counts is 'Cache of object counts';
 comment on column object_counts.object_type is 'Object type (''content'', ''directory'', ''revision'',
  ''release'', ''snapshot'')';
 comment on column object_counts.value is 'Count of objects in the table';
 comment on column object_counts.last_update is 'Last update for object count';
 comment on column object_counts.single_update is 'standalone (true) or bucketed counts (false)';
 
 
 create table object_counts_bucketed
 (
     line serial not null,       -- PK
     object_type text not null,  -- table for which we're counting objects
     identifier text not null,   -- identifier across which we're bucketing objects
     bucket_start bytea,         -- lower bound (inclusive) for the bucket
     bucket_end bytea,           -- upper bound (exclusive) for the bucket
     value bigint,               -- count of objects in the bucket
     last_update timestamptz     -- last update for the object count in this bucket
 );
 
 comment on table object_counts_bucketed is 'Bucketed count for objects ordered by type';
 comment on column object_counts_bucketed.line is 'Auto incremented idenitfier value';
 comment on column object_counts_bucketed.object_type is 'Object type (''content'', ''directory'', ''revision'',
  ''release'', ''snapshot'')';
 comment on column object_counts_bucketed.identifier is 'Common identifier for bucketed objects';
 comment on column object_counts_bucketed.bucket_start is 'Lower bound (inclusive) for the bucket';
 comment on column object_counts_bucketed.bucket_end is 'Upper bound (exclusive) for the bucket';
 comment on column object_counts_bucketed.value is 'Count of objects in the bucket';
 comment on column object_counts_bucketed.last_update is 'Last update for the object count in this bucket';
diff --git a/swh/storage/sql/40-swh-func.sql b/swh/storage/sql/40-swh-func.sql
index a2030a7e..0a8a0a7f 100644
--- a/swh/storage/sql/40-swh-func.sql
+++ b/swh/storage/sql/40-swh-func.sql
@@ -1,948 +1,949 @@
 create or replace function hash_sha1(text)
        returns text
 as $$
    select encode(digest($1, 'sha1'), 'hex')
 $$ language sql strict immutable;
 
 comment on function hash_sha1(text) is 'Compute SHA1 hash as text';
 
 -- create a temporary table called tmp_TBLNAME, mimicking existing table
 -- TBLNAME
 --
 -- Args:
 --     tblname: name of the table to mimic
 create or replace function swh_mktemp(tblname regclass)
     returns void
     language plpgsql
 as $$
 begin
     execute format('
 	create temporary table if not exists tmp_%1$I
 	    (like %1$I including defaults)
 	    on commit delete rows;
       alter table tmp_%1$I drop column if exists object_id;
 	', tblname);
     return;
 end
 $$;
 
 -- create a temporary table for directory entries called tmp_TBLNAME,
 -- mimicking existing table TBLNAME with an extra dir_id (sha1_git)
 -- column, and dropping the id column.
 --
 -- This is used to create the tmp_directory_entry_<foo> tables.
 --
 -- Args:
 --     tblname: name of the table to mimic
 create or replace function swh_mktemp_dir_entry(tblname regclass)
     returns void
     language plpgsql
 as $$
 begin
     execute format('
 	create temporary table if not exists tmp_%1$I
 	    (like %1$I including defaults, dir_id sha1_git)
 	    on commit delete rows;
         alter table tmp_%1$I drop column if exists id;
 	', tblname);
     return;
 end
 $$;
 
 -- create a temporary table for revisions called tmp_revisions,
 -- mimicking existing table revision, replacing the foreign keys to
 -- people with an email and name field
 --
 create or replace function swh_mktemp_revision()
     returns void
     language sql
 as $$
     create temporary table if not exists tmp_revision (
         like revision including defaults,
         author_fullname bytea,
         author_name bytea,
         author_email bytea,
         committer_fullname bytea,
         committer_name bytea,
         committer_email bytea
     ) on commit delete rows;
     alter table tmp_revision drop column if exists author;
     alter table tmp_revision drop column if exists committer;
     alter table tmp_revision drop column if exists object_id;
 $$;
 
 -- create a temporary table for releases called tmp_release,
 -- mimicking existing table release, replacing the foreign keys to
 -- people with an email and name field
 --
 create or replace function swh_mktemp_release()
     returns void
     language sql
 as $$
     create temporary table if not exists tmp_release (
         like release including defaults,
         author_fullname bytea,
         author_name bytea,
         author_email bytea
     ) on commit delete rows;
     alter table tmp_release drop column if exists author;
     alter table tmp_release drop column if exists object_id;
 $$;
 
 -- create a temporary table for the branches of a snapshot
 create or replace function swh_mktemp_snapshot_branch()
     returns void
     language sql
 as $$
   create temporary table if not exists tmp_snapshot_branch (
       name bytea not null,
       target bytea,
       target_type snapshot_target
   ) on commit delete rows;
 $$;
 
 -- a content signature is a set of cryptographic checksums that we use to
 -- uniquely identify content, for the purpose of verifying if we already have
 -- some content or not during content injection
 create type content_signature as (
     sha1       sha1,
     sha1_git   sha1_git,
     sha256     sha256,
     blake2s256 blake2s256
 );
 
 
 -- check which entries of tmp_skipped_content are missing from skipped_content
 --
 -- operates in bulk: 0. swh_mktemp(skipped_content), 1. COPY to tmp_skipped_content,
 -- 2. call this function
 create or replace function swh_skipped_content_missing()
     returns setof content_signature
     language plpgsql
 as $$
 begin
     return query
 	select sha1, sha1_git, sha256, blake2s256 from tmp_skipped_content t
 	where not exists
 	(select 1 from skipped_content s where
 	    s.sha1 is not distinct from t.sha1 and
 	    s.sha1_git is not distinct from t.sha1_git and
 	    s.sha256 is not distinct from t.sha256);
     return;
 end
 $$;
 
 
 -- add tmp_content entries to content, skipping duplicates
 --
 -- operates in bulk: 0. swh_mktemp(content), 1. COPY to tmp_content,
 -- 2. call this function
 create or replace function swh_content_add()
     returns void
     language plpgsql
 as $$
 begin
     insert into content (sha1, sha1_git, sha256, blake2s256, length, status, ctime)
         select distinct sha1, sha1_git, sha256, blake2s256, length, status, ctime from tmp_content;
     return;
 end
 $$;
 
 
 -- add tmp_skipped_content entries to skipped_content, skipping duplicates
 --
 -- operates in bulk: 0. swh_mktemp(skipped_content), 1. COPY to tmp_skipped_content,
 -- 2. call this function
 create or replace function swh_skipped_content_add()
     returns void
     language plpgsql
 as $$
 begin
     insert into skipped_content (sha1, sha1_git, sha256, blake2s256, length, status, reason, origin)
         select distinct sha1, sha1_git, sha256, blake2s256, length, status, reason, origin
 	from tmp_skipped_content
 	where (coalesce(sha1, ''), coalesce(sha1_git, ''), coalesce(sha256, '')) in (
             select coalesce(sha1, ''), coalesce(sha1_git, ''), coalesce(sha256, '')
             from swh_skipped_content_missing()
         );
         -- TODO XXX use postgres 9.5 "UPSERT" support here, when available.
         -- Specifically, using "INSERT .. ON CONFLICT IGNORE" we can avoid
         -- the extra swh_skipped_content_missing() query here.
     return;
 end
 $$;
 
 -- Update content entries from temporary table.
 -- (columns are potential new columns added to the schema, this cannot be empty)
 --
 create or replace function swh_content_update(columns_update text[])
     returns void
     language plpgsql
 as $$
 declare
    query text;
    tmp_array text[];
 begin
     if array_length(columns_update, 1) = 0 then
         raise exception 'Please, provide the list of column names to update.';
     end if;
 
     tmp_array := array(select format('%1$s=t.%1$s', unnest) from unnest(columns_update));
 
     query = format('update content set %s
                     from tmp_content t where t.sha1 = content.sha1',
                     array_to_string(tmp_array, ', '));
 
     execute query;
 
     return;
 end
 $$;
 
 comment on function swh_content_update(text[]) IS 'Update existing content''s columns';
 
 
 create type directory_entry_type as enum('file', 'dir', 'rev');
 
 
 -- Add tmp_directory_entry_* entries to directory_entry_* and directory,
 -- skipping duplicates in directory_entry_*.  This is a generic function that
 -- works on all kind of directory entries.
 --
 -- operates in bulk: 0. swh_mktemp_dir_entry('directory_entry_*'), 1 COPY to
 -- tmp_directory_entry_*, 2. call this function
 --
 -- Assumption: this function is used in the same transaction that inserts the
 -- context directory in table "directory".
 create or replace function swh_directory_entry_add(typ directory_entry_type)
     returns void
     language plpgsql
 as $$
 begin
     execute format('
     insert into directory_entry_%1$s (target, name, perms)
     select distinct t.target, t.name, t.perms
     from tmp_directory_entry_%1$s t
     where not exists (
     select 1
     from directory_entry_%1$s i
     where t.target = i.target and t.name = i.name and t.perms = i.perms)
    ', typ);
 
     execute format('
     with new_entries as (
 	select t.dir_id, array_agg(i.id) as entries
 	from tmp_directory_entry_%1$s t
 	inner join directory_entry_%1$s i
 	using (target, name, perms)
 	group by t.dir_id
     )
     update tmp_directory as d
     set %1$s_entries = new_entries.entries
     from new_entries
     where d.id = new_entries.dir_id
     ', typ);
 
     return;
 end
 $$;
 
 -- Insert the data from tmp_directory, tmp_directory_entry_file,
 -- tmp_directory_entry_dir, tmp_directory_entry_rev into their final
 -- tables.
 --
 -- Prerequisites:
 --  directory ids in tmp_directory
 --  entries in tmp_directory_entry_{file,dir,rev}
 --
 create or replace function swh_directory_add()
     returns void
     language plpgsql
 as $$
 begin
     perform swh_directory_entry_add('file');
     perform swh_directory_entry_add('dir');
     perform swh_directory_entry_add('rev');
 
     insert into directory
     select * from tmp_directory t
     where not exists (
         select 1 from directory d
 	where d.id = t.id);
 
     return;
 end
 $$;
 
 -- a directory listing entry with all the metadata
 --
 -- can be used to list a directory, and retrieve all the data in one go.
 create type directory_entry as
 (
   dir_id   sha1_git,     -- id of the parent directory
   type     directory_entry_type,  -- type of entry
   target   sha1_git,     -- id of target
   name     unix_path,    -- path name, relative to containing dir
   perms    file_perms,   -- unix-like permissions
   status   content_status,  -- visible or absent
   sha1     sha1,            -- content if sha1 if type is not dir
   sha1_git sha1_git,        -- content's sha1 git if type is not dir
   sha256   sha256,          -- content's sha256 if type is not dir
   length   bigint           -- content length if type is not dir
 );
 
 
 -- List a single level of directory walked_dir_id
 -- FIXME: order by name is not correct. For git, we need to order by
 -- lexicographic order but as if a trailing / is present in directory
 -- name
 create or replace function swh_directory_walk_one(walked_dir_id sha1_git)
     returns setof directory_entry
     language sql
     stable
 as $$
     with dir as (
 	select id as dir_id, dir_entries, file_entries, rev_entries
 	from directory
 	where id = walked_dir_id),
     ls_d as (select dir_id, unnest(dir_entries) as entry_id from dir),
     ls_f as (select dir_id, unnest(file_entries) as entry_id from dir),
     ls_r as (select dir_id, unnest(rev_entries) as entry_id from dir)
     (select dir_id, 'dir'::directory_entry_type as type,
             e.target, e.name, e.perms, NULL::content_status,
             NULL::sha1, NULL::sha1_git, NULL::sha256, NULL::bigint
      from ls_d
      left join directory_entry_dir e on ls_d.entry_id = e.id)
     union
     (select dir_id, 'file'::directory_entry_type as type,
             e.target, e.name, e.perms, c.status,
             c.sha1, c.sha1_git, c.sha256, c.length
      from ls_f
      left join directory_entry_file e on ls_f.entry_id = e.id
      left join content c on e.target = c.sha1_git)
     union
     (select dir_id, 'rev'::directory_entry_type as type,
             e.target, e.name, e.perms, NULL::content_status,
             NULL::sha1, NULL::sha1_git, NULL::sha256, NULL::bigint
      from ls_r
      left join directory_entry_rev e on ls_r.entry_id = e.id)
     order by name;
 $$;
 
 -- List recursively the revision directory arborescence
 create or replace function swh_directory_walk(walked_dir_id sha1_git)
     returns setof directory_entry
     language sql
     stable
 as $$
     with recursive entries as (
         select dir_id, type, target, name, perms, status, sha1, sha1_git,
                sha256, length
         from swh_directory_walk_one(walked_dir_id)
         union all
         select dir_id, type, target, (dirname || '/' || name)::unix_path as name,
                perms, status, sha1, sha1_git, sha256, length
         from (select (swh_directory_walk_one(dirs.target)).*, dirs.name as dirname
               from (select target, name from entries where type = 'dir') as dirs) as with_parent
     )
     select dir_id, type, target, name, perms, status, sha1, sha1_git, sha256, length
     from entries
 $$;
 
 -- Find a directory entry by its path
 create or replace function swh_find_directory_entry_by_path(
     walked_dir_id sha1_git,
     dir_or_content_path bytea[])
     returns directory_entry
     language plpgsql
 as $$
 declare
     end_index integer;
     paths bytea default '';
     path bytea;
     res bytea[];
     r record;
 begin
     end_index := array_upper(dir_or_content_path, 1);
     res[1] := walked_dir_id;
 
     for i in 1..end_index
     loop
         path := dir_or_content_path[i];
         -- concatenate path for patching the name in the result record (if we found it)
         if i = 1 then
             paths = path;
         else
             paths := paths || '/' || path;  -- concatenate paths
         end if;
 
         if i <> end_index then
             select *
             from swh_directory_walk_one(res[i] :: sha1_git)
             where name=path
             and type = 'dir'
             limit 1 into r;
         else
             select *
             from swh_directory_walk_one(res[i] :: sha1_git)
             where name=path
             limit 1 into r;
         end if;
 
         -- find the path
         if r is null then
            return null;
         else
             -- store the next dir to lookup the next local path from
             res[i+1] := r.target;
         end if;
     end loop;
 
     -- at this moment, r is the result. Patch its 'name' with the full path before returning it.
     r.name := paths;
     return r;
 end
 $$;
 
 -- List all revision IDs starting from a given revision, going back in time
 --
 -- TODO ordering: should be breadth-first right now (what do we want?)
 -- TODO ordering: ORDER BY parent_rank somewhere?
 create or replace function swh_revision_list(root_revisions bytea[], num_revs bigint default NULL)
     returns table (id sha1_git, parents bytea[])
     language sql
     stable
 as $$
     with recursive full_rev_list(id) as (
         (select id from revision where id = ANY(root_revisions))
         union
         (select h.parent_id
          from revision_history as h
          join full_rev_list on h.id = full_rev_list.id)
     ),
     rev_list as (select id from full_rev_list limit num_revs)
     select rev_list.id as id,
            array(select rh.parent_id::bytea
                  from revision_history rh
                  where rh.id = rev_list.id
                  order by rh.parent_rank
                 ) as parent
     from rev_list;
 $$;
 
 
 -- Detailed entry for a revision
 create type revision_entry as
 (
   id                             sha1_git,
   date                           timestamptz,
   date_offset                    smallint,
   date_neg_utc_offset            boolean,
   committer_date                 timestamptz,
   committer_date_offset          smallint,
   committer_date_neg_utc_offset  boolean,
   type                           revision_type,
   directory                      sha1_git,
   message                        bytea,
   author_id                      bigint,
   author_fullname                bytea,
   author_name                    bytea,
   author_email                   bytea,
   committer_id                   bigint,
   committer_fullname             bytea,
   committer_name                 bytea,
   committer_email                bytea,
   metadata                       jsonb,
   synthetic                      boolean,
   parents                        bytea[],
   object_id                      bigint
 );
 
 
 -- "git style" revision log. Similar to swh_revision_list(), but returning all
 -- information associated to each revision, and expanding authors/committers
 create or replace function swh_revision_log(root_revisions bytea[], num_revs bigint default NULL)
     returns setof revision_entry
     language sql
     stable
 as $$
     select t.id, r.date, r.date_offset, r.date_neg_utc_offset,
            r.committer_date, r.committer_date_offset, r.committer_date_neg_utc_offset,
            r.type, r.directory, r.message,
            a.id, a.fullname, a.name, a.email,
            c.id, c.fullname, c.name, c.email,
            r.metadata, r.synthetic, t.parents, r.object_id
     from swh_revision_list(root_revisions, num_revs) as t
     left join revision r on t.id = r.id
     left join person a on a.id = r.author
     left join person c on c.id = r.committer;
 $$;
 
 
 -- Detailed entry for a release
 create type release_entry as
 (
   id                   sha1_git,
   target               sha1_git,
   target_type          object_type,
   date                 timestamptz,
   date_offset          smallint,
   date_neg_utc_offset  boolean,
   name                 bytea,
   comment              bytea,
   synthetic            boolean,
   author_id            bigint,
   author_fullname      bytea,
   author_name          bytea,
   author_email         bytea,
   object_id            bigint
 );
 
 -- Create entries in person from tmp_revision
 create or replace function swh_person_add_from_revision()
     returns void
     language plpgsql
 as $$
 begin
     with t as (
         select author_fullname as fullname, author_name as name, author_email as email from tmp_revision
     union
         select committer_fullname as fullname, committer_name as name, committer_email as email from tmp_revision
     ) insert into person (fullname, name, email)
     select distinct on (fullname) fullname, name, email from t
     where not exists (
         select 1
         from person p
         where t.fullname = p.fullname
     );
     return;
 end
 $$;
 
 
 -- Create entries in revision from tmp_revision
 create or replace function swh_revision_add()
     returns void
     language plpgsql
 as $$
 begin
     perform swh_person_add_from_revision();
 
     insert into revision (id, date, date_offset, date_neg_utc_offset, committer_date, committer_date_offset, committer_date_neg_utc_offset, type, directory, message, author, committer, metadata, synthetic)
     select t.id, t.date, t.date_offset, t.date_neg_utc_offset, t.committer_date, t.committer_date_offset, t.committer_date_neg_utc_offset, t.type, t.directory, t.message, a.id, c.id, t.metadata, t.synthetic
     from tmp_revision t
     left join person a on a.fullname = t.author_fullname
     left join person c on c.fullname = t.committer_fullname;
     return;
 end
 $$;
 
 
 -- Create entries in person from tmp_release
 create or replace function swh_person_add_from_release()
     returns void
     language plpgsql
 as $$
 begin
     with t as (
         select distinct author_fullname as fullname, author_name as name, author_email as email from tmp_release
         where author_fullname is not null
     ) insert into person (fullname, name, email)
     select distinct on (fullname) fullname, name, email from t
     where not exists (
         select 1
         from person p
         where t.fullname = p.fullname
     );
     return;
 end
 $$;
 
 
 -- Create entries in release from tmp_release
 create or replace function swh_release_add()
     returns void
     language plpgsql
 as $$
 begin
     perform swh_person_add_from_release();
 
     insert into release (id, target, target_type, date, date_offset, date_neg_utc_offset, name, comment, author, synthetic)
-    select t.id, t.target, t.target_type, t.date, t.date_offset, t.date_neg_utc_offset, t.name, t.comment, a.id, t.synthetic
-    from tmp_release t
-    left join person a on a.fullname = t.author_fullname;
+      select distinct t.id, t.target, t.target_type, t.date, t.date_offset, t.date_neg_utc_offset, t.name, t.comment, a.id, t.synthetic
+        from tmp_release t
+        left join person a on a.fullname = t.author_fullname
+        where not exists (select 1 from release where t.id = release.id);
     return;
 end
 $$;
 
 
 -- add a new origin_visit for origin origin_id at date.
 --
 -- Returns the new visit id.
 create or replace function swh_origin_visit_add(origin_url text, date timestamptz, type text)
     returns bigint
     language sql
 as $$
   with origin_id as (
     select id
     from origin
     where url = origin_url
   ), last_known_visit as (
     select coalesce(max(visit), 0) as visit
     from origin_visit
     where origin = (select id from origin_id)
   )
   insert into origin_visit (origin, date, type, visit)
   values ((select id from origin_id), date, type,
           (select visit from last_known_visit) + 1)
   returning visit;
 $$;
 
 create or replace function swh_snapshot_add(snapshot_id sha1_git)
   returns void
   language plpgsql
 as $$
 declare
   snapshot_object_id snapshot.object_id%type;
 begin
   select object_id from snapshot where id = snapshot_id into snapshot_object_id;
   if snapshot_object_id is null then
      insert into snapshot (id) values (snapshot_id) returning object_id into snapshot_object_id;
      insert into snapshot_branch (name, target_type, target)
        select name, target_type, target from tmp_snapshot_branch tmp
        where not exists (
          select 1
          from snapshot_branch sb
          where sb.name = tmp.name
            and sb.target = tmp.target
            and sb.target_type = tmp.target_type
        )
        on conflict do nothing;
      insert into snapshot_branches (snapshot_id, branch_id)
      select snapshot_object_id, sb.object_id as branch_id
        from tmp_snapshot_branch tmp
        join snapshot_branch sb
        using (name, target, target_type)
        where tmp.target is not null and tmp.target_type is not null
      union
      select snapshot_object_id, sb.object_id as branch_id
        from tmp_snapshot_branch tmp
        join snapshot_branch sb
        using (name)
        where tmp.target is null and tmp.target_type is null
          and sb.target is null and sb.target_type is null;
   end if;
   truncate table tmp_snapshot_branch;
 end;
 $$;
 
 create type snapshot_result as (
   snapshot_id  sha1_git,
   name         bytea,
   target       bytea,
   target_type  snapshot_target
 );
 
 create or replace function swh_snapshot_get_by_id(id sha1_git,
     branches_from bytea default '', branches_count bigint default null,
     target_types snapshot_target[] default NULL)
   returns setof snapshot_result
   language sql
   stable
 as $$
   -- with small limits, the "naive" version of this query can degenerate into
   -- using the deduplication index on snapshot_branch (name, target,
   -- target_type); The planner happily scans several hundred million rows.
 
   -- Do the query in two steps: first pull the relevant branches for the given
   -- snapshot (filtering them by type), then do the limiting. This two-step
   -- process guides the planner into using the proper index.
   with filtered_snapshot_branches as (
     select swh_snapshot_get_by_id.id as snapshot_id, name, target, target_type
       from snapshot_branches
       inner join snapshot_branch on snapshot_branches.branch_id = snapshot_branch.object_id
       where snapshot_id = (select object_id from snapshot where snapshot.id = swh_snapshot_get_by_id.id)
         and (target_types is null or target_type = any(target_types))
       order by name
   )
   select snapshot_id, name, target, target_type
     from filtered_snapshot_branches
     where name >= branches_from
     order by name limit branches_count;
 $$;
 
 create type snapshot_size as (
   target_type snapshot_target,
   count bigint
 );
 
 create or replace function swh_snapshot_count_branches(id sha1_git)
   returns setof snapshot_size
   language sql
   stable
 as $$
   SELECT target_type, count(name)
   from swh_snapshot_get_by_id(swh_snapshot_count_branches.id)
   group by target_type;
 $$;
 
 -- Absolute path: directory reference + complete path relative to it
 create type content_dir as (
     directory  sha1_git,
     path       unix_path
 );
 
 
 -- Find the containing directory of a given content, specified by sha1
 -- (note: *not* sha1_git).
 --
 -- Return a pair (dir_it, path) where path is a UNIX path that, from the
 -- directory root, reach down to a file with the desired content. Return NULL
 -- if no match is found.
 --
 -- In case of multiple paths (i.e., pretty much always), an arbitrary one is
 -- chosen.
 create or replace function swh_content_find_directory(content_id sha1)
     returns content_dir
     language sql
     stable
 as $$
     with recursive path as (
 	-- Recursively build a path from the requested content to a root
 	-- directory. Each iteration returns a pair (dir_id, filename) where
 	-- filename is relative to dir_id. Stops when no parent directory can
 	-- be found.
 	(select dir.id as dir_id, dir_entry_f.name as name, 0 as depth
 	 from directory_entry_file as dir_entry_f
 	 join content on content.sha1_git = dir_entry_f.target
 	 join directory as dir on dir.file_entries @> array[dir_entry_f.id]
 	 where content.sha1 = content_id
 	 limit 1)
 	union all
 	(select dir.id as dir_id,
 		(dir_entry_d.name || '/' || path.name)::unix_path as name,
 		path.depth + 1
 	 from path
 	 join directory_entry_dir as dir_entry_d on dir_entry_d.target = path.dir_id
 	 join directory as dir on dir.dir_entries @> array[dir_entry_d.id]
 	 limit 1)
     )
     select dir_id, name from path order by depth desc limit 1;
 $$;
 
 -- Find the visit of origin closest to date visit_date
 -- Breaks ties by selecting the largest visit id
 create or replace function swh_visit_find_by_date(origin_url text, visit_date timestamptz default NOW())
     returns setof origin_visit
     language plpgsql
     stable
 as $$
 declare
   origin_id bigint;
 begin
   select id into origin_id from origin where url=origin_url;
   return query
   with closest_two_visits as ((
     select ov, (date - visit_date), visit as interval
     from origin_visit ov
     where ov.origin = origin_id
           and ov.date >= visit_date
     order by ov.date asc, ov.visit desc
     limit 1
   ) union (
     select ov, (visit_date - date), visit as interval
     from origin_visit ov
     where ov.origin = origin_id
           and ov.date < visit_date
     order by ov.date desc, ov.visit desc
     limit 1
   )) select (ov).* from closest_two_visits order by interval, visit limit 1;
 end
 $$;
 
 -- Object listing by object_id
 
 create or replace function swh_content_list_by_object_id(
     min_excl bigint,
     max_incl bigint
 )
     returns setof content
     language sql
     stable
 as $$
     select * from content
     where object_id > min_excl and object_id <= max_incl
     order by object_id;
 $$;
 
 create or replace function swh_revision_list_by_object_id(
     min_excl bigint,
     max_incl bigint
 )
     returns setof revision_entry
     language sql
     stable
 as $$
     with revs as (
         select * from revision
         where object_id > min_excl and object_id <= max_incl
     )
     select r.id, r.date, r.date_offset, r.date_neg_utc_offset,
            r.committer_date, r.committer_date_offset, r.committer_date_neg_utc_offset,
            r.type, r.directory, r.message,
            a.id, a.fullname, a.name, a.email, c.id, c.fullname, c.name, c.email, r.metadata, r.synthetic,
            array(select rh.parent_id::bytea from revision_history rh where rh.id = r.id order by rh.parent_rank)
                as parents, r.object_id
     from revs r
     left join person a on a.id = r.author
     left join person c on c.id = r.committer
     order by r.object_id;
 $$;
 
 create or replace function swh_release_list_by_object_id(
     min_excl bigint,
     max_incl bigint
 )
     returns setof release_entry
     language sql
     stable
 as $$
     with rels as (
         select * from release
         where object_id > min_excl and object_id <= max_incl
     )
     select r.id, r.target, r.target_type, r.date, r.date_offset, r.date_neg_utc_offset, r.name, r.comment,
            r.synthetic, p.id as author_id, p.fullname as author_fullname, p.name as author_name, p.email as author_email, r.object_id
     from rels r
     left join person p on p.id = r.author
     order by r.object_id;
 $$;
 
 
 -- simple counter mapping a textual label to an integer value
 create type counter as (
     label  text,
     value  bigint
 );
 
 -- return statistics about the number of tuples in various SWH tables
 --
 -- Note: the returned values are based on postgres internal statistics
 -- (pg_class table), which are only updated daily (by autovacuum) or so
 create or replace function swh_stat_counters()
     returns setof counter
     language sql
     stable
 as $$
     select object_type as label, value as value
     from object_counts
     where object_type in (
         'content',
         'directory',
         'directory_entry_dir',
         'directory_entry_file',
         'directory_entry_rev',
         'origin',
         'origin_visit',
         'person',
         'release',
         'revision',
         'revision_history',
         'skipped_content',
         'snapshot'
     );
 $$;
 
 create or replace function swh_update_counter(object_type text)
     returns void
     language plpgsql
 as $$
 begin
     execute format('
 	insert into object_counts
     (value, last_update, object_type)
   values
     ((select count(*) from %1$I), NOW(), %1$L)
   on conflict (object_type) do update set
     value = excluded.value,
     last_update = excluded.last_update',
   object_type);
     return;
 end;
 $$;
 
 create or replace function swh_update_counter_bucketed()
     returns void
     language plpgsql
 as $$
 declare
   query text;
   line_to_update int;
   new_value bigint;
 begin
   select
     object_counts_bucketed.line,
     format(
       'select count(%I) from %I where %s',
       coalesce(identifier, '*'),
       object_type,
       coalesce(
         concat_ws(
           ' and ',
           case when bucket_start is not null then
             format('%I >= %L', identifier, bucket_start) -- lower bound condition, inclusive
           end,
           case when bucket_end is not null then
             format('%I < %L', identifier, bucket_end) -- upper bound condition, exclusive
           end
         ),
         'true'
       )
     )
     from object_counts_bucketed
     order by coalesce(last_update, now() - '1 month'::interval) asc
     limit 1
     into line_to_update, query;
 
   execute query into new_value;
 
   update object_counts_bucketed
     set value = new_value,
         last_update = now()
     where object_counts_bucketed.line = line_to_update;
 
 END
 $$;
 
 create or replace function swh_update_counters_from_buckets()
   returns trigger
   language plpgsql
 as $$
 begin
 with to_update as (
   select object_type, sum(value) as value, max(last_update) as last_update
   from object_counts_bucketed ob1
   where not exists (
     select 1 from object_counts_bucketed ob2
     where ob1.object_type = ob2.object_type
     and value is null
     )
   group by object_type
 ) update object_counts
   set
     value = to_update.value,
     last_update = to_update.last_update
   from to_update
   where
     object_counts.object_type = to_update.object_type
     and object_counts.value != to_update.value;
 return null;
 end
 $$;
 
 create trigger update_counts_from_bucketed
   after insert or update
   on object_counts_bucketed
   for each row
   when (NEW.line % 256 = 0)
   execute procedure swh_update_counters_from_buckets();
diff --git a/swh/storage/tests/test_storage.py b/swh/storage/tests/test_storage.py
index 00187a3c..83a594dc 100644
--- a/swh/storage/tests/test_storage.py
+++ b/swh/storage/tests/test_storage.py
@@ -1,3966 +1,3970 @@
 # Copyright (C) 2015-2020  The Software Heritage developers
 # See the AUTHORS file at the top-level directory of this distribution
 # License: GNU General Public License version 3, or any later version
 # See top-level LICENSE file for more information
 
 import copy
 from contextlib import contextmanager
 import datetime
 import inspect
 import itertools
 import math
 import queue
 import random
 import threading
 
 from collections import defaultdict
 from datetime import timedelta
 from unittest.mock import Mock
 
 import psycopg2
 import pytest
 
 from hypothesis import given, strategies, settings, HealthCheck
 
 from typing import ClassVar, Optional
 
 from swh.model import from_disk, identifiers
 from swh.model.hashutil import hash_to_bytes
 from swh.model.model import (
     Content,
     Directory,
     Origin,
     OriginVisit,
     OriginVisitStatus,
     Release,
     Revision,
     Snapshot,
 )
 from swh.model.hypothesis_strategies import objects
 from swh.model.hashutil import hash_to_hex
 from swh.storage import get_storage
 from swh.storage.converters import origin_url_to_sha1 as sha1
 from swh.storage.exc import HashCollision, StorageArgumentException
 from swh.storage.interface import StorageInterface
 from swh.storage.utils import content_hex_hashes, now
 
 from .storage_data import data
 
 
 @contextmanager
 def db_transaction(storage):
     with storage.db() as db:
         with db.transaction() as cur:
             yield db, cur
 
 
 def normalize_entity(entity):
     entity = copy.deepcopy(entity)
     for key in ("date", "committer_date"):
         if key in entity:
             entity[key] = identifiers.normalize_timestamp(entity[key])
     return entity
 
 
 def transform_entries(dir_, *, prefix=b""):
     for ent in dir_["entries"]:
         yield {
             "dir_id": dir_["id"],
             "type": ent["type"],
             "target": ent["target"],
             "name": prefix + ent["name"],
             "perms": ent["perms"],
             "status": None,
             "sha1": None,
             "sha1_git": None,
             "sha256": None,
             "length": None,
         }
 
 
 def cmpdir(directory):
     return (directory["type"], directory["dir_id"])
 
 
 def short_revision(revision):
     return [revision["id"], revision["parents"]]
 
 
 def assert_contents_ok(
     expected_contents, actual_contents, keys_to_check={"sha1", "data"}
 ):
     """Assert that a given list of contents matches on a given set of keys.
 
     """
     for k in keys_to_check:
         expected_list = set([c.get(k) for c in expected_contents])
         actual_list = set([c.get(k) for c in actual_contents])
         assert actual_list == expected_list, k
 
 
 def round_to_milliseconds(date):
     """Round datetime to milliseconds before insertion, so equality doesn't fail after a
     round-trip through a DB (eg. Cassandra)
 
     """
     return date.replace(microsecond=(date.microsecond // 1000) * 1000)
 
 
 def test_round_to_milliseconds():
     date = now()
 
     for (ms, expected_ms) in [(0, 0), (1000, 1000), (555555, 555000), (999500, 999000)]:
         date = date.replace(microsecond=ms)
         actual_date = round_to_milliseconds(date)
         assert actual_date.microsecond == expected_ms
 
 
 class LazyContent(Content):
     def with_data(self):
         return Content.from_dict({**self.to_dict(), "data": data.cont["data"]})
 
 
 class TestStorage:
     """Main class for Storage testing.
 
     This class is used as-is to test local storage (see TestLocalStorage
     below) and remote storage (see TestRemoteStorage in
     test_remote_storage.py.
 
     We need to have the two classes inherit from this base class
     separately to avoid nosetests running the tests from the base
     class twice.
     """
 
     maxDiff = None  # type: ClassVar[Optional[int]]
 
     def test_types(self, swh_storage_backend_config):
         """Checks all methods of StorageInterface are implemented by this
         backend, and that they have the same signature."""
         # Create an instance of the protocol (which cannot be instantiated
         # directly, so this creates a subclass, then instantiates it)
         interface = type("_", (StorageInterface,), {})()
         storage = get_storage(**swh_storage_backend_config)
 
         assert "content_add" in dir(interface)
 
         missing_methods = []
 
         for meth_name in dir(interface):
             if meth_name.startswith("_"):
                 continue
             interface_meth = getattr(interface, meth_name)
             try:
                 concrete_meth = getattr(storage, meth_name)
             except AttributeError:
                 if not getattr(interface_meth, "deprecated_endpoint", False):
                     # The backend is missing a (non-deprecated) endpoint
                     missing_methods.append(meth_name)
                 continue
 
             expected_signature = inspect.signature(interface_meth)
             actual_signature = inspect.signature(concrete_meth)
 
             assert expected_signature == actual_signature, meth_name
 
         assert missing_methods == []
 
     def test_check_config(self, swh_storage):
         assert swh_storage.check_config(check_write=True)
         assert swh_storage.check_config(check_write=False)
 
     def test_content_add(self, swh_storage):
         cont = data.cont
 
         insertion_start_time = now()
         actual_result = swh_storage.content_add([cont])
         insertion_end_time = now()
 
         assert actual_result == {
             "content:add": 1,
             "content:add:bytes": cont["length"],
         }
 
         assert list(swh_storage.content_get([cont["sha1"]])) == [
             {"sha1": cont["sha1"], "data": cont["data"]}
         ]
 
         expected_cont = data.cont
         del expected_cont["data"]
         contents = [
             obj
             for (obj_type, obj) in swh_storage.journal_writer.journal.objects
             if obj_type == "content"
         ]
         assert len(contents) == 1
         for obj in contents:
             assert insertion_start_time <= obj.ctime
             assert obj.ctime <= insertion_end_time
             obj_d = obj.to_dict()
             del obj_d["ctime"]
             assert obj_d == expected_cont
 
         swh_storage.refresh_stat_counters()
         assert swh_storage.stat_counters()["content"] == 1
 
     def test_content_add_from_generator(self, swh_storage):
         def _cnt_gen():
             yield data.cont
 
         actual_result = swh_storage.content_add(_cnt_gen())
 
         assert actual_result == {
             "content:add": 1,
             "content:add:bytes": data.cont["length"],
         }
 
         swh_storage.refresh_stat_counters()
         assert swh_storage.stat_counters()["content"] == 1
 
     def test_content_add_from_lazy_content(self, swh_storage):
 
         lazy_content = LazyContent.from_dict({**data.cont, "data": b"nope",})
 
         insertion_start_time = now()
 
         # bypass the validation proxy for now, to directly put a dict
         actual_result = swh_storage.storage.content_add([lazy_content])
 
         insertion_end_time = now()
 
         assert actual_result == {
             "content:add": 1,
             "content:add:bytes": data.cont["length"],
         }
 
         # the fact that we retrieve the content object from the storage with
         # the correct 'data' field ensures it has been 'called'
         assert list(swh_storage.content_get([data.cont["sha1"]])) == [
             {"sha1": data.cont["sha1"], "data": data.cont["data"]}
         ]
 
         expected_cont = data.cont
         del expected_cont["data"]
         contents = [
             obj
             for (obj_type, obj) in swh_storage.journal_writer.journal.objects
             if obj_type == "content"
         ]
         assert len(contents) == 1
         for obj in contents:
             assert insertion_start_time <= obj.ctime
             assert obj.ctime <= insertion_end_time
             obj_d = obj.to_dict()
             del obj_d["ctime"]
             assert obj_d == expected_cont
 
         swh_storage.refresh_stat_counters()
         assert swh_storage.stat_counters()["content"] == 1
 
     def test_content_add_validation(self, swh_storage):
         cont = data.cont
 
         with pytest.raises(StorageArgumentException, match="status"):
             swh_storage.content_add([{**cont, "status": "absent"}])
 
         with pytest.raises(StorageArgumentException, match="status"):
             swh_storage.content_add([{**cont, "status": "foobar"}])
 
         with pytest.raises(StorageArgumentException, match="(?i)length"):
             swh_storage.content_add([{**cont, "length": -2}])
 
         with pytest.raises(StorageArgumentException, match="reason"):
             swh_storage.content_add([{**cont, "reason": "foobar"}])
 
     def test_skipped_content_add_validation(self, swh_storage):
         cont = data.cont.copy()
         del cont["data"]
 
         with pytest.raises(StorageArgumentException, match="status"):
             swh_storage.skipped_content_add([{**cont, "status": "visible"}])
 
         with pytest.raises(StorageArgumentException, match="reason") as cm:
             swh_storage.skipped_content_add([{**cont, "status": "absent"}])
 
         if type(cm.value) == psycopg2.IntegrityError:
             assert cm.exception.pgcode == psycopg2.errorcodes.NOT_NULL_VIOLATION
 
     def test_content_get_missing(self, swh_storage):
         cont = data.cont
 
         swh_storage.content_add([cont])
 
         # Query a single missing content
         results = list(swh_storage.content_get([data.cont2["sha1"]]))
         assert results == [None]
 
         # Check content_get does not abort after finding a missing content
         results = list(swh_storage.content_get([data.cont["sha1"], data.cont2["sha1"]]))
         assert results == [{"sha1": cont["sha1"], "data": cont["data"]}, None]
 
         # Check content_get does not discard found countent when it finds
         # a missing content.
         results = list(swh_storage.content_get([data.cont2["sha1"], data.cont["sha1"]]))
         assert results == [None, {"sha1": cont["sha1"], "data": cont["data"]}]
 
     def test_content_add_different_input(self, swh_storage):
         cont = data.cont
         cont2 = data.cont2
 
         actual_result = swh_storage.content_add([cont, cont2])
         assert actual_result == {
             "content:add": 2,
             "content:add:bytes": cont["length"] + cont2["length"],
         }
 
     def test_content_add_twice(self, swh_storage):
         actual_result = swh_storage.content_add([data.cont])
         assert actual_result == {
             "content:add": 1,
             "content:add:bytes": data.cont["length"],
         }
         assert len(swh_storage.journal_writer.journal.objects) == 1
 
         actual_result = swh_storage.content_add([data.cont, data.cont2])
         assert actual_result == {
             "content:add": 1,
             "content:add:bytes": data.cont2["length"],
         }
         assert 2 <= len(swh_storage.journal_writer.journal.objects) <= 3
 
         assert len(swh_storage.content_find(data.cont)) == 1
         assert len(swh_storage.content_find(data.cont2)) == 1
 
     def test_content_add_collision(self, swh_storage):
         cont1 = data.cont
 
         # create (corrupted) content with same sha1{,_git} but != sha256
         cont1b = cont1.copy()
         sha256_array = bytearray(cont1b["sha256"])
         sha256_array[0] += 1
         cont1b["sha256"] = bytes(sha256_array)
 
         with pytest.raises(HashCollision) as cm:
             swh_storage.content_add([cont1, cont1b])
 
         exc = cm.value
         actual_algo = exc.algo
         assert actual_algo in ["sha1", "sha1_git", "blake2s256"]
         actual_id = exc.hash_id
         assert actual_id == hash_to_hex(cont1[actual_algo])
         collisions = exc.args[2]
         assert len(collisions) == 2
         assert collisions == [
             content_hex_hashes(Content.from_dict(cont1).hashes()),
             content_hex_hashes(Content.from_dict(cont1b).hashes()),
         ]
         assert exc.colliding_content_hashes() == [
             Content.from_dict(cont1).hashes(),
             Content.from_dict(cont1b).hashes(),
         ]
 
     def test_content_add_duplicate(self, swh_storage):
         swh_storage.content_add([data.cont, data.cont])
 
         assert list(swh_storage.content_get([data.cont["sha1"]])) == [
             {"sha1": data.cont["sha1"], "data": data.cont["data"]}
         ]
 
     def test_content_update(self, swh_storage):
         if hasattr(swh_storage, "storage"):
             swh_storage.journal_writer.journal = None  # TODO, not supported
 
         cont = copy.deepcopy(data.cont)
 
         swh_storage.content_add([cont])
         # alter the sha1_git for example
         cont["sha1_git"] = hash_to_bytes("3a60a5275d0333bf13468e8b3dcab90f4046e654")
 
         swh_storage.content_update([cont], keys=["sha1_git"])
 
         results = swh_storage.content_get_metadata([cont["sha1"]])
         del cont["data"]
         assert tuple(results[cont["sha1"]]) == (cont,)
 
     def test_content_add_metadata(self, swh_storage):
         cont = data.cont
         del cont["data"]
         cont["ctime"] = now()
 
         actual_result = swh_storage.content_add_metadata([cont])
         assert actual_result == {
             "content:add": 1,
         }
 
         expected_cont = cont.copy()
         del expected_cont["ctime"]
         assert tuple(
             swh_storage.content_get_metadata([cont["sha1"]])[cont["sha1"]]
         ) == (expected_cont,)
         contents = [
             obj
             for (obj_type, obj) in swh_storage.journal_writer.journal.objects
             if obj_type == "content"
         ]
         assert len(contents) == 1
         for obj in contents:
             obj_d = obj.to_dict()
             del obj_d["ctime"]
             assert obj_d == expected_cont
 
     def test_content_add_metadata_different_input(self, swh_storage):
         cont = data.cont
         del cont["data"]
         cont["ctime"] = now()
         cont2 = data.cont2
         del cont2["data"]
         cont2["ctime"] = now()
 
         actual_result = swh_storage.content_add_metadata([cont, cont2])
         assert actual_result == {
             "content:add": 2,
         }
 
     def test_content_add_metadata_collision(self, swh_storage):
         cont1 = data.cont
         del cont1["data"]
         cont1["ctime"] = now()
 
         # create (corrupted) content with same sha1{,_git} but != sha256
         cont1b = cont1.copy()
         sha1_git_array = bytearray(cont1b["sha256"])
         sha1_git_array[0] += 1
         cont1b["sha256"] = bytes(sha1_git_array)
 
         with pytest.raises(HashCollision) as cm:
             swh_storage.content_add_metadata([cont1, cont1b])
 
         exc = cm.value
         actual_algo = exc.algo
         assert actual_algo in ["sha1", "sha1_git", "blake2s256"]
         actual_id = exc.hash_id
         assert actual_id == hash_to_hex(cont1[actual_algo])
         collisions = exc.args[2]
         assert len(collisions) == 2
         assert collisions == [
             content_hex_hashes(Content.from_dict(cont1).hashes()),
             content_hex_hashes(Content.from_dict(cont1b).hashes()),
         ]
         assert exc.colliding_content_hashes() == [
             Content.from_dict(cont1).hashes(),
             Content.from_dict(cont1b).hashes(),
         ]
 
     def test_skipped_content_add(self, swh_storage):
         cont = data.skipped_cont
         cont2 = data.skipped_cont2
         cont2["blake2s256"] = None
 
         missing = list(swh_storage.skipped_content_missing([cont, cont2]))
 
         assert missing == [
             {
                 "sha1": cont["sha1"],
                 "sha1_git": cont["sha1_git"],
                 "blake2s256": cont["blake2s256"],
                 "sha256": cont["sha256"],
             },
             {
                 "sha1": cont2["sha1"],
                 "sha1_git": cont2["sha1_git"],
                 "blake2s256": cont2["blake2s256"],
                 "sha256": cont2["sha256"],
             },
         ]
 
         actual_result = swh_storage.skipped_content_add([cont, cont, cont2])
 
         assert 2 <= actual_result.pop("skipped_content:add") <= 3
         assert actual_result == {}
 
         missing = list(swh_storage.skipped_content_missing([cont, cont2]))
 
         assert missing == []
 
     def test_skipped_content_add_missing_hashes(self, swh_storage):
         cont = data.skipped_cont
         cont2 = data.skipped_cont2
         cont["sha1_git"] = cont2["sha1_git"] = None
 
         missing = list(swh_storage.skipped_content_missing([cont, cont2]))
 
         assert len(missing) == 2
 
         actual_result = swh_storage.skipped_content_add([cont, cont, cont2])
 
         assert 2 <= actual_result.pop("skipped_content:add") <= 3
         assert actual_result == {}
 
         missing = list(swh_storage.skipped_content_missing([cont, cont2]))
 
         assert missing == []
 
     def test_skipped_content_missing_partial_hash(self, swh_storage):
         cont = data.skipped_cont
         cont2 = cont.copy()
         cont2["sha1_git"] = None
 
         missing = list(swh_storage.skipped_content_missing([cont, cont2]))
 
         assert len(missing) == 2
 
         actual_result = swh_storage.skipped_content_add([cont])
 
         assert actual_result.pop("skipped_content:add") == 1
         assert actual_result == {}
 
         missing = list(swh_storage.skipped_content_missing([cont, cont2]))
 
         assert missing == [
             {
                 "sha1": cont2["sha1"],
                 "sha1_git": cont2["sha1_git"],
                 "blake2s256": cont2["blake2s256"],
                 "sha256": cont2["sha256"],
             }
         ]
 
     @pytest.mark.property_based
     @settings(deadline=None)  # this test is very slow
     @given(
         strategies.sets(
             elements=strategies.sampled_from(["sha256", "sha1_git", "blake2s256"]),
             min_size=0,
         )
     )
     def test_content_missing(self, swh_storage, algos):
         algos |= {"sha1"}
         cont2 = data.cont2
         missing_cont = data.missing_cont
         swh_storage.content_add([cont2])
         test_contents = [cont2]
         missing_per_hash = defaultdict(list)
         for i in range(256):
             test_content = missing_cont.copy()
             for hash in algos:
                 test_content[hash] = bytes([i]) + test_content[hash][1:]
                 missing_per_hash[hash].append(test_content[hash])
             test_contents.append(test_content)
 
         assert set(swh_storage.content_missing(test_contents)) == set(
             missing_per_hash["sha1"]
         )
 
         for hash in algos:
             assert set(
                 swh_storage.content_missing(test_contents, key_hash=hash)
             ) == set(missing_per_hash[hash])
 
     @pytest.mark.property_based
     @given(
         strategies.sets(
             elements=strategies.sampled_from(["sha256", "sha1_git", "blake2s256"]),
             min_size=0,
         )
     )
     def test_content_missing_unknown_algo(self, swh_storage, algos):
         algos |= {"sha1"}
         cont2 = data.cont2
         missing_cont = data.missing_cont
         swh_storage.content_add([cont2])
         test_contents = [cont2]
         missing_per_hash = defaultdict(list)
         for i in range(16):
             test_content = missing_cont.copy()
             for hash in algos:
                 test_content[hash] = bytes([i]) + test_content[hash][1:]
                 missing_per_hash[hash].append(test_content[hash])
             test_content["nonexisting_algo"] = b"\x00"
             test_contents.append(test_content)
 
         assert set(swh_storage.content_missing(test_contents)) == set(
             missing_per_hash["sha1"]
         )
 
         for hash in algos:
             assert set(
                 swh_storage.content_missing(test_contents, key_hash=hash)
             ) == set(missing_per_hash[hash])
 
     def test_content_missing_per_sha1(self, swh_storage):
         # given
         cont2 = data.cont2
         missing_cont = data.missing_cont
         swh_storage.content_add([cont2])
         # when
         gen = swh_storage.content_missing_per_sha1(
             [cont2["sha1"], missing_cont["sha1"]]
         )
         # then
         assert list(gen) == [missing_cont["sha1"]]
 
     def test_content_missing_per_sha1_git(self, swh_storage):
         cont = data.cont
         cont2 = data.cont2
         missing_cont = data.missing_cont
 
         swh_storage.content_add([cont, cont2])
 
         contents = [cont["sha1_git"], cont2["sha1_git"], missing_cont["sha1_git"]]
 
         missing_contents = swh_storage.content_missing_per_sha1_git(contents)
         assert list(missing_contents) == [missing_cont["sha1_git"]]
 
     def test_content_get_partition(self, swh_storage, swh_contents):
         """content_get_partition paginates results if limit exceeded"""
         expected_contents = [c for c in swh_contents if c["status"] != "absent"]
 
         actual_contents = []
         for i in range(16):
             actual_result = swh_storage.content_get_partition(i, 16)
             assert actual_result["next_page_token"] is None
             actual_contents.extend(actual_result["contents"])
 
         assert_contents_ok(expected_contents, actual_contents, ["sha1"])
 
     def test_content_get_partition_full(self, swh_storage, swh_contents):
         """content_get_partition for a single partition returns all available
         contents"""
         expected_contents = [c for c in swh_contents if c["status"] != "absent"]
 
         actual_result = swh_storage.content_get_partition(0, 1)
         assert actual_result["next_page_token"] is None
 
         actual_contents = actual_result["contents"]
         assert_contents_ok(expected_contents, actual_contents, ["sha1"])
 
     def test_content_get_partition_empty(self, swh_storage, swh_contents):
         """content_get_partition when at least one of the partitions is
         empty"""
         expected_contents = {
             cont["sha1"] for cont in swh_contents if cont["status"] != "absent"
         }
         # nb_partitions = smallest power of 2 such that at least one of
         # the partitions is empty
         nb_partitions = 1 << math.floor(math.log2(len(swh_contents)) + 1)
 
         seen_sha1s = []
 
         for i in range(nb_partitions):
             actual_result = swh_storage.content_get_partition(
                 i, nb_partitions, limit=len(swh_contents) + 1
             )
 
             for cont in actual_result["contents"]:
                 seen_sha1s.append(cont["sha1"])
 
             # Limit is higher than the max number of results
             assert actual_result["next_page_token"] is None
 
         assert set(seen_sha1s) == expected_contents
 
     def test_content_get_partition_limit_none(self, swh_storage):
         """content_get_partition call with wrong limit input should fail"""
         with pytest.raises(StorageArgumentException) as e:
             swh_storage.content_get_partition(1, 16, limit=None)
 
         assert e.value.args == ("limit should not be None",)
 
     def test_generate_content_get_partition_pagination(self, swh_storage, swh_contents):
         """content_get_partition returns contents within range provided"""
         expected_contents = [c for c in swh_contents if c["status"] != "absent"]
 
         # retrieve contents
         actual_contents = []
         for i in range(4):
             page_token = None
             while True:
                 actual_result = swh_storage.content_get_partition(
                     i, 4, limit=3, page_token=page_token
                 )
                 actual_contents.extend(actual_result["contents"])
                 page_token = actual_result["next_page_token"]
 
                 if page_token is None:
                     break
 
         assert_contents_ok(expected_contents, actual_contents, ["sha1"])
 
     def test_content_get_metadata(self, swh_storage):
         cont1 = data.cont
         cont2 = data.cont2
 
         swh_storage.content_add([cont1, cont2])
 
         actual_md = swh_storage.content_get_metadata([cont1["sha1"], cont2["sha1"]])
 
         # we only retrieve the metadata
         cont1.pop("data")
         cont2.pop("data")
 
         assert tuple(actual_md[cont1["sha1"]]) == (cont1,)
         assert tuple(actual_md[cont2["sha1"]]) == (cont2,)
         assert len(actual_md.keys()) == 2
 
     def test_content_get_metadata_missing_sha1(self, swh_storage):
         cont1 = data.cont
         cont2 = data.cont2
         missing_cont = data.missing_cont
 
         swh_storage.content_add([cont1, cont2])
 
         actual_contents = swh_storage.content_get_metadata([missing_cont["sha1"]])
 
         assert len(actual_contents) == 1
         assert tuple(actual_contents[missing_cont["sha1"]]) == ()
 
     def test_content_get_random(self, swh_storage):
         swh_storage.content_add([data.cont, data.cont2, data.cont3])
 
         assert swh_storage.content_get_random() in {
             data.cont["sha1_git"],
             data.cont2["sha1_git"],
             data.cont3["sha1_git"],
         }
 
     def test_directory_add(self, swh_storage):
         init_missing = list(swh_storage.directory_missing([data.dir["id"]]))
         assert [data.dir["id"]] == init_missing
 
         actual_result = swh_storage.directory_add([data.dir])
         assert actual_result == {"directory:add": 1}
 
         assert list(swh_storage.journal_writer.journal.objects) == [
             ("directory", Directory.from_dict(data.dir))
         ]
 
         actual_data = list(swh_storage.directory_ls(data.dir["id"]))
         expected_data = list(transform_entries(data.dir))
 
         assert sorted(expected_data, key=cmpdir) == sorted(actual_data, key=cmpdir)
 
         after_missing = list(swh_storage.directory_missing([data.dir["id"]]))
         assert after_missing == []
 
         swh_storage.refresh_stat_counters()
         assert swh_storage.stat_counters()["directory"] == 1
 
     def test_directory_add_from_generator(self, swh_storage):
         def _dir_gen():
             yield data.dir
 
         actual_result = swh_storage.directory_add(directories=_dir_gen())
         assert actual_result == {"directory:add": 1}
 
         assert list(swh_storage.journal_writer.journal.objects) == [
             ("directory", Directory.from_dict(data.dir))
         ]
 
         swh_storage.refresh_stat_counters()
         assert swh_storage.stat_counters()["directory"] == 1
 
     def test_directory_add_validation(self, swh_storage):
         dir_ = copy.deepcopy(data.dir)
         dir_["entries"][0]["type"] = "foobar"
 
         with pytest.raises(StorageArgumentException, match="type.*foobar"):
             swh_storage.directory_add([dir_])
 
         dir_ = copy.deepcopy(data.dir)
         del dir_["entries"][0]["target"]
 
         with pytest.raises(StorageArgumentException, match="target") as cm:
             swh_storage.directory_add([dir_])
 
         if type(cm.value) == psycopg2.IntegrityError:
             assert cm.value.pgcode == psycopg2.errorcodes.NOT_NULL_VIOLATION
 
     def test_directory_add_twice(self, swh_storage):
         actual_result = swh_storage.directory_add([data.dir])
         assert actual_result == {"directory:add": 1}
 
         assert list(swh_storage.journal_writer.journal.objects) == [
             ("directory", Directory.from_dict(data.dir))
         ]
 
         actual_result = swh_storage.directory_add([data.dir])
         assert actual_result == {"directory:add": 0}
 
         assert list(swh_storage.journal_writer.journal.objects) == [
             ("directory", Directory.from_dict(data.dir))
         ]
 
     def test_directory_get_recursive(self, swh_storage):
         init_missing = list(swh_storage.directory_missing([data.dir["id"]]))
         assert init_missing == [data.dir["id"]]
 
         actual_result = swh_storage.directory_add([data.dir, data.dir2, data.dir3])
         assert actual_result == {"directory:add": 3}
 
         assert list(swh_storage.journal_writer.journal.objects) == [
             ("directory", Directory.from_dict(data.dir)),
             ("directory", Directory.from_dict(data.dir2)),
             ("directory", Directory.from_dict(data.dir3)),
         ]
 
         # List directory containing a file and an unknown subdirectory
         actual_data = list(swh_storage.directory_ls(data.dir["id"], recursive=True))
         expected_data = list(transform_entries(data.dir))
         assert sorted(expected_data, key=cmpdir) == sorted(actual_data, key=cmpdir)
 
         # List directory containing a file and an unknown subdirectory
         actual_data = list(swh_storage.directory_ls(data.dir2["id"], recursive=True))
         expected_data = list(transform_entries(data.dir2))
         assert sorted(expected_data, key=cmpdir) == sorted(actual_data, key=cmpdir)
 
         # List directory containing a known subdirectory, entries should
         # be both those of the directory and of the subdir
         actual_data = list(swh_storage.directory_ls(data.dir3["id"], recursive=True))
         expected_data = list(
             itertools.chain(
                 transform_entries(data.dir3),
                 transform_entries(data.dir, prefix=b"subdir/"),
             )
         )
         assert sorted(expected_data, key=cmpdir) == sorted(actual_data, key=cmpdir)
 
     def test_directory_get_non_recursive(self, swh_storage):
         init_missing = list(swh_storage.directory_missing([data.dir["id"]]))
         assert init_missing == [data.dir["id"]]
 
         actual_result = swh_storage.directory_add([data.dir, data.dir2, data.dir3])
         assert actual_result == {"directory:add": 3}
 
         assert list(swh_storage.journal_writer.journal.objects) == [
             ("directory", Directory.from_dict(data.dir)),
             ("directory", Directory.from_dict(data.dir2)),
             ("directory", Directory.from_dict(data.dir3)),
         ]
 
         # List directory containing a file and an unknown subdirectory
         actual_data = list(swh_storage.directory_ls(data.dir["id"]))
         expected_data = list(transform_entries(data.dir))
         assert sorted(expected_data, key=cmpdir) == sorted(actual_data, key=cmpdir)
 
         # List directory contaiining a single file
         actual_data = list(swh_storage.directory_ls(data.dir2["id"]))
         expected_data = list(transform_entries(data.dir2))
         assert sorted(expected_data, key=cmpdir) == sorted(actual_data, key=cmpdir)
 
         # List directory containing a known subdirectory, entries should
         # only be those of the parent directory, not of the subdir
         actual_data = list(swh_storage.directory_ls(data.dir3["id"]))
         expected_data = list(transform_entries(data.dir3))
         assert sorted(expected_data, key=cmpdir) == sorted(actual_data, key=cmpdir)
 
     def test_directory_entry_get_by_path(self, swh_storage):
         # given
         init_missing = list(swh_storage.directory_missing([data.dir3["id"]]))
         assert [data.dir3["id"]] == init_missing
 
         actual_result = swh_storage.directory_add([data.dir3, data.dir4])
         assert actual_result == {"directory:add": 2}
 
         expected_entries = [
             {
                 "dir_id": data.dir3["id"],
                 "name": b"foo",
                 "type": "file",
                 "target": data.cont["sha1_git"],
                 "sha1": None,
                 "sha1_git": None,
                 "sha256": None,
                 "status": None,
                 "perms": from_disk.DentryPerms.content,
                 "length": None,
             },
             {
                 "dir_id": data.dir3["id"],
                 "name": b"subdir",
                 "type": "dir",
                 "target": data.dir["id"],
                 "sha1": None,
                 "sha1_git": None,
                 "sha256": None,
                 "status": None,
                 "perms": from_disk.DentryPerms.directory,
                 "length": None,
             },
             {
                 "dir_id": data.dir3["id"],
                 "name": b"hello",
                 "type": "file",
                 "target": b"12345678901234567890",
                 "sha1": None,
                 "sha1_git": None,
                 "sha256": None,
                 "status": None,
                 "perms": from_disk.DentryPerms.content,
                 "length": None,
             },
         ]
 
         # when (all must be found here)
         for entry, expected_entry in zip(data.dir3["entries"], expected_entries):
             actual_entry = swh_storage.directory_entry_get_by_path(
                 data.dir3["id"], [entry["name"]]
             )
             assert actual_entry == expected_entry
 
         # same, but deeper
         for entry, expected_entry in zip(data.dir3["entries"], expected_entries):
             actual_entry = swh_storage.directory_entry_get_by_path(
                 data.dir4["id"], [b"subdir1", entry["name"]]
             )
             expected_entry = expected_entry.copy()
             expected_entry["name"] = b"subdir1/" + expected_entry["name"]
             assert actual_entry == expected_entry
 
         # when (nothing should be found here since data.dir is not persisted.)
         for entry in data.dir["entries"]:
             actual_entry = swh_storage.directory_entry_get_by_path(
                 data.dir["id"], [entry["name"]]
             )
             assert actual_entry is None
 
     def test_directory_get_random(self, swh_storage):
         swh_storage.directory_add([data.dir, data.dir2, data.dir3])
 
         assert swh_storage.directory_get_random() in {
             data.dir["id"],
             data.dir2["id"],
             data.dir3["id"],
         }
 
     def test_revision_add(self, swh_storage):
         init_missing = swh_storage.revision_missing([data.revision["id"]])
         assert list(init_missing) == [data.revision["id"]]
 
         actual_result = swh_storage.revision_add([data.revision])
         assert actual_result == {"revision:add": 1}
 
         end_missing = swh_storage.revision_missing([data.revision["id"]])
         assert list(end_missing) == []
 
         assert list(swh_storage.journal_writer.journal.objects) == [
             ("revision", Revision.from_dict(data.revision))
         ]
 
         # already there so nothing added
         actual_result = swh_storage.revision_add([data.revision])
         assert actual_result == {"revision:add": 0}
 
         swh_storage.refresh_stat_counters()
         assert swh_storage.stat_counters()["revision"] == 1
 
     def test_revision_add_from_generator(self, swh_storage):
         def _rev_gen():
             yield data.revision
 
         actual_result = swh_storage.revision_add(_rev_gen())
         assert actual_result == {"revision:add": 1}
 
         swh_storage.refresh_stat_counters()
         assert swh_storage.stat_counters()["revision"] == 1
 
     def test_revision_add_validation(self, swh_storage):
         rev = copy.deepcopy(data.revision)
         rev["date"]["offset"] = 2 ** 16
 
         with pytest.raises(StorageArgumentException, match="offset") as cm:
             swh_storage.revision_add([rev])
 
         if type(cm.value) == psycopg2.DataError:
             assert cm.value.pgcode == psycopg2.errorcodes.NUMERIC_VALUE_OUT_OF_RANGE
 
         rev = copy.deepcopy(data.revision)
         rev["committer_date"]["offset"] = 2 ** 16
 
         with pytest.raises(StorageArgumentException, match="offset") as cm:
             swh_storage.revision_add([rev])
 
         if type(cm.value) == psycopg2.DataError:
             assert cm.value.pgcode == psycopg2.errorcodes.NUMERIC_VALUE_OUT_OF_RANGE
 
         rev = copy.deepcopy(data.revision)
         rev["type"] = "foobar"
 
         with pytest.raises(StorageArgumentException, match="(?i)type") as cm:
             swh_storage.revision_add([rev])
 
         if type(cm.value) == psycopg2.DataError:
             assert cm.value.pgcode == psycopg2.errorcodes.INVALID_TEXT_REPRESENTATION
 
     def test_revision_add_twice(self, swh_storage):
         actual_result = swh_storage.revision_add([data.revision])
         assert actual_result == {"revision:add": 1}
 
         assert list(swh_storage.journal_writer.journal.objects) == [
             ("revision", Revision.from_dict(data.revision))
         ]
 
         actual_result = swh_storage.revision_add([data.revision, data.revision2])
         assert actual_result == {"revision:add": 1}
 
         assert list(swh_storage.journal_writer.journal.objects) == [
             ("revision", Revision.from_dict(data.revision)),
             ("revision", Revision.from_dict(data.revision2)),
         ]
 
     def test_revision_add_name_clash(self, swh_storage):
         revision1 = data.revision
         revision2 = data.revision2
 
         revision1["author"] = {
             "fullname": b"John Doe <john.doe@example.com>",
             "name": b"John Doe",
             "email": b"john.doe@example.com",
         }
         revision2["author"] = {
             "fullname": b"John Doe <john.doe@example.com>",
             "name": b"John Doe ",
             "email": b"john.doe@example.com ",
         }
         actual_result = swh_storage.revision_add([revision1, revision2])
         assert actual_result == {"revision:add": 2}
 
     def test_revision_get_order(self, swh_storage):
         add_result = swh_storage.revision_add([data.revision, data.revision2])
         assert add_result == {"revision:add": 2}
 
         # order 1
         res1 = swh_storage.revision_get([data.revision["id"], data.revision2["id"]])
         assert list(res1) == [data.revision, data.revision2]
 
         # order 2
         res2 = swh_storage.revision_get([data.revision2["id"], data.revision["id"]])
         assert list(res2) == [data.revision2, data.revision]
 
     def test_revision_log(self, swh_storage):
         # given
         # data.revision4 -is-child-of-> data.revision3
         swh_storage.revision_add([data.revision3, data.revision4])
 
         # when
         actual_results = list(swh_storage.revision_log([data.revision4["id"]]))
 
         # hack: ids generated
         for actual_result in actual_results:
             if "id" in actual_result["author"]:
                 del actual_result["author"]["id"]
             if "id" in actual_result["committer"]:
                 del actual_result["committer"]["id"]
 
         assert len(actual_results) == 2  # rev4 -child-> rev3
         assert actual_results[0] == normalize_entity(data.revision4)
         assert actual_results[1] == normalize_entity(data.revision3)
 
         assert list(swh_storage.journal_writer.journal.objects) == [
             ("revision", Revision.from_dict(data.revision3)),
             ("revision", Revision.from_dict(data.revision4)),
         ]
 
     def test_revision_log_with_limit(self, swh_storage):
         # given
         # data.revision4 -is-child-of-> data.revision3
         swh_storage.revision_add([data.revision3, data.revision4])
         actual_results = list(swh_storage.revision_log([data.revision4["id"]], 1))
 
         # hack: ids generated
         for actual_result in actual_results:
             if "id" in actual_result["author"]:
                 del actual_result["author"]["id"]
             if "id" in actual_result["committer"]:
                 del actual_result["committer"]["id"]
 
         assert len(actual_results) == 1
         assert actual_results[0] == data.revision4
 
     def test_revision_log_unknown_revision(self, swh_storage):
         rev_log = list(swh_storage.revision_log([data.revision["id"]]))
         assert rev_log == []
 
     def test_revision_shortlog(self, swh_storage):
         # given
         # data.revision4 -is-child-of-> data.revision3
         swh_storage.revision_add([data.revision3, data.revision4])
 
         # when
         actual_results = list(swh_storage.revision_shortlog([data.revision4["id"]]))
 
         assert len(actual_results) == 2  # rev4 -child-> rev3
         assert list(actual_results[0]) == short_revision(data.revision4)
         assert list(actual_results[1]) == short_revision(data.revision3)
 
     def test_revision_shortlog_with_limit(self, swh_storage):
         # given
         # data.revision4 -is-child-of-> data.revision3
         swh_storage.revision_add([data.revision3, data.revision4])
         actual_results = list(swh_storage.revision_shortlog([data.revision4["id"]], 1))
 
         assert len(actual_results) == 1
         assert list(actual_results[0]) == short_revision(data.revision4)
 
     def test_revision_get(self, swh_storage):
         swh_storage.revision_add([data.revision])
 
         actual_revisions = list(
             swh_storage.revision_get([data.revision["id"], data.revision2["id"]])
         )
 
         # when
         if "id" in actual_revisions[0]["author"]:
             del actual_revisions[0]["author"]["id"]  # hack: ids are generated
         if "id" in actual_revisions[0]["committer"]:
             del actual_revisions[0]["committer"]["id"]
 
         assert len(actual_revisions) == 2
         assert actual_revisions[0] == normalize_entity(data.revision)
         assert actual_revisions[1] is None
 
     def test_revision_get_no_parents(self, swh_storage):
         swh_storage.revision_add([data.revision3])
 
         get = list(swh_storage.revision_get([data.revision3["id"]]))
 
         assert len(get) == 1
         assert get[0]["parents"] == ()  # no parents on this one
 
     def test_revision_get_random(self, swh_storage):
         swh_storage.revision_add([data.revision, data.revision2, data.revision3])
 
         assert swh_storage.revision_get_random() in {
             data.revision["id"],
             data.revision2["id"],
             data.revision3["id"],
         }
 
     def test_release_add(self, swh_storage):
         init_missing = swh_storage.release_missing(
             [data.release["id"], data.release2["id"]]
         )
         assert [data.release["id"], data.release2["id"]] == list(init_missing)
 
         actual_result = swh_storage.release_add([data.release, data.release2])
         assert actual_result == {"release:add": 2}
 
         end_missing = swh_storage.release_missing(
             [data.release["id"], data.release2["id"]]
         )
         assert list(end_missing) == []
 
         assert list(swh_storage.journal_writer.journal.objects) == [
             ("release", Release.from_dict(data.release)),
             ("release", Release.from_dict(data.release2)),
         ]
 
         # already present so nothing added
         actual_result = swh_storage.release_add([data.release, data.release2])
         assert actual_result == {"release:add": 0}
 
         swh_storage.refresh_stat_counters()
         assert swh_storage.stat_counters()["release"] == 2
 
     def test_release_add_from_generator(self, swh_storage):
         def _rel_gen():
             yield data.release
             yield data.release2
 
         actual_result = swh_storage.release_add(_rel_gen())
         assert actual_result == {"release:add": 2}
 
         assert list(swh_storage.journal_writer.journal.objects) == [
             ("release", Release.from_dict(data.release)),
             ("release", Release.from_dict(data.release2)),
         ]
 
         swh_storage.refresh_stat_counters()
         assert swh_storage.stat_counters()["release"] == 2
 
     def test_release_add_no_author_date(self, swh_storage):
         release = data.release
 
         release["author"] = None
         release["date"] = None
 
         actual_result = swh_storage.release_add([release])
         assert actual_result == {"release:add": 1}
 
         end_missing = swh_storage.release_missing([data.release["id"]])
         assert list(end_missing) == []
 
         assert list(swh_storage.journal_writer.journal.objects) == [
             ("release", Release.from_dict(release))
         ]
 
     def test_release_add_validation(self, swh_storage):
         rel = copy.deepcopy(data.release)
         rel["date"]["offset"] = 2 ** 16
 
         with pytest.raises(StorageArgumentException, match="offset") as cm:
             swh_storage.release_add([rel])
 
         if type(cm.value) == psycopg2.DataError:
             assert cm.value.pgcode == psycopg2.errorcodes.NUMERIC_VALUE_OUT_OF_RANGE
 
         rel = copy.deepcopy(data.release)
         rel["author"] = None
 
         with pytest.raises(StorageArgumentException, match="date") as cm:
             swh_storage.release_add([rel])
 
         if type(cm.value) == psycopg2.IntegrityError:
             assert cm.value.pgcode == psycopg2.errorcodes.CHECK_VIOLATION
 
     def test_release_add_validation_type(self, swh_storage):
         rel = copy.deepcopy(data.release)
 
         rel["date"]["offset"] = "toto"
         with pytest.raises(StorageArgumentException):
             swh_storage.release_add([rel])
 
     def test_release_add_twice(self, swh_storage):
         actual_result = swh_storage.release_add([data.release])
         assert actual_result == {"release:add": 1}
 
         assert list(swh_storage.journal_writer.journal.objects) == [
             ("release", Release.from_dict(data.release))
         ]
 
-        actual_result = swh_storage.release_add([data.release, data.release2])
+        actual_result = swh_storage.release_add(
+            [data.release, data.release2, data.release, data.release2]
+        )
         assert actual_result == {"release:add": 1}
 
-        assert list(swh_storage.journal_writer.journal.objects) == [
-            ("release", Release.from_dict(data.release)),
-            ("release", Release.from_dict(data.release2)),
-        ]
+        assert set(swh_storage.journal_writer.journal.objects) == set(
+            [
+                ("release", Release.from_dict(data.release)),
+                ("release", Release.from_dict(data.release2)),
+            ]
+        )
 
     def test_release_add_name_clash(self, swh_storage):
         release1 = data.release.copy()
         release2 = data.release2.copy()
 
         release1["author"] = {
             "fullname": b"John Doe <john.doe@example.com>",
             "name": b"John Doe",
             "email": b"john.doe@example.com",
         }
         release2["author"] = {
             "fullname": b"John Doe <john.doe@example.com>",
             "name": b"John Doe ",
             "email": b"john.doe@example.com ",
         }
         actual_result = swh_storage.release_add([release1, release2])
         assert actual_result == {"release:add": 2}
 
     def test_release_get(self, swh_storage):
         # given
         swh_storage.release_add([data.release, data.release2])
 
         # when
         actual_releases = list(
             swh_storage.release_get([data.release["id"], data.release2["id"]])
         )
 
         # then
         for actual_release in actual_releases:
             if "id" in actual_release["author"]:
                 del actual_release["author"]["id"]  # hack: ids are generated
 
         assert [normalize_entity(data.release), normalize_entity(data.release2)] == [
             actual_releases[0],
             actual_releases[1],
         ]
 
         unknown_releases = list(swh_storage.release_get([data.release3["id"]]))
 
         assert unknown_releases[0] is None
 
     def test_release_get_order(self, swh_storage):
         add_result = swh_storage.release_add([data.release, data.release2])
         assert add_result == {"release:add": 2}
 
         # order 1
         res1 = swh_storage.release_get([data.release["id"], data.release2["id"]])
         assert list(res1) == [data.release, data.release2]
 
         # order 2
         res2 = swh_storage.release_get([data.release2["id"], data.release["id"]])
         assert list(res2) == [data.release2, data.release]
 
     def test_release_get_random(self, swh_storage):
         swh_storage.release_add([data.release, data.release2, data.release3])
 
         assert swh_storage.release_get_random() in {
             data.release["id"],
             data.release2["id"],
             data.release3["id"],
         }
 
     def test_origin_add_one(self, swh_storage):
         origin0 = swh_storage.origin_get(data.origin)
         assert origin0 is None
 
         id = swh_storage.origin_add_one(data.origin)
 
         actual_origin = swh_storage.origin_get({"url": data.origin["url"]})
         assert actual_origin["url"] == data.origin["url"]
 
         id2 = swh_storage.origin_add_one(data.origin)
 
         assert id == id2
 
     def test_origin_add(self, swh_storage):
         origin0 = swh_storage.origin_get([data.origin])[0]
         assert origin0 is None
 
         stats = swh_storage.origin_add([data.origin, data.origin2])
         assert stats == {"origin:add": 2}
 
         actual_origin = swh_storage.origin_get([{"url": data.origin["url"],}])[0]
         assert actual_origin["url"] == data.origin["url"]
 
         actual_origin2 = swh_storage.origin_get([{"url": data.origin2["url"],}])[0]
         assert actual_origin2["url"] == data.origin2["url"]
 
         assert set(swh_storage.journal_writer.journal.objects) == set(
             [
                 ("origin", Origin.from_dict(actual_origin)),
                 ("origin", Origin.from_dict(actual_origin2)),
             ]
         )
 
         swh_storage.refresh_stat_counters()
         assert swh_storage.stat_counters()["origin"] == 2
 
     def test_origin_add_from_generator(self, swh_storage):
         def _ori_gen():
             yield data.origin
             yield data.origin2
 
         stats = swh_storage.origin_add(_ori_gen())
         assert stats == {"origin:add": 2}
 
         actual_origin = swh_storage.origin_get([{"url": data.origin["url"],}])[0]
         assert actual_origin["url"] == data.origin["url"]
 
         actual_origin2 = swh_storage.origin_get([{"url": data.origin2["url"],}])[0]
         assert actual_origin2["url"] == data.origin2["url"]
 
         if "id" in actual_origin:
             del actual_origin["id"]
             del actual_origin2["id"]
 
         assert set(swh_storage.journal_writer.journal.objects) == set(
             [
                 ("origin", Origin.from_dict(actual_origin)),
                 ("origin", Origin.from_dict(actual_origin2)),
             ]
         )
 
         swh_storage.refresh_stat_counters()
         assert swh_storage.stat_counters()["origin"] == 2
 
     def test_origin_add_twice(self, swh_storage):
         add1 = swh_storage.origin_add([data.origin, data.origin2])
         assert set(swh_storage.journal_writer.journal.objects) == set(
             [
                 ("origin", Origin.from_dict(data.origin)),
                 ("origin", Origin.from_dict(data.origin2)),
             ]
         )
         assert add1 == {"origin:add": 2}
 
         add2 = swh_storage.origin_add([data.origin, data.origin2])
         assert set(swh_storage.journal_writer.journal.objects) == set(
             [
                 ("origin", Origin.from_dict(data.origin)),
                 ("origin", Origin.from_dict(data.origin2)),
             ]
         )
         assert add2 == {"origin:add": 0}
 
     def test_origin_add_validation(self, swh_storage):
         """Incorrect formatted origin should fail the validation
 
         """
         with pytest.raises(StorageArgumentException, match="url"):
             swh_storage.origin_add([{}])
         with pytest.raises(
             StorageArgumentException, match="unexpected keyword argument"
         ):
             swh_storage.origin_add([{"ul": "mistyped url key"}])
 
     def test_origin_get_legacy(self, swh_storage):
         assert swh_storage.origin_get(data.origin) is None
         swh_storage.origin_add_one(data.origin)
 
         actual_origin0 = swh_storage.origin_get({"url": data.origin["url"]})
         assert actual_origin0["url"] == data.origin["url"]
 
     def test_origin_get(self, swh_storage):
         assert swh_storage.origin_get(data.origin) is None
         assert swh_storage.origin_get([data.origin]) == [None]
         swh_storage.origin_add_one(data.origin)
 
         actual_origin0 = swh_storage.origin_get([{"url": data.origin["url"]}])
         assert len(actual_origin0) == 1
         assert actual_origin0[0]["url"] == data.origin["url"]
 
         actual_origins = swh_storage.origin_get(
             [{"url": data.origin["url"]}, {"url": "not://exists"}]
         )
         assert actual_origins == [{"url": data.origin["url"]}, None]
 
     def _generate_random_visits(self, nb_visits=100, start=0, end=7):
         """Generate random visits within the last 2 months (to avoid
         computations)
 
         """
         visits = []
         today = now()
         for weeks in range(nb_visits, 0, -1):
             hours = random.randint(0, 24)
             minutes = random.randint(0, 60)
             seconds = random.randint(0, 60)
             days = random.randint(0, 28)
             weeks = random.randint(start, end)
             date_visit = today - timedelta(
                 weeks=weeks, hours=hours, minutes=minutes, seconds=seconds, days=days
             )
             visits.append(date_visit)
         return visits
 
     def test_origin_visit_get_all(self, swh_storage):
         origin = Origin.from_dict(data.origin)
         swh_storage.origin_add_one(origin)
         visits = swh_storage.origin_visit_add(
             [
                 OriginVisit(
                     origin=origin.url, date=data.date_visit1, type=data.type_visit1,
                 ),
                 OriginVisit(
                     origin=origin.url, date=data.date_visit2, type=data.type_visit2,
                 ),
                 OriginVisit(
                     origin=origin.url, date=data.date_visit2, type=data.type_visit2,
                 ),
             ]
         )
         ov1, ov2, ov3 = [
             {**v.to_dict(), "status": "created", "snapshot": None, "metadata": None,}
             for v in visits
         ]
 
         # order asc, no pagination, no limit
         all_visits = list(swh_storage.origin_visit_get(origin.url))
         assert all_visits == [ov1, ov2, ov3]
 
         # order asc, no pagination, limit
         all_visits2 = list(swh_storage.origin_visit_get(origin.url, limit=2))
         assert all_visits2 == [ov1, ov2]
 
         # order asc, pagination, no limit
         all_visits3 = list(
             swh_storage.origin_visit_get(origin.url, last_visit=ov1["visit"])
         )
         assert all_visits3 == [ov2, ov3]
 
         # order asc, pagination, limit
         all_visits4 = list(
             swh_storage.origin_visit_get(origin.url, last_visit=ov2["visit"], limit=1)
         )
         assert all_visits4 == [ov3]
 
         # order desc, no pagination, no limit
         all_visits5 = list(swh_storage.origin_visit_get(origin.url, order="desc"))
         assert all_visits5 == [ov3, ov2, ov1]
 
         # order desc, no pagination, limit
         all_visits6 = list(
             swh_storage.origin_visit_get(origin.url, limit=2, order="desc")
         )
         assert all_visits6 == [ov3, ov2]
 
         # order desc, pagination, no limit
         all_visits7 = list(
             swh_storage.origin_visit_get(
                 origin.url, last_visit=ov3["visit"], order="desc"
             )
         )
         assert all_visits7 == [ov2, ov1]
 
         # order desc, pagination, limit
         all_visits8 = list(
             swh_storage.origin_visit_get(
                 origin.url, last_visit=ov3["visit"], order="desc", limit=1
             )
         )
         assert all_visits8 == [ov2]
 
     def test_origin_visit_get__unknown_origin(self, swh_storage):
         assert [] == list(swh_storage.origin_visit_get("foo"))
 
     def test_origin_visit_get_random(self, swh_storage):
         swh_storage.origin_add(data.origins)
         # Add some random visits within the selection range
         visits = self._generate_random_visits()
         visit_type = "git"
 
         # Add visits to those origins
         for origin in data.origins:
             origin_url = origin["url"]
             for date_visit in visits:
                 visit = swh_storage.origin_visit_add(
                     [OriginVisit(origin=origin_url, date=date_visit, type=visit_type,)]
                 )[0]
                 swh_storage.origin_visit_status_add(
                     [
                         OriginVisitStatus(
                             origin=origin_url,
                             visit=visit.visit,
                             date=now(),
                             status="full",
                             snapshot=None,
                         )
                     ]
                 )
 
         swh_storage.refresh_stat_counters()
 
         stats = swh_storage.stat_counters()
         assert stats["origin"] == len(data.origins)
         assert stats["origin_visit"] == len(data.origins) * len(visits)
 
         random_origin_visit = swh_storage.origin_visit_get_random(visit_type)
         assert random_origin_visit
         assert random_origin_visit["origin"] is not None
         original_urls = [o["url"] for o in data.origins]
         assert random_origin_visit["origin"] in original_urls
 
     def test_origin_visit_get_random_nothing_found(self, swh_storage):
         swh_storage.origin_add(data.origins)
         visit_type = "hg"
         # Add some visits outside of the random generation selection so nothing
         # will be found by the random selection
         visits = self._generate_random_visits(nb_visits=3, start=13, end=24)
         for origin in data.origins:
             origin_url = origin["url"]
             for date_visit in visits:
                 visit = swh_storage.origin_visit_add(
                     [OriginVisit(origin=origin_url, date=date_visit, type=visit_type,)]
                 )[0]
                 swh_storage.origin_visit_status_add(
                     [
                         OriginVisitStatus(
                             origin=origin_url,
                             visit=visit.visit,
                             date=now(),
                             status="full",
                             snapshot=None,
                         )
                     ]
                 )
 
         random_origin_visit = swh_storage.origin_visit_get_random(visit_type)
         assert random_origin_visit is None
 
     def test_origin_get_by_sha1(self, swh_storage):
         assert swh_storage.origin_get(data.origin) is None
         swh_storage.origin_add_one(data.origin)
 
         origins = list(swh_storage.origin_get_by_sha1([sha1(data.origin["url"])]))
         assert len(origins) == 1
         assert origins[0]["url"] == data.origin["url"]
 
     def test_origin_get_by_sha1_not_found(self, swh_storage):
         assert swh_storage.origin_get(data.origin) is None
         origins = list(swh_storage.origin_get_by_sha1([sha1(data.origin["url"])]))
         assert len(origins) == 1
         assert origins[0] is None
 
     def test_origin_search_single_result(self, swh_storage):
         found_origins = list(swh_storage.origin_search(data.origin["url"]))
         assert len(found_origins) == 0
 
         found_origins = list(swh_storage.origin_search(data.origin["url"], regexp=True))
         assert len(found_origins) == 0
 
         swh_storage.origin_add_one(data.origin)
         origin_data = {"url": data.origin["url"]}
         found_origins = list(swh_storage.origin_search(data.origin["url"]))
         assert len(found_origins) == 1
         if "id" in found_origins[0]:
             del found_origins[0]["id"]
         assert found_origins[0] == origin_data
 
         found_origins = list(
             swh_storage.origin_search("." + data.origin["url"][1:-1] + ".", regexp=True)
         )
         assert len(found_origins) == 1
         if "id" in found_origins[0]:
             del found_origins[0]["id"]
         assert found_origins[0] == origin_data
 
         swh_storage.origin_add_one(data.origin2)
         origin2_data = {"url": data.origin2["url"]}
         found_origins = list(swh_storage.origin_search(data.origin2["url"]))
         assert len(found_origins) == 1
         if "id" in found_origins[0]:
             del found_origins[0]["id"]
         assert found_origins[0] == origin2_data
 
         found_origins = list(
             swh_storage.origin_search(
                 "." + data.origin2["url"][1:-1] + ".", regexp=True
             )
         )
         assert len(found_origins) == 1
         if "id" in found_origins[0]:
             del found_origins[0]["id"]
         assert found_origins[0] == origin2_data
 
     def test_origin_search_no_regexp(self, swh_storage):
         swh_storage.origin_add_one(data.origin)
         swh_storage.origin_add_one(data.origin2)
 
         origin = swh_storage.origin_get({"url": data.origin["url"]})
         origin2 = swh_storage.origin_get({"url": data.origin2["url"]})
 
         # no pagination
         found_origins = list(swh_storage.origin_search("/"))
         assert len(found_origins) == 2
 
         # offset=0
         found_origins0 = list(swh_storage.origin_search("/", offset=0, limit=1))  # noqa
         assert len(found_origins0) == 1
         assert found_origins0[0] in [origin, origin2]
 
         # offset=1
         found_origins1 = list(swh_storage.origin_search("/", offset=1, limit=1))  # noqa
         assert len(found_origins1) == 1
         assert found_origins1[0] in [origin, origin2]
 
         # check both origins were returned
         assert found_origins0 != found_origins1
 
     def test_origin_search_regexp_substring(self, swh_storage):
         swh_storage.origin_add_one(data.origin)
         swh_storage.origin_add_one(data.origin2)
 
         origin = swh_storage.origin_get({"url": data.origin["url"]})
         origin2 = swh_storage.origin_get({"url": data.origin2["url"]})
 
         # no pagination
         found_origins = list(swh_storage.origin_search("/", regexp=True))
         assert len(found_origins) == 2
 
         # offset=0
         found_origins0 = list(
             swh_storage.origin_search("/", offset=0, limit=1, regexp=True)
         )  # noqa
         assert len(found_origins0) == 1
         assert found_origins0[0] in [origin, origin2]
 
         # offset=1
         found_origins1 = list(
             swh_storage.origin_search("/", offset=1, limit=1, regexp=True)
         )  # noqa
         assert len(found_origins1) == 1
         assert found_origins1[0] in [origin, origin2]
 
         # check both origins were returned
         assert found_origins0 != found_origins1
 
     def test_origin_search_regexp_fullstring(self, swh_storage):
         swh_storage.origin_add_one(data.origin)
         swh_storage.origin_add_one(data.origin2)
 
         origin = swh_storage.origin_get({"url": data.origin["url"]})
         origin2 = swh_storage.origin_get({"url": data.origin2["url"]})
 
         # no pagination
         found_origins = list(swh_storage.origin_search(".*/.*", regexp=True))
         assert len(found_origins) == 2
 
         # offset=0
         found_origins0 = list(
             swh_storage.origin_search(".*/.*", offset=0, limit=1, regexp=True)
         )  # noqa
         assert len(found_origins0) == 1
         assert found_origins0[0] in [origin, origin2]
 
         # offset=1
         found_origins1 = list(
             swh_storage.origin_search(".*/.*", offset=1, limit=1, regexp=True)
         )  # noqa
         assert len(found_origins1) == 1
         assert found_origins1[0] in [origin, origin2]
 
         # check both origins were returned
         assert found_origins0 != found_origins1
 
     def test_origin_visit_add(self, swh_storage):
         origin1 = Origin.from_dict(data.origin2)
         swh_storage.origin_add_one(origin1)
 
         date_visit = now()
         date_visit2 = date_visit + datetime.timedelta(minutes=1)
 
         date_visit = round_to_milliseconds(date_visit)
         date_visit2 = round_to_milliseconds(date_visit2)
 
         visit1 = OriginVisit(
             origin=origin1.url, date=date_visit, type=data.type_visit1,
         )
         visit2 = OriginVisit(
             origin=origin1.url, date=date_visit2, type=data.type_visit2,
         )
 
         # add once
         ov1, ov2 = swh_storage.origin_visit_add([visit1, visit2])
         # then again (will be ignored as they already exist)
         origin_visit1, origin_visit2 = swh_storage.origin_visit_add([ov1, ov2])
         assert ov1 == origin_visit1
         assert ov2 == origin_visit2
 
         ovs1 = OriginVisitStatus(
             origin=origin1.url,
             visit=ov1.visit,
             date=date_visit,
             status="created",
             snapshot=None,
         )
         ovs2 = OriginVisitStatus(
             origin=origin1.url,
             visit=ov2.visit,
             date=date_visit2,
             status="created",
             snapshot=None,
         )
 
         actual_origin_visits = list(swh_storage.origin_visit_get(origin1.url))
         expected_visits = [
             {**ovs1.to_dict(), "type": ov1.type},
             {**ovs2.to_dict(), "type": ov2.type},
         ]
 
         assert len(expected_visits) == len(actual_origin_visits)
 
         for visit in expected_visits:
             assert visit in actual_origin_visits
 
         actual_objects = list(swh_storage.journal_writer.journal.objects)
         expected_objects = list(
             [("origin", origin1)]
             + [("origin_visit", visit) for visit in [ov1, ov2]] * 2
             + [("origin_visit_status", ovs) for ovs in [ovs1, ovs2]]
         )
 
         for obj in expected_objects:
             assert obj in actual_objects
 
     def test_origin_visit_add_validation(self, swh_storage):
         """Unknown origin when adding visits should raise"""
         visit = OriginVisit(
             origin="something-unknown", date=now(), type=data.type_visit1,
         )
         with pytest.raises(StorageArgumentException, match="Unknown origin"):
             swh_storage.origin_visit_add([visit])
 
         objects = list(swh_storage.journal_writer.journal.objects)
         assert not objects
 
     def test_origin_visit_status_add_validation(self, swh_storage):
         """Wrong origin_visit_status input should raise storage argument error"""
         date_visit = now()
         visit_status1 = OriginVisitStatus(
             origin="unknown-origin-url",
             visit=10,
             date=date_visit,
             status="full",
             snapshot=None,
         )
         with pytest.raises(StorageArgumentException, match="Unknown origin"):
             swh_storage.origin_visit_status_add([visit_status1])
 
         objects = list(swh_storage.journal_writer.journal.objects)
         assert not objects
 
     def test_origin_visit_status_add(self, swh_storage):
         """Correct origin visit statuses should add a new visit status
 
         """
         origin1 = Origin.from_dict(data.origin2)
         origin2 = Origin(url="new-origin")
         swh_storage.origin_add([origin1, origin2])
 
         ov1, ov2 = swh_storage.origin_visit_add(
             [
                 OriginVisit(
                     origin=origin1.url, date=data.date_visit1, type=data.type_visit1,
                 ),
                 OriginVisit(
                     origin=origin2.url, date=data.date_visit2, type=data.type_visit2,
                 ),
             ]
         )
 
         ovs1 = OriginVisitStatus(
             origin=origin1.url,
             visit=ov1.visit,
             date=data.date_visit1,
             status="created",
             snapshot=None,
         )
         ovs2 = OriginVisitStatus(
             origin=origin2.url,
             visit=ov2.visit,
             date=data.date_visit2,
             status="created",
             snapshot=None,
         )
 
         snapshot_id = data.snapshot["id"]
         date_visit_now = now()
         visit_status1 = OriginVisitStatus(
             origin=ov1.origin,
             visit=ov1.visit,
             date=date_visit_now,
             status="full",
             snapshot=snapshot_id,
         )
 
         date_visit_now = now()
         visit_status2 = OriginVisitStatus(
             origin=ov2.origin,
             visit=ov2.visit,
             date=date_visit_now,
             status="ongoing",
             snapshot=None,
             metadata={"intrinsic": "something"},
         )
         swh_storage.origin_visit_status_add([visit_status1, visit_status2])
 
         origin_visit1 = swh_storage.origin_visit_get_latest(
             origin1.url, require_snapshot=True
         )
         assert origin_visit1
         assert origin_visit1["status"] == "full"
         assert origin_visit1["snapshot"] == snapshot_id
 
         origin_visit2 = swh_storage.origin_visit_get_latest(
             origin2.url, require_snapshot=False
         )
         assert origin2.url != origin1.url
         assert origin_visit2
         assert origin_visit2["status"] == "ongoing"
         assert origin_visit2["snapshot"] is None
         assert origin_visit2["metadata"] == {"intrinsic": "something"}
 
         actual_objects = list(swh_storage.journal_writer.journal.objects)
 
         expected_origins = [origin1, origin2]
         expected_visits = [ov1, ov2]
         expected_visit_statuses = [ovs1, ovs2, visit_status1, visit_status2]
 
         expected_objects = (
             [("origin", o) for o in expected_origins]
             + [("origin_visit", v) for v in expected_visits]
             + [("origin_visit_status", ovs) for ovs in expected_visit_statuses]
         )
 
         for obj in expected_objects:
             assert obj in actual_objects
 
     def test_origin_visit_status_add_twice(self, swh_storage):
         """Correct origin visit statuses should add a new visit status
 
         """
         origin1 = Origin.from_dict(data.origin2)
         swh_storage.origin_add([origin1])
         ov1 = swh_storage.origin_visit_add(
             [
                 OriginVisit(
                     origin=origin1.url, date=data.date_visit1, type=data.type_visit1,
                 ),
             ]
         )[0]
 
         ovs1 = OriginVisitStatus(
             origin=origin1.url,
             visit=ov1.visit,
             date=data.date_visit1,
             status="created",
             snapshot=None,
         )
         snapshot_id = data.snapshot["id"]
         date_visit_now = now()
         visit_status1 = OriginVisitStatus(
             origin=ov1.origin,
             visit=ov1.visit,
             date=date_visit_now,
             status="full",
             snapshot=snapshot_id,
         )
 
         swh_storage.origin_visit_status_add([visit_status1])
         # second call will ignore existing entries (will send to storage though)
         swh_storage.origin_visit_status_add([visit_status1])
 
         origin_visits = list(swh_storage.origin_visit_get(ov1.origin))
 
         assert len(origin_visits) == 1
         origin_visit1 = origin_visits[0]
         assert origin_visit1
         assert origin_visit1["status"] == "full"
         assert origin_visit1["snapshot"] == snapshot_id
 
         actual_objects = list(swh_storage.journal_writer.journal.objects)
 
         expected_origins = [origin1]
         expected_visits = [ov1]
         expected_visit_statuses = [ovs1, visit_status1, visit_status1]
 
         # write twice in the journal
         expected_objects = (
             [("origin", o) for o in expected_origins]
             + [("origin_visit", v) for v in expected_visits]
             + [("origin_visit_status", ovs) for ovs in expected_visit_statuses]
         )
 
         for obj in expected_objects:
             assert obj in actual_objects
 
     def test_origin_visit_find_by_date(self, swh_storage):
         # given
         origin = Origin.from_dict(data.origin)
         swh_storage.origin_add_one(data.origin)
         visit1 = OriginVisit(
             origin=origin.url, date=data.date_visit2, type=data.type_visit1,
         )
         visit2 = OriginVisit(
             origin=origin.url, date=data.date_visit3, type=data.type_visit2,
         )
         visit3 = OriginVisit(
             origin=origin.url, date=data.date_visit2, type=data.type_visit3,
         )
         ov1, ov2, ov3 = swh_storage.origin_visit_add([visit1, visit2, visit3])
 
         ovs1 = OriginVisitStatus(
             origin=origin.url,
             visit=ov1.visit,
             date=data.date_visit2,
             status="ongoing",
             snapshot=None,
         )
         ovs2 = OriginVisitStatus(
             origin=origin.url,
             visit=ov2.visit,
             date=data.date_visit3,
             status="ongoing",
             snapshot=None,
         )
         ovs3 = OriginVisitStatus(
             origin=origin.url,
             visit=ov3.visit,
             date=data.date_visit2,
             status="ongoing",
             snapshot=None,
         )
         swh_storage.origin_visit_status_add([ovs1, ovs2, ovs3])
 
         # Simple case
         visit = swh_storage.origin_visit_find_by_date(origin.url, data.date_visit3)
         assert visit["visit"] == ov2.visit
 
         # There are two visits at the same date, the latest must be returned
         visit = swh_storage.origin_visit_find_by_date(origin.url, data.date_visit2)
         assert visit["visit"] == ov3.visit
 
     def test_origin_visit_find_by_date__unknown_origin(self, swh_storage):
         swh_storage.origin_visit_find_by_date("foo", data.date_visit2)
 
     def test_origin_visit_get_by(self, swh_storage):
         origin_url = swh_storage.origin_add_one(data.origin)
         origin_url2 = swh_storage.origin_add_one(data.origin2)
         visit = OriginVisit(
             origin=origin_url, date=data.date_visit2, type=data.type_visit2,
         )
         origin_visit1 = swh_storage.origin_visit_add([visit])[0]
 
         swh_storage.snapshot_add([data.snapshot])
         swh_storage.origin_visit_status_add(
             [
                 OriginVisitStatus(
                     origin=origin_url,
                     visit=origin_visit1.visit,
                     date=now(),
                     status="ongoing",
                     snapshot=data.snapshot["id"],
                 )
             ]
         )
 
         # Add some other {origin, visit} entries
         visit2 = OriginVisit(
             origin=origin_url, date=data.date_visit3, type=data.type_visit3,
         )
         visit3 = OriginVisit(
             origin=origin_url2, date=data.date_visit3, type=data.type_visit3,
         )
         swh_storage.origin_visit_add([visit2, visit3])
 
         # when
         visit1_metadata = {
             "contents": 42,
             "directories": 22,
         }
 
         swh_storage.origin_visit_status_add(
             [
                 OriginVisitStatus(
                     origin=origin_url,
                     visit=origin_visit1.visit,
                     date=now(),
                     status="full",
                     snapshot=data.snapshot["id"],
                     metadata=visit1_metadata,
                 )
             ]
         )
 
         expected_origin_visit = origin_visit1.to_dict()
         expected_origin_visit.update(
             {
                 "origin": origin_url,
                 "visit": origin_visit1.visit,
                 "date": data.date_visit2,
                 "type": data.type_visit2,
                 "metadata": visit1_metadata,
                 "status": "full",
                 "snapshot": data.snapshot["id"],
             }
         )
 
         # when
         actual_origin_visit1 = swh_storage.origin_visit_get_by(
             origin_url, origin_visit1.visit
         )
 
         # then
         assert actual_origin_visit1 == expected_origin_visit
 
     def test_origin_visit_get_by__unknown_origin(self, swh_storage):
         assert swh_storage.origin_visit_get_by("foo", 10) is None
 
     def test_origin_visit_get_by_no_result(self, swh_storage):
         swh_storage.origin_add([data.origin])
         actual_origin_visit = swh_storage.origin_visit_get_by(data.origin["url"], 999)
         assert actual_origin_visit is None
 
     def test_origin_visit_get_latest_none(self, swh_storage):
         """Origin visit get latest on unknown objects should return nothing
 
         """
         # unknown origin so no result
         assert swh_storage.origin_visit_get_latest("unknown-origin") is None
 
         # unknown type
         origin = Origin.from_dict(data.origin)
         swh_storage.origin_add_one(origin)
         assert swh_storage.origin_visit_get_latest(origin.url, type="unknown") is None
 
     def test_origin_visit_get_latest_filter_type(self, swh_storage):
         """Filtering origin visit get latest with filter type should be ok
 
         """
         origin = Origin.from_dict(data.origin)
         swh_storage.origin_add_one(origin)
         visit1 = OriginVisit(
             origin=origin.url, date=data.date_visit1, type=data.type_visit1,
         )
         visit2 = OriginVisit(
             origin=origin.url, date=data.date_visit2, type=data.type_visit2,
         )
         # Add a visit with the same date as the previous one
         visit3 = OriginVisit(
             origin=origin.url, date=data.date_visit2, type=data.type_visit2,
         )
         assert data.type_visit1 != data.type_visit2
         assert data.date_visit1 < data.date_visit2
 
         ov1, ov2, ov3 = swh_storage.origin_visit_add([visit1, visit2, visit3])
 
         origin_visit1 = swh_storage.origin_visit_get_by(origin.url, ov1.visit)
         origin_visit3 = swh_storage.origin_visit_get_by(origin.url, ov3.visit)
 
         assert data.type_visit1 != data.type_visit2
 
         # Check type filter is ok
         actual_ov1 = swh_storage.origin_visit_get_latest(
             origin.url, type=data.type_visit1,
         )
         assert actual_ov1 == origin_visit1
 
         actual_ov3 = swh_storage.origin_visit_get_latest(
             origin.url, type=data.type_visit2,
         )
         assert actual_ov3 == origin_visit3
 
         new_type = "npm"
         assert new_type not in [data.type_visit1, data.type_visit2]
 
         assert (
             swh_storage.origin_visit_get_latest(
                 origin.url, type=new_type,  # no visit matching that type
             )
             is None
         )
 
     def test_origin_visit_get_latest(self, swh_storage):
         origin = Origin.from_dict(data.origin)
         swh_storage.origin_add_one(origin)
         visit1 = OriginVisit(
             origin=origin.url, date=data.date_visit1, type=data.type_visit1,
         )
         visit2 = OriginVisit(
             origin=origin.url, date=data.date_visit2, type=data.type_visit2,
         )
         # Add a visit with the same date as the previous one
         visit3 = OriginVisit(
             origin=origin.url, date=data.date_visit2, type=data.type_visit2,
         )
 
         ov1, ov2, ov3 = swh_storage.origin_visit_add([visit1, visit2, visit3])
 
         origin_visit1 = swh_storage.origin_visit_get_by(origin.url, ov1.visit)
         origin_visit2 = swh_storage.origin_visit_get_by(origin.url, ov2.visit)
         origin_visit3 = swh_storage.origin_visit_get_by(origin.url, ov3.visit)
         # Two visits, both with no snapshot
         assert origin_visit3 == swh_storage.origin_visit_get_latest(origin.url)
         assert (
             swh_storage.origin_visit_get_latest(origin.url, require_snapshot=True)
             is None
         )
 
         # Add snapshot to visit1; require_snapshot=True makes it return
         # visit1 and require_snapshot=False still returns visit2
         complete_snapshot = Snapshot.from_dict(data.complete_snapshot)
         swh_storage.snapshot_add([complete_snapshot])
         swh_storage.origin_visit_status_add(
             [
                 OriginVisitStatus(
                     origin=origin.url,
                     visit=ov1.visit,
                     date=now(),
                     status="ongoing",
                     snapshot=complete_snapshot.id,
                 )
             ]
         )
         actual_visit = swh_storage.origin_visit_get_latest(
             origin.url, require_snapshot=True
         )
         assert actual_visit == {
             **origin_visit1,
             "snapshot": complete_snapshot.id,
             "status": "ongoing",  # visit1 has status created now
         }
 
         assert origin_visit3 == swh_storage.origin_visit_get_latest(origin.url)
 
         # Status filter: all three visits are status=ongoing, so no visit
         # returned
         assert (
             swh_storage.origin_visit_get_latest(origin.url, allowed_statuses=["full"])
             is None
         )
 
         # Mark the first visit as completed and check status filter again
         swh_storage.origin_visit_status_add(
             [
                 OriginVisitStatus(
                     origin=origin.url,
                     visit=ov1.visit,
                     date=now(),
                     status="full",
                     snapshot=complete_snapshot.id,
                 )
             ]
         )
 
         assert {
             **origin_visit1,
             "snapshot": complete_snapshot.id,
             "status": "full",
         } == swh_storage.origin_visit_get_latest(origin.url, allowed_statuses=["full"])
 
         assert origin_visit3 == swh_storage.origin_visit_get_latest(origin.url)
 
         # Add snapshot to visit2 and check that the new snapshot is returned
         empty_snapshot = Snapshot.from_dict(data.empty_snapshot)
         swh_storage.snapshot_add([empty_snapshot])
 
         swh_storage.origin_visit_status_add(
             [
                 OriginVisitStatus(
                     origin=origin.url,
                     visit=ov2.visit,
                     date=now(),
                     status="ongoing",
                     snapshot=empty_snapshot.id,
                 )
             ]
         )
         assert {
             **origin_visit2,
             "snapshot": empty_snapshot.id,
             "status": "ongoing",
         } == swh_storage.origin_visit_get_latest(origin.url, require_snapshot=True)
 
         assert origin_visit3 == swh_storage.origin_visit_get_latest(origin.url)
 
         # Check that the status filter is still working
         assert {
             **origin_visit1,
             "snapshot": complete_snapshot.id,
             "status": "full",
         } == swh_storage.origin_visit_get_latest(origin.url, allowed_statuses=["full"])
 
         # Add snapshot to visit3 (same date as visit2)
         swh_storage.origin_visit_status_add(
             [
                 OriginVisitStatus(
                     origin=origin.url,
                     visit=ov3.visit,
                     date=now(),
                     status="ongoing",
                     snapshot=complete_snapshot.id,
                 )
             ]
         )
         assert {
             **origin_visit1,
             "snapshot": complete_snapshot.id,
             "status": "full",
         } == swh_storage.origin_visit_get_latest(origin.url, allowed_statuses=["full"])
         assert {
             **origin_visit1,
             "snapshot": complete_snapshot.id,
             "status": "full",
         } == swh_storage.origin_visit_get_latest(
             origin.url, allowed_statuses=["full"], require_snapshot=True
         )
         assert {
             **origin_visit3,
             "snapshot": complete_snapshot.id,
             "status": "ongoing",
         } == swh_storage.origin_visit_get_latest(origin.url)
 
         assert {
             **origin_visit3,
             "snapshot": complete_snapshot.id,
             "status": "ongoing",
         } == swh_storage.origin_visit_get_latest(origin.url, require_snapshot=True)
 
     def test_origin_visit_status_get_latest(self, swh_storage):
         origin1 = Origin.from_dict(data.origin)
         swh_storage.origin_add_one(data.origin)
 
         # to have some reference visits
 
         ov1, ov2 = swh_storage.origin_visit_add(
             [
                 OriginVisit(
                     origin=origin1.url, date=data.date_visit1, type=data.type_visit1,
                 ),
                 OriginVisit(
                     origin=origin1.url, date=data.date_visit2, type=data.type_visit2,
                 ),
             ]
         )
 
         snapshot = Snapshot.from_dict(data.complete_snapshot)
         swh_storage.snapshot_add([snapshot])
 
         date_now = now()
         date_now = round_to_milliseconds(date_now)
         assert data.date_visit1 < data.date_visit2
         assert data.date_visit2 < date_now
 
         ovs1 = OriginVisitStatus(
             origin=origin1.url,
             visit=ov1.visit,
             date=data.date_visit1,
             status="partial",
             snapshot=None,
         )
         ovs2 = OriginVisitStatus(
             origin=origin1.url,
             visit=ov1.visit,
             date=data.date_visit2,
             status="ongoing",
             snapshot=None,
         )
         ovs3 = OriginVisitStatus(
             origin=origin1.url,
             visit=ov2.visit,
             date=data.date_visit2 + datetime.timedelta(minutes=1),  # to not be ignored
             status="ongoing",
             snapshot=None,
         )
         ovs4 = OriginVisitStatus(
             origin=origin1.url,
             visit=ov2.visit,
             date=date_now,
             status="full",
             snapshot=snapshot.id,
             metadata={"something": "wicked"},
         )
 
         swh_storage.origin_visit_status_add([ovs1, ovs2, ovs3, ovs4])
 
         # unknown origin so no result
         actual_origin_visit = swh_storage.origin_visit_status_get_latest(
             "unknown-origin", ov1.visit
         )
         assert actual_origin_visit is None
 
         # unknown visit so no result
         actual_origin_visit = swh_storage.origin_visit_status_get_latest(
             ov1.origin, ov1.visit + 10
         )
         assert actual_origin_visit is None
 
         # Two visits, both with no snapshot, take the most recent
         actual_origin_visit2 = swh_storage.origin_visit_status_get_latest(
             origin1.url, ov1.visit
         )
         assert isinstance(actual_origin_visit2, OriginVisitStatus)
         assert actual_origin_visit2 == ovs2
         assert ovs2.origin == origin1.url
         assert ovs2.visit == ov1.visit
 
         actual_origin_visit = swh_storage.origin_visit_status_get_latest(
             origin1.url, ov1.visit, require_snapshot=True
         )
         # there is no visit with snapshot yet for that visit
         assert actual_origin_visit is None
 
         actual_origin_visit2 = swh_storage.origin_visit_status_get_latest(
             origin1.url, ov1.visit, allowed_statuses=["partial", "ongoing"]
         )
         # visit status with partial status visit elected
         assert actual_origin_visit2 == ovs2
         assert actual_origin_visit2.status == "ongoing"
 
         actual_origin_visit4 = swh_storage.origin_visit_status_get_latest(
             origin1.url, ov2.visit, require_snapshot=True
         )
         assert actual_origin_visit4 == ovs4
         assert actual_origin_visit4.snapshot == snapshot.id
 
         actual_origin_visit = swh_storage.origin_visit_status_get_latest(
             origin1.url, ov2.visit, require_snapshot=True, allowed_statuses=["ongoing"]
         )
         # nothing matches so nothing
         assert actual_origin_visit is None  # there is no visit with status full
 
         actual_origin_visit3 = swh_storage.origin_visit_status_get_latest(
             origin1.url, ov2.visit, allowed_statuses=["ongoing"]
         )
         assert actual_origin_visit3 == ovs3
 
     def test_person_fullname_unicity(self, swh_storage):
         # given (person injection through revisions for example)
         revision = data.revision
 
         # create a revision with same committer fullname but wo name and email
         revision2 = copy.deepcopy(data.revision2)
         revision2["committer"] = dict(revision["committer"])
         revision2["committer"]["email"] = None
         revision2["committer"]["name"] = None
 
         swh_storage.revision_add([revision])
         swh_storage.revision_add([revision2])
 
         # when getting added revisions
         revisions = list(swh_storage.revision_get([revision["id"], revision2["id"]]))
 
         # then
         # check committers are the same
         assert revisions[0]["committer"] == revisions[1]["committer"]
 
     def test_snapshot_add_get_empty(self, swh_storage):
         origin_url = swh_storage.origin_add_one(data.origin)
         ov1 = swh_storage.origin_visit_add(
             [
                 OriginVisit(
                     origin=origin_url, date=data.date_visit1, type=data.type_visit1,
                 )
             ]
         )[0]
 
         actual_result = swh_storage.snapshot_add([data.empty_snapshot])
         assert actual_result == {"snapshot:add": 1}
 
         date_now = now()
 
         swh_storage.origin_visit_status_add(
             [
                 OriginVisitStatus(
                     origin=origin_url,
                     visit=ov1.visit,
                     date=date_now,
                     status="full",
                     snapshot=data.empty_snapshot["id"],
                 )
             ]
         )
 
         by_id = swh_storage.snapshot_get(data.empty_snapshot["id"])
         assert by_id == {**data.empty_snapshot, "next_branch": None}
 
         by_ov = swh_storage.snapshot_get_by_origin_visit(origin_url, ov1.visit)
         assert by_ov == {**data.empty_snapshot, "next_branch": None}
 
         ovs1 = OriginVisitStatus.from_dict(
             {
                 "origin": origin_url,
                 "date": data.date_visit1,
                 "visit": ov1.visit,
                 "status": "created",
                 "snapshot": None,
                 "metadata": None,
             }
         )
         ovs2 = OriginVisitStatus.from_dict(
             {
                 "origin": origin_url,
                 "date": date_now,
                 "visit": ov1.visit,
                 "status": "full",
                 "metadata": None,
                 "snapshot": data.empty_snapshot["id"],
             }
         )
         actual_objects = list(swh_storage.journal_writer.journal.objects)
 
         expected_objects = [
             ("origin", Origin.from_dict(data.origin)),
             ("origin_visit", ov1),
             ("origin_visit_status", ovs1,),
             ("snapshot", Snapshot.from_dict(data.empty_snapshot)),
             ("origin_visit_status", ovs2,),
         ]
         for obj in expected_objects:
             assert obj in actual_objects
 
     def test_snapshot_add_get_complete(self, swh_storage):
         origin_url = data.origin["url"]
         origin_url = swh_storage.origin_add_one(data.origin)
         visit = OriginVisit(
             origin=origin_url, date=data.date_visit1, type=data.type_visit1,
         )
         origin_visit1 = swh_storage.origin_visit_add([visit])[0]
         visit_id = origin_visit1.visit
 
         actual_result = swh_storage.snapshot_add([data.complete_snapshot])
         swh_storage.origin_visit_status_add(
             [
                 OriginVisitStatus(
                     origin=origin_url,
                     visit=origin_visit1.visit,
                     date=now(),
                     status="ongoing",
                     snapshot=data.complete_snapshot["id"],
                 )
             ]
         )
         assert actual_result == {"snapshot:add": 1}
 
         by_id = swh_storage.snapshot_get(data.complete_snapshot["id"])
         assert by_id == {**data.complete_snapshot, "next_branch": None}
 
         by_ov = swh_storage.snapshot_get_by_origin_visit(origin_url, visit_id)
         assert by_ov == {**data.complete_snapshot, "next_branch": None}
 
     def test_snapshot_add_many(self, swh_storage):
         actual_result = swh_storage.snapshot_add(
             [data.snapshot, data.complete_snapshot]
         )
         assert actual_result == {"snapshot:add": 2}
 
         assert {
             **data.complete_snapshot,
             "next_branch": None,
         } == swh_storage.snapshot_get(data.complete_snapshot["id"])
 
         assert {**data.snapshot, "next_branch": None} == swh_storage.snapshot_get(
             data.snapshot["id"]
         )
 
         swh_storage.refresh_stat_counters()
         assert swh_storage.stat_counters()["snapshot"] == 2
 
     def test_snapshot_add_many_from_generator(self, swh_storage):
         def _snp_gen():
             yield data.snapshot
             yield data.complete_snapshot
 
         actual_result = swh_storage.snapshot_add(_snp_gen())
         assert actual_result == {"snapshot:add": 2}
 
         swh_storage.refresh_stat_counters()
         assert swh_storage.stat_counters()["snapshot"] == 2
 
     def test_snapshot_add_many_incremental(self, swh_storage):
         actual_result = swh_storage.snapshot_add([data.complete_snapshot])
         assert actual_result == {"snapshot:add": 1}
 
         actual_result2 = swh_storage.snapshot_add(
             [data.snapshot, data.complete_snapshot]
         )
         assert actual_result2 == {"snapshot:add": 1}
 
         assert {
             **data.complete_snapshot,
             "next_branch": None,
         } == swh_storage.snapshot_get(data.complete_snapshot["id"])
 
         assert {**data.snapshot, "next_branch": None} == swh_storage.snapshot_get(
             data.snapshot["id"]
         )
 
     def test_snapshot_add_twice(self, swh_storage):
         actual_result = swh_storage.snapshot_add([data.empty_snapshot])
         assert actual_result == {"snapshot:add": 1}
 
         assert list(swh_storage.journal_writer.journal.objects) == [
             ("snapshot", Snapshot.from_dict(data.empty_snapshot))
         ]
 
         actual_result = swh_storage.snapshot_add([data.snapshot])
         assert actual_result == {"snapshot:add": 1}
 
         assert list(swh_storage.journal_writer.journal.objects) == [
             ("snapshot", Snapshot.from_dict(data.empty_snapshot)),
             ("snapshot", Snapshot.from_dict(data.snapshot)),
         ]
 
     def test_snapshot_add_validation(self, swh_storage):
         snap = copy.deepcopy(data.snapshot)
         snap["branches"][b"foo"] = {"target_type": "revision"}
 
         with pytest.raises(StorageArgumentException, match="target"):
             swh_storage.snapshot_add([snap])
 
         snap = copy.deepcopy(data.snapshot)
         snap["branches"][b"foo"] = {"target": b"\x42" * 20}
 
         with pytest.raises(StorageArgumentException, match="target_type"):
             swh_storage.snapshot_add([snap])
 
     def test_snapshot_add_count_branches(self, swh_storage):
         actual_result = swh_storage.snapshot_add([data.complete_snapshot])
         assert actual_result == {"snapshot:add": 1}
 
         snp_id = data.complete_snapshot["id"]
         snp_size = swh_storage.snapshot_count_branches(snp_id)
 
         expected_snp_size = {
             "alias": 1,
             "content": 1,
             "directory": 2,
             "release": 1,
             "revision": 1,
             "snapshot": 1,
             None: 1,
         }
         assert snp_size == expected_snp_size
 
     def test_snapshot_add_get_paginated(self, swh_storage):
         swh_storage.snapshot_add([data.complete_snapshot])
 
         snp_id = data.complete_snapshot["id"]
         branches = data.complete_snapshot["branches"]
         branch_names = list(sorted(branches))
 
         # Test branch_from
         snapshot = swh_storage.snapshot_get_branches(snp_id, branches_from=b"release")
 
         rel_idx = branch_names.index(b"release")
         expected_snapshot = {
             "id": snp_id,
             "branches": {name: branches[name] for name in branch_names[rel_idx:]},
             "next_branch": None,
         }
 
         assert snapshot == expected_snapshot
 
         # Test branches_count
         snapshot = swh_storage.snapshot_get_branches(snp_id, branches_count=1)
 
         expected_snapshot = {
             "id": snp_id,
             "branches": {branch_names[0]: branches[branch_names[0]],},
             "next_branch": b"content",
         }
         assert snapshot == expected_snapshot
 
         # test branch_from + branches_count
 
         snapshot = swh_storage.snapshot_get_branches(
             snp_id, branches_from=b"directory", branches_count=3
         )
 
         dir_idx = branch_names.index(b"directory")
         expected_snapshot = {
             "id": snp_id,
             "branches": {
                 name: branches[name] for name in branch_names[dir_idx : dir_idx + 3]
             },
             "next_branch": branch_names[dir_idx + 3],
         }
 
         assert snapshot == expected_snapshot
 
     def test_snapshot_add_get_filtered(self, swh_storage):
         origin_url = swh_storage.origin_add_one(data.origin)
         visit = OriginVisit(
             origin=origin_url, date=data.date_visit1, type=data.type_visit1,
         )
         origin_visit1 = swh_storage.origin_visit_add([visit])[0]
 
         swh_storage.snapshot_add([data.complete_snapshot])
         swh_storage.origin_visit_status_add(
             [
                 OriginVisitStatus(
                     origin=origin_url,
                     visit=origin_visit1.visit,
                     date=now(),
                     status="ongoing",
                     snapshot=data.complete_snapshot["id"],
                 )
             ]
         )
 
         snp_id = data.complete_snapshot["id"]
         branches = data.complete_snapshot["branches"]
 
         snapshot = swh_storage.snapshot_get_branches(
             snp_id, target_types=["release", "revision"]
         )
 
         expected_snapshot = {
             "id": snp_id,
             "branches": {
                 name: tgt
                 for name, tgt in branches.items()
                 if tgt and tgt["target_type"] in ["release", "revision"]
             },
             "next_branch": None,
         }
 
         assert snapshot == expected_snapshot
 
         snapshot = swh_storage.snapshot_get_branches(snp_id, target_types=["alias"])
 
         expected_snapshot = {
             "id": snp_id,
             "branches": {
                 name: tgt
                 for name, tgt in branches.items()
                 if tgt and tgt["target_type"] == "alias"
             },
             "next_branch": None,
         }
 
         assert snapshot == expected_snapshot
 
     def test_snapshot_add_get_filtered_and_paginated(self, swh_storage):
         swh_storage.snapshot_add([data.complete_snapshot])
 
         snp_id = data.complete_snapshot["id"]
         branches = data.complete_snapshot["branches"]
         branch_names = list(sorted(branches))
 
         # Test branch_from
 
         snapshot = swh_storage.snapshot_get_branches(
             snp_id, target_types=["directory", "release"], branches_from=b"directory2"
         )
 
         expected_snapshot = {
             "id": snp_id,
             "branches": {name: branches[name] for name in (b"directory2", b"release")},
             "next_branch": None,
         }
 
         assert snapshot == expected_snapshot
 
         # Test branches_count
 
         snapshot = swh_storage.snapshot_get_branches(
             snp_id, target_types=["directory", "release"], branches_count=1
         )
 
         expected_snapshot = {
             "id": snp_id,
             "branches": {b"directory": branches[b"directory"]},
             "next_branch": b"directory2",
         }
         assert snapshot == expected_snapshot
 
         # Test branches_count
 
         snapshot = swh_storage.snapshot_get_branches(
             snp_id, target_types=["directory", "release"], branches_count=2
         )
 
         expected_snapshot = {
             "id": snp_id,
             "branches": {
                 name: branches[name] for name in (b"directory", b"directory2")
             },
             "next_branch": b"release",
         }
         assert snapshot == expected_snapshot
 
         # test branch_from + branches_count
 
         snapshot = swh_storage.snapshot_get_branches(
             snp_id,
             target_types=["directory", "release"],
             branches_from=b"directory2",
             branches_count=1,
         )
 
         dir_idx = branch_names.index(b"directory2")
         expected_snapshot = {
             "id": snp_id,
             "branches": {branch_names[dir_idx]: branches[branch_names[dir_idx]],},
             "next_branch": b"release",
         }
 
         assert snapshot == expected_snapshot
 
     def test_snapshot_add_get(self, swh_storage):
         origin_url = swh_storage.origin_add_one(data.origin)
         visit = OriginVisit(
             origin=origin_url, date=data.date_visit1, type=data.type_visit1,
         )
         origin_visit1 = swh_storage.origin_visit_add([visit])[0]
         visit_id = origin_visit1.visit
 
         swh_storage.snapshot_add([data.snapshot])
         swh_storage.origin_visit_status_add(
             [
                 OriginVisitStatus(
                     origin=origin_url,
                     visit=origin_visit1.visit,
                     date=now(),
                     status="ongoing",
                     snapshot=data.snapshot["id"],
                 )
             ]
         )
 
         by_id = swh_storage.snapshot_get(data.snapshot["id"])
         assert by_id == {**data.snapshot, "next_branch": None}
 
         by_ov = swh_storage.snapshot_get_by_origin_visit(origin_url, visit_id)
         assert by_ov == {**data.snapshot, "next_branch": None}
 
         origin_visit_info = swh_storage.origin_visit_get_by(origin_url, visit_id)
         assert origin_visit_info["snapshot"] == data.snapshot["id"]
 
     def test_snapshot_add_twice__by_origin_visit(self, swh_storage):
         origin_url = swh_storage.origin_add_one(data.origin)
         ov1 = swh_storage.origin_visit_add(
             [
                 OriginVisit(
                     origin=origin_url, date=data.date_visit1, type=data.type_visit1,
                 )
             ]
         )[0]
         swh_storage.snapshot_add([data.snapshot])
         date_now2 = now()
 
         swh_storage.origin_visit_status_add(
             [
                 OriginVisitStatus(
                     origin=origin_url,
                     visit=ov1.visit,
                     date=date_now2,
                     status="ongoing",
                     snapshot=data.snapshot["id"],
                 )
             ]
         )
 
         by_ov1 = swh_storage.snapshot_get_by_origin_visit(origin_url, ov1.visit)
         assert by_ov1 == {**data.snapshot, "next_branch": None}
 
         ov2 = swh_storage.origin_visit_add(
             [
                 OriginVisit(
                     origin=origin_url, date=data.date_visit2, type=data.type_visit2,
                 )
             ]
         )[0]
 
         swh_storage.snapshot_add([data.snapshot])
         date_now4 = now()
         swh_storage.origin_visit_status_add(
             [
                 OriginVisitStatus(
                     origin=origin_url,
                     visit=ov2.visit,
                     date=date_now4,
                     status="ongoing",
                     snapshot=data.snapshot["id"],
                 )
             ]
         )
 
         by_ov2 = swh_storage.snapshot_get_by_origin_visit(origin_url, ov2.visit)
         assert by_ov2 == {**data.snapshot, "next_branch": None}
 
         ovs1 = OriginVisitStatus.from_dict(
             {
                 "origin": origin_url,
                 "date": data.date_visit1,
                 "visit": ov1.visit,
                 "status": "created",
                 "metadata": None,
                 "snapshot": None,
             }
         )
         ovs2 = OriginVisitStatus.from_dict(
             {
                 "origin": origin_url,
                 "date": date_now2,
                 "visit": ov1.visit,
                 "status": "ongoing",
                 "metadata": None,
                 "snapshot": data.snapshot["id"],
             }
         )
         ovs3 = OriginVisitStatus.from_dict(
             {
                 "origin": origin_url,
                 "date": data.date_visit2,
                 "visit": ov2.visit,
                 "status": "created",
                 "metadata": None,
                 "snapshot": None,
             }
         )
         ovs4 = OriginVisitStatus.from_dict(
             {
                 "origin": origin_url,
                 "date": date_now4,
                 "visit": ov2.visit,
                 "status": "ongoing",
                 "metadata": None,
                 "snapshot": data.snapshot["id"],
             }
         )
         actual_objects = list(swh_storage.journal_writer.journal.objects)
         expected_objects = [
             ("origin", Origin.from_dict(data.origin)),
             ("origin_visit", ov1),
             ("origin_visit_status", ovs1),
             ("snapshot", Snapshot.from_dict(data.snapshot)),
             ("origin_visit_status", ovs2),
             ("origin_visit", ov2),
             ("origin_visit_status", ovs3),
             ("origin_visit_status", ovs4),
         ]
 
         for obj in expected_objects:
             assert obj in actual_objects
 
     def test_snapshot_get_random(self, swh_storage):
         swh_storage.snapshot_add(
             [data.snapshot, data.empty_snapshot, data.complete_snapshot]
         )
 
         assert swh_storage.snapshot_get_random() in {
             data.snapshot["id"],
             data.empty_snapshot["id"],
             data.complete_snapshot["id"],
         }
 
     def test_snapshot_missing(self, swh_storage):
         snap = data.snapshot
         missing_snap = data.empty_snapshot
         snapshots = [snap["id"], missing_snap["id"]]
         swh_storage.snapshot_add([snap])
 
         missing_snapshots = swh_storage.snapshot_missing(snapshots)
 
         assert list(missing_snapshots) == [missing_snap["id"]]
 
     def test_stat_counters(self, swh_storage):
         expected_keys = ["content", "directory", "origin", "revision"]
 
         # Initially, all counters are 0
 
         swh_storage.refresh_stat_counters()
         counters = swh_storage.stat_counters()
         assert set(expected_keys) <= set(counters)
         for key in expected_keys:
             assert counters[key] == 0
 
         # Add a content. Only the content counter should increase.
 
         swh_storage.content_add([data.cont])
 
         swh_storage.refresh_stat_counters()
         counters = swh_storage.stat_counters()
 
         assert set(expected_keys) <= set(counters)
         for key in expected_keys:
             if key != "content":
                 assert counters[key] == 0
         assert counters["content"] == 1
 
         # Add other objects. Check their counter increased as well.
 
         origin_url = swh_storage.origin_add_one(data.origin2)
         visit = OriginVisit(
             origin=origin_url, date=data.date_visit2, type=data.type_visit2,
         )
         origin_visit1 = swh_storage.origin_visit_add([visit])[0]
 
         swh_storage.snapshot_add([data.snapshot])
         swh_storage.origin_visit_status_add(
             [
                 OriginVisitStatus(
                     origin=origin_url,
                     visit=origin_visit1.visit,
                     date=now(),
                     status="ongoing",
                     snapshot=data.snapshot["id"],
                 )
             ]
         )
         swh_storage.directory_add([data.dir])
         swh_storage.revision_add([data.revision])
         swh_storage.release_add([data.release])
 
         swh_storage.refresh_stat_counters()
         counters = swh_storage.stat_counters()
         assert counters["content"] == 1
         assert counters["directory"] == 1
         assert counters["snapshot"] == 1
         assert counters["origin"] == 1
         assert counters["origin_visit"] == 1
         assert counters["revision"] == 1
         assert counters["release"] == 1
         assert counters["snapshot"] == 1
         if "person" in counters:
             assert counters["person"] == 3
 
     def test_content_find_ctime(self, swh_storage):
         cont = data.cont.copy()
         del cont["data"]
         ctime = now()
         cont["ctime"] = ctime
         swh_storage.content_add_metadata([cont])
 
         actually_present = swh_storage.content_find({"sha1": cont["sha1"]})
 
         # check ctime up to one second
         dt = actually_present[0]["ctime"] - ctime
         assert abs(dt.total_seconds()) <= 1
         del actually_present[0]["ctime"]
 
         assert actually_present[0] == {
             "sha1": cont["sha1"],
             "sha256": cont["sha256"],
             "sha1_git": cont["sha1_git"],
             "blake2s256": cont["blake2s256"],
             "length": cont["length"],
             "status": "visible",
         }
 
     def test_content_find_with_present_content(self, swh_storage):
         # 1. with something to find
         cont = data.cont
         swh_storage.content_add([cont, data.cont2])
 
         actually_present = swh_storage.content_find({"sha1": cont["sha1"]})
         assert 1 == len(actually_present)
         actually_present[0].pop("ctime")
 
         assert actually_present[0] == {
             "sha1": cont["sha1"],
             "sha256": cont["sha256"],
             "sha1_git": cont["sha1_git"],
             "blake2s256": cont["blake2s256"],
             "length": cont["length"],
             "status": "visible",
         }
 
         # 2. with something to find
         actually_present = swh_storage.content_find({"sha1_git": cont["sha1_git"]})
         assert 1 == len(actually_present)
 
         actually_present[0].pop("ctime")
         assert actually_present[0] == {
             "sha1": cont["sha1"],
             "sha256": cont["sha256"],
             "sha1_git": cont["sha1_git"],
             "blake2s256": cont["blake2s256"],
             "length": cont["length"],
             "status": "visible",
         }
 
         # 3. with something to find
         actually_present = swh_storage.content_find({"sha256": cont["sha256"]})
         assert 1 == len(actually_present)
 
         actually_present[0].pop("ctime")
         assert actually_present[0] == {
             "sha1": cont["sha1"],
             "sha256": cont["sha256"],
             "sha1_git": cont["sha1_git"],
             "blake2s256": cont["blake2s256"],
             "length": cont["length"],
             "status": "visible",
         }
 
         # 4. with something to find
         actually_present = swh_storage.content_find(
             {
                 "sha1": cont["sha1"],
                 "sha1_git": cont["sha1_git"],
                 "sha256": cont["sha256"],
                 "blake2s256": cont["blake2s256"],
             }
         )
         assert 1 == len(actually_present)
 
         actually_present[0].pop("ctime")
         assert actually_present[0] == {
             "sha1": cont["sha1"],
             "sha256": cont["sha256"],
             "sha1_git": cont["sha1_git"],
             "blake2s256": cont["blake2s256"],
             "length": cont["length"],
             "status": "visible",
         }
 
     def test_content_find_with_non_present_content(self, swh_storage):
         # 1. with something that does not exist
         missing_cont = data.missing_cont
 
         actually_present = swh_storage.content_find({"sha1": missing_cont["sha1"]})
 
         assert actually_present == []
 
         # 2. with something that does not exist
         actually_present = swh_storage.content_find(
             {"sha1_git": missing_cont["sha1_git"]}
         )
 
         assert actually_present == []
 
         # 3. with something that does not exist
         actually_present = swh_storage.content_find({"sha256": missing_cont["sha256"]})
 
         assert actually_present == []
 
     def test_content_find_with_duplicate_input(self, swh_storage):
         cont1 = data.cont
         duplicate_cont = cont1.copy()
 
         # Create fake data with colliding sha256 and blake2s256
         sha1_array = bytearray(duplicate_cont["sha1"])
         sha1_array[0] += 1
         duplicate_cont["sha1"] = bytes(sha1_array)
         sha1git_array = bytearray(duplicate_cont["sha1_git"])
         sha1git_array[0] += 1
         duplicate_cont["sha1_git"] = bytes(sha1git_array)
         # Inject the data
         swh_storage.content_add([cont1, duplicate_cont])
         finder = {
             "blake2s256": duplicate_cont["blake2s256"],
             "sha256": duplicate_cont["sha256"],
         }
         actual_result = list(swh_storage.content_find(finder))
 
         cont1.pop("data")
         duplicate_cont.pop("data")
         actual_result[0].pop("ctime")
         actual_result[1].pop("ctime")
 
         expected_result = [cont1, duplicate_cont]
         for result in expected_result:
             assert result in actual_result
 
     def test_content_find_with_duplicate_sha256(self, swh_storage):
         cont1 = data.cont
         duplicate_cont = cont1.copy()
 
         # Create fake data with colliding sha256
         for hashalgo in ("sha1", "sha1_git", "blake2s256"):
             value = bytearray(duplicate_cont[hashalgo])
             value[0] += 1
             duplicate_cont[hashalgo] = bytes(value)
         swh_storage.content_add([cont1, duplicate_cont])
 
         finder = {"sha256": duplicate_cont["sha256"]}
         actual_result = list(swh_storage.content_find(finder))
         assert len(actual_result) == 2
 
         cont1.pop("data")
         duplicate_cont.pop("data")
         actual_result[0].pop("ctime")
         actual_result[1].pop("ctime")
         expected_result = [cont1, duplicate_cont]
         assert expected_result == sorted(actual_result, key=lambda x: x["sha1"])
 
         # Find with both sha256 and blake2s256
         finder = {
             "sha256": duplicate_cont["sha256"],
             "blake2s256": duplicate_cont["blake2s256"],
         }
         actual_result = list(swh_storage.content_find(finder))
         assert len(actual_result) == 1
         actual_result[0].pop("ctime")
 
         expected_result = [duplicate_cont]
         assert actual_result[0] == duplicate_cont
 
     def test_content_find_with_duplicate_blake2s256(self, swh_storage):
         cont1 = data.cont
         duplicate_cont = cont1.copy()
 
         # Create fake data with colliding sha256 and blake2s256
         sha1_array = bytearray(duplicate_cont["sha1"])
         sha1_array[0] += 1
         duplicate_cont["sha1"] = bytes(sha1_array)
         sha1git_array = bytearray(duplicate_cont["sha1_git"])
         sha1git_array[0] += 1
         duplicate_cont["sha1_git"] = bytes(sha1git_array)
         sha256_array = bytearray(duplicate_cont["sha256"])
         sha256_array[0] += 1
         duplicate_cont["sha256"] = bytes(sha256_array)
         swh_storage.content_add([cont1, duplicate_cont])
         finder = {"blake2s256": duplicate_cont["blake2s256"]}
         actual_result = list(swh_storage.content_find(finder))
 
         cont1.pop("data")
         duplicate_cont.pop("data")
         actual_result[0].pop("ctime")
         actual_result[1].pop("ctime")
         expected_result = [cont1, duplicate_cont]
         for result in expected_result:
             assert result in actual_result
 
         # Find with both sha256 and blake2s256
         finder = {
             "sha256": duplicate_cont["sha256"],
             "blake2s256": duplicate_cont["blake2s256"],
         }
         actual_result = list(swh_storage.content_find(finder))
 
         actual_result[0].pop("ctime")
 
         expected_result = [duplicate_cont]
         assert expected_result == actual_result
 
     def test_content_find_bad_input(self, swh_storage):
         # 1. with bad input
         with pytest.raises(StorageArgumentException):
             swh_storage.content_find({})  # empty is bad
 
         # 2. with bad input
         with pytest.raises(StorageArgumentException):
             swh_storage.content_find({"unknown-sha1": "something"})  # not the right key
 
     def test_object_find_by_sha1_git(self, swh_storage):
         sha1_gits = [b"00000000000000000000"]
         expected = {
             b"00000000000000000000": [],
         }
 
         swh_storage.content_add([data.cont])
         sha1_gits.append(data.cont["sha1_git"])
         expected[data.cont["sha1_git"]] = [
             {"sha1_git": data.cont["sha1_git"], "type": "content",}
         ]
 
         swh_storage.directory_add([data.dir])
         sha1_gits.append(data.dir["id"])
         expected[data.dir["id"]] = [{"sha1_git": data.dir["id"], "type": "directory",}]
 
         swh_storage.revision_add([data.revision])
         sha1_gits.append(data.revision["id"])
         expected[data.revision["id"]] = [
             {"sha1_git": data.revision["id"], "type": "revision",}
         ]
 
         swh_storage.release_add([data.release])
         sha1_gits.append(data.release["id"])
         expected[data.release["id"]] = [
             {"sha1_git": data.release["id"], "type": "release",}
         ]
 
         ret = swh_storage.object_find_by_sha1_git(sha1_gits)
 
         assert expected == ret
 
     def test_metadata_fetcher_add_get(self, swh_storage):
         actual_fetcher = swh_storage.metadata_fetcher_get(
             data.metadata_fetcher["name"], data.metadata_fetcher["version"]
         )
         assert actual_fetcher is None  # does not exist
 
         swh_storage.metadata_fetcher_add(**data.metadata_fetcher)
 
         res = swh_storage.metadata_fetcher_get(
             data.metadata_fetcher["name"], data.metadata_fetcher["version"]
         )
 
         assert res is not data.metadata_fetcher
         assert res == data.metadata_fetcher
 
     def test_metadata_authority_add_get(self, swh_storage):
         actual_authority = swh_storage.metadata_authority_get(
             data.metadata_authority["type"], data.metadata_authority["url"]
         )
         assert actual_authority is None  # does not exist
 
         swh_storage.metadata_authority_add(**data.metadata_authority)
 
         res = swh_storage.metadata_authority_get(
             data.metadata_authority["type"], data.metadata_authority["url"]
         )
 
         assert res is not data.metadata_authority
         assert res == data.metadata_authority
 
     def test_origin_metadata_add(self, swh_storage):
         origin = data.origin
         fetcher = data.metadata_fetcher
         authority = data.metadata_authority
         assert swh_storage.origin_add([origin]) == {"origin:add": 1}
 
         swh_storage.metadata_fetcher_add(**fetcher)
         swh_storage.metadata_authority_add(**authority)
 
         swh_storage.origin_metadata_add(**data.origin_metadata)
         swh_storage.origin_metadata_add(**data.origin_metadata2)
 
         result = swh_storage.origin_metadata_get(origin["url"], authority)
         assert result["next_page_token"] is None
         assert [data.origin_metadata, data.origin_metadata2] == list(
             sorted(result["results"], key=lambda x: x["discovery_date"],)
         )
 
     def test_origin_metadata_add_duplicate(self, swh_storage):
         """Duplicates should be silently updated."""
         origin = data.origin
         fetcher = data.metadata_fetcher
         authority = data.metadata_authority
         assert swh_storage.origin_add([origin]) == {"origin:add": 1}
 
         new_origin_metadata2 = {
             **data.origin_metadata2,
             "format": "new-format",
             "metadata": b"new-metadata",
         }
 
         swh_storage.metadata_fetcher_add(**fetcher)
         swh_storage.metadata_authority_add(**authority)
 
         swh_storage.origin_metadata_add(**data.origin_metadata)
         swh_storage.origin_metadata_add(**data.origin_metadata2)
         swh_storage.origin_metadata_add(**new_origin_metadata2)
 
         result = swh_storage.origin_metadata_get(origin["url"], authority)
         assert result["next_page_token"] is None
         assert [data.origin_metadata, new_origin_metadata2] == list(
             sorted(result["results"], key=lambda x: x["discovery_date"],)
         )
 
     def test_origin_metadata_add_dict(self, swh_storage):
         origin = data.origin
         fetcher = data.metadata_fetcher
         authority = data.metadata_authority
         assert swh_storage.origin_add([origin]) == {"origin:add": 1}
 
         swh_storage.metadata_fetcher_add(**fetcher)
         swh_storage.metadata_authority_add(**authority)
 
         kwargs = data.origin_metadata.copy()
         kwargs["metadata"] = {"foo": "bar"}
 
         with pytest.raises(StorageArgumentException):
             swh_storage.origin_metadata_add(**kwargs)
 
     def test_origin_metadata_get(self, swh_storage):
         authority = data.metadata_authority
         fetcher = data.metadata_fetcher
         authority2 = data.metadata_authority2
         fetcher2 = data.metadata_fetcher2
         origin_url1 = data.origin["url"]
         origin_url2 = data.origin2["url"]
         assert swh_storage.origin_add([data.origin, data.origin2]) == {"origin:add": 2}
 
         origin1_metadata1 = data.origin_metadata
         origin1_metadata2 = data.origin_metadata2
         origin1_metadata3 = data.origin_metadata3
         origin2_metadata = {**data.origin_metadata2, "origin_url": origin_url2}
 
         swh_storage.metadata_authority_add(**authority)
         swh_storage.metadata_fetcher_add(**fetcher)
         swh_storage.metadata_authority_add(**authority2)
         swh_storage.metadata_fetcher_add(**fetcher2)
 
         swh_storage.origin_metadata_add(**origin1_metadata1)
         swh_storage.origin_metadata_add(**origin1_metadata2)
         swh_storage.origin_metadata_add(**origin1_metadata3)
         swh_storage.origin_metadata_add(**origin2_metadata)
 
         result = swh_storage.origin_metadata_get(origin_url1, authority)
         assert result["next_page_token"] is None
         assert [origin1_metadata1, origin1_metadata2] == list(
             sorted(result["results"], key=lambda x: x["discovery_date"],)
         )
 
         result = swh_storage.origin_metadata_get(origin_url1, authority2)
         assert result["next_page_token"] is None
         assert [origin1_metadata3] == list(
             sorted(result["results"], key=lambda x: x["discovery_date"],)
         )
 
         result = swh_storage.origin_metadata_get(origin_url2, authority)
         assert result["next_page_token"] is None
         assert [origin2_metadata] == list(result["results"],)
 
     def test_origin_metadata_get_after(self, swh_storage):
         origin = data.origin
         fetcher = data.metadata_fetcher
         authority = data.metadata_authority
         assert swh_storage.origin_add([origin]) == {"origin:add": 1}
 
         swh_storage.metadata_fetcher_add(**fetcher)
         swh_storage.metadata_authority_add(**authority)
 
         swh_storage.origin_metadata_add(**data.origin_metadata)
         swh_storage.origin_metadata_add(**data.origin_metadata2)
 
         result = swh_storage.origin_metadata_get(
             origin["url"],
             authority,
             after=data.origin_metadata["discovery_date"] - timedelta(seconds=1),
         )
         assert result["next_page_token"] is None
         assert [data.origin_metadata, data.origin_metadata2] == list(
             sorted(result["results"], key=lambda x: x["discovery_date"],)
         )
 
         result = swh_storage.origin_metadata_get(
             origin["url"], authority, after=data.origin_metadata["discovery_date"]
         )
         assert result["next_page_token"] is None
         assert [data.origin_metadata2] == result["results"]
 
         result = swh_storage.origin_metadata_get(
             origin["url"], authority, after=data.origin_metadata2["discovery_date"]
         )
         assert result["next_page_token"] is None
         assert [] == result["results"]
 
     def test_origin_metadata_get_paginate(self, swh_storage):
         origin = data.origin
         fetcher = data.metadata_fetcher
         authority = data.metadata_authority
         assert swh_storage.origin_add([origin]) == {"origin:add": 1}
 
         swh_storage.metadata_fetcher_add(**fetcher)
         swh_storage.metadata_authority_add(**authority)
 
         swh_storage.origin_metadata_add(**data.origin_metadata)
         swh_storage.origin_metadata_add(**data.origin_metadata2)
 
         swh_storage.origin_metadata_get(origin["url"], authority)
 
         result = swh_storage.origin_metadata_get(origin["url"], authority, limit=1)
         assert result["next_page_token"] is not None
         assert [data.origin_metadata] == result["results"]
 
         result = swh_storage.origin_metadata_get(
             origin["url"], authority, limit=1, page_token=result["next_page_token"]
         )
         assert result["next_page_token"] is None
         assert [data.origin_metadata2] == result["results"]
 
     def test_origin_metadata_get_paginate_same_date(self, swh_storage):
         origin = data.origin
         fetcher1 = data.metadata_fetcher
         fetcher2 = data.metadata_fetcher2
         authority = data.metadata_authority
         assert swh_storage.origin_add([origin]) == {"origin:add": 1}
 
         swh_storage.metadata_fetcher_add(**fetcher1)
         swh_storage.metadata_fetcher_add(**fetcher2)
         swh_storage.metadata_authority_add(**authority)
 
         origin_metadata2 = {
             **data.origin_metadata2,
             "discovery_date": data.origin_metadata2["discovery_date"],
             "fetcher": {"name": fetcher2["name"], "version": fetcher2["version"],},
         }
 
         swh_storage.origin_metadata_add(**data.origin_metadata)
         swh_storage.origin_metadata_add(**origin_metadata2)
 
         result = swh_storage.origin_metadata_get(origin["url"], authority, limit=1)
         assert result["next_page_token"] is not None
         assert [data.origin_metadata] == result["results"]
 
         result = swh_storage.origin_metadata_get(
             origin["url"], authority, limit=1, page_token=result["next_page_token"]
         )
         assert result["next_page_token"] is None
         assert [origin_metadata2] == result["results"]
 
 
 class TestStorageGeneratedData:
     def test_generate_content_get(self, swh_storage, swh_contents):
         contents_with_data = [c for c in swh_contents if c["status"] != "absent"]
         # input the list of sha1s we want from storage
         get_sha1s = [c["sha1"] for c in contents_with_data]
 
         # retrieve contents
         actual_contents = list(swh_storage.content_get(get_sha1s))
         assert None not in actual_contents
         assert_contents_ok(contents_with_data, actual_contents)
 
     def test_generate_content_get_metadata(self, swh_storage, swh_contents):
         # input the list of sha1s we want from storage
         expected_contents = [c for c in swh_contents if c["status"] != "absent"]
         get_sha1s = [c["sha1"] for c in expected_contents]
 
         # retrieve contents
         meta_contents = swh_storage.content_get_metadata(get_sha1s)
 
         assert len(list(meta_contents)) == len(get_sha1s)
 
         actual_contents = []
         for contents in meta_contents.values():
             actual_contents.extend(contents)
 
         keys_to_check = {"length", "status", "sha1", "sha1_git", "sha256", "blake2s256"}
 
         assert_contents_ok(
             expected_contents, actual_contents, keys_to_check=keys_to_check
         )
 
     def test_generate_content_get_range(self, swh_storage, swh_contents):
         """content_get_range returns complete range"""
         present_contents = [c for c in swh_contents if c["status"] != "absent"]
 
         get_sha1s = sorted([c["sha1"] for c in swh_contents if c["status"] != "absent"])
         start = get_sha1s[2]
         end = get_sha1s[-2]
         actual_result = swh_storage.content_get_range(start, end)
 
         assert actual_result["next"] is None
 
         actual_contents = actual_result["contents"]
         expected_contents = [c for c in present_contents if start <= c["sha1"] <= end]
         if expected_contents:
             assert_contents_ok(expected_contents, actual_contents, ["sha1"])
         else:
             assert actual_contents == []
 
     def test_generate_content_get_range_full(self, swh_storage, swh_contents):
         """content_get_range for a full range returns all available contents"""
         present_contents = [c for c in swh_contents if c["status"] != "absent"]
 
         start = b"0" * 40
         end = b"f" * 40
         actual_result = swh_storage.content_get_range(start, end)
         assert actual_result["next"] is None
 
         actual_contents = actual_result["contents"]
         expected_contents = [c for c in present_contents if start <= c["sha1"] <= end]
         if expected_contents:
             assert_contents_ok(expected_contents, actual_contents, ["sha1"])
         else:
             assert actual_contents == []
 
     def test_generate_content_get_range_empty(self, swh_storage, swh_contents):
         """content_get_range for an empty range returns nothing"""
         start = b"0" * 40
         end = b"f" * 40
         actual_result = swh_storage.content_get_range(end, start)
         assert actual_result["next"] is None
         assert len(actual_result["contents"]) == 0
 
     def test_generate_content_get_range_limit_none(self, swh_storage):
         """content_get_range call with wrong limit input should fail"""
         with pytest.raises(StorageArgumentException) as e:
             swh_storage.content_get_range(start=None, end=None, limit=None)
 
         assert e.value.args == ("limit should not be None",)
 
     def test_generate_content_get_range_no_limit(self, swh_storage, swh_contents):
         """content_get_range returns contents within range provided"""
         # input the list of sha1s we want from storage
         get_sha1s = sorted([c["sha1"] for c in swh_contents if c["status"] != "absent"])
         start = get_sha1s[0]
         end = get_sha1s[-1]
 
         # retrieve contents
         actual_result = swh_storage.content_get_range(start, end)
 
         actual_contents = actual_result["contents"]
         assert actual_result["next"] is None
         assert len(actual_contents) == len(get_sha1s)
 
         expected_contents = [c for c in swh_contents if c["status"] != "absent"]
         assert_contents_ok(expected_contents, actual_contents, ["sha1"])
 
     def test_generate_content_get_range_limit(self, swh_storage, swh_contents):
         """content_get_range paginates results if limit exceeded"""
         contents_map = {c["sha1"]: c for c in swh_contents}
 
         # input the list of sha1s we want from storage
         get_sha1s = sorted([c["sha1"] for c in swh_contents if c["status"] != "absent"])
         start = get_sha1s[0]
         end = get_sha1s[-1]
 
         # retrieve contents limited to n-1 results
         limited_results = len(get_sha1s) - 1
         actual_result = swh_storage.content_get_range(start, end, limit=limited_results)
 
         actual_contents = actual_result["contents"]
         assert actual_result["next"] == get_sha1s[-1]
         assert len(actual_contents) == limited_results
 
         expected_contents = [contents_map[sha1] for sha1 in get_sha1s[:-1]]
         assert_contents_ok(expected_contents, actual_contents, ["sha1"])
 
         # retrieve next part
         actual_results2 = swh_storage.content_get_range(start=end, end=end)
         assert actual_results2["next"] is None
         actual_contents2 = actual_results2["contents"]
         assert len(actual_contents2) == 1
 
         assert_contents_ok([contents_map[get_sha1s[-1]]], actual_contents2, ["sha1"])
 
     def test_origin_get_range_from_zero(self, swh_storage, swh_origins):
         actual_origins = list(
             swh_storage.origin_get_range(origin_from=0, origin_count=0)
         )
         assert len(actual_origins) == 0
 
         actual_origins = list(
             swh_storage.origin_get_range(origin_from=0, origin_count=1)
         )
         assert len(actual_origins) == 1
         assert actual_origins[0]["id"] == 1
         assert actual_origins[0]["url"] == swh_origins[0]["url"]
 
     @pytest.mark.parametrize(
         "origin_from,origin_count",
         [(1, 1), (1, 10), (1, 20), (1, 101), (11, 0), (11, 10), (91, 11)],
     )
     def test_origin_get_range(
         self, swh_storage, swh_origins, origin_from, origin_count
     ):
         actual_origins = list(
             swh_storage.origin_get_range(
                 origin_from=origin_from, origin_count=origin_count
             )
         )
 
         origins_with_id = list(enumerate(swh_origins, start=1))
         expected_origins = [
             {"url": origin["url"], "id": origin_id,}
             for (origin_id, origin) in origins_with_id[
                 origin_from - 1 : origin_from + origin_count - 1
             ]
         ]
 
         assert actual_origins == expected_origins
 
     @pytest.mark.parametrize("limit", [1, 7, 10, 100, 1000])
     def test_origin_list(self, swh_storage, swh_origins, limit):
         returned_origins = []
 
         page_token = None
         i = 0
         while True:
             result = swh_storage.origin_list(page_token=page_token, limit=limit)
             assert len(result["origins"]) <= limit
 
             returned_origins.extend(origin["url"] for origin in result["origins"])
 
             i += 1
             page_token = result.get("next_page_token")
 
             if page_token is None:
                 assert i * limit >= len(swh_origins)
                 break
             else:
                 assert len(result["origins"]) == limit
 
         expected_origins = [origin["url"] for origin in swh_origins]
         assert sorted(returned_origins) == sorted(expected_origins)
 
     ORIGINS = [
         "https://github.com/user1/repo1",
         "https://github.com/user2/repo1",
         "https://github.com/user3/repo1",
         "https://gitlab.com/user1/repo1",
         "https://gitlab.com/user2/repo1",
         "https://forge.softwareheritage.org/source/repo1",
     ]
 
     def test_origin_count(self, swh_storage):
         swh_storage.origin_add([{"url": url} for url in self.ORIGINS])
 
         assert swh_storage.origin_count("github") == 3
         assert swh_storage.origin_count("gitlab") == 2
         assert swh_storage.origin_count(".*user.*", regexp=True) == 5
         assert swh_storage.origin_count(".*user.*", regexp=False) == 0
         assert swh_storage.origin_count(".*user1.*", regexp=True) == 2
         assert swh_storage.origin_count(".*user1.*", regexp=False) == 0
 
     def test_origin_count_with_visit_no_visits(self, swh_storage):
         swh_storage.origin_add([{"url": url} for url in self.ORIGINS])
 
         # none of them have visits, so with_visit=True => 0
         assert swh_storage.origin_count("github", with_visit=True) == 0
         assert swh_storage.origin_count("gitlab", with_visit=True) == 0
         assert swh_storage.origin_count(".*user.*", regexp=True, with_visit=True) == 0
         assert swh_storage.origin_count(".*user.*", regexp=False, with_visit=True) == 0
         assert swh_storage.origin_count(".*user1.*", regexp=True, with_visit=True) == 0
         assert swh_storage.origin_count(".*user1.*", regexp=False, with_visit=True) == 0
 
     def test_origin_count_with_visit_with_visits_no_snapshot(self, swh_storage):
         swh_storage.origin_add([{"url": url} for url in self.ORIGINS])
 
         origin_url = "https://github.com/user1/repo1"
         visit = OriginVisit(origin=origin_url, date=now(), type="git",)
         swh_storage.origin_visit_add([visit])
 
         assert swh_storage.origin_count("github", with_visit=False) == 3
         # it has a visit, but no snapshot, so with_visit=True => 0
         assert swh_storage.origin_count("github", with_visit=True) == 0
 
         assert swh_storage.origin_count("gitlab", with_visit=False) == 2
         # these gitlab origins have no visit
         assert swh_storage.origin_count("gitlab", with_visit=True) == 0
 
         assert (
             swh_storage.origin_count("github.*user1", regexp=True, with_visit=False)
             == 1
         )
         assert (
             swh_storage.origin_count("github.*user1", regexp=True, with_visit=True) == 0
         )
         assert swh_storage.origin_count("github", regexp=True, with_visit=True) == 0
 
     def test_origin_count_with_visit_with_visits_and_snapshot(self, swh_storage):
         swh_storage.origin_add([{"url": url} for url in self.ORIGINS])
 
         swh_storage.snapshot_add([data.snapshot])
         origin_url = "https://github.com/user1/repo1"
         visit = OriginVisit(origin=origin_url, date=now(), type="git",)
         visit = swh_storage.origin_visit_add([visit])[0]
         swh_storage.origin_visit_status_add(
             [
                 OriginVisitStatus(
                     origin=origin_url,
                     visit=visit.visit,
                     date=now(),
                     status="ongoing",
                     snapshot=data.snapshot["id"],
                 )
             ]
         )
 
         assert swh_storage.origin_count("github", with_visit=False) == 3
         # github/user1 has a visit and a snapshot, so with_visit=True => 1
         assert swh_storage.origin_count("github", with_visit=True) == 1
 
         assert (
             swh_storage.origin_count("github.*user1", regexp=True, with_visit=False)
             == 1
         )
         assert (
             swh_storage.origin_count("github.*user1", regexp=True, with_visit=True) == 1
         )
         assert swh_storage.origin_count("github", regexp=True, with_visit=True) == 1
 
     @settings(suppress_health_check=[HealthCheck.too_slow])
     @given(strategies.lists(objects(), max_size=2))
     def test_add_arbitrary(self, swh_storage, objects):
         for (obj_type, obj) in objects:
             obj = obj.to_dict()
             if obj_type == "origin_visit":
                 origin_url = obj.pop("origin")
                 swh_storage.origin_add_one({"url": origin_url})
                 if "visit" in obj:
                     del obj["visit"]
                 visit = OriginVisit(
                     origin=origin_url, date=obj["date"], type=obj["type"],
                 )
                 swh_storage.origin_visit_add([visit])
             else:
                 if obj_type == "content" and obj["status"] == "absent":
                     obj_type = "skipped_content"
                 method = getattr(swh_storage, obj_type + "_add")
                 try:
                     method([obj])
                 except HashCollision:
                     pass
 
 
 @pytest.mark.db
 class TestLocalStorage:
     """Test the local storage"""
 
     # This test is only relevant on the local storage, with an actual
     # objstorage raising an exception
     def test_content_add_objstorage_exception(self, swh_storage):
         swh_storage.objstorage.content_add = Mock(
             side_effect=Exception("mocked broken objstorage")
         )
 
         with pytest.raises(Exception) as e:
             swh_storage.content_add([data.cont])
 
         assert e.value.args == ("mocked broken objstorage",)
         missing = list(swh_storage.content_missing([data.cont]))
         assert missing == [data.cont["sha1"]]
 
 
 @pytest.mark.db
 class TestStorageRaceConditions:
     @pytest.mark.xfail
     def test_content_add_race(self, swh_storage):
 
         results = queue.Queue()
 
         def thread():
             try:
                 with db_transaction(swh_storage) as (db, cur):
                     ret = swh_storage.content_add([data.cont], db=db, cur=cur)
                 results.put((threading.get_ident(), "data", ret))
             except Exception as e:
                 results.put((threading.get_ident(), "exc", e))
 
         t1 = threading.Thread(target=thread)
         t2 = threading.Thread(target=thread)
         t1.start()
         # this avoids the race condition
         # import time
         # time.sleep(1)
         t2.start()
         t1.join()
         t2.join()
 
         r1 = results.get(block=False)
         r2 = results.get(block=False)
 
         with pytest.raises(queue.Empty):
             results.get(block=False)
         assert r1[0] != r2[0]
         assert r1[1] == "data", "Got exception %r in Thread%s" % (r1[2], r1[0])
         assert r2[1] == "data", "Got exception %r in Thread%s" % (r2[2], r2[0])
 
 
 @pytest.mark.db
 class TestPgStorage:
     """This class is dedicated for the rare case where the schema needs to
        be altered dynamically.
 
        Otherwise, the tests could be blocking when ran altogether.
 
     """
 
     def test_content_update_with_new_cols(self, swh_storage):
         swh_storage.journal_writer.journal = None  # TODO, not supported
 
         with db_transaction(swh_storage) as (_, cur):
             cur.execute(
                 """alter table content
                            add column test text default null,
                            add column test2 text default null"""
             )
 
         cont = copy.deepcopy(data.cont2)
         swh_storage.content_add([cont])
         cont["test"] = "value-1"
         cont["test2"] = "value-2"
 
         swh_storage.content_update([cont], keys=["test", "test2"])
         with db_transaction(swh_storage) as (_, cur):
             cur.execute(
                 """SELECT sha1, sha1_git, sha256, length, status,
                    test, test2
                    FROM content WHERE sha1 = %s""",
                 (cont["sha1"],),
             )
 
             datum = cur.fetchone()
 
         assert datum == (
             cont["sha1"],
             cont["sha1_git"],
             cont["sha256"],
             cont["length"],
             "visible",
             cont["test"],
             cont["test2"],
         )
 
         with db_transaction(swh_storage) as (_, cur):
             cur.execute(
                 """alter table content drop column test,
                                                drop column test2"""
             )
 
     def test_content_add_db(self, swh_storage):
         cont = data.cont
 
         actual_result = swh_storage.content_add([cont])
 
         assert actual_result == {
             "content:add": 1,
             "content:add:bytes": cont["length"],
         }
 
         if hasattr(swh_storage, "objstorage"):
             assert cont["sha1"] in swh_storage.objstorage.objstorage
 
         with db_transaction(swh_storage) as (_, cur):
             cur.execute(
                 "SELECT sha1, sha1_git, sha256, length, status"
                 " FROM content WHERE sha1 = %s",
                 (cont["sha1"],),
             )
             datum = cur.fetchone()
 
         assert datum == (
             cont["sha1"],
             cont["sha1_git"],
             cont["sha256"],
             cont["length"],
             "visible",
         )
 
         expected_cont = cont.copy()
         del expected_cont["data"]
         contents = [
             obj
             for (obj_type, obj) in swh_storage.journal_writer.journal.objects
             if obj_type == "content"
         ]
         assert len(contents) == 1
         for obj in contents:
             obj_d = obj.to_dict()
             del obj_d["ctime"]
             assert obj_d == expected_cont
 
     def test_content_add_metadata_db(self, swh_storage):
         cont = data.cont
         del cont["data"]
         cont["ctime"] = now()
 
         actual_result = swh_storage.content_add_metadata([cont])
 
         assert actual_result == {
             "content:add": 1,
         }
 
         if hasattr(swh_storage, "objstorage"):
             assert cont["sha1"] not in swh_storage.objstorage.objstorage
         with db_transaction(swh_storage) as (_, cur):
             cur.execute(
                 "SELECT sha1, sha1_git, sha256, length, status"
                 " FROM content WHERE sha1 = %s",
                 (cont["sha1"],),
             )
             datum = cur.fetchone()
         assert datum == (
             cont["sha1"],
             cont["sha1_git"],
             cont["sha256"],
             cont["length"],
             "visible",
         )
 
         contents = [
             obj
             for (obj_type, obj) in swh_storage.journal_writer.journal.objects
             if obj_type == "content"
         ]
         assert len(contents) == 1
         for obj in contents:
             obj_d = obj.to_dict()
             assert obj_d == cont
 
     def test_skipped_content_add_db(self, swh_storage):
         cont = data.skipped_cont
         cont2 = data.skipped_cont2
         cont2["blake2s256"] = None
 
         actual_result = swh_storage.skipped_content_add([cont, cont, cont2])
 
         assert 2 <= actual_result.pop("skipped_content:add") <= 3
         assert actual_result == {}
 
         with db_transaction(swh_storage) as (_, cur):
             cur.execute(
                 "SELECT sha1, sha1_git, sha256, blake2s256, "
                 "length, status, reason "
                 "FROM skipped_content ORDER BY sha1_git"
             )
 
             dbdata = cur.fetchall()
 
         assert len(dbdata) == 2
         assert dbdata[0] == (
             cont["sha1"],
             cont["sha1_git"],
             cont["sha256"],
             cont["blake2s256"],
             cont["length"],
             "absent",
             "Content too long",
         )
 
         assert dbdata[1] == (
             cont2["sha1"],
             cont2["sha1_git"],
             cont2["sha256"],
             cont2["blake2s256"],
             cont2["length"],
             "absent",
             "Content too long",
         )
 
     def test_clear_buffers(self, swh_storage):
         """Calling clear buffers on real storage does nothing
 
         """
         assert swh_storage.clear_buffers() is None
 
     def test_flush(self, swh_storage):
         """Calling clear buffers on real storage does nothing
 
         """
         assert swh_storage.flush() == {}