diff --git a/swh/storage/postgresql/storage.py b/swh/storage/postgresql/storage.py index c40aea3a..a14eccef 100644 --- a/swh/storage/postgresql/storage.py +++ b/swh/storage/postgresql/storage.py @@ -1,1702 +1,1702 @@ # Copyright (C) 2015-2022 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 base64 from collections import defaultdict import contextlib from contextlib import contextmanager import datetime import itertools import logging import operator from typing import Any, Counter, Dict, Iterable, List, Optional, Sequence, Tuple import attr import psycopg2 import psycopg2.errors import psycopg2.pool from swh.core.api.serializers import msgpack_dumps, msgpack_loads from swh.core.db.common import db_transaction, db_transaction_generator from swh.core.db.db_utils import swh_db_flavor, swh_db_version from swh.model.hashutil import DEFAULT_ALGORITHMS, hash_to_bytes, hash_to_hex from swh.model.model import ( SHA1_SIZE, Content, Directory, DirectoryEntry, ExtID, MetadataAuthority, MetadataAuthorityType, MetadataFetcher, Origin, OriginVisit, OriginVisitStatus, RawExtrinsicMetadata, Release, Revision, Sha1, Sha1Git, SkippedContent, Snapshot, SnapshotBranch, TargetType, ) from swh.model.swhids import ExtendedObjectType, ExtendedSWHID, ObjectType from swh.storage.exc import HashCollision, StorageArgumentException, StorageDBError from swh.storage.interface import ( VISIT_STATUSES, ListOrder, OriginVisitWithStatuses, PagedResult, PartialBranches, ) from swh.storage.objstorage import ObjStorage from swh.storage.utils import ( extract_collision_hash, get_partition_bounds_bytes, map_optional, now, ) from swh.storage.writer import JournalWriter from . import converters from .db import Db logger = logging.getLogger(__name__) # Max block size of contents to return BULK_BLOCK_CONTENT_LEN_MAX = 10000 EMPTY_SNAPSHOT_ID = hash_to_bytes("1a8893e6a86f444e8be8e7bda6cb34fb1735a00e") """Identifier for the empty snapshot""" VALIDATION_EXCEPTIONS = ( KeyError, TypeError, ValueError, psycopg2.errors.CheckViolation, psycopg2.errors.IntegrityError, psycopg2.errors.InvalidTextRepresentation, psycopg2.errors.NotNullViolation, psycopg2.errors.NumericValueOutOfRange, psycopg2.errors.UndefinedFunction, # (raised on wrong argument typs) ) """Exceptions raised by postgresql when validation of the arguments failed.""" @contextlib.contextmanager def convert_validation_exceptions(): """Catches postgresql errors related to invalid arguments, and re-raises a StorageArgumentException.""" try: yield except psycopg2.errors.UniqueViolation: # This only happens because of concurrent insertions, but it is # a subclass of IntegrityError; so we need to catch and reraise it # before the next clause converts it to StorageArgumentException. raise except VALIDATION_EXCEPTIONS as e: raise StorageArgumentException(str(e)) class Storage: """SWH storage datastore proxy, encompassing DB and object storage""" - current_version: int = 183 + current_version: int = 184 def __init__( self, db, objstorage, min_pool_conns=1, max_pool_conns=10, journal_writer=None, query_options=None, ): """Instantiate a storage instance backed by a PostgreSQL database and an objstorage. When ``db`` is passed as a connection string, then this module automatically manages a connection pool between ``min_pool_conns`` and ``max_pool_conns``. When ``db`` is an explicit psycopg2 connection, then ``min_pool_conns`` and ``max_pool_conns`` are ignored and the connection is used directly. Args: db: either a libpq connection string, or a psycopg2 connection objstorage: configuration for the backend :class:`ObjStorage` min_pool_conns: min number of connections in the psycopg2 pool max_pool_conns: max number of connections in the psycopg2 pool journal_writer: configuration for the :class:`JournalWriter` query_options: configuration for the sql connections; keys of the dict are the method names decorated with :func:`db_transaction` or :func:`db_transaction_generator` (eg. :func:`content_find`), and values are dicts (config_name, config_value) used to configure the sql connection for the method_name. For example, using:: {"content_get": {"statement_timeout": 5000}} will override the default statement timeout for the :func:`content_get` endpoint from 500ms to 5000ms. See :mod:`swh.core.db.common` for more details. """ try: if isinstance(db, psycopg2.extensions.connection): self._pool = None self._db = Db(db) # See comment below self._db.cursor().execute("SET TIME ZONE 'UTC'") else: self._pool = psycopg2.pool.ThreadedConnectionPool( min_pool_conns, max_pool_conns, db ) self._db = None except psycopg2.OperationalError as e: raise StorageDBError(e) self.journal_writer = JournalWriter(journal_writer) self.objstorage = ObjStorage(objstorage) self.query_options = query_options self._flavor: Optional[str] = None def get_db(self): if self._db: return self._db else: db = Db.from_pool(self._pool) # Workaround for psycopg2 < 2.9.0 not handling fractional timezones, # which may happen on old revision/release dates on systems configured # with non-UTC timezones. # https://www.psycopg.org/docs/usage.html#time-zones-handling db.cursor().execute("SET TIME ZONE 'UTC'") return db def put_db(self, db): if db is not self._db: db.put_conn() @contextmanager def db(self): db = None try: db = self.get_db() yield db finally: if db: self.put_db(db) @db_transaction() def get_flavor(self, *, db: Db, cur=None) -> str: flavor = swh_db_flavor(db.conn.dsn) assert flavor is not None return flavor @property def flavor(self) -> str: if self._flavor is None: self._flavor = self.get_flavor() assert self._flavor is not None return self._flavor @db_transaction() def check_config(self, *, check_write: bool, db: Db, cur=None) -> bool: if not self.objstorage.check_config(check_write=check_write): return False dbversion = swh_db_version(db.conn.dsn) if dbversion != self.current_version: logger.warning( "database dbversion (%s) != %s current_version (%s)", dbversion, __name__, self.current_version, ) return False # Check permissions on one of the tables check = "INSERT" if check_write else "SELECT" cur.execute("select has_table_privilege(current_user, 'content', %s)", (check,)) return cur.fetchone()[0] def _content_unique_key(self, hash, db): """Given a hash (tuple or dict), return a unique key from the aggregation of keys. """ keys = db.content_hash_keys if isinstance(hash, tuple): return hash return tuple([hash[k] for k in keys]) def _content_add_metadata(self, db, cur, content): """Add content to the postgresql database but not the object storage.""" # create temporary table for metadata injection db.mktemp("content", cur) db.copy_to( (c.to_dict() for c in content), "tmp_content", db.content_add_keys, cur ) # move metadata in place try: db.content_add_from_temp(cur) except psycopg2.IntegrityError as e: if e.diag.sqlstate == "23505" and e.diag.table_name == "content": message_detail = e.diag.message_detail if message_detail: hash_name, hash_id = extract_collision_hash(message_detail) collision_contents_hashes = [ c.hashes() for c in content if c.get_hash(hash_name) == hash_id ] else: constraint_to_hash_name = { "content_pkey": "sha1", "content_sha1_git_idx": "sha1_git", "content_sha256_idx": "sha256", } hash_name = constraint_to_hash_name.get(e.diag.constraint_name) hash_id = None collision_contents_hashes = None raise HashCollision( hash_name, hash_id, collision_contents_hashes ) from None else: raise def content_add(self, content: List[Content]) -> Dict[str, int]: ctime = now() contents = [attr.evolve(c, ctime=ctime) for c in content] # Must add to the objstorage before the DB and journal. Otherwise: # 1. in case of a crash the DB may "believe" we have the content, but # we didn't have time to write to the objstorage before the crash # 2. the objstorage mirroring, which reads from the journal, may attempt to # read from the objstorage before we finished writing it objstorage_summary = self.objstorage.content_add(contents) with self.db() as db: with db.transaction() as cur: missing = list( self.content_missing( map(Content.to_dict, contents), key_hash="sha1_git", db=db, cur=cur, ) ) contents = [c for c in contents if c.sha1_git in missing] self.journal_writer.content_add(contents) self._content_add_metadata(db, cur, contents) return { "content:add": len(contents), "content:add:bytes": objstorage_summary["content:add:bytes"], } @db_transaction() def content_update( self, contents: List[Dict[str, Any]], keys: List[str] = [], *, db: Db, cur=None ) -> None: # TODO: Add a check on input keys. How to properly implement # this? We don't know yet the new columns. self.journal_writer.content_update(contents) db.mktemp("content", cur) select_keys = list(set(db.content_get_metadata_keys).union(set(keys))) with convert_validation_exceptions(): db.copy_to(contents, "tmp_content", select_keys, cur) db.content_update_from_temp(keys_to_update=keys, cur=cur) @db_transaction() def content_add_metadata( self, content: List[Content], *, db: Db, cur=None ) -> Dict[str, int]: missing = self.content_missing( (c.to_dict() for c in content), key_hash="sha1_git", db=db, cur=cur, ) contents = [c for c in content if c.sha1_git in missing] self.journal_writer.content_add_metadata(contents) self._content_add_metadata(db, cur, contents) return { "content:add": len(contents), } def content_get_data(self, content: Sha1) -> Optional[bytes]: # FIXME: Make this method support slicing the `data` return self.objstorage.content_get(content) @db_transaction() def content_get_partition( self, partition_id: int, nb_partitions: int, page_token: Optional[str] = None, limit: int = 1000, *, db: Db, cur=None, ) -> PagedResult[Content]: 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 next_page_token: Optional[str] = None contents = [] for counter, row in enumerate(db.content_get_range(start, end, limit + 1, cur)): row_d = dict(zip(db.content_get_metadata_keys, row)) content = Content(**row_d) if counter >= limit: # take the last content for the next page starting from this next_page_token = hash_to_hex(content.sha1) break contents.append(content) assert len(contents) <= limit return PagedResult(results=contents, next_page_token=next_page_token) @db_transaction(statement_timeout=500) def content_get( self, contents: List[bytes], algo: str = "sha1", *, db: Db, cur=None ) -> List[Optional[Content]]: contents_by_hash: Dict[bytes, Optional[Content]] = {} if algo not in DEFAULT_ALGORITHMS: raise StorageArgumentException( "algo should be one of {','.join(DEFAULT_ALGORITHMS)}" ) rows = db.content_get_metadata_from_hashes(contents, algo, cur) key = operator.attrgetter(algo) for row in rows: row_d = dict(zip(db.content_get_metadata_keys, row)) content = Content(**row_d) contents_by_hash[key(content)] = content return [contents_by_hash.get(sha1) for sha1 in contents] @db_transaction_generator() def content_missing( self, contents: List[Dict[str, Any]], key_hash: str = "sha1", *, db: Db, cur=None, ) -> Iterable[bytes]: if key_hash not in DEFAULT_ALGORITHMS: raise StorageArgumentException( "key_hash should be one of {','.join(DEFAULT_ALGORITHMS)}" ) keys = db.content_hash_keys key_hash_idx = keys.index(key_hash) for obj in db.content_missing_from_list(contents, cur): yield obj[key_hash_idx] @db_transaction_generator() def content_missing_per_sha1( self, contents: List[bytes], *, db: Db, cur=None ) -> Iterable[bytes]: for obj in db.content_missing_per_sha1(contents, cur): yield obj[0] @db_transaction_generator() def content_missing_per_sha1_git( self, contents: List[bytes], *, db: Db, cur=None ) -> Iterable[Sha1Git]: for obj in db.content_missing_per_sha1_git(contents, cur): yield obj[0] @db_transaction() def content_find( self, content: Dict[str, Any], *, db: Db, cur=None ) -> List[Content]: if not set(content).intersection(DEFAULT_ALGORITHMS): raise StorageArgumentException( "content keys must contain at least one " f"of: {', '.join(sorted(DEFAULT_ALGORITHMS))}" ) rows = db.content_find( sha1=content.get("sha1"), sha1_git=content.get("sha1_git"), sha256=content.get("sha256"), blake2s256=content.get("blake2s256"), cur=cur, ) contents = [] for row in rows: row_d = dict(zip(db.content_find_cols, row)) contents.append(Content(**row_d)) return contents @db_transaction() def content_get_random(self, *, db: Db, cur=None) -> Sha1Git: return db.content_get_random(cur) @staticmethod def _skipped_content_normalize(d): d = d.copy() if d.get("status") is None: d["status"] = "absent" if d.get("length") is None: d["length"] = -1 return d def _skipped_content_add_metadata(self, db, cur, content: List[SkippedContent]): origin_ids = db.origin_id_get_by_url([cont.origin for cont in content], cur=cur) content = [ attr.evolve(c, origin=origin_id) for (c, origin_id) in zip(content, origin_ids) ] db.mktemp("skipped_content", cur) db.copy_to( [c.to_dict() for c in content], "tmp_skipped_content", db.skipped_content_keys, cur, ) # move metadata in place db.skipped_content_add_from_temp(cur) @db_transaction() def skipped_content_add( self, content: List[SkippedContent], *, db: Db, cur=None ) -> Dict[str, int]: ctime = now() content = [attr.evolve(c, ctime=ctime) for c in content] missing_contents = self.skipped_content_missing( (c.to_dict() for c in content), db=db, cur=cur, ) content = [ c for c in content if any( all( c.get_hash(algo) == missing_content.get(algo) for algo in DEFAULT_ALGORITHMS ) for missing_content in missing_contents ) ] self.journal_writer.skipped_content_add(content) self._skipped_content_add_metadata(db, cur, content) return { "skipped_content:add": len(content), } @db_transaction_generator() def skipped_content_missing( self, contents: List[Dict[str, Any]], *, db: Db, cur=None ) -> Iterable[Dict[str, Any]]: contents = list(contents) for content in db.skipped_content_missing(contents, cur): yield dict(zip(db.content_hash_keys, content)) @db_transaction() def directory_add( self, directories: List[Directory], *, db: Db, cur=None ) -> Dict[str, int]: summary = {"directory:add": 0} dirs = set() dir_entries: Dict[str, defaultdict] = { "file": defaultdict(list), "dir": defaultdict(list), "rev": defaultdict(list), } for cur_dir in directories: dir_id = cur_dir.id dirs.add(dir_id) for src_entry in cur_dir.entries: entry = src_entry.to_dict() entry["dir_id"] = dir_id dir_entries[entry["type"]][dir_id].append(entry) dirs_missing = set(self.directory_missing(dirs, db=db, cur=cur)) if not dirs_missing: return summary self.journal_writer.directory_add( dir_ for dir_ in directories if dir_.id in dirs_missing ) # Copy directory metadata dirs_missing_dict = ( {"id": dir_.id, "raw_manifest": dir_.raw_manifest} for dir_ in directories if dir_.id in dirs_missing ) db.mktemp("directory", cur) db.copy_to(dirs_missing_dict, "tmp_directory", ["id", "raw_manifest"], cur) # Copy entries for entry_type, entry_list in dir_entries.items(): entries = itertools.chain.from_iterable( entries_for_dir for dir_id, entries_for_dir in entry_list.items() if dir_id in dirs_missing ) db.mktemp_dir_entry(entry_type) db.copy_to( entries, "tmp_directory_entry_%s" % entry_type, ["target", "name", "perms", "dir_id"], cur, ) # Do the final copy db.directory_add_from_temp(cur) summary["directory:add"] = len(dirs_missing) return summary @db_transaction_generator() def directory_missing( self, directories: List[Sha1Git], *, db: Db, cur=None ) -> Iterable[Sha1Git]: for obj in db.directory_missing_from_list(directories, cur): yield obj[0] @db_transaction_generator(statement_timeout=20000) def directory_ls( self, directory: Sha1Git, recursive: bool = False, *, db: Db, cur=None ) -> Iterable[Dict[str, Any]]: if recursive: res_gen = db.directory_walk(directory, cur=cur) else: res_gen = db.directory_walk_one(directory, cur=cur) for line in res_gen: yield dict(zip(db.directory_ls_cols, line)) @db_transaction(statement_timeout=4000) def directory_entry_get_by_path( self, directory: Sha1Git, paths: List[bytes], *, db: Db, cur=None ) -> Optional[Dict[str, Any]]: res = db.directory_entry_get_by_path(directory, paths, cur) return dict(zip(db.directory_ls_cols, res)) if res else None @db_transaction() def directory_get_random(self, *, db: Db, cur=None) -> Sha1Git: return db.directory_get_random(cur) @db_transaction() def directory_get_entries( self, directory_id: Sha1Git, page_token: Optional[bytes] = None, limit: int = 1000, *, db: Db, cur=None, ) -> Optional[PagedResult[DirectoryEntry]]: if list(self.directory_missing([directory_id], db=db, cur=cur)): return None if page_token is not None: raise StorageArgumentException("Unsupported page token") # TODO: actually paginate rows = db.directory_get_entries(directory_id, cur=cur) return PagedResult( results=[ DirectoryEntry(**dict(zip(db.directory_get_entries_cols, row))) for row in rows ], next_page_token=None, ) @db_transaction() def directory_get_raw_manifest( self, directory_ids: List[Sha1Git], *, db: Db, cur=None ) -> Dict[Sha1Git, Optional[bytes]]: return dict(db.directory_get_raw_manifest(directory_ids, cur=cur)) @db_transaction() def revision_add( self, revisions: List[Revision], *, db: Db, cur=None ) -> Dict[str, int]: summary = {"revision:add": 0} revisions_missing = set( self.revision_missing( set(revision.id for revision in revisions), db=db, cur=cur ) ) if not revisions_missing: return summary db.mktemp_revision(cur) revisions_filtered = [ revision for revision in revisions if revision.id in revisions_missing ] self.journal_writer.revision_add(revisions_filtered) db_revisions_filtered = list(map(converters.revision_to_db, revisions_filtered)) parents_filtered: List[Dict[str, Any]] = [] with convert_validation_exceptions(): db.copy_to( db_revisions_filtered, "tmp_revision", db.revision_add_cols, cur, lambda rev: parents_filtered.extend(rev["parents"]), ) db.revision_add_from_temp(cur) db.copy_to( parents_filtered, "revision_history", ["id", "parent_id", "parent_rank"], cur, ) return {"revision:add": len(revisions_missing)} @db_transaction_generator() def revision_missing( self, revisions: List[Sha1Git], *, db: Db, cur=None ) -> Iterable[Sha1Git]: if not revisions: return None for obj in db.revision_missing_from_list(revisions, cur): yield obj[0] @db_transaction(statement_timeout=2000) def revision_get( self, revision_ids: List[Sha1Git], ignore_displayname: bool = False, *, db: Db, cur=None, ) -> List[Optional[Revision]]: revisions = [] for line in db.revision_get_from_list(revision_ids, ignore_displayname, cur): revision = converters.db_to_revision(dict(zip(db.revision_get_cols, line))) revisions.append(revision) return revisions @db_transaction_generator(statement_timeout=2000) def revision_log( self, revisions: List[Sha1Git], ignore_displayname: bool = False, limit: Optional[int] = None, *, db: Db, cur=None, ) -> Iterable[Optional[Dict[str, Any]]]: for line in db.revision_log( revisions, ignore_displayname=ignore_displayname, limit=limit, cur=cur ): data = converters.db_to_revision(dict(zip(db.revision_get_cols, line))) if not data: yield None continue yield data.to_dict() @db_transaction_generator(statement_timeout=2000) def revision_shortlog( self, revisions: List[Sha1Git], limit: Optional[int] = None, *, db: Db, cur=None ) -> Iterable[Optional[Tuple[Sha1Git, Tuple[Sha1Git, ...]]]]: yield from db.revision_shortlog(revisions, limit, cur) @db_transaction() def revision_get_random(self, *, db: Db, cur=None) -> Sha1Git: return db.revision_get_random(cur) @db_transaction() def extid_get_from_extid( self, id_type: str, ids: List[bytes], version: Optional[int] = None, *, db: Db, cur=None, ) -> List[ExtID]: extids = [] for row in db.extid_get_from_extid_list(id_type, ids, version=version, cur=cur): if row[0] is not None: extids.append(converters.db_to_extid(dict(zip(db.extid_cols, row)))) return extids @db_transaction() def extid_get_from_target( self, target_type: ObjectType, ids: List[Sha1Git], extid_type: Optional[str] = None, extid_version: Optional[int] = None, *, db: Db, cur=None, ) -> List[ExtID]: extids = [] if (extid_version is not None and extid_type is None) or ( extid_version is None and extid_type is not None ): raise ValueError("You must provide both extid_type and extid_version") for row in db.extid_get_from_swhid_list( target_type.value, ids, extid_version=extid_version, extid_type=extid_type, cur=cur, ): if row[0] is not None: extids.append(converters.db_to_extid(dict(zip(db.extid_cols, row)))) return extids @db_transaction() def extid_add(self, ids: List[ExtID], *, db: Db, cur=None) -> Dict[str, int]: extid = [ { "extid": extid.extid, "extid_type": extid.extid_type, "extid_version": getattr(extid, "extid_version", 0), "target": extid.target.object_id, "target_type": extid.target.object_type.name.lower(), # arghh } for extid in ids ] db.mktemp("extid", cur) self.journal_writer.extid_add(ids) db.copy_to(extid, "tmp_extid", db.extid_cols, cur) # move metadata in place db.extid_add_from_temp(cur) return {"extid:add": len(extid)} @db_transaction() def release_add( self, releases: List[Release], *, db: Db, cur=None ) -> Dict[str, int]: summary = {"release:add": 0} release_ids = set(release.id for release in releases) releases_missing = set(self.release_missing(release_ids, db=db, cur=cur)) if not releases_missing: return summary db.mktemp_release(cur) releases_filtered = [ release for release in releases if release.id in releases_missing ] self.journal_writer.release_add(releases_filtered) db_releases_filtered = list(map(converters.release_to_db, releases_filtered)) with convert_validation_exceptions(): db.copy_to(db_releases_filtered, "tmp_release", db.release_add_cols, cur) db.release_add_from_temp(cur) return {"release:add": len(releases_missing)} @db_transaction_generator() def release_missing( self, releases: List[Sha1Git], *, db: Db, cur=None ) -> Iterable[Sha1Git]: if not releases: return for obj in db.release_missing_from_list(releases, cur): yield obj[0] @db_transaction(statement_timeout=1000) def release_get( self, releases: List[Sha1Git], ignore_displayname: bool = False, *, db: Db, cur=None, ) -> List[Optional[Release]]: rels = [] for release in db.release_get_from_list(releases, ignore_displayname, cur): data = converters.db_to_release(dict(zip(db.release_get_cols, release))) rels.append(data if data else None) return rels @db_transaction() def release_get_random(self, *, db: Db, cur=None) -> Sha1Git: return db.release_get_random(cur) @db_transaction() def snapshot_add( self, snapshots: List[Snapshot], *, db: Db, cur=None ) -> Dict[str, int]: created_temp_table = False count = 0 for snapshot in snapshots: if not db.snapshot_exists(snapshot.id, cur): if not created_temp_table: db.mktemp_snapshot_branch(cur) created_temp_table = True with convert_validation_exceptions(): db.copy_to( ( { "name": name, "target": info.target if info else None, "target_type": ( info.target_type.value if info else None ), } for name, info in snapshot.branches.items() ), "tmp_snapshot_branch", ["name", "target", "target_type"], cur, ) self.journal_writer.snapshot_add([snapshot]) db.snapshot_add(snapshot.id, cur) count += 1 return {"snapshot:add": count} @db_transaction_generator() def snapshot_missing( self, snapshots: List[Sha1Git], *, db: Db, cur=None ) -> Iterable[Sha1Git]: for obj in db.snapshot_missing_from_list(snapshots, cur): yield obj[0] @db_transaction(statement_timeout=2000) def snapshot_get( self, snapshot_id: Sha1Git, *, db: Db, cur=None ) -> Optional[Dict[str, Any]]: d = self.snapshot_get_branches(snapshot_id) if d is None: return d return { "id": d["id"], "branches": { name: branch.to_dict() if branch else None for (name, branch) in d["branches"].items() }, "next_branch": d["next_branch"], } @db_transaction(statement_timeout=2000) def snapshot_count_branches( self, snapshot_id: Sha1Git, branch_name_exclude_prefix: Optional[bytes] = None, *, db: Db, cur=None, ) -> Optional[Dict[Optional[str], int]]: return dict( [ bc for bc in db.snapshot_count_branches( snapshot_id, branch_name_exclude_prefix, cur, ) ] ) @db_transaction(statement_timeout=2000) def snapshot_get_branches( self, snapshot_id: Sha1Git, branches_from: bytes = b"", branches_count: int = 1000, target_types: Optional[List[str]] = None, branch_name_include_substring: Optional[bytes] = None, branch_name_exclude_prefix: Optional[bytes] = None, *, db: Db, cur=None, ) -> Optional[PartialBranches]: if snapshot_id == EMPTY_SNAPSHOT_ID: return PartialBranches( id=snapshot_id, branches={}, next_branch=None, ) if list(self.snapshot_missing([snapshot_id])): return None branches = {} next_branch = None fetched_branches = list( db.snapshot_get_by_id( snapshot_id, branches_from=branches_from, # the underlying SQL query can be quite expensive to execute for small # branches_count value, so we ensure a minimum branches limit of 10 for # optimal performances branches_count=max(branches_count + 1, 10), target_types=target_types, branch_name_include_substring=branch_name_include_substring, branch_name_exclude_prefix=branch_name_exclude_prefix, cur=cur, ) ) for row in fetched_branches[:branches_count]: branch_d = dict(zip(db.snapshot_get_cols, row)) del branch_d["snapshot_id"] name = branch_d.pop("name") if branch_d["target"] is None and branch_d["target_type"] is None: branch = None else: assert branch_d["target_type"] is not None branch = SnapshotBranch( target=branch_d["target"], target_type=TargetType(branch_d["target_type"]), ) branches[name] = branch if len(fetched_branches) > branches_count: next_branch = dict( zip(db.snapshot_get_cols, fetched_branches[branches_count]) )["name"] return PartialBranches( id=snapshot_id, branches=branches, next_branch=next_branch, ) @db_transaction() def snapshot_get_random(self, *, db: Db, cur=None) -> Sha1Git: return db.snapshot_get_random(cur) @db_transaction() def origin_visit_add( self, visits: List[OriginVisit], *, db: Db, cur=None ) -> Iterable[OriginVisit]: for visit in visits: origin = self.origin_get([visit.origin], db=db, cur=cur)[0] if not origin: # Cannot add a visit without an origin raise StorageArgumentException("Unknown origin %s", visit.origin) all_visits = [] for visit in visits: if visit.visit: self.journal_writer.origin_visit_add([visit]) db.origin_visit_add_with_id(visit, cur=cur) else: # visit_id is not given, it needs to be set by the db with convert_validation_exceptions(): visit_id = db.origin_visit_add( visit.origin, visit.date, visit.type, cur=cur ) visit = attr.evolve(visit, visit=visit_id) # Forced to write in the journal after the db (since its the db # call that set the visit id) self.journal_writer.origin_visit_add([visit]) # In this case, we also want to create the initial OVS object visit_status = OriginVisitStatus( origin=visit.origin, visit=visit_id, date=visit.date, type=visit.type, status="created", snapshot=None, ) self._origin_visit_status_add(visit_status, db=db, cur=cur) assert visit.visit is not None all_visits.append(visit) return all_visits def _origin_visit_status_add( self, visit_status: OriginVisitStatus, db, cur ) -> None: """Add an origin visit status""" self.journal_writer.origin_visit_status_add([visit_status]) db.origin_visit_status_add(visit_status, cur=cur) @db_transaction() def origin_visit_status_add( self, visit_statuses: List[OriginVisitStatus], *, db: Db, cur=None, ) -> Dict[str, int]: visit_statuses_ = [] # First round to check existence (fail early if any is ko) for visit_status in visit_statuses: origin_url = self.origin_get([visit_status.origin], db=db, cur=cur)[0] if not origin_url: raise StorageArgumentException(f"Unknown origin {visit_status.origin}") if visit_status.type is None: origin_visit = self.origin_visit_get_by( visit_status.origin, visit_status.visit, db=db, cur=cur ) if origin_visit is None: raise StorageArgumentException( f"Unknown origin visit {visit_status.visit} " f"of origin {visit_status.origin}" ) origin_visit_status = attr.evolve(visit_status, type=origin_visit.type) else: origin_visit_status = visit_status visit_statuses_.append(origin_visit_status) for visit_status in visit_statuses_: self._origin_visit_status_add(visit_status, db, cur) return {"origin_visit_status:add": len(visit_statuses_)} @db_transaction() def origin_visit_status_get_latest( self, origin_url: str, visit: int, allowed_statuses: Optional[List[str]] = None, require_snapshot: bool = False, *, db: Db, cur=None, ) -> Optional[OriginVisitStatus]: if allowed_statuses and not set(allowed_statuses).intersection(VISIT_STATUSES): raise StorageArgumentException( f"Unknown allowed statuses {','.join(allowed_statuses)}, only " f"{','.join(VISIT_STATUSES)} authorized" ) row_d = db.origin_visit_status_get_latest( origin_url, visit, allowed_statuses, require_snapshot, cur=cur ) if not row_d: return None return OriginVisitStatus(**row_d) @db_transaction(statement_timeout=500) def origin_visit_get( self, origin: str, page_token: Optional[str] = None, order: ListOrder = ListOrder.ASC, limit: int = 10, *, db: Db, cur=None, ) -> PagedResult[OriginVisit]: page_token = page_token or "0" if not isinstance(order, ListOrder): raise StorageArgumentException("order must be a ListOrder value") if not isinstance(page_token, str): raise StorageArgumentException("page_token must be a string.") next_page_token = None visit_from = int(page_token) visits: List[OriginVisit] = [] extra_limit = limit + 1 for row in db.origin_visit_get_range( origin, visit_from=visit_from, order=order, limit=extra_limit, cur=cur ): row_d = dict(zip(db.origin_visit_cols, row)) visits.append( OriginVisit( origin=row_d["origin"], visit=row_d["visit"], date=row_d["date"], type=row_d["type"], ) ) assert len(visits) <= extra_limit if len(visits) == extra_limit: visits = visits[:limit] next_page_token = str(visits[-1].visit) return PagedResult(results=visits, next_page_token=next_page_token) @db_transaction(statement_timeout=2000) def origin_visit_get_with_statuses( self, origin: str, allowed_statuses: Optional[List[str]] = None, require_snapshot: bool = False, page_token: Optional[str] = None, order: ListOrder = ListOrder.ASC, limit: int = 10, *, db: Db, cur=None, ) -> PagedResult[OriginVisitWithStatuses]: page_token = page_token or "0" if not isinstance(order, ListOrder): raise StorageArgumentException("order must be a ListOrder value") if not isinstance(page_token, str): raise StorageArgumentException("page_token must be a string.") # First get visits (plus one so we can use it as the next page token if any) visits_page = self.origin_visit_get( origin=origin, page_token=page_token, order=order, limit=limit, db=db, cur=cur, ) visits = visits_page.results next_page_token = visits_page.next_page_token if visits: visit_from = min(visits[0].visit, visits[-1].visit) visit_to = max(visits[0].visit, visits[-1].visit) # Then, fetch all statuses associated to these visits visit_statuses: Dict[int, List[OriginVisitStatus]] = defaultdict(list) for row in db.origin_visit_status_get_all_in_range( origin, allowed_statuses, require_snapshot, visit_from=visit_from, visit_to=visit_to, cur=cur, ): row_d = dict(zip(db.origin_visit_status_cols, row)) visit_statuses[row_d["visit"]].append(OriginVisitStatus(**row_d)) results = [ OriginVisitWithStatuses(visit=visit, statuses=visit_statuses[visit.visit]) for visit in visits ] return PagedResult(results=results, next_page_token=next_page_token) @db_transaction(statement_timeout=2000) def origin_visit_find_by_date( self, origin: str, visit_date: datetime.datetime, *, db: Db, cur=None ) -> Optional[OriginVisit]: row_d = db.origin_visit_find_by_date(origin, visit_date, cur=cur) if not row_d: return None return OriginVisit( origin=row_d["origin"], visit=row_d["visit"], date=row_d["date"], type=row_d["type"], ) @db_transaction(statement_timeout=500) def origin_visit_get_by( self, origin: str, visit: int, *, db: Db, cur=None ) -> Optional[OriginVisit]: row = db.origin_visit_get(origin, visit, cur) if row: row_d = dict(zip(db.origin_visit_get_cols, row)) return OriginVisit( origin=row_d["origin"], visit=row_d["visit"], date=row_d["date"], type=row_d["type"], ) return None @db_transaction(statement_timeout=4000) def origin_visit_get_latest( self, origin: str, type: Optional[str] = None, allowed_statuses: Optional[List[str]] = None, require_snapshot: bool = False, *, db: Db, cur=None, ) -> Optional[OriginVisit]: if allowed_statuses and not set(allowed_statuses).intersection(VISIT_STATUSES): raise StorageArgumentException( f"Unknown allowed statuses {','.join(allowed_statuses)}, only " f"{','.join(VISIT_STATUSES)} authorized" ) row = db.origin_visit_get_latest( origin, type=type, allowed_statuses=allowed_statuses, require_snapshot=require_snapshot, cur=cur, ) if row: row_d = dict(zip(db.origin_visit_get_cols, row)) visit = OriginVisit( origin=row_d["origin"], visit=row_d["visit"], date=row_d["date"], type=row_d["type"], ) return visit return None @db_transaction(statement_timeout=500) def origin_visit_status_get( self, origin: str, visit: int, page_token: Optional[str] = None, order: ListOrder = ListOrder.ASC, limit: int = 10, *, db: Db, cur=None, ) -> PagedResult[OriginVisitStatus]: next_page_token = None date_from = None if page_token is not None: date_from = datetime.datetime.fromisoformat(page_token) visit_statuses: List[OriginVisitStatus] = [] # Take one more visit status so we can reuse it as the next page token if any for row in db.origin_visit_status_get_range( origin, visit, date_from=date_from, order=order, limit=limit + 1, cur=cur, ): row_d = dict(zip(db.origin_visit_status_cols, row)) visit_statuses.append(OriginVisitStatus(**row_d)) if len(visit_statuses) > limit: # last visit status date is the next page token next_page_token = str(visit_statuses[-1].date) # excluding that visit status from the result to respect the limit size visit_statuses = visit_statuses[:limit] return PagedResult(results=visit_statuses, next_page_token=next_page_token) @db_transaction() def origin_visit_status_get_random( self, type: str, *, db: Db, cur=None ) -> Optional[OriginVisitStatus]: row = db.origin_visit_get_random(type, cur) if row is not None: row_d = dict(zip(db.origin_visit_status_cols, row)) return OriginVisitStatus(**row_d) return None @db_transaction(statement_timeout=2000) def object_find_by_sha1_git( self, ids: List[Sha1Git], *, db: Db, cur=None ) -> Dict[Sha1Git, List[Dict]]: ret: Dict[Sha1Git, List[Dict]] = {id: [] for id in ids} for retval in db.object_find_by_sha1_git(ids, cur=cur): if retval[1]: ret[retval[0]].append( dict(zip(db.object_find_by_sha1_git_cols, retval)) ) return ret @db_transaction(statement_timeout=1000) def origin_get( self, origins: List[str], *, db: Db, cur=None ) -> Iterable[Optional[Origin]]: rows = db.origin_get_by_url(origins, cur) result: List[Optional[Origin]] = [] for row in rows: origin_d = dict(zip(db.origin_cols, row)) url = origin_d["url"] result.append(None if url is None else Origin(url=url)) return result @db_transaction(statement_timeout=1000) def origin_get_by_sha1( self, sha1s: List[bytes], *, db: Db, cur=None ) -> List[Optional[Dict[str, Any]]]: return [ dict(zip(db.origin_cols, row)) if row[0] else None for row in db.origin_get_by_sha1(sha1s, cur) ] @db_transaction_generator() def origin_get_range(self, origin_from=1, origin_count=100, *, db: Db, cur=None): for origin in db.origin_get_range(origin_from, origin_count, cur): yield dict(zip(db.origin_get_range_cols, origin)) @db_transaction() def origin_list( self, page_token: Optional[str] = None, limit: int = 100, *, db: Db, cur=None ) -> PagedResult[Origin]: page_token = page_token or "0" if not isinstance(page_token, str): raise StorageArgumentException("page_token must be a string.") origin_from = int(page_token) next_page_token = None origins: List[Origin] = [] # Take one more origin so we can reuse it as the next page token if any for row_d in self.origin_get_range(origin_from, limit + 1, db=db, cur=cur): origins.append(Origin(url=row_d["url"])) # keep the last_id for the pagination if needed last_id = row_d["id"] if len(origins) > limit: # data left for subsequent call # last origin id is the next page token next_page_token = str(last_id) # excluding that origin from the result to respect the limit size origins = origins[:limit] assert len(origins) <= limit return PagedResult(results=origins, next_page_token=next_page_token) @db_transaction() def origin_search( self, url_pattern: str, page_token: Optional[str] = None, limit: int = 50, regexp: bool = False, with_visit: bool = False, visit_types: Optional[List[str]] = None, *, db: Db, cur=None, ) -> PagedResult[Origin]: next_page_token = None offset = int(page_token) if page_token else 0 origins = [] # Take one more origin so we can reuse it as the next page token if any for origin in db.origin_search( url_pattern, offset, limit + 1, regexp, with_visit, visit_types, cur ): row_d = dict(zip(db.origin_cols, origin)) origins.append(Origin(url=row_d["url"])) if len(origins) > limit: # next offset next_page_token = str(offset + limit) # excluding that origin from the result to respect the limit size origins = origins[:limit] assert len(origins) <= limit return PagedResult(results=origins, next_page_token=next_page_token) @db_transaction() def origin_count( self, url_pattern: str, regexp: bool = False, with_visit: bool = False, *, db: Db, cur=None, ) -> int: return db.origin_count(url_pattern, regexp, with_visit, cur) @db_transaction() def origin_snapshot_get_all( self, origin_url: str, *, db: Db, cur=None ) -> List[Sha1Git]: return list(db.origin_snapshot_get_all(origin_url, cur)) @db_transaction() def origin_add(self, origins: List[Origin], *, db: Db, cur=None) -> Dict[str, int]: urls = [o.url for o in origins] known_origins = set(url for (url,) in db.origin_get_by_url(urls, cur)) # keep only one occurrence of each given origin while keeping the list # sorted as originally given to_add = sorted(set(urls) - known_origins, key=urls.index) self.journal_writer.origin_add([Origin(url=url) for url in to_add]) added = 0 for url in to_add: if db.origin_add(url, cur): added += 1 return {"origin:add": added} @db_transaction(statement_timeout=500) def stat_counters(self, *, db: Db, cur=None): return {k: v for (k, v) in db.stat_counters()} @db_transaction() def refresh_stat_counters(self, *, db: Db, cur=None): keys = [ "content", "directory", "directory_entry_dir", "directory_entry_file", "directory_entry_rev", "origin", "origin_visit", "person", "release", "revision", "revision_history", "skipped_content", "snapshot", ] for key in keys: cur.execute("select * from swh_update_counter(%s)", (key,)) @db_transaction() def raw_extrinsic_metadata_add( self, metadata: List[RawExtrinsicMetadata], db, cur, ) -> Dict[str, int]: metadata = list(metadata) self.journal_writer.raw_extrinsic_metadata_add(metadata) counter = Counter[ExtendedObjectType]() for metadata_entry in metadata: authority_id = self._get_authority_id(metadata_entry.authority, db, cur) fetcher_id = self._get_fetcher_id(metadata_entry.fetcher, db, cur) db.raw_extrinsic_metadata_add( id=metadata_entry.id, type=metadata_entry.target.object_type.name.lower(), target=str(metadata_entry.target), discovery_date=metadata_entry.discovery_date, authority_id=authority_id, fetcher_id=fetcher_id, format=metadata_entry.format, metadata=metadata_entry.metadata, origin=metadata_entry.origin, visit=metadata_entry.visit, snapshot=map_optional(str, metadata_entry.snapshot), release=map_optional(str, metadata_entry.release), revision=map_optional(str, metadata_entry.revision), path=metadata_entry.path, directory=map_optional(str, metadata_entry.directory), cur=cur, ) counter[metadata_entry.target.object_type] += 1 return { f"{type.value}_metadata:add": count for (type, count) in counter.items() } @db_transaction() def raw_extrinsic_metadata_get( self, target: ExtendedSWHID, authority: MetadataAuthority, after: Optional[datetime.datetime] = None, page_token: Optional[bytes] = None, limit: int = 1000, *, db: Db, cur=None, ) -> PagedResult[RawExtrinsicMetadata]: if page_token: (after_time, after_fetcher) = msgpack_loads(base64.b64decode(page_token)) if after and after_time < after: raise StorageArgumentException( "page_token is inconsistent with the value of 'after'." ) else: after_time = after after_fetcher = None authority_id = self._get_authority_id(authority, db, cur) if not authority_id: return PagedResult( next_page_token=None, results=[], ) rows = db.raw_extrinsic_metadata_get( str(target), authority_id, after_time, after_fetcher, limit + 1, cur, ) rows = [dict(zip(db.raw_extrinsic_metadata_get_cols, row)) for row in rows] results = [] for row in rows: assert str(target) == row["raw_extrinsic_metadata.target"] results.append(converters.db_to_raw_extrinsic_metadata(row)) if len(results) > limit: results.pop() assert len(results) == limit last_returned_row = rows[-2] # rows[-1] corresponds to the popped result next_page_token: Optional[str] = base64.b64encode( msgpack_dumps( ( last_returned_row["discovery_date"], last_returned_row["metadata_fetcher.id"], ) ) ).decode() else: next_page_token = None return PagedResult( next_page_token=next_page_token, results=results, ) @db_transaction() def raw_extrinsic_metadata_get_by_ids( self, ids: List[Sha1Git], *, db: Db, cur=None, ) -> List[RawExtrinsicMetadata]: return [ converters.db_to_raw_extrinsic_metadata( dict(zip(db.raw_extrinsic_metadata_get_cols, row)) ) for row in db.raw_extrinsic_metadata_get_by_ids(ids) ] @db_transaction() def raw_extrinsic_metadata_get_authorities( self, target: ExtendedSWHID, *, db: Db, cur=None, ) -> List[MetadataAuthority]: return [ MetadataAuthority( type=MetadataAuthorityType(authority_type), url=authority_url ) for ( authority_type, authority_url, ) in db.raw_extrinsic_metadata_get_authorities(str(target), cur) ] @db_transaction() def metadata_fetcher_add( self, fetchers: List[MetadataFetcher], *, db: Db, cur=None ) -> Dict[str, int]: fetchers = list(fetchers) self.journal_writer.metadata_fetcher_add(fetchers) count = 0 for fetcher in fetchers: db.metadata_fetcher_add(fetcher.name, fetcher.version, cur=cur) count += 1 return {"metadata_fetcher:add": count} @db_transaction(statement_timeout=500) def metadata_fetcher_get( self, name: str, version: str, *, db: Db, cur=None ) -> Optional[MetadataFetcher]: row = db.metadata_fetcher_get(name, version, cur=cur) if not row: return None return MetadataFetcher.from_dict(dict(zip(db.metadata_fetcher_cols, row))) @db_transaction() def metadata_authority_add( self, authorities: List[MetadataAuthority], *, db: Db, cur=None ) -> Dict[str, int]: authorities = list(authorities) self.journal_writer.metadata_authority_add(authorities) count = 0 for authority in authorities: db.metadata_authority_add(authority.type.value, authority.url, cur=cur) count += 1 return {"metadata_authority:add": count} @db_transaction() def metadata_authority_get( self, type: MetadataAuthorityType, url: str, *, db: Db, cur=None ) -> Optional[MetadataAuthority]: row = db.metadata_authority_get(type.value, url, cur=cur) if not row: return None return MetadataAuthority.from_dict(dict(zip(db.metadata_authority_cols, row))) def clear_buffers(self, object_types: Sequence[str] = ()) -> None: """Do nothing""" return None def flush(self, object_types: Sequence[str] = ()) -> Dict[str, int]: return {} def _get_authority_id(self, authority: MetadataAuthority, db, cur): authority_id = db.metadata_authority_get_id( authority.type.value, authority.url, cur ) if not authority_id: raise StorageArgumentException(f"Unknown authority {authority}") return authority_id def _get_fetcher_id(self, fetcher: MetadataFetcher, db, cur): fetcher_id = db.metadata_fetcher_get_id(fetcher.name, fetcher.version, cur) if not fetcher_id: raise StorageArgumentException(f"Unknown fetcher {fetcher}") return fetcher_id diff --git a/swh/storage/sql/40-funcs.sql b/swh/storage/sql/40-funcs.sql index f87caaae..626f0b3d 100644 --- a/swh/storage/sql/40-funcs.sql +++ b/swh/storage/sql/40-funcs.sql @@ -1,1024 +1,1031 @@ 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_ 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 $$ begin perform swh_directory_entry_add('file'); perform swh_directory_entry_add('dir'); perform swh_directory_entry_add('rev'); insert into directory (id, dir_entries, file_entries, rev_entries, raw_manifest) select id, dir_entries, file_entries, rev_entries, raw_manifest from tmp_directory t where not exists ( select 1 from directory d where d.id = t.id); return; end $$; -- 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 $$ with dir as ( select id as dir_id, dir_entries, file_entries, rev_entries from directory where id = walked_dir_id), ls_d as (select dir_id, unnest(dir_entries) as entry_id from dir), ls_f as (select dir_id, unnest(file_entries) as entry_id from dir), ls_r as (select dir_id, unnest(rev_entries) as entry_id from dir) (select dir_id, 'dir'::directory_entry_type as type, e.target, e.name, e.perms, NULL::content_status, NULL::sha1, NULL::sha1_git, NULL::sha256, NULL::bigint from ls_d left join directory_entry_dir e on ls_d.entry_id = e.id) union (with known_contents as (select dir_id, 'file'::directory_entry_type as type, e.target, e.name, e.perms, c.status, c.sha1, c.sha1_git, c.sha256, c.length from ls_f left join directory_entry_file e on ls_f.entry_id = e.id inner join content c on e.target = c.sha1_git) select * from known_contents union (select dir_id, 'file'::directory_entry_type as type, e.target, e.name, e.perms, c.status, c.sha1, c.sha1_git, c.sha256, c.length from ls_f left join directory_entry_file e on ls_f.entry_id = e.id left join skipped_content c on e.target = c.sha1_git where not exists (select 1 from known_contents where known_contents.sha1_git=e.target))) union (select dir_id, 'rev'::directory_entry_type as type, e.target, e.name, e.perms, NULL::content_status, NULL::sha1, NULL::sha1_git, NULL::sha256, NULL::bigint from ls_r left join directory_entry_rev e on ls_r.entry_id = e.id) order by name; $$; -- 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 $$ declare end_index integer; paths bytea default ''; path bytea; res bytea[]; r record; begin end_index := array_upper(dir_or_content_path, 1); res[1] := walked_dir_id; for i in 1..end_index loop path := dir_or_content_path[i]; -- concatenate path for patching the name in the result record (if we found it) if i = 1 then paths = path; else paths := paths || '/' || path; -- concatenate paths end if; if i <> end_index then select * from swh_directory_walk_one(res[i] :: sha1_git) where name=path and type = 'dir' limit 1 into r; else select * from swh_directory_walk_one(res[i] :: sha1_git) where name=path limit 1 into r; end if; -- find the path if r is null then return null; else -- store the next dir to lookup the next local path from res[i+1] := r.target; end if; end loop; -- at this moment, r is the result. Patch its 'name' with the full path before returning it. r.name := paths; return r; end $$; -- Returns the entries in a directory, without joining with their target tables create or replace function swh_directory_get_entries(dir_id sha1_git) returns table ( dir_id directory_entry_type, target sha1_git, name unix_path, perms file_perms ) language sql stable as $$ with dir as ( select id as dir_id, dir_entries, file_entries, rev_entries from directory where id = dir_id), ls_d as (select dir_id, unnest(dir_entries) as entry_id from dir), ls_f as (select dir_id, unnest(file_entries) as entry_id from dir), ls_r as (select dir_id, unnest(rev_entries) as entry_id from dir) (select 'dir'::directory_entry_type, e.target, e.name, e.perms from ls_d left join directory_entry_dir e on ls_d.entry_id = e.id) union (select 'file'::directory_entry_type, e.target, e.name, e.perms from ls_f left join directory_entry_file e on ls_f.entry_id = e.id) union (select 'rev'::directory_entry_type, e.target, e.name, e.perms from ls_r left join directory_entry_rev e on ls_r.entry_id = e.id) $$; -- 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, date_offset_bytes bytea, committer_date timestamptz, committer_date_offset smallint, committer_date_neg_utc_offset boolean, committer_date_offset_bytes bytea, 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, extra_headers bytea[][], raw_manifest bytea ); create or replace function swh_revision_log(root_revisions bytea[], num_revs bigint default NULL, ignore_displayname boolean default false) returns setof revision_entry language sql stable as $$ -- when name and email are null, swh.storage.postgresql.converters.db_to_author() -- parses fullname to populate them, so we can just drop them here select t.id, r.date, r.date_offset, r.date_neg_utc_offset, r.date_offset_bytes, r.committer_date, r.committer_date_offset, r.committer_date_neg_utc_offset, r.committer_date_offset_bytes, r.type, r.directory, r.message, a.id, case when ignore_displayname or a.displayname is null then a.fullname else a.displayname end, case when ignore_displayname or a.displayname is null then a.name else null end, case when ignore_displayname or a.displayname is null then a.email else null end, c.id, case when ignore_displayname or c.displayname is null then c.fullname else c.displayname end, case when ignore_displayname or c.displayname is null then c.name else null end, case when ignore_displayname or c.displayname is null then c.email else null end, r.metadata, r.synthetic, t.parents, r.object_id, r.extra_headers, r.raw_manifest from swh_revision_list(root_revisions, num_revs) as t left join revision r on t.id = r.id left join person a on a.id = r.author left join person c on c.id = r.committer; $$; comment on function swh_revision_log(root_revisions bytea[], num_revs bigint, ignore_displayname boolean) is '"git style" revision log. Similar to swh_revision_list(), but returning ' 'all information associated to each revision, and expanding authors/committers'; -- 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 $$ begin with t as ( select author_fullname as fullname, author_name as name, author_email as email from tmp_revision union select committer_fullname as fullname, committer_name as name, committer_email as email from tmp_revision ) insert into person (fullname, name, email) select distinct on (fullname) fullname, name, email from t where t.fullname is not null and not exists ( select 1 from person p where t.fullname = p.fullname ); return; end $$; -- Create entries in revision from tmp_revision create or replace function swh_revision_add() returns void language plpgsql as $$ begin perform swh_person_add_from_revision(); insert into revision (id, date, date_offset, date_neg_utc_offset, date_offset_bytes, committer_date, committer_date_offset, committer_date_neg_utc_offset, committer_date_offset_bytes, type, directory, message, author, committer, metadata, synthetic, extra_headers, raw_manifest) select t.id, t.date, t.date_offset, t.date_neg_utc_offset, t.date_offset_bytes, t.committer_date, t.committer_date_offset, t.committer_date_neg_utc_offset, t.committer_date_offset_bytes, t.type, t.directory, t.message, a.id, c.id, t.metadata, t.synthetic, t.extra_headers, t.raw_manifest from tmp_revision t left join person a on a.fullname = t.author_fullname left join person c on c.fullname = t.committer_fullname; return; end $$; -- Create entries in extid from tmp_extid -- operates in bulk: 0. swh_mktemp(extid), 1. COPY to tmp_extid, -- 2. call this function create or replace function swh_extid_add() returns void language plpgsql as $$ begin insert into extid (extid_type, extid, extid_version, target_type, target) select distinct t.extid_type, t.extid, t.extid_version, t.target_type, t.target from tmp_extid t on conflict do nothing; return; end $$; -- 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 $$ begin perform swh_person_add_from_release(); insert into release (id, target, target_type, date, date_offset, date_neg_utc_offset, date_offset_bytes, name, comment, author, synthetic, raw_manifest) select distinct t.id, t.target, t.target_type, t.date, t.date_offset, t.date_neg_utc_offset, t.date_offset_bytes, t.name, t.comment, a.id, t.synthetic, t.raw_manifest from tmp_release t left join person a on a.fullname = t.author_fullname where not exists (select 1 from release where t.id = release.id); return; end $$; -- 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 $$ with origin_id as ( select id from origin where url = origin_url ), last_known_visit as ( - select coalesce(max(visit), 0) as visit - from origin_visit - where origin = (select id from origin_id) + select coalesce( + ( + select visit + from origin_visit + where origin = (select id from origin_id) + order by visit desc + limit 1 + for update + ), + 0) as visit ) insert into origin_visit (origin, date, type, visit) values ((select id from origin_id), date, type, (select visit from last_known_visit) + 1) returning visit; $$; 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, branch_name_include_substring bytea default NULL, branch_name_exclude_prefix bytea default NULL) returns setof snapshot_result language sql stable as $$ -- with small limits, the "naive" version of this query can degenerate into -- using the deduplication index on snapshot_branch (name, target, -- target_type); The planner happily scans several hundred million rows. -- Do the query in two steps: first pull the relevant branches for the given -- snapshot (filtering them by type), then do the limiting. This two-step -- process guides the planner into using the proper index. with filtered_snapshot_branches as ( select swh_snapshot_get_by_id.id as snapshot_id, name, target, target_type from snapshot_branches inner join snapshot_branch on snapshot_branches.branch_id = snapshot_branch.object_id where snapshot_id = (select object_id from snapshot where snapshot.id = swh_snapshot_get_by_id.id) and (target_types is null or target_type = any(target_types)) order by name ) select snapshot_id, name, target, target_type from filtered_snapshot_branches where name >= branches_from and (branch_name_include_substring is null or name like '%'||branch_name_include_substring||'%') and (branch_name_exclude_prefix is null or name not like branch_name_exclude_prefix||'%') order by name limit branches_count; $$; create type snapshot_size as ( target_type snapshot_target, count bigint ); create or replace function swh_snapshot_count_branches(id sha1_git, branch_name_exclude_prefix bytea default NULL) returns setof snapshot_size language sql stable as $$ SELECT target_type, count(name) from swh_snapshot_get_by_id(swh_snapshot_count_branches.id, branch_name_exclude_prefix => swh_snapshot_count_branches.branch_name_exclude_prefix) group by target_type; $$; -- 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 $$ with revs as ( select * from revision where object_id > min_excl and object_id <= max_incl ) select r.id, r.date, r.date_offset, r.date_neg_utc_offset, r.date_offset_bytes, r.committer_date, r.committer_date_offset, r.committer_date_neg_utc_offset, r.committer_date_offset_bytes, r.type, r.directory, r.message, a.id, a.fullname, a.name, a.email, c.id, c.fullname, c.name, c.email, r.metadata, r.synthetic, array(select rh.parent_id::bytea from revision_history rh where rh.id = r.id order by rh.parent_rank) as parents, r.object_id, r.extra_headers, r.raw_manifest from revs r left join person a on a.id = r.author left join person c on c.id = r.committer order by r.object_id; $$; 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/sql/upgrades/184.sql b/swh/storage/sql/upgrades/184.sql new file mode 100644 index 00000000..42150172 --- /dev/null +++ b/swh/storage/sql/upgrades/184.sql @@ -0,0 +1,33 @@ +-- SWH DB schema upgrade +-- from_version: 183 +-- to_version: 184 +-- description: origin_visit_add: Fix crash when adding multiple visits to the same origin simultaneously + +insert into dbversion(version, release, description) + values(184, now(), 'Work In Progress'); + +create or replace function swh_origin_visit_add(origin_url text, date timestamptz, type text) + returns bigint + language sql +as $$ + with origin_id as ( + select id + from origin + where url = origin_url + ), last_known_visit as ( + select coalesce( + ( + select visit + from origin_visit + where origin = (select id from origin_id) + order by visit desc + limit 1 + for update + ), + 0) as visit + ) + insert into origin_visit (origin, date, type, visit) + values ((select id from origin_id), date, type, + (select visit from last_known_visit) + 1) + returning visit; +$$;