diff --git a/swh/storage/cassandra/cql.py b/swh/storage/cassandra/cql.py
index 05416d6e..99f9ccf1 100644
--- a/swh/storage/cassandra/cql.py
+++ b/swh/storage/cassandra/cql.py
@@ -1,1162 +1,1194 @@
 # 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
 
 from collections import Counter
 import dataclasses
 import datetime
 import functools
 import logging
 import random
 from typing import (
     Any,
     Callable,
     Dict,
     Iterable,
     Iterator,
     List,
     Optional,
     Tuple,
     Type,
     TypeVar,
     Union,
 )
 
 from cassandra import CoordinationFailure
 from cassandra.cluster import EXEC_PROFILE_DEFAULT, Cluster, ExecutionProfile, ResultSet
 from cassandra.policies import DCAwareRoundRobinPolicy, TokenAwarePolicy
 from cassandra.query import BoundStatement, PreparedStatement, dict_factory
 from mypy_extensions import NamedArg
 from tenacity import (
     retry,
     retry_if_exception_type,
     stop_after_attempt,
     wait_random_exponential,
 )
 
 from swh.model.identifiers import CoreSWHID
 from swh.model.model import (
     Content,
     Person,
     Sha1Git,
     SkippedContent,
     Timestamp,
     TimestampWithTimezone,
 )
 from swh.storage.interface import ListOrder
 
 from ..utils import remove_keys
 from .common import TOKEN_BEGIN, TOKEN_END, hash_url
 from .model import (
     MAGIC_NULL_PK,
     BaseRow,
     ContentRow,
     DirectoryEntryRow,
     DirectoryRow,
+    ExtIDByTargetRow,
     ExtIDRow,
     MetadataAuthorityRow,
     MetadataFetcherRow,
     ObjectCountRow,
     OriginRow,
     OriginVisitRow,
     OriginVisitStatusRow,
     RawExtrinsicMetadataRow,
     ReleaseRow,
     RevisionParentRow,
     RevisionRow,
     SkippedContentRow,
     SnapshotBranchRow,
     SnapshotRow,
+    content_index_table_name,
 )
 from .schema import CREATE_TABLES_QUERIES, HASH_ALGORITHMS
 
 logger = logging.getLogger(__name__)
 
 
 _execution_profiles = {
     EXEC_PROFILE_DEFAULT: ExecutionProfile(
         load_balancing_policy=TokenAwarePolicy(DCAwareRoundRobinPolicy()),
         row_factory=dict_factory,
     ),
 }
 # Configuration for cassandra-driver's access to servers:
 # * hit the right server directly when sending a query (TokenAwarePolicy),
 # * if there's more than one, then pick one at random that's in the same
 #   datacenter as the client (DCAwareRoundRobinPolicy)
 
 
 def create_keyspace(
     hosts: List[str], keyspace: str, port: int = 9042, *, durable_writes=True
 ):
     cluster = Cluster(hosts, port=port, execution_profiles=_execution_profiles)
     session = cluster.connect()
     extra_params = ""
     if not durable_writes:
         extra_params = "AND durable_writes = false"
     session.execute(
         """CREATE KEYSPACE IF NOT EXISTS "%s"
                        WITH REPLICATION = {
                            'class' : 'SimpleStrategy',
                            'replication_factor' : 1
                        } %s;
                     """
         % (keyspace, extra_params)
     )
     session.execute('USE "%s"' % keyspace)
     for query in CREATE_TABLES_QUERIES:
         session.execute(query)
 
 
 TRet = TypeVar("TRet")
 
 
 def _prepared_statement(
     query: str,
 ) -> Callable[[Callable[..., TRet]], Callable[..., TRet]]:
     """Returns a decorator usable on methods of CqlRunner, to
     inject them with a 'statement' argument, that is a prepared
     statement corresponding to the query.
 
     This only works on methods of CqlRunner, as preparing a
     statement requires a connection to a Cassandra server."""
 
     def decorator(f):
         @functools.wraps(f)
         def newf(self, *args, **kwargs) -> TRet:
             if f.__name__ not in self._prepared_statements:
                 statement: PreparedStatement = self._session.prepare(query)
                 self._prepared_statements[f.__name__] = statement
             return f(
                 self, *args, **kwargs, statement=self._prepared_statements[f.__name__]
             )
 
         return newf
 
     return decorator
 
 
 TArg = TypeVar("TArg")
 TSelf = TypeVar("TSelf")
 
 
 def _prepared_insert_statement(
     row_class: Type[BaseRow],
 ) -> Callable[
     [Callable[[TSelf, TArg, NamedArg(Any, "statement")], TRet]],  # noqa
     Callable[[TSelf, TArg], TRet],
 ]:
     """Shorthand for using `_prepared_statement` for `INSERT INTO`
     statements."""
     columns = row_class.cols()
     return _prepared_statement(
         "INSERT INTO %s (%s) VALUES (%s)"
         % (row_class.TABLE, ", ".join(columns), ", ".join("?" for _ in columns),)
     )
 
 
 def _prepared_exists_statement(
     table_name: str,
 ) -> Callable[
     [Callable[[TSelf, TArg, NamedArg(Any, "statement")], TRet]],  # noqa
     Callable[[TSelf, TArg], TRet],
 ]:
     """Shorthand for using `_prepared_statement` for queries that only
     check which ids in a list exist in the table."""
     return _prepared_statement(f"SELECT id FROM {table_name} WHERE id IN ?")
 
 
 def _prepared_select_statement(
     row_class: Type[BaseRow], clauses: str = "", cols: Optional[List[str]] = None,
 ) -> Callable[[Callable[..., TRet]], Callable[..., TRet]]:
     if cols is None:
         cols = row_class.cols()
 
     return _prepared_statement(
         f"SELECT {', '.join(cols)} FROM {row_class.TABLE} {clauses}"
     )
 
 
 def _prepared_select_statements(
     row_class: Type[BaseRow], queries: Dict[Any, str],
 ) -> Callable[[Callable[..., TRet]], Callable[..., TRet]]:
     """Like _prepared_statement, but supports multiple statements, passed a dict,
     and passes a dict of prepared statements to the decorated method"""
     cols = row_class.cols()
 
     statement_start = f"SELECT {', '.join(cols)} FROM {row_class.TABLE} "
 
     def decorator(f):
         @functools.wraps(f)
         def newf(self, *args, **kwargs) -> TRet:
             if f.__name__ not in self._prepared_statements:
                 self._prepared_statements[f.__name__] = {
                     key: self._session.prepare(statement_start + query)
                     for (key, query) in queries.items()
                 }
             return f(
                 self, *args, **kwargs, statements=self._prepared_statements[f.__name__]
             )
 
         return newf
 
     return decorator
 
 
 def _next_bytes_value(value: bytes) -> bytes:
     """Returns the next bytes value by incrementing the integer
     representation of the provided value and converting it back
     to bytes.
 
     For instance when prefix is b"abcd", it returns b"abce".
     """
     next_value_int = int.from_bytes(value, byteorder="big") + 1
     return next_value_int.to_bytes(
         (next_value_int.bit_length() + 7) // 8, byteorder="big"
     )
 
 
 class CqlRunner:
     """Class managing prepared statements and building queries to be sent
     to Cassandra."""
 
     def __init__(self, hosts: List[str], keyspace: str, port: int):
         self._cluster = Cluster(
             hosts, port=port, execution_profiles=_execution_profiles
         )
         self._session = self._cluster.connect(keyspace)
         self._cluster.register_user_type(
             keyspace, "microtimestamp_with_timezone", TimestampWithTimezone
         )
         self._cluster.register_user_type(keyspace, "microtimestamp", Timestamp)
         self._cluster.register_user_type(keyspace, "person", Person)
 
         # directly a PreparedStatement for methods decorated with
         # @_prepared_statements (and its wrappers, _prepared_insert_statement,
         # _prepared_exists_statement, and _prepared_select_statement);
         # and a dict of PreparedStatements with @_prepared_select_statements
         self._prepared_statements: Dict[
             str, Union[PreparedStatement, Dict[Any, PreparedStatement]]
         ] = {}
 
     ##########################
     # Common utility functions
     ##########################
 
     MAX_RETRIES = 3
 
     @retry(
         wait=wait_random_exponential(multiplier=1, max=10),
         stop=stop_after_attempt(MAX_RETRIES),
         retry=retry_if_exception_type(CoordinationFailure),
     )
     def _execute_with_retries(self, statement, args) -> ResultSet:
         return self._session.execute(statement, args, timeout=1000.0)
 
     @_prepared_statement(
         "UPDATE object_count SET count = count + ? "
         "WHERE partition_key = 0 AND object_type = ?"
     )
     def _increment_counter(
         self, object_type: str, nb: int, *, statement: PreparedStatement
     ) -> None:
         self._execute_with_retries(statement, [nb, object_type])
 
     def _add_one(self, statement, obj: BaseRow) -> None:
         self._increment_counter(obj.TABLE, 1)
         self._execute_with_retries(statement, dataclasses.astuple(obj))
 
     _T = TypeVar("_T", bound=BaseRow)
 
     def _get_random_row(self, row_class: Type[_T], statement) -> Optional[_T]:  # noqa
         """Takes a prepared statement of the form
         "SELECT * FROM <table> WHERE token(<keys>) > ? LIMIT 1"
         and uses it to return a random row"""
         token = random.randint(TOKEN_BEGIN, TOKEN_END)
         rows = self._execute_with_retries(statement, [token])
         if not rows:
             # There are no row with a greater token; wrap around to get
             # the row with the smallest token
             rows = self._execute_with_retries(statement, [TOKEN_BEGIN])
         if rows:
             return row_class.from_dict(rows.one())  # type: ignore
         else:
             return None
 
     def _missing(self, statement, ids):
         rows = self._execute_with_retries(statement, [ids])
         found_ids = {row["id"] for row in rows}
         return [id_ for id_ in ids if id_ not in found_ids]
 
     ##########################
     # 'content' table
     ##########################
 
     def _content_add_finalize(self, statement: BoundStatement) -> None:
         """Returned currified by content_add_prepare, to be called when the
         content row should be added to the primary table."""
         self._execute_with_retries(statement, None)
         self._increment_counter("content", 1)
 
     @_prepared_insert_statement(ContentRow)
     def content_add_prepare(
         self, content: ContentRow, *, statement
     ) -> Tuple[int, Callable[[], None]]:
         """Prepares insertion of a Content to the main 'content' table.
         Returns a token (to be used in secondary tables), and a function to be
         called to perform the insertion in the main table."""
         statement = statement.bind(dataclasses.astuple(content))
 
         # Type used for hashing keys (usually, it will be
         # cassandra.metadata.Murmur3Token)
         token_class = self._cluster.metadata.token_map.token_class
 
         # Token of the row when it will be inserted. This is equivalent to
         # "SELECT token({', '.join(ContentRow.PARTITION_KEY)}) FROM content WHERE ..."
         # after the row is inserted; but we need the token to insert in the
         # index tables *before* inserting to the main 'content' table
         token = token_class.from_key(statement.routing_key).value
         assert TOKEN_BEGIN <= token <= TOKEN_END
 
         # Function to be called after the indexes contain their respective
         # row
         finalizer = functools.partial(self._content_add_finalize, statement)
 
         return (token, finalizer)
 
     @_prepared_select_statement(
         ContentRow, f"WHERE {' AND '.join(map('%s = ?'.__mod__, HASH_ALGORITHMS))}"
     )
     def content_get_from_pk(
         self, content_hashes: Dict[str, bytes], *, statement
     ) -> Optional[ContentRow]:
         rows = list(
             self._execute_with_retries(
                 statement, [content_hashes[algo] for algo in HASH_ALGORITHMS]
             )
         )
         assert len(rows) <= 1
         if rows:
             return ContentRow(**rows[0])
         else:
             return None
 
     @_prepared_select_statement(
         ContentRow, f"WHERE token({', '.join(ContentRow.PARTITION_KEY)}) = ?"
     )
     def content_get_from_token(self, token, *, statement) -> Iterable[ContentRow]:
         return map(ContentRow.from_dict, self._execute_with_retries(statement, [token]))
 
     @_prepared_select_statement(
         ContentRow, f"WHERE token({', '.join(ContentRow.PARTITION_KEY)}) > ? LIMIT 1"
     )
     def content_get_random(self, *, statement) -> Optional[ContentRow]:
         return self._get_random_row(ContentRow, statement)
 
     @_prepared_statement(
-        (
-            "SELECT token({0}) AS tok, {1} FROM content "
-            "WHERE token({0}) >= ? AND token({0}) <= ? LIMIT ?"
-        ).format(", ".join(ContentRow.PARTITION_KEY), ", ".join(ContentRow.cols()))
+        """
+        SELECT token({pk}) AS tok, {cols} FROM {table}
+        WHERE token({pk}) >= ? AND token({pk}) <= ? LIMIT ?
+        """.format(
+            pk=", ".join(ContentRow.PARTITION_KEY),
+            cols=", ".join(ContentRow.cols()),
+            table=ContentRow.TABLE,
+        )
     )
     def content_get_token_range(
         self, start: int, end: int, limit: int, *, statement
     ) -> Iterable[Tuple[int, ContentRow]]:
         """Returns an iterable of (token, row)"""
         return (
             (row["tok"], ContentRow.from_dict(remove_keys(row, ("tok",))))
             for row in self._execute_with_retries(statement, [start, end, limit])
         )
 
     ##########################
     # 'content_by_*' tables
     ##########################
 
     @_prepared_statement(
-        "SELECT sha1_git AS id FROM content_by_sha1_git WHERE sha1_git IN ?"
+        f"""
+        SELECT sha1_git AS id
+        FROM {content_index_table_name("sha1_git", skipped_content=False)}
+        WHERE sha1_git IN ?
+        """
     )
     def content_missing_by_sha1_git(
         self, ids: List[bytes], *, statement
     ) -> List[bytes]:
         return self._missing(statement, ids)
 
     def content_index_add_one(self, algo: str, content: Content, token: int) -> None:
         """Adds a row mapping content[algo] to the token of the Content in
         the main 'content' table."""
-        query = (
-            f"INSERT INTO content_by_{algo} ({algo}, target_token) " f"VALUES (%s, %s)"
-        )
+        query = f"""
+            INSERT INTO {content_index_table_name(algo, skipped_content=False)}
+            ({algo}, target_token)
+            VALUES (%s, %s)
+        """
         self._execute_with_retries(query, [content.get_hash(algo), token])
 
     def content_get_tokens_from_single_hash(
         self, algo: str, hash_: bytes
     ) -> Iterable[int]:
         assert algo in HASH_ALGORITHMS
-        query = f"SELECT target_token FROM content_by_{algo} WHERE {algo} = %s"
+        query = f"""
+            SELECT target_token
+            FROM {content_index_table_name(algo, skipped_content=False)}
+            WHERE {algo} = %s
+        """
         return (
             row["target_token"] for row in self._execute_with_retries(query, [hash_])
         )
 
     ##########################
     # 'skipped_content' table
     ##########################
 
     def _skipped_content_add_finalize(self, statement: BoundStatement) -> None:
         """Returned currified by skipped_content_add_prepare, to be called
         when the content row should be added to the primary table."""
         self._execute_with_retries(statement, None)
         self._increment_counter("skipped_content", 1)
 
     @_prepared_insert_statement(SkippedContentRow)
     def skipped_content_add_prepare(
         self, content, *, statement
     ) -> Tuple[int, Callable[[], None]]:
         """Prepares insertion of a Content to the main 'skipped_content' table.
         Returns a token (to be used in secondary tables), and a function to be
         called to perform the insertion in the main table."""
 
         # Replace NULLs (which are not allowed in the partition key) with
         # an empty byte string
         for key in SkippedContentRow.PARTITION_KEY:
             if getattr(content, key) is None:
                 setattr(content, key, MAGIC_NULL_PK)
 
         statement = statement.bind(dataclasses.astuple(content))
 
         # Type used for hashing keys (usually, it will be
         # cassandra.metadata.Murmur3Token)
         token_class = self._cluster.metadata.token_map.token_class
 
         # Token of the row when it will be inserted. This is equivalent to
         # "SELECT token({', '.join(SkippedContentRow.PARTITION_KEY)})
         #  FROM skipped_content WHERE ..."
         # after the row is inserted; but we need the token to insert in the
         # index tables *before* inserting to the main 'skipped_content' table
         token = token_class.from_key(statement.routing_key).value
         assert TOKEN_BEGIN <= token <= TOKEN_END
 
         # Function to be called after the indexes contain their respective
         # row
         finalizer = functools.partial(self._skipped_content_add_finalize, statement)
 
         return (token, finalizer)
 
     @_prepared_select_statement(
         SkippedContentRow,
         f"WHERE {' AND '.join(map('%s = ?'.__mod__, HASH_ALGORITHMS))}",
     )
     def skipped_content_get_from_pk(
         self, content_hashes: Dict[str, bytes], *, statement
     ) -> Optional[SkippedContentRow]:
         rows = list(
             self._execute_with_retries(
                 statement,
                 [content_hashes[algo] or MAGIC_NULL_PK for algo in HASH_ALGORITHMS],
             )
         )
         assert len(rows) <= 1
         if rows:
             return SkippedContentRow.from_dict(rows[0])
         else:
             return None
 
     @_prepared_select_statement(
         SkippedContentRow,
         f"WHERE token({', '.join(SkippedContentRow.PARTITION_KEY)}) = ?",
     )
     def skipped_content_get_from_token(
         self, token, *, statement
     ) -> Iterable[SkippedContentRow]:
         return map(
             SkippedContentRow.from_dict, self._execute_with_retries(statement, [token])
         )
 
     ##########################
     # 'skipped_content_by_*' tables
     ##########################
 
     def skipped_content_index_add_one(
         self, algo: str, content: SkippedContent, token: int
     ) -> None:
         """Adds a row mapping content[algo] to the token of the SkippedContent
         in the main 'skipped_content' table."""
         query = (
             f"INSERT INTO skipped_content_by_{algo} ({algo}, target_token) "
             f"VALUES (%s, %s)"
         )
         self._execute_with_retries(
             query, [content.get_hash(algo) or MAGIC_NULL_PK, token]
         )
 
     def skipped_content_get_tokens_from_single_hash(
         self, algo: str, hash_: bytes
     ) -> Iterable[int]:
         assert algo in HASH_ALGORITHMS
-        query = f"SELECT target_token FROM skipped_content_by_{algo} WHERE {algo} = %s"
+        query = f"""
+            SELECT target_token
+            FROM {content_index_table_name(algo, skipped_content=True)}
+            WHERE {algo} = %s
+        """
         return (
             row["target_token"] for row in self._execute_with_retries(query, [hash_])
         )
 
     ##########################
     # 'revision' table
     ##########################
 
     @_prepared_exists_statement("revision")
     def revision_missing(self, ids: List[bytes], *, statement) -> List[bytes]:
         return self._missing(statement, ids)
 
     @_prepared_insert_statement(RevisionRow)
     def revision_add_one(self, revision: RevisionRow, *, statement) -> None:
         self._add_one(statement, revision)
 
-    @_prepared_statement("SELECT id FROM revision WHERE id IN ?")
+    @_prepared_statement(f"SELECT id FROM {RevisionRow.TABLE} WHERE id IN ?")
     def revision_get_ids(self, revision_ids, *, statement) -> Iterable[int]:
         return (
             row["id"] for row in self._execute_with_retries(statement, [revision_ids])
         )
 
     @_prepared_select_statement(RevisionRow, "WHERE id IN ?")
     def revision_get(
         self, revision_ids: List[Sha1Git], *, statement
     ) -> Iterable[RevisionRow]:
         return map(
             RevisionRow.from_dict, self._execute_with_retries(statement, [revision_ids])
         )
 
     @_prepared_select_statement(RevisionRow, "WHERE token(id) > ? LIMIT 1")
     def revision_get_random(self, *, statement) -> Optional[RevisionRow]:
         return self._get_random_row(RevisionRow, statement)
 
     ##########################
     # 'revision_parent' table
     ##########################
 
     @_prepared_insert_statement(RevisionParentRow)
     def revision_parent_add_one(
         self, revision_parent: RevisionParentRow, *, statement
     ) -> None:
         self._add_one(statement, revision_parent)
 
-    @_prepared_statement("SELECT parent_id FROM revision_parent WHERE id = ?")
+    @_prepared_statement(
+        f"SELECT parent_id FROM {RevisionParentRow.TABLE} WHERE id = ?"
+    )
     def revision_parent_get(
         self, revision_id: Sha1Git, *, statement
     ) -> Iterable[bytes]:
         return (
             row["parent_id"]
             for row in self._execute_with_retries(statement, [revision_id])
         )
 
     ##########################
     # 'release' table
     ##########################
 
     @_prepared_exists_statement("release")
     def release_missing(self, ids: List[bytes], *, statement) -> List[bytes]:
         return self._missing(statement, ids)
 
     @_prepared_insert_statement(ReleaseRow)
     def release_add_one(self, release: ReleaseRow, *, statement) -> None:
         self._add_one(statement, release)
 
     @_prepared_select_statement(ReleaseRow, "WHERE id in ?")
     def release_get(self, release_ids: List[str], *, statement) -> Iterable[ReleaseRow]:
         return map(
             ReleaseRow.from_dict, self._execute_with_retries(statement, [release_ids])
         )
 
     @_prepared_select_statement(ReleaseRow, "WHERE token(id) > ? LIMIT 1")
     def release_get_random(self, *, statement) -> Optional[ReleaseRow]:
         return self._get_random_row(ReleaseRow, statement)
 
     ##########################
     # 'directory' table
     ##########################
 
     @_prepared_exists_statement("directory")
     def directory_missing(self, ids: List[bytes], *, statement) -> List[bytes]:
         return self._missing(statement, ids)
 
     @_prepared_insert_statement(DirectoryRow)
     def directory_add_one(self, directory: DirectoryRow, *, statement) -> None:
         """Called after all calls to directory_entry_add_one, to
         commit/finalize the directory."""
         self._add_one(statement, directory)
 
     @_prepared_select_statement(DirectoryRow, "WHERE token(id) > ? LIMIT 1")
     def directory_get_random(self, *, statement) -> Optional[DirectoryRow]:
         return self._get_random_row(DirectoryRow, statement)
 
     ##########################
     # 'directory_entry' table
     ##########################
 
     @_prepared_insert_statement(DirectoryEntryRow)
     def directory_entry_add_one(self, entry: DirectoryEntryRow, *, statement) -> None:
         self._add_one(statement, entry)
 
     @_prepared_select_statement(DirectoryEntryRow, "WHERE directory_id IN ?")
     def directory_entry_get(
         self, directory_ids, *, statement
     ) -> Iterable[DirectoryEntryRow]:
         return map(
             DirectoryEntryRow.from_dict,
             self._execute_with_retries(statement, [directory_ids]),
         )
 
     ##########################
     # 'snapshot' table
     ##########################
 
     @_prepared_exists_statement("snapshot")
     def snapshot_missing(self, ids: List[bytes], *, statement) -> List[bytes]:
         return self._missing(statement, ids)
 
     @_prepared_insert_statement(SnapshotRow)
     def snapshot_add_one(self, snapshot: SnapshotRow, *, statement) -> None:
         self._add_one(statement, snapshot)
 
     @_prepared_select_statement(SnapshotRow, "WHERE token(id) > ? LIMIT 1")
     def snapshot_get_random(self, *, statement) -> Optional[SnapshotRow]:
         return self._get_random_row(SnapshotRow, statement)
 
     ##########################
     # 'snapshot_branch' table
     ##########################
 
     @_prepared_insert_statement(SnapshotBranchRow)
     def snapshot_branch_add_one(self, branch: SnapshotBranchRow, *, statement) -> None:
         self._add_one(statement, branch)
 
     @_prepared_statement(
-        "SELECT ascii_bins_count(target_type) AS counts "
-        "FROM snapshot_branch "
-        "WHERE snapshot_id = ? AND name >= ?"
+        f"""
+        SELECT ascii_bins_count(target_type) AS counts
+        FROM {SnapshotBranchRow.TABLE}
+        WHERE snapshot_id = ? AND name >= ?
+        """
     )
     def snapshot_count_branches_from_name(
         self, snapshot_id: Sha1Git, from_: bytes, *, statement
     ) -> Dict[Optional[str], int]:
         row = self._execute_with_retries(statement, [snapshot_id, from_]).one()
         (nb_none, counts) = row["counts"]
         return {None: nb_none, **counts}
 
     @_prepared_statement(
-        "SELECT ascii_bins_count(target_type) AS counts "
-        "FROM snapshot_branch "
-        "WHERE snapshot_id = ? AND name < ?"
+        f"""
+        SELECT ascii_bins_count(target_type) AS counts
+        FROM {SnapshotBranchRow.TABLE}
+        WHERE snapshot_id = ? AND name < ?
+        """
     )
     def snapshot_count_branches_before_name(
         self, snapshot_id: Sha1Git, before: bytes, *, statement,
     ) -> Dict[Optional[str], int]:
         row = self._execute_with_retries(statement, [snapshot_id, before]).one()
         (nb_none, counts) = row["counts"]
         return {None: nb_none, **counts}
 
     def snapshot_count_branches(
         self, snapshot_id: Sha1Git, branch_name_exclude_prefix: Optional[bytes] = None,
     ) -> Dict[Optional[str], int]:
         """Returns a dictionary from type names to the number of branches
         of that type."""
         prefix = branch_name_exclude_prefix
         if prefix is None:
             return self.snapshot_count_branches_from_name(snapshot_id, b"")
         else:
             # counts branches before exclude prefix
             counts = Counter(
                 self.snapshot_count_branches_before_name(snapshot_id, prefix)
             )
 
             # no need to execute that part if each bit of the prefix equals 1
             if prefix.replace(b"\xff", b"") != b"":
                 # counts branches after exclude prefix and update counters
                 counts.update(
                     self.snapshot_count_branches_from_name(
                         snapshot_id, _next_bytes_value(prefix)
                     )
                 )
             return counts
 
     @_prepared_select_statement(
         SnapshotBranchRow, "WHERE snapshot_id = ? AND name >= ? LIMIT ?"
     )
     def snapshot_branch_get_from_name(
         self, snapshot_id: Sha1Git, from_: bytes, limit: int, *, statement
     ) -> Iterable[SnapshotBranchRow]:
         return map(
             SnapshotBranchRow.from_dict,
             self._execute_with_retries(statement, [snapshot_id, from_, limit]),
         )
 
     @_prepared_select_statement(
         SnapshotBranchRow, "WHERE snapshot_id = ? AND name >= ? AND name < ? LIMIT ?"
     )
     def snapshot_branch_get_range(
         self,
         snapshot_id: Sha1Git,
         from_: bytes,
         before: bytes,
         limit: int,
         *,
         statement,
     ) -> Iterable[SnapshotBranchRow]:
         return map(
             SnapshotBranchRow.from_dict,
             self._execute_with_retries(statement, [snapshot_id, from_, before, limit]),
         )
 
     def snapshot_branch_get(
         self,
         snapshot_id: Sha1Git,
         from_: bytes,
         limit: int,
         branch_name_exclude_prefix: Optional[bytes] = None,
     ) -> Iterable[SnapshotBranchRow]:
         prefix = branch_name_exclude_prefix
         if prefix is None:
             return self.snapshot_branch_get_from_name(snapshot_id, from_, limit)
         else:
             # get branches before the exclude prefix
             branches = list(
                 self.snapshot_branch_get_range(snapshot_id, from_, prefix, limit)
             )
             nb_branches = len(branches)
             # no need to execute that part if limit is reached
             # or if each bit of the prefix equals 1
             if nb_branches < limit and prefix.replace(b"\xff", b"") != b"":
                 # get branches after the exclude prefix and update list to return
                 branches.extend(
                     self.snapshot_branch_get_from_name(
                         snapshot_id, _next_bytes_value(prefix), limit - nb_branches
                     )
                 )
             return branches
 
     ##########################
     # 'origin' table
     ##########################
 
     @_prepared_insert_statement(OriginRow)
     def origin_add_one(self, origin: OriginRow, *, statement) -> None:
         self._add_one(statement, origin)
 
     @_prepared_select_statement(OriginRow, "WHERE sha1 = ?")
     def origin_get_by_sha1(self, sha1: bytes, *, statement) -> Iterable[OriginRow]:
         return map(OriginRow.from_dict, self._execute_with_retries(statement, [sha1]))
 
     def origin_get_by_url(self, url: str) -> Iterable[OriginRow]:
         return self.origin_get_by_sha1(hash_url(url))
 
     @_prepared_statement(
-        f'SELECT token(sha1) AS tok, {", ".join(OriginRow.cols())} '
-        f"FROM origin WHERE token(sha1) >= ? LIMIT ?"
+        f"""
+        SELECT token(sha1) AS tok, {", ".join(OriginRow.cols())}
+        FROM {OriginRow.TABLE}
+        WHERE token(sha1) >= ? LIMIT ?
+        """
     )
     def origin_list(
         self, start_token: int, limit: int, *, statement
     ) -> Iterable[Tuple[int, OriginRow]]:
         """Returns an iterable of (token, origin)"""
         return (
             (row["tok"], OriginRow.from_dict(remove_keys(row, ("tok",))))
             for row in self._execute_with_retries(statement, [start_token, limit])
         )
 
     @_prepared_select_statement(OriginRow)
     def origin_iter_all(self, *, statement) -> Iterable[OriginRow]:
         return map(OriginRow.from_dict, self._execute_with_retries(statement, []))
 
-    @_prepared_statement("SELECT next_visit_id FROM origin WHERE sha1 = ?")
+    @_prepared_statement(f"SELECT next_visit_id FROM {OriginRow.TABLE} WHERE sha1 = ?")
     def _origin_get_next_visit_id(self, origin_sha1: bytes, *, statement) -> int:
         rows = list(self._execute_with_retries(statement, [origin_sha1]))
         assert len(rows) == 1  # TODO: error handling
         return rows[0]["next_visit_id"]
 
     @_prepared_statement(
-        "UPDATE origin SET next_visit_id=? WHERE sha1 = ? IF next_visit_id=?"
+        f"""
+        UPDATE {OriginRow.TABLE}
+        SET next_visit_id=?
+        WHERE sha1 = ? IF next_visit_id=?
+        """
     )
     def origin_generate_unique_visit_id(self, origin_url: str, *, statement) -> int:
         origin_sha1 = hash_url(origin_url)
         next_id = self._origin_get_next_visit_id(origin_sha1)
         while True:
             res = list(
                 self._execute_with_retries(
                     statement, [next_id + 1, origin_sha1, next_id]
                 )
             )
             assert len(res) == 1
             if res[0]["[applied]"]:
                 # No data race
                 return next_id
             else:
                 # Someone else updated it before we did, let's try again
                 next_id = res[0]["next_visit_id"]
                 # TODO: abort after too many attempts
 
         return next_id
 
     ##########################
     # 'origin_visit' table
     ##########################
 
     @_prepared_select_statements(
         OriginVisitRow,
         {
             (True, ListOrder.ASC): (
                 "WHERE origin = ? AND visit > ? ORDER BY visit ASC LIMIT ?"
             ),
             (True, ListOrder.DESC): (
                 "WHERE origin = ? AND visit < ? ORDER BY visit DESC LIMIT ?"
             ),
             (False, ListOrder.ASC): "WHERE origin = ? ORDER BY visit ASC LIMIT ?",
             (False, ListOrder.DESC): "WHERE origin = ? ORDER BY visit DESC LIMIT ?",
         },
     )
     def origin_visit_get(
         self,
         origin_url: str,
         last_visit: Optional[int],
         limit: int,
         order: ListOrder,
         *,
         statements,
     ) -> Iterable[OriginVisitRow]:
         args: List[Any] = [origin_url]
 
         if last_visit is not None:
             args.append(last_visit)
 
         args.append(limit)
 
         statement = statements[(last_visit is not None, order)]
         return map(
             OriginVisitRow.from_dict, self._execute_with_retries(statement, args)
         )
 
     @_prepared_insert_statement(OriginVisitRow)
     def origin_visit_add_one(self, visit: OriginVisitRow, *, statement) -> None:
         self._add_one(statement, visit)
 
     @_prepared_select_statement(OriginVisitRow, "WHERE origin = ? AND visit = ?")
     def origin_visit_get_one(
         self, origin_url: str, visit_id: int, *, statement
     ) -> Optional[OriginVisitRow]:
         # TODO: error handling
         rows = list(self._execute_with_retries(statement, [origin_url, visit_id]))
         if rows:
             return OriginVisitRow.from_dict(rows[0])
         else:
             return None
 
     @_prepared_select_statement(OriginVisitRow, "WHERE origin = ?")
     def origin_visit_get_all(
         self, origin_url: str, *, statement
     ) -> Iterable[OriginVisitRow]:
         return map(
             OriginVisitRow.from_dict,
             self._execute_with_retries(statement, [origin_url]),
         )
 
     @_prepared_select_statement(OriginVisitRow, "WHERE token(origin) >= ?")
     def _origin_visit_iter_from(
         self, min_token: int, *, statement
     ) -> Iterable[OriginVisitRow]:
         return map(
             OriginVisitRow.from_dict, self._execute_with_retries(statement, [min_token])
         )
 
     @_prepared_select_statement(OriginVisitRow, "WHERE token(origin) < ?")
     def _origin_visit_iter_to(
         self, max_token: int, *, statement
     ) -> Iterable[OriginVisitRow]:
         return map(
             OriginVisitRow.from_dict, self._execute_with_retries(statement, [max_token])
         )
 
     def origin_visit_iter(self, start_token: int) -> Iterator[OriginVisitRow]:
         """Returns all origin visits in order from this token,
         and wraps around the token space."""
         yield from self._origin_visit_iter_from(start_token)
         yield from self._origin_visit_iter_to(start_token)
 
     ##########################
     # 'origin_visit_status' table
     ##########################
 
     @_prepared_select_statements(
         OriginVisitStatusRow,
         {
             (True, ListOrder.ASC): (
                 "WHERE origin = ? AND visit = ? AND date >= ? "
                 "ORDER BY visit ASC LIMIT ?"
             ),
             (True, ListOrder.DESC): (
                 "WHERE origin = ? AND visit = ? AND date <= ? "
                 "ORDER BY visit DESC LIMIT ?"
             ),
             (False, ListOrder.ASC): (
                 "WHERE origin = ? AND visit = ? ORDER BY visit ASC LIMIT ?"
             ),
             (False, ListOrder.DESC): (
                 "WHERE origin = ? AND visit = ? ORDER BY visit DESC LIMIT ?"
             ),
         },
     )
     def origin_visit_status_get_range(
         self,
         origin: str,
         visit: int,
         date_from: Optional[datetime.datetime],
         limit: int,
         order: ListOrder,
         *,
         statements,
     ) -> Iterable[OriginVisitStatusRow]:
         args: List[Any] = [origin, visit]
 
         if date_from is not None:
             args.append(date_from)
 
         args.append(limit)
 
         statement = statements[(date_from is not None, order)]
 
         return map(
             OriginVisitStatusRow.from_dict, self._execute_with_retries(statement, args)
         )
 
     @_prepared_insert_statement(OriginVisitStatusRow)
     def origin_visit_status_add_one(
         self, visit_update: OriginVisitStatusRow, *, statement
     ) -> None:
         self._add_one(statement, visit_update)
 
     def origin_visit_status_get_latest(
         self, origin: str, visit: int,
     ) -> Optional[OriginVisitStatusRow]:
         """Given an origin visit id, return its latest origin_visit_status
 
          """
         return next(self.origin_visit_status_get(origin, visit), None)
 
     @_prepared_select_statement(
         OriginVisitStatusRow, "WHERE origin = ? AND visit = ? ORDER BY date DESC"
     )
     def origin_visit_status_get(
         self, origin: str, visit: int, *, statement,
     ) -> Iterator[OriginVisitStatusRow]:
         """Return all origin visit statuses for a given visit
 
         """
         return map(
             OriginVisitStatusRow.from_dict,
             self._execute_with_retries(statement, [origin, visit]),
         )
 
     ##########################
     # 'metadata_authority' table
     ##########################
 
     @_prepared_insert_statement(MetadataAuthorityRow)
     def metadata_authority_add(self, authority: MetadataAuthorityRow, *, statement):
         self._add_one(statement, authority)
 
     @_prepared_select_statement(MetadataAuthorityRow, "WHERE type = ? AND url = ?")
     def metadata_authority_get(
         self, type, url, *, statement
     ) -> Optional[MetadataAuthorityRow]:
         rows = list(self._execute_with_retries(statement, [type, url]))
         if rows:
             return MetadataAuthorityRow.from_dict(rows[0])
         else:
             return None
 
     ##########################
     # 'metadata_fetcher' table
     ##########################
 
     @_prepared_insert_statement(MetadataFetcherRow)
     def metadata_fetcher_add(self, fetcher, *, statement):
         self._add_one(statement, fetcher)
 
     @_prepared_select_statement(MetadataFetcherRow, "WHERE name = ? AND version = ?")
     def metadata_fetcher_get(
         self, name, version, *, statement
     ) -> Optional[MetadataFetcherRow]:
         rows = list(self._execute_with_retries(statement, [name, version]))
         if rows:
             return MetadataFetcherRow.from_dict(rows[0])
         else:
             return None
 
     #########################
     # 'raw_extrinsic_metadata' table
     #########################
 
     @_prepared_insert_statement(RawExtrinsicMetadataRow)
     def raw_extrinsic_metadata_add(self, raw_extrinsic_metadata, *, statement):
         self._add_one(statement, raw_extrinsic_metadata)
 
     @_prepared_select_statement(
         RawExtrinsicMetadataRow,
         "WHERE target=? AND authority_url=? AND discovery_date>? AND authority_type=?",
     )
     def raw_extrinsic_metadata_get_after_date(
         self,
         target: str,
         authority_type: str,
         authority_url: str,
         after: datetime.datetime,
         *,
         statement,
     ) -> Iterable[RawExtrinsicMetadataRow]:
         return map(
             RawExtrinsicMetadataRow.from_dict,
             self._execute_with_retries(
                 statement, [target, authority_url, after, authority_type]
             ),
         )
 
     @_prepared_select_statement(
         RawExtrinsicMetadataRow,
         "WHERE target=? AND authority_type=? AND authority_url=? "
         "AND (discovery_date, id) > (?, ?)",
     )
     def raw_extrinsic_metadata_get_after_date_and_id(
         self,
         target: str,
         authority_type: str,
         authority_url: str,
         after_date: datetime.datetime,
         after_id: bytes,
         *,
         statement,
     ) -> Iterable[RawExtrinsicMetadataRow]:
         return map(
             RawExtrinsicMetadataRow.from_dict,
             self._execute_with_retries(
                 statement,
                 [target, authority_type, authority_url, after_date, after_id,],
             ),
         )
 
     @_prepared_select_statement(
         RawExtrinsicMetadataRow,
         "WHERE target=? AND authority_url=? AND authority_type=?",
     )
     def raw_extrinsic_metadata_get(
         self, target: str, authority_type: str, authority_url: str, *, statement
     ) -> Iterable[RawExtrinsicMetadataRow]:
         return map(
             RawExtrinsicMetadataRow.from_dict,
             self._execute_with_retries(
                 statement, [target, authority_url, authority_type]
             ),
         )
 
     ##########################
     # 'extid' table
     ##########################
     def _extid_add_finalize(self, statement: BoundStatement) -> None:
         """Returned currified by extid_add_prepare, to be called when the
         extid row should be added to the primary table."""
         self._execute_with_retries(statement, None)
         self._increment_counter("extid", 1)
 
     @_prepared_insert_statement(ExtIDRow)
     def extid_add_prepare(
         self, extid: ExtIDRow, *, statement
     ) -> Tuple[int, Callable[[], None]]:
         statement = statement.bind(dataclasses.astuple(extid))
         token_class = self._cluster.metadata.token_map.token_class
         token = token_class.from_key(statement.routing_key).value
         assert TOKEN_BEGIN <= token <= TOKEN_END
 
         # Function to be called after the indexes contain their respective
         # row
         finalizer = functools.partial(self._extid_add_finalize, statement)
 
         return (token, finalizer)
 
     @_prepared_select_statement(
         ExtIDRow, "WHERE extid_type=? AND extid=? AND target_type=? AND target=?",
     )
     def extid_get_from_pk(
         self, extid_type: str, extid: bytes, target: CoreSWHID, *, statement,
     ) -> Optional[ExtIDRow]:
         rows = list(
             self._execute_with_retries(
                 statement,
                 [extid_type, extid, target.object_type.value, target.object_id],
             ),
         )
         assert len(rows) <= 1
         if rows:
             return ExtIDRow(**rows[0])
         else:
             return None
 
     @_prepared_select_statement(
         ExtIDRow, "WHERE token(extid_type, extid) = ?",
     )
     def extid_get_from_token(self, token: int, *, statement) -> Iterable[ExtIDRow]:
         return map(ExtIDRow.from_dict, self._execute_with_retries(statement, [token]),)
 
     @_prepared_select_statement(
         ExtIDRow, "WHERE extid_type=? AND extid=?",
     )
     def extid_get_from_extid(
         self, extid_type: str, extid: bytes, *, statement
     ) -> Iterable[ExtIDRow]:
         return map(
             ExtIDRow.from_dict,
             self._execute_with_retries(statement, [extid_type, extid]),
         )
 
     def extid_get_from_target(
         self, target_type: str, target: bytes
     ) -> Iterable[ExtIDRow]:
         for token in self._extid_get_tokens_from_target(target_type, target):
             if token is not None:
                 for extid in self.extid_get_from_token(token):
                     # re-check the extid against target (in case of murmur3 collision)
                     if (
                         extid is not None
                         and extid.target_type == target_type
                         and extid.target == target
                     ):
                         yield extid
 
     ##########################
     # 'extid_by_target' table
     ##########################
 
-    @_prepared_statement(
-        "INSERT INTO extid_by_target (target_type, target, target_token) "
-        "VALUES (?, ?, ?)"
-    )
-    def extid_index_add_one(self, extid: ExtIDRow, token: int, *, statement) -> None:
+    @_prepared_insert_statement(ExtIDByTargetRow)
+    def extid_index_add_one(self, row: ExtIDByTargetRow, *, statement) -> None:
         """Adds a row mapping extid[target_type, target] to the token of the ExtID in
         the main 'extid' table."""
-        self._execute_with_retries(statement, [extid.target_type, extid.target, token])
+        self._add_one(statement, row)
 
     @_prepared_statement(
-        "SELECT target_token "
-        "FROM extid_by_target "
-        "WHERE target_type = ? AND target = ?"
+        f"""
+        SELECT target_token
+        FROM {ExtIDByTargetRow.TABLE}
+        WHERE target_type = ? AND target = ?
+        """
     )
     def _extid_get_tokens_from_target(
         self, target_type: str, target: bytes, *, statement
     ) -> Iterable[int]:
         return (
             row["target_token"]
             for row in self._execute_with_retries(statement, [target_type, target])
         )
 
     ##########################
     # Miscellaneous
     ##########################
 
     @_prepared_statement("SELECT uuid() FROM revision LIMIT 1;")
     def check_read(self, *, statement):
         self._execute_with_retries(statement, [])
 
     @_prepared_select_statement(ObjectCountRow, "WHERE partition_key=0")
     def stat_counters(self, *, statement) -> Iterable[ObjectCountRow]:
         return map(ObjectCountRow.from_dict, self._execute_with_retries(statement, []))
diff --git a/swh/storage/cassandra/model.py b/swh/storage/cassandra/model.py
index f842178a..a7803ae7 100644
--- a/swh/storage/cassandra/model.py
+++ b/swh/storage/cassandra/model.py
@@ -1,294 +1,318 @@
 # Copyright (C) 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
 
 """Classes representing tables in the Cassandra database.
 
 They are very close to classes found in swh.model.model, but most of
 them are subtly different:
 
 * Large objects are split into other classes (eg. RevisionRow has no
   'parents' field, because parents are stored in a different table,
   represented by RevisionParentRow)
 * They have a "cols" field, which returns the list of column names
   of the table
 * They only use types that map directly to Cassandra's schema (ie. no enums)
 
 Therefore, this model doesn't reuse swh.model.model, except for types
 that can be mapped to UDTs (Person and TimestampWithTimezone).
 """
 
 import dataclasses
 import datetime
 from typing import Any, ClassVar, Dict, List, Optional, Tuple, Type, TypeVar
 
 from swh.model.model import Person, TimestampWithTimezone
 
 MAGIC_NULL_PK = b"<null>"
 """
 NULLs (or all-empty blobs) are not allowed in primary keys; instead we use a
 special value that can't possibly be a valid hash.
 """
 
 
 T = TypeVar("T", bound="BaseRow")
 
 
+def content_index_table_name(algo: str, skipped_content: bool) -> str:
+    """Given an algorithm name, returns the name of one of the 'content_by_*'
+    and 'skipped_content_by_*' tables that serve as index for the 'content'
+    and 'skipped_content' tables based on this algorithm's hashes.
+
+    For now it is a simple substitution, but future versions may append a version
+    number to it, if needed for schema updates."""
+    if skipped_content:
+        return f"skipped_content_by_{algo}"
+    else:
+        return f"content_by_{algo}"
+
+
 class BaseRow:
     TABLE: ClassVar[str]
     PARTITION_KEY: ClassVar[Tuple[str, ...]]
     CLUSTERING_KEY: ClassVar[Tuple[str, ...]] = ()
 
     @classmethod
     def from_dict(cls: Type[T], d: Dict[str, Any]) -> T:
         return cls(**d)  # type: ignore
 
     @classmethod
     def cols(cls) -> List[str]:
         return [field.name for field in dataclasses.fields(cls)]
 
     def to_dict(self) -> Dict[str, Any]:
         return dataclasses.asdict(self)
 
 
 @dataclasses.dataclass
 class ContentRow(BaseRow):
     TABLE = "content"
     PARTITION_KEY = ("sha1", "sha1_git", "sha256", "blake2s256")
 
     sha1: bytes
     sha1_git: bytes
     sha256: bytes
     blake2s256: bytes
     length: int
     ctime: datetime.datetime
     status: str
 
 
 @dataclasses.dataclass
 class SkippedContentRow(BaseRow):
     TABLE = "skipped_content"
     PARTITION_KEY = ("sha1", "sha1_git", "sha256", "blake2s256")
 
     sha1: Optional[bytes]
     sha1_git: Optional[bytes]
     sha256: Optional[bytes]
     blake2s256: Optional[bytes]
     length: Optional[int]
     ctime: Optional[datetime.datetime]
     status: str
     reason: str
     origin: str
 
     @classmethod
     def from_dict(cls, d: Dict[str, Any]) -> "SkippedContentRow":
         d = d.copy()
         for k in ("sha1", "sha1_git", "sha256", "blake2s256"):
             if d[k] == MAGIC_NULL_PK:
                 d[k] = None
         return super().from_dict(d)
 
 
 @dataclasses.dataclass
 class DirectoryRow(BaseRow):
     TABLE = "directory"
     PARTITION_KEY = ("id",)
 
     id: bytes
 
 
 @dataclasses.dataclass
 class DirectoryEntryRow(BaseRow):
     TABLE = "directory_entry"
     PARTITION_KEY = ("directory_id",)
     CLUSTERING_KEY = ("name",)
 
     directory_id: bytes
     name: bytes
     target: bytes
     perms: int
     type: str
 
 
 @dataclasses.dataclass
 class RevisionRow(BaseRow):
     TABLE = "revision"
     PARTITION_KEY = ("id",)
 
     id: bytes
     date: Optional[TimestampWithTimezone]
     committer_date: Optional[TimestampWithTimezone]
     type: str
     directory: bytes
     message: bytes
     author: Person
     committer: Person
     synthetic: bool
     metadata: str
     extra_headers: dict
 
 
 @dataclasses.dataclass
 class RevisionParentRow(BaseRow):
     TABLE = "revision_parent"
     PARTITION_KEY = ("id",)
     CLUSTERING_KEY = ("parent_rank",)
 
     id: bytes
     parent_rank: int
     parent_id: bytes
 
 
 @dataclasses.dataclass
 class ReleaseRow(BaseRow):
     TABLE = "release"
     PARTITION_KEY = ("id",)
 
     id: bytes
     target_type: str
     target: bytes
     date: TimestampWithTimezone
     name: bytes
     message: bytes
     author: Person
     synthetic: bool
 
 
 @dataclasses.dataclass
 class SnapshotRow(BaseRow):
     TABLE = "snapshot"
     PARTITION_KEY = ("id",)
 
     id: bytes
 
 
 @dataclasses.dataclass
 class SnapshotBranchRow(BaseRow):
     TABLE = "snapshot_branch"
     PARTITION_KEY = ("snapshot_id",)
     CLUSTERING_KEY = ("name",)
 
     snapshot_id: bytes
     name: bytes
     target_type: Optional[str]
     target: Optional[bytes]
 
 
 @dataclasses.dataclass
 class OriginVisitRow(BaseRow):
     TABLE = "origin_visit"
     PARTITION_KEY = ("origin",)
     CLUSTERING_KEY = ("visit",)
 
     origin: str
     visit: int
     date: datetime.datetime
     type: str
 
 
 @dataclasses.dataclass
 class OriginVisitStatusRow(BaseRow):
     TABLE = "origin_visit_status"
     PARTITION_KEY = ("origin",)
     CLUSTERING_KEY = ("visit", "date")
 
     origin: str
     visit: int
     date: datetime.datetime
     type: str
     status: str
     metadata: str
     snapshot: bytes
 
     @classmethod
     def from_dict(cls: Type[T], d: Dict[str, Any]) -> T:
         return cls(**d)  # type: ignore
 
 
 @dataclasses.dataclass
 class OriginRow(BaseRow):
     TABLE = "origin"
     PARTITION_KEY = ("sha1",)
 
     sha1: bytes
     url: str
     next_visit_id: int
 
 
 @dataclasses.dataclass
 class MetadataAuthorityRow(BaseRow):
     TABLE = "metadata_authority"
     PARTITION_KEY = ("url",)
     CLUSTERING_KEY = ("type",)
 
     url: str
     type: str
     metadata: str
 
 
 @dataclasses.dataclass
 class MetadataFetcherRow(BaseRow):
     TABLE = "metadata_fetcher"
     PARTITION_KEY = ("name",)
     CLUSTERING_KEY = ("version",)
 
     name: str
     version: str
     metadata: str
 
 
 @dataclasses.dataclass
 class RawExtrinsicMetadataRow(BaseRow):
     TABLE = "raw_extrinsic_metadata"
     PARTITION_KEY = ("target",)
     CLUSTERING_KEY = (
         "authority_type",
         "authority_url",
         "discovery_date",
         "id",
     )
 
     id: bytes
 
     type: str
     target: str
 
     authority_type: str
     authority_url: str
     discovery_date: datetime.datetime
     fetcher_name: str
     fetcher_version: str
 
     format: str
     metadata: bytes
 
     origin: Optional[str]
     visit: Optional[int]
     snapshot: Optional[str]
     release: Optional[str]
     revision: Optional[str]
     path: Optional[bytes]
     directory: Optional[str]
 
 
 @dataclasses.dataclass
 class ObjectCountRow(BaseRow):
     TABLE = "object_count"
     PARTITION_KEY = ("partition_key",)
     CLUSTERING_KEY = ("object_type",)
 
     partition_key: int
     object_type: str
     count: int
 
 
 @dataclasses.dataclass
 class ExtIDRow(BaseRow):
     TABLE = "extid"
     PARTITION_KEY = ("target", "target_type", "extid", "extid_type")
 
     extid_type: str
     extid: bytes
     target_type: str
     target: bytes
+
+
+@dataclasses.dataclass
+class ExtIDByTargetRow(BaseRow):
+    TABLE = "extid_by_target"
+    PARTITION_KEY = ("target_type", "target")
+    CLUSTERING_KEY = ("target_token",)
+
+    target_type: str
+    target: bytes
+    target_token: int
diff --git a/swh/storage/cassandra/storage.py b/swh/storage/cassandra/storage.py
index cb33ab46..ad8e18a4 100644
--- a/swh/storage/cassandra/storage.py
+++ b/swh/storage/cassandra/storage.py
@@ -1,1414 +1,1420 @@
 # 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 base64
 import datetime
 import itertools
 import json
 import random
 import re
 from typing import (
     Any,
     Callable,
     Dict,
     Iterable,
     List,
     Optional,
     Sequence,
     Set,
     Tuple,
     Union,
 )
 
 import attr
 
 from swh.core.api.classes import stream_results
 from swh.core.api.serializers import msgpack_dumps, msgpack_loads
 from swh.model.hashutil import DEFAULT_ALGORITHMS
 from swh.model.identifiers import CoreSWHID, ExtendedSWHID
 from swh.model.identifiers import ObjectType as SwhidObjectType
 from swh.model.model import (
     Content,
     Directory,
     DirectoryEntry,
     ExtID,
     MetadataAuthority,
     MetadataAuthorityType,
     MetadataFetcher,
     Origin,
     OriginVisit,
     OriginVisitStatus,
     RawExtrinsicMetadata,
     Release,
     Revision,
     Sha1Git,
     SkippedContent,
     Snapshot,
     SnapshotBranch,
     TargetType,
 )
 from swh.storage.interface import (
     VISIT_STATUSES,
     ListOrder,
     PagedResult,
     PartialBranches,
     Sha1,
 )
 from swh.storage.objstorage import ObjStorage
 from swh.storage.utils import map_optional, now
 from swh.storage.writer import JournalWriter
 
 from . import converters
 from ..exc import HashCollision, StorageArgumentException
 from ..utils import remove_keys
 from .common import TOKEN_BEGIN, TOKEN_END, hash_url
 from .cql import CqlRunner
 from .model import (
     ContentRow,
     DirectoryEntryRow,
     DirectoryRow,
+    ExtIDByTargetRow,
     ExtIDRow,
     MetadataAuthorityRow,
     MetadataFetcherRow,
     OriginRow,
     OriginVisitRow,
     OriginVisitStatusRow,
     RawExtrinsicMetadataRow,
     RevisionParentRow,
     SkippedContentRow,
     SnapshotBranchRow,
     SnapshotRow,
 )
 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 = CqlRunner(hosts, keyspace, port)
         self.journal_writer: JournalWriter = JournalWriter(journal_writer)
         self.objstorage: ObjStorage = ObjStorage(objstorage)
 
     def check_config(self, *, check_write: bool) -> bool:
         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:
             assert isinstance(token, int), 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())
         ]
 
         if with_data:
             # First insert to the objstorage, if the endpoint is
             # `content_add` (as opposed to `content_add_metadata`).
 
             # 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
             summary = self.objstorage.content_add(
                 c for c in contents if c.status != "absent"
             )
             content_add_bytes = summary["content:add:bytes"]
 
         self.journal_writer.content_add(contents)
 
         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 other_algo in HASH_ALGORITHMS:
                         if getattr(row, other_algo) != content.get_hash(other_algo):
                             # This hash didn't match; discard the row.
                             collisions.append(
                                 {k: getattr(row, k) for k 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(
                 ContentRow(**remove_keys(content.to_dict(), ("data",)))
             )
 
             # 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: List[Content]) -> Dict:
         to_add = {
             (c.sha1, c.sha1_git, c.sha256, c.blake2s256): c for c in content
         }.values()
         contents = [attr.evolve(c, ctime=now()) for c in to_add]
         return self._content_add(list(contents), with_data=True)
 
     def content_update(
         self, contents: List[Dict[str, Any]], keys: List[str] = []
     ) -> None:
         raise NotImplementedError(
             "content_update is not supported by the Cassandra backend"
         )
 
     def content_add_metadata(self, content: List[Content]) -> Dict:
         return self._content_add(content, with_data=False)
 
     def content_get_data(self, content: Sha1) -> Optional[bytes]:
         # FIXME: Make this method support slicing the `data`
         return self.objstorage.content_get(content)
 
     def content_get_partition(
         self,
         partition_id: int,
         nb_partitions: int,
         page_token: Optional[str] = None,
         limit: int = 1000,
     ) -> PagedResult[Content]:
         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)
 
         next_page_token: Optional[str] = None
 
         rows = self._cql_runner.content_get_token_range(
             range_start, range_end, limit + 1
         )
         contents = []
         for counter, (tok, row) in enumerate(rows):
             if row.status == "absent":
                 continue
             row_d = row.to_dict()
             if counter >= limit:
                 next_page_token = str(tok)
                 break
             row_d.pop("ctime")
             contents.append(Content(**row_d))
 
         assert len(contents) <= limit
         return PagedResult(results=contents, next_page_token=next_page_token)
 
     def content_get(self, contents: List[Sha1]) -> List[Optional[Content]]:
         contents_by_sha1: Dict[Sha1, Optional[Content]] = {}
         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):
                 row_d = row.to_dict()
                 row_d.pop("ctime")
                 content = Content(**row_d)
                 contents_by_sha1[content.sha1] = content
         return [contents_by_sha1.get(sha1) for sha1 in contents]
 
     def content_find(self, content: Dict[str, Any]) -> List[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 "
                 f"of: {', '.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.
                 row_d = row.to_dict()
                 row_d["ctime"] = row.ctime.replace(tzinfo=datetime.timezone.utc)
                 results.append(Content(**row_d))
         return results
 
     def content_missing(
         self, contents: List[Dict[str, Any]], key_hash: str = "sha1"
     ) -> Iterable[bytes]:
         if key_hash not in DEFAULT_ALGORITHMS:
             raise StorageArgumentException(
                 "key_hash should be one of {','.join(DEFAULT_ALGORITHMS)}"
             )
 
         for content in contents:
             res = self.content_find(content)
             if not res:
                 yield content[key_hash]
 
     def content_missing_per_sha1(self, contents: List[bytes]) -> Iterable[bytes]:
         return self.content_missing([{"sha1": c} for c in contents])
 
     def content_missing_per_sha1_git(
         self, contents: List[Sha1Git]
     ) -> Iterable[Sha1Git]:
         return self.content_missing(
             [{"sha1_git": c} for c in contents], key_hash="sha1_git"
         )
 
     def content_get_random(self) -> Sha1Git:
         content = self._cql_runner.content_get_random()
         assert content, "Could not find any content"
         return content.sha1_git
 
     def _skipped_content_add(self, contents: List[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(
                 SkippedContentRow.from_dict({"origin": None, **content.to_dict()})
             )
 
             # 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: List[SkippedContent]) -> Dict:
         contents = [attr.evolve(c, ctime=now()) for c in content]
         return self._skipped_content_add(contents)
 
     def skipped_content_missing(
         self, contents: List[Dict[str, Any]]
     ) -> Iterable[Dict[str, Any]]:
         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: List[Directory]) -> Dict:
         to_add = {d.id: d for d in directories}.values()
         # Filter out directories that are already inserted.
         missing = self.directory_missing([dir_.id for dir_ in to_add])
         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(
                     DirectoryEntryRow(directory_id=directory.id, **entry.to_dict())
                 )
 
             # 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(DirectoryRow(id=directory.id))
 
         return {"directory:add": len(directories)}
 
     def directory_missing(self, directories: List[Sha1Git]) -> Iterable[Sha1Git]:
         return self._cql_runner.directory_missing(directories)
 
     def _join_dentry_to_content(self, dentry: DirectoryEntry) -> Dict[str, Any]:
         contents: Union[List[Content], List[SkippedContentRow]]
         keys = (
             "status",
             "sha1",
             "sha1_git",
             "sha256",
             "length",
         )
         ret = dict.fromkeys(keys)
         ret.update(dentry.to_dict())
         if ret["type"] == "file":
             contents = self.content_find({"sha1_git": ret["target"]})
             if not contents:
                 tokens = list(
                     self._cql_runner.skipped_content_get_tokens_from_single_hash(
                         "sha1_git", ret["target"]
                     )
                 )
                 if tokens:
                     contents = list(
                         self._cql_runner.skipped_content_get_from_token(tokens[0])
                     )
             if contents:
                 content = contents[0]
                 for key in keys:
                     ret[key] = getattr(content, key)
         return ret
 
     def _directory_ls(
         self, directory_id: Sha1Git, recursive: bool, prefix: bytes = b""
     ) -> Iterable[Dict[str, Any]]:
         if self.directory_missing([directory_id]):
             return
         rows = list(self._cql_runner.directory_entry_get([directory_id]))
 
         for row in rows:
             entry_d = row.to_dict()
             # Build and yield the directory entry dict
             del entry_d["directory_id"]
             entry = DirectoryEntry.from_dict(entry_d)
             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: Sha1Git, paths: List[bytes]
     ) -> Optional[Dict[str, Any]]:
         return self._directory_entry_get_by_path(directory, paths, b"")
 
     def _directory_entry_get_by_path(
         self, directory: Sha1Git, paths: List[bytes], prefix: bytes
     ) -> Optional[Dict[str, Any]]:
         if not paths:
             return None
 
         contents = list(self.directory_ls(directory))
 
         if not contents:
             return None
 
         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 None
 
         return self._directory_entry_get_by_path(
             first_item["target"], paths[1:], prefix + paths[0] + b"/"
         )
 
     def directory_ls(
         self, directory: Sha1Git, recursive: bool = False
     ) -> Iterable[Dict[str, Any]]:
         yield from self._directory_ls(directory, recursive)
 
     def directory_get_random(self) -> Sha1Git:
         directory = self._cql_runner.directory_get_random()
         assert directory, "Could not find any directory"
         return directory.id
 
     def revision_add(self, revisions: List[Revision]) -> Dict:
         # Filter-out revisions already in the database
         to_add = {r.id: r for r in revisions}.values()
         missing = self.revision_missing([rev.id for rev in to_add])
         revisions = [rev for rev in revisions if rev.id in missing]
         self.journal_writer.revision_add(revisions)
 
         for revision in revisions:
             revobject = converters.revision_to_db(revision)
             if revobject:
                 # Add parents first
                 for (rank, parent) in enumerate(revision.parents):
                     self._cql_runner.revision_parent_add_one(
                         RevisionParentRow(
                             id=revobject.id, parent_rank=rank, parent_id=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: List[Sha1Git]) -> Iterable[Sha1Git]:
         return self._cql_runner.revision_missing(revisions)
 
     def revision_get(self, revision_ids: List[Sha1Git]) -> List[Optional[Revision]]:
         rows = self._cql_runner.revision_get(revision_ids)
         revisions: Dict[Sha1Git, Revision] = {}
         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)
             parents = tuple(self._cql_runner.revision_parent_get(row.id))
             # parent_rank is the clustering key, so results are already
             # sorted by rank.
             rev = converters.revision_from_db(row, parents=parents)
             revisions[rev.id] = rev
 
         return [revisions.get(rev_id) for rev_id in revision_ids]
 
     def _get_parent_revs(
         self,
         rev_ids: Iterable[Sha1Git],
         seen: Set[Sha1Git],
         limit: Optional[int],
         short: bool,
     ) -> Union[
         Iterable[Dict[str, Any]], Iterable[Tuple[Sha1Git, Tuple[Sha1Git, ...]]],
     ]:
         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:
             ids = self._cql_runner.revision_get_ids(rev_ids)
             for id_ in ids:
                 # 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)
                 parents = tuple(self._cql_runner.revision_parent_get(id_))
 
                 # parent_rank is the clustering key, so results are already
                 # sorted by rank.
 
                 yield (id_, parents)
                 yield from self._get_parent_revs(parents, seen, limit, short)
         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)
                 parents = tuple(self._cql_runner.revision_parent_get(row.id))
 
                 # parent_rank is the clustering key, so results are already
                 # sorted by rank.
 
                 rev = converters.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: List[Sha1Git], limit: Optional[int] = None
     ) -> Iterable[Optional[Dict[str, Any]]]:
         seen: Set[Sha1Git] = set()
         yield from self._get_parent_revs(revisions, seen, limit, False)
 
     def revision_shortlog(
         self, revisions: List[Sha1Git], limit: Optional[int] = None
     ) -> Iterable[Optional[Tuple[Sha1Git, Tuple[Sha1Git, ...]]]]:
         seen: Set[Sha1Git] = set()
         yield from self._get_parent_revs(revisions, seen, limit, True)
 
     def revision_get_random(self) -> Sha1Git:
         revision = self._cql_runner.revision_get_random()
         assert revision, "Could not find any revision"
         return revision.id
 
     def release_add(self, releases: List[Release]) -> Dict:
         to_add = {r.id: r for r in releases}.values()
         missing = set(self.release_missing([rel.id for rel in to_add]))
         releases = [rel for rel in to_add if rel.id in missing]
         self.journal_writer.release_add(releases)
 
         for release in releases:
             if release:
                 self._cql_runner.release_add_one(converters.release_to_db(release))
 
         return {"release:add": len(releases)}
 
     def release_missing(self, releases: List[Sha1Git]) -> Iterable[Sha1Git]:
         return self._cql_runner.release_missing(releases)
 
     def release_get(self, releases: List[Sha1Git]) -> List[Optional[Release]]:
         rows = self._cql_runner.release_get(releases)
         rels: Dict[Sha1Git, Release] = {}
         for row in rows:
             release = converters.release_from_db(row)
             rels[row.id] = release
 
         return [rels.get(rel_id) for rel_id in releases]
 
     def release_get_random(self) -> Sha1Git:
         release = self._cql_runner.release_get_random()
         assert release, "Could not find any release"
         return release.id
 
     def snapshot_add(self, snapshots: List[Snapshot]) -> Dict:
         to_add = {s.id: s for s in snapshots}.values()
         missing = self._cql_runner.snapshot_missing([snp.id for snp in to_add])
         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: Optional[str] = None
                     target: Optional[bytes] = None
                 else:
                     target_type = branch.target_type.value
                     target = branch.target
                 self._cql_runner.snapshot_branch_add_one(
                     SnapshotBranchRow(
                         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(SnapshotRow(id=snapshot.id))
 
         return {"snapshot:add": len(snapshots)}
 
     def snapshot_missing(self, snapshots: List[Sha1Git]) -> Iterable[Sha1Git]:
         return self._cql_runner.snapshot_missing(snapshots)
 
     def snapshot_get(self, snapshot_id: Sha1Git) -> Optional[Dict[str, Any]]:
         d = self.snapshot_get_branches(snapshot_id)
         if d is None:
             return None
         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"],
         }
 
     def snapshot_count_branches(
         self, snapshot_id: Sha1Git, branch_name_exclude_prefix: Optional[bytes] = None,
     ) -> Optional[Dict[Optional[str], int]]:
         if self._cql_runner.snapshot_missing([snapshot_id]):
             # Makes sure we don't fetch branches for a snapshot that is
             # being added.
             return None
 
         return self._cql_runner.snapshot_count_branches(
             snapshot_id, branch_name_exclude_prefix
         )
 
     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,
     ) -> Optional[PartialBranches]:
         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: List = []
         while len(branches) < branches_count + 1:
             new_branches = list(
                 self._cql_runner.snapshot_branch_get(
                     snapshot_id,
                     branches_from,
                     branches_count + 1,
                     branch_name_exclude_prefix,
                 )
             )
 
             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
                 ]
 
             # Filter by branches_name_pattern
             if branch_name_include_substring:
                 new_branches_filtered = [
                     branch
                     for branch in new_branches_filtered
                     if branch.name is not None
                     and (
                         branch_name_include_substring is None
                         or branch_name_include_substring in branch.name
                     )
                 ]
 
             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
 
         return PartialBranches(
             id=snapshot_id,
             branches={
                 branch.name: None
                 if branch.target is None
                 else SnapshotBranch(
                     target=branch.target, target_type=TargetType(branch.target_type)
                 )
                 for branch in branches
             },
             next_branch=last_branch,
         )
 
     def snapshot_get_random(self) -> Sha1Git:
         snapshot = self._cql_runner.snapshot_get_random()
         assert snapshot, "Could not find any snapshot"
         return snapshot.id
 
     def object_find_by_sha1_git(self, ids: List[Sha1Git]) -> Dict[Sha1Git, List[Dict]]:
         results: Dict[Sha1Git, List[Dict]] = {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: List[Tuple[str, Callable[[List[Sha1Git]], List[Sha1Git]]]] = [
             ("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(list(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: List[str]) -> Iterable[Optional[Origin]]:
         return [self.origin_get_one(origin) for origin in origins]
 
     def origin_get_one(self, origin_url: str) -> Optional[Origin]:
         """Given an origin url, return the origin if it exists, None otherwise
 
         """
         rows = list(self._cql_runner.origin_get_by_url(origin_url))
         if rows:
             assert len(rows) == 1
             return Origin(url=rows[0].url)
         else:
             return None
 
     def origin_get_by_sha1(self, sha1s: List[bytes]) -> List[Optional[Dict[str, Any]]]:
         results = []
         for sha1 in sha1s:
             rows = list(self._cql_runner.origin_get_by_sha1(sha1))
             origin = {"url": rows[0].url} if rows else None
             results.append(origin)
         return results
 
     def origin_list(
         self, page_token: Optional[str] = None, limit: int = 100
     ) -> PagedResult[Origin]:
         # 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.")
         next_page_token = None
 
         origins = []
         # Take one more origin so we can reuse it as the next page token if any
         for (tok, row) in self._cql_runner.origin_list(start_token, limit + 1):
             origins.append(Origin(url=row.url))
             # keep reference of the last id for pagination purposes
             last_id = tok
 
         if len(origins) > limit:
             # 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)
 
     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,
     ) -> PagedResult[Origin]:
         # TODO: remove this endpoint, swh-search should be used instead.
         next_page_token = None
         offset = int(page_token) if page_token else 0
 
         origin_rows = [row for row in self._cql_runner.origin_iter_all()]
         if regexp:
             pat = re.compile(url_pattern)
             origin_rows = [row for row in origin_rows if pat.search(row.url)]
         else:
             origin_rows = [row for row in origin_rows if url_pattern in row.url]
 
         if with_visit:
             origin_rows = [row for row in origin_rows if row.next_visit_id > 1]
 
         if visit_types:
 
             def _has_visit_types(origin, visit_types):
                 for origin_visit in stream_results(self.origin_visit_get, origin):
                     if origin_visit.type in visit_types:
                         return True
                 return False
 
             origin_rows = [
                 row for row in origin_rows if _has_visit_types(row.url, visit_types)
             ]
 
         origins = [Origin(url=row.url) for row in origin_rows]
 
         origins = origins[offset : offset + limit + 1]
         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)
 
     def origin_count(
         self, url_pattern: str, regexp: bool = False, with_visit: bool = False
     ) -> int:
         raise NotImplementedError(
             "The Cassandra backend does not implement origin_count"
         )
 
     def origin_add(self, origins: List[Origin]) -> Dict[str, int]:
         to_add = {o.url: o for o in origins}.values()
         origins = [ori for ori in to_add if self.origin_get_one(ori.url) is None]
 
         self.journal_writer.origin_add(origins)
         for origin in origins:
             self._cql_runner.origin_add_one(
                 OriginRow(sha1=hash_url(origin.url), url=origin.url, next_visit_id=1)
             )
         return {"origin:add": len(origins)}
 
     def origin_visit_add(self, visits: List[OriginVisit]) -> Iterable[OriginVisit]:
         for visit in visits:
             origin = self.origin_get_one(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(OriginVisitRow(**visit.to_dict()))
             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,
                     type=visit.type,
                     status="created",
                     snapshot=None,
                 )
             )
 
         return all_visits
 
     def _origin_visit_status_add(self, visit_status: OriginVisitStatus) -> None:
         """Add an origin visit status"""
         if visit_status.type is None:
             visit_row = self._cql_runner.origin_visit_get_one(
                 visit_status.origin, visit_status.visit
             )
             if visit_row is None:
                 raise StorageArgumentException(
                     f"Unknown origin visit {visit_status.visit} "
                     f"of origin {visit_status.origin}"
                 )
             visit_status = attr.evolve(visit_status, type=visit_row.type)
 
         self.journal_writer.origin_visit_status_add([visit_status])
         self._cql_runner.origin_visit_status_add_one(
             converters.visit_status_to_row(visit_status)
         )
 
     def origin_visit_status_add(self, visit_statuses: List[OriginVisitStatus]) -> None:
         # First round to check existence (fail early if any is ko)
         for visit_status in visit_statuses:
             origin_url = self.origin_get_one(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_apply_status(
         self, visit: Dict[str, Any], visit_status: OriginVisitStatusRow
     ) -> Dict[str, Any]:
         """Retrieve the latest visit status information for the origin visit.
         Then merge it with the visit and return it.
 
         """
         return {
             # 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"],
             # We use the visit type from origin visit
             # if it's not present on the origin visit status
             "type": visit_status.type or visit["type"],
         }
 
     def _origin_visit_get_latest_status(self, visit: OriginVisit) -> OriginVisitStatus:
         """Retrieve the latest visit status information for the origin visit object.
 
         """
         assert visit.visit
         row = self._cql_runner.origin_visit_status_get_latest(visit.origin, visit.visit)
         assert row is not None
         visit_status = converters.row_to_visit_status(row)
         return attr.evolve(visit_status, origin=visit.origin)
 
     @staticmethod
     def _format_origin_visit_row(visit):
         return {
             **visit.to_dict(),
             "origin": visit.origin,
             "date": visit.date.replace(tzinfo=datetime.timezone.utc),
         }
 
     def origin_visit_get(
         self,
         origin: str,
         page_token: Optional[str] = None,
         order: ListOrder = ListOrder.ASC,
         limit: int = 10,
     ) -> PagedResult[OriginVisit]:
         if not isinstance(order, ListOrder):
             raise StorageArgumentException("order must be a ListOrder value")
         if page_token and not isinstance(page_token, str):
             raise StorageArgumentException("page_token must be a string.")
 
         next_page_token = None
         visit_from = None if page_token is None else int(page_token)
         visits: List[OriginVisit] = []
         extra_limit = limit + 1
 
         rows = self._cql_runner.origin_visit_get(origin, visit_from, extra_limit, order)
         for row in rows:
             visits.append(converters.row_to_visit(row))
 
         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)
 
     def origin_visit_status_get(
         self,
         origin: str,
         visit: int,
         page_token: Optional[str] = None,
         order: ListOrder = ListOrder.ASC,
         limit: int = 10,
     ) -> PagedResult[OriginVisitStatus]:
         next_page_token = None
         date_from = None
         if page_token is not None:
             date_from = datetime.datetime.fromisoformat(page_token)
 
         # Take one more visit status so we can reuse it as the next page token if any
         rows = self._cql_runner.origin_visit_status_get_range(
             origin, visit, date_from, limit + 1, order
         )
         visit_statuses = [converters.row_to_visit_status(row) for row in rows]
         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)
 
     def origin_visit_find_by_date(
         self, origin: str, visit_date: datetime.datetime
     ) -> Optional[OriginVisit]:
         # 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:
             return converters.row_to_visit(min(rows, key=key))
         return None
 
     def origin_visit_get_by(self, origin: str, visit: int) -> Optional[OriginVisit]:
         row = self._cql_runner.origin_visit_get_one(origin, visit)
         if row:
             return converters.row_to_visit(row)
         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[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"
             )
         # 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)
             for status_row in self._cql_runner.origin_visit_status_get(
                 origin, visit["visit"]
             ):
                 updated_visit = self._origin_visit_apply_status(visit, status_row)
                 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
 
         if latest_visit is None:
             return None
         return OriginVisit(
             origin=latest_visit["origin"],
             visit=latest_visit["visit"],
             date=latest_visit["date"],
             type=latest_visit["type"],
         )
 
     def origin_visit_status_get_latest(
         self,
         origin_url: str,
         visit: int,
         allowed_statuses: Optional[List[str]] = None,
         require_snapshot: bool = False,
     ) -> 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"
             )
         rows = list(self._cql_runner.origin_visit_status_get(origin_url, visit))
         # 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 converters.row_to_visit_status(rows[0])
 
     def origin_visit_status_get_random(self, type: str) -> Optional[OriginVisitStatus]:
         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 = converters.row_to_visit(row)
             visit_status = self._origin_visit_get_latest_status(visit)
             if visit.date > back_in_the_day and visit_status.status == "full":
                 return visit_status
         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 raw_extrinsic_metadata_add(self, metadata: List[RawExtrinsicMetadata]) -> None:
         self.journal_writer.raw_extrinsic_metadata_add(metadata)
         for metadata_entry in metadata:
             if not self._cql_runner.metadata_authority_get(
                 metadata_entry.authority.type.value, metadata_entry.authority.url
             ):
                 raise StorageArgumentException(
                     f"Unknown authority {metadata_entry.authority}"
                 )
             if not self._cql_runner.metadata_fetcher_get(
                 metadata_entry.fetcher.name, metadata_entry.fetcher.version
             ):
                 raise StorageArgumentException(
                     f"Unknown fetcher {metadata_entry.fetcher}"
                 )
 
             try:
                 row = RawExtrinsicMetadataRow(
                     id=metadata_entry.id,
                     type=metadata_entry.target.object_type.name.lower(),
                     target=str(metadata_entry.target),
                     authority_type=metadata_entry.authority.type.value,
                     authority_url=metadata_entry.authority.url,
                     discovery_date=metadata_entry.discovery_date,
                     fetcher_name=metadata_entry.fetcher.name,
                     fetcher_version=metadata_entry.fetcher.version,
                     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),
                 )
                 self._cql_runner.raw_extrinsic_metadata_add(row)
             except TypeError as e:
                 raise StorageArgumentException(*e.args)
 
     def raw_extrinsic_metadata_get(
         self,
         target: ExtendedSWHID,
         authority: MetadataAuthority,
         after: Optional[datetime.datetime] = None,
         page_token: Optional[bytes] = None,
         limit: int = 1000,
     ) -> PagedResult[RawExtrinsicMetadata]:
         if page_token is not None:
             (after_date, id_) = msgpack_loads(base64.b64decode(page_token))
             if after and after_date < after:
                 raise StorageArgumentException(
                     "page_token is inconsistent with the value of 'after'."
                 )
             entries = self._cql_runner.raw_extrinsic_metadata_get_after_date_and_id(
                 str(target), authority.type.value, authority.url, after_date, id_,
             )
         elif after is not None:
             entries = self._cql_runner.raw_extrinsic_metadata_get_after_date(
                 str(target), authority.type.value, authority.url, after
             )
         else:
             entries = self._cql_runner.raw_extrinsic_metadata_get(
                 str(target), authority.type.value, 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)
 
             assert str(target) == entry.target
 
             result = RawExtrinsicMetadata(
                 target=target,
                 authority=MetadataAuthority(
                     type=MetadataAuthorityType(entry.authority_type),
                     url=entry.authority_url,
                 ),
                 fetcher=MetadataFetcher(
                     name=entry.fetcher_name, version=entry.fetcher_version,
                 ),
                 discovery_date=discovery_date,
                 format=entry.format,
                 metadata=entry.metadata,
                 origin=entry.origin,
                 visit=entry.visit,
                 snapshot=map_optional(CoreSWHID.from_string, entry.snapshot),
                 release=map_optional(CoreSWHID.from_string, entry.release),
                 revision=map_optional(CoreSWHID.from_string, entry.revision),
                 path=entry.path,
                 directory=map_optional(CoreSWHID.from_string, entry.directory),
             )
 
             results.append(result)
 
         if len(results) > limit:
             results.pop()
             assert len(results) == limit
             last_result = results[-1]
             next_page_token: Optional[str] = base64.b64encode(
                 msgpack_dumps((last_result.discovery_date, last_result.id,))
             ).decode()
         else:
             next_page_token = None
 
         return PagedResult(next_page_token=next_page_token, results=results,)
 
     def metadata_fetcher_add(self, fetchers: List[MetadataFetcher]) -> None:
         self.journal_writer.metadata_fetcher_add(fetchers)
         for fetcher in fetchers:
             self._cql_runner.metadata_fetcher_add(
                 MetadataFetcherRow(
                     name=fetcher.name,
                     version=fetcher.version,
                     metadata=json.dumps(map_optional(dict, fetcher.metadata)),
                 )
             )
 
     def metadata_fetcher_get(
         self, name: str, version: str
     ) -> Optional[MetadataFetcher]:
         fetcher = self._cql_runner.metadata_fetcher_get(name, version)
         if fetcher:
             return MetadataFetcher(
                 name=fetcher.name,
                 version=fetcher.version,
                 metadata=json.loads(fetcher.metadata),
             )
         else:
             return None
 
     def metadata_authority_add(self, authorities: List[MetadataAuthority]) -> None:
         self.journal_writer.metadata_authority_add(authorities)
         for authority in authorities:
             self._cql_runner.metadata_authority_add(
                 MetadataAuthorityRow(
                     url=authority.url,
                     type=authority.type.value,
                     metadata=json.dumps(map_optional(dict, authority.metadata)),
                 )
             )
 
     def metadata_authority_get(
         self, type: MetadataAuthorityType, url: str
     ) -> Optional[MetadataAuthority]:
         authority = self._cql_runner.metadata_authority_get(type.value, url)
         if authority:
             return MetadataAuthority(
                 type=MetadataAuthorityType(authority.type),
                 url=authority.url,
                 metadata=json.loads(authority.metadata),
             )
         else:
             return None
 
     # ExtID tables
     def extid_add(self, ids: List[ExtID]) -> Dict[str, int]:
         extids = [
             extid
             for extid in ids
             if not self._cql_runner.extid_get_from_pk(
                 extid_type=extid.extid_type, extid=extid.extid, target=extid.target,
             )
         ]
 
         self.journal_writer.extid_add(extids)
 
         inserted = 0
         for extid in extids:
+            target_type = extid.target.object_type.value
+            target = extid.target.object_id
             extidrow = ExtIDRow(
                 extid_type=extid.extid_type,
                 extid=extid.extid,
-                target_type=extid.target.object_type.value,
-                target=extid.target.object_id,
+                target_type=target_type,
+                target=target,
             )
             (token, insertion_finalizer) = self._cql_runner.extid_add_prepare(extidrow)
-            self._cql_runner.extid_index_add_one(extidrow, token)
+            indexrow = ExtIDByTargetRow(
+                target_type=target_type, target=target, target_token=token,
+            )
+            self._cql_runner.extid_index_add_one(indexrow)
             insertion_finalizer()
             inserted += 1
         return {"extid:add": inserted}
 
     def extid_get_from_extid(self, id_type: str, ids: List[bytes]) -> List[ExtID]:
         result: List[ExtID] = []
         for extid in ids:
             extidrows = list(self._cql_runner.extid_get_from_extid(id_type, extid))
             result.extend(
                 ExtID(
                     extid_type=extidrow.extid_type,
                     extid=extidrow.extid,
                     target=CoreSWHID(
                         object_type=extidrow.target_type, object_id=extidrow.target,
                     ),
                 )
                 for extidrow in extidrows
             )
         return result
 
     def extid_get_from_target(
         self, target_type: SwhidObjectType, ids: List[Sha1Git]
     ) -> List[ExtID]:
         result: List[ExtID] = []
         for target in ids:
             extidrows = list(
                 self._cql_runner.extid_get_from_target(target_type.value, target)
             )
             result.extend(
                 ExtID(
                     extid_type=extidrow.extid_type,
                     extid=extidrow.extid,
                     target=CoreSWHID(
                         object_type=SwhidObjectType(extidrow.target_type),
                         object_id=extidrow.target,
                     ),
                 )
                 for extidrow in extidrows
             )
         return result
 
     # Misc
     def clear_buffers(self, object_types: Sequence[str] = ()) -> None:
         """Do nothing
 
         """
         return None
 
     def flush(self, object_types: Sequence[str] = ()) -> Dict[str, int]:
         return {}
diff --git a/swh/storage/in_memory.py b/swh/storage/in_memory.py
index d69b7340..b227394b 100644
--- a/swh/storage/in_memory.py
+++ b/swh/storage/in_memory.py
@@ -1,686 +1,687 @@
 # 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
 
 from collections import defaultdict
 import datetime
 import functools
 import random
 from typing import (
     Any,
     Dict,
     Generic,
     Iterable,
     Iterator,
     List,
     Optional,
     Tuple,
     Type,
     TypeVar,
     Union,
 )
 
 from swh.model.identifiers import ExtendedSWHID
 from swh.model.model import Content, Sha1Git, SkippedContent
 from swh.storage.cassandra import CassandraStorage
 from swh.storage.cassandra.model import (
     BaseRow,
     ContentRow,
     DirectoryEntryRow,
     DirectoryRow,
+    ExtIDByTargetRow,
     ExtIDRow,
     MetadataAuthorityRow,
     MetadataFetcherRow,
     ObjectCountRow,
     OriginRow,
     OriginVisitRow,
     OriginVisitStatusRow,
     RawExtrinsicMetadataRow,
     ReleaseRow,
     RevisionParentRow,
     RevisionRow,
     SkippedContentRow,
     SnapshotBranchRow,
     SnapshotRow,
 )
 from swh.storage.interface import ListOrder
 from swh.storage.objstorage import ObjStorage
 
 from .common import origin_url_to_sha1
 from .writer import JournalWriter
 
 TRow = TypeVar("TRow", bound=BaseRow)
 
 
 class Table(Generic[TRow]):
     def __init__(self, row_class: Type[TRow]):
         self.row_class = row_class
         self.primary_key_cols = row_class.PARTITION_KEY + row_class.CLUSTERING_KEY
 
         # Map from tokens to clustering keys to rows
         # These are not actually partitions (or rather, there is one partition
         # for each token) and they aren't sorted.
         # But it is good enough if we don't care about performance;
         # and makes the code a lot simpler.
         self.data: Dict[int, Dict[Tuple, TRow]] = defaultdict(dict)
 
     def __repr__(self):
         return f"<__module__.Table[{self.row_class.__name__}] object>"
 
     def partition_key(self, row: Union[TRow, Dict[str, Any]]) -> Tuple:
         """Returns the partition key of a row (ie. the cells which get hashed
         into the token."""
         if isinstance(row, dict):
             row_d = row
         else:
             row_d = row.to_dict()
         return tuple(row_d[col] for col in self.row_class.PARTITION_KEY)
 
     def clustering_key(self, row: Union[TRow, Dict[str, Any]]) -> Tuple:
         """Returns the clustering key of a row (ie. the cells which are used
         for sorting rows within a partition."""
         if isinstance(row, dict):
             row_d = row
         else:
             row_d = row.to_dict()
         return tuple(row_d[col] for col in self.row_class.CLUSTERING_KEY)
 
     def primary_key(self, row):
         return self.partition_key(row) + self.clustering_key(row)
 
     def primary_key_from_dict(self, d: Dict[str, Any]) -> Tuple:
         """Returns the primary key (ie. concatenation of partition key and
         clustering key) of the given dictionary interpreted as a row."""
         return tuple(d[col] for col in self.primary_key_cols)
 
     def token(self, key: Tuple):
         """Returns the token of a row (ie. the hash of its partition key)."""
         return hash(key)
 
     def get_partition(self, token: int) -> Dict[Tuple, TRow]:
         """Returns the partition that contains this token."""
         return self.data[token]
 
     def insert(self, row: TRow):
         partition = self.data[self.token(self.partition_key(row))]
         partition[self.clustering_key(row)] = row
 
     def split_primary_key(self, key: Tuple) -> Tuple[Tuple, Tuple]:
         """Returns (partition_key, clustering_key) from a partition key"""
         assert len(key) == len(self.primary_key_cols)
 
         partition_key = key[0 : len(self.row_class.PARTITION_KEY)]
         clustering_key = key[len(self.row_class.PARTITION_KEY) :]
 
         return (partition_key, clustering_key)
 
     def get_from_partition_key(self, partition_key: Tuple) -> Iterable[TRow]:
         """Returns at most one row, from its partition key."""
         token = self.token(partition_key)
         for row in self.get_from_token(token):
             if self.partition_key(row) == partition_key:
                 yield row
 
     def get_from_primary_key(self, primary_key: Tuple) -> Optional[TRow]:
         """Returns at most one row, from its primary key."""
         (partition_key, clustering_key) = self.split_primary_key(primary_key)
 
         token = self.token(partition_key)
         partition = self.get_partition(token)
 
         return partition.get(clustering_key)
 
     def get_from_token(self, token: int) -> Iterable[TRow]:
         """Returns all rows whose token (ie. non-cryptographic hash of the
         partition key) is the one passed as argument."""
         return (v for (k, v) in sorted(self.get_partition(token).items()))
 
     def iter_all(self) -> Iterator[Tuple[Tuple, TRow]]:
         return (
             (self.primary_key(row), row)
             for (token, partition) in self.data.items()
             for (clustering_key, row) in partition.items()
         )
 
     def get_random(self) -> Optional[TRow]:
         return random.choice([row for (pk, row) in self.iter_all()])
 
 
 class InMemoryCqlRunner:
     def __init__(self):
         self._contents = Table(ContentRow)
         self._content_indexes = defaultdict(lambda: defaultdict(set))
         self._skipped_contents = Table(ContentRow)
         self._skipped_content_indexes = defaultdict(lambda: defaultdict(set))
         self._directories = Table(DirectoryRow)
         self._directory_entries = Table(DirectoryEntryRow)
         self._revisions = Table(RevisionRow)
         self._revision_parents = Table(RevisionParentRow)
         self._releases = Table(ReleaseRow)
         self._snapshots = Table(SnapshotRow)
         self._snapshot_branches = Table(SnapshotBranchRow)
         self._origins = Table(OriginRow)
         self._origin_visits = Table(OriginVisitRow)
         self._origin_visit_statuses = Table(OriginVisitStatusRow)
         self._metadata_authorities = Table(MetadataAuthorityRow)
         self._metadata_fetchers = Table(MetadataFetcherRow)
         self._raw_extrinsic_metadata = Table(RawExtrinsicMetadataRow)
         self._extid = Table(ExtIDRow)
         self._stat_counters = defaultdict(int)
 
     def increment_counter(self, object_type: str, nb: int):
         self._stat_counters[object_type] += nb
 
     def stat_counters(self) -> Iterable[ObjectCountRow]:
         for (object_type, count) in self._stat_counters.items():
             yield ObjectCountRow(partition_key=0, object_type=object_type, count=count)
 
     ##########################
     # 'content' table
     ##########################
 
     def _content_add_finalize(self, content: ContentRow) -> None:
         self._contents.insert(content)
         self.increment_counter("content", 1)
 
     def content_add_prepare(self, content: ContentRow):
         finalizer = functools.partial(self._content_add_finalize, content)
         return (self._contents.token(self._contents.partition_key(content)), finalizer)
 
     def content_get_from_pk(
         self, content_hashes: Dict[str, bytes]
     ) -> Optional[ContentRow]:
         primary_key = self._contents.primary_key_from_dict(content_hashes)
         return self._contents.get_from_primary_key(primary_key)
 
     def content_get_from_token(self, token: int) -> Iterable[ContentRow]:
         return self._contents.get_from_token(token)
 
     def content_get_random(self) -> Optional[ContentRow]:
         return self._contents.get_random()
 
     def content_get_token_range(
         self, start: int, end: int, limit: int,
     ) -> Iterable[Tuple[int, ContentRow]]:
         matches = [
             (token, row)
             for (token, partition) in self._contents.data.items()
             for (clustering_key, row) in partition.items()
             if start <= token <= end
         ]
         matches.sort()
         return matches[0:limit]
 
     ##########################
     # 'content_by_*' tables
     ##########################
 
     def content_missing_by_sha1_git(self, ids: List[bytes]) -> List[bytes]:
         missing = []
         for id_ in ids:
             if id_ not in self._content_indexes["sha1_git"]:
                 missing.append(id_)
         return missing
 
     def content_index_add_one(self, algo: str, content: Content, token: int) -> None:
         self._content_indexes[algo][content.get_hash(algo)].add(token)
 
     def content_get_tokens_from_single_hash(
         self, algo: str, hash_: bytes
     ) -> Iterable[int]:
         return self._content_indexes[algo][hash_]
 
     ##########################
     # 'skipped_content' table
     ##########################
 
     def _skipped_content_add_finalize(self, content: SkippedContentRow) -> None:
         self._skipped_contents.insert(content)
         self.increment_counter("skipped_content", 1)
 
     def skipped_content_add_prepare(self, content: SkippedContentRow):
         finalizer = functools.partial(self._skipped_content_add_finalize, content)
         return (
             self._skipped_contents.token(self._contents.partition_key(content)),
             finalizer,
         )
 
     def skipped_content_get_from_pk(
         self, content_hashes: Dict[str, bytes]
     ) -> Optional[SkippedContentRow]:
         primary_key = self._skipped_contents.primary_key_from_dict(content_hashes)
         return self._skipped_contents.get_from_primary_key(primary_key)
 
     def skipped_content_get_from_token(self, token: int) -> Iterable[SkippedContentRow]:
         return self._skipped_contents.get_from_token(token)
 
     ##########################
     # 'skipped_content_by_*' tables
     ##########################
 
     def skipped_content_index_add_one(
         self, algo: str, content: SkippedContent, token: int
     ) -> None:
         self._skipped_content_indexes[algo][content.get_hash(algo)].add(token)
 
     def skipped_content_get_tokens_from_single_hash(
         self, algo: str, hash_: bytes
     ) -> Iterable[int]:
         return self._skipped_content_indexes[algo][hash_]
 
     ##########################
     # 'directory' table
     ##########################
 
     def directory_missing(self, ids: List[bytes]) -> List[bytes]:
         missing = []
         for id_ in ids:
             if self._directories.get_from_primary_key((id_,)) is None:
                 missing.append(id_)
         return missing
 
     def directory_add_one(self, directory: DirectoryRow) -> None:
         self._directories.insert(directory)
         self.increment_counter("directory", 1)
 
     def directory_get_random(self) -> Optional[DirectoryRow]:
         return self._directories.get_random()
 
     ##########################
     # 'directory_entry' table
     ##########################
 
     def directory_entry_add_one(self, entry: DirectoryEntryRow) -> None:
         self._directory_entries.insert(entry)
 
     def directory_entry_get(
         self, directory_ids: List[Sha1Git]
     ) -> Iterable[DirectoryEntryRow]:
         for id_ in directory_ids:
             yield from self._directory_entries.get_from_partition_key((id_,))
 
     ##########################
     # 'revision' table
     ##########################
 
     def revision_missing(self, ids: List[bytes]) -> Iterable[bytes]:
         missing = []
         for id_ in ids:
             if self._revisions.get_from_primary_key((id_,)) is None:
                 missing.append(id_)
         return missing
 
     def revision_add_one(self, revision: RevisionRow) -> None:
         self._revisions.insert(revision)
         self.increment_counter("revision", 1)
 
     def revision_get_ids(self, revision_ids) -> Iterable[int]:
         for id_ in revision_ids:
             if self._revisions.get_from_primary_key((id_,)) is not None:
                 yield id_
 
     def revision_get(self, revision_ids: List[Sha1Git]) -> Iterable[RevisionRow]:
         for id_ in revision_ids:
             row = self._revisions.get_from_primary_key((id_,))
             if row:
                 yield row
 
     def revision_get_random(self) -> Optional[RevisionRow]:
         return self._revisions.get_random()
 
     ##########################
     # 'revision_parent' table
     ##########################
 
     def revision_parent_add_one(self, revision_parent: RevisionParentRow) -> None:
         self._revision_parents.insert(revision_parent)
 
     def revision_parent_get(self, revision_id: Sha1Git) -> Iterable[bytes]:
         for parent in self._revision_parents.get_from_partition_key((revision_id,)):
             yield parent.parent_id
 
     ##########################
     # 'release' table
     ##########################
 
     def release_missing(self, ids: List[bytes]) -> List[bytes]:
         missing = []
         for id_ in ids:
             if self._releases.get_from_primary_key((id_,)) is None:
                 missing.append(id_)
         return missing
 
     def release_add_one(self, release: ReleaseRow) -> None:
         self._releases.insert(release)
         self.increment_counter("release", 1)
 
     def release_get(self, release_ids: List[str]) -> Iterable[ReleaseRow]:
         for id_ in release_ids:
             row = self._releases.get_from_primary_key((id_,))
             if row:
                 yield row
 
     def release_get_random(self) -> Optional[ReleaseRow]:
         return self._releases.get_random()
 
     ##########################
     # 'snapshot' table
     ##########################
 
     def snapshot_missing(self, ids: List[bytes]) -> List[bytes]:
         missing = []
         for id_ in ids:
             if self._snapshots.get_from_primary_key((id_,)) is None:
                 missing.append(id_)
         return missing
 
     def snapshot_add_one(self, snapshot: SnapshotRow) -> None:
         self._snapshots.insert(snapshot)
         self.increment_counter("snapshot", 1)
 
     def snapshot_get_random(self) -> Optional[SnapshotRow]:
         return self._snapshots.get_random()
 
     ##########################
     # 'snapshot_branch' table
     ##########################
 
     def snapshot_branch_add_one(self, branch: SnapshotBranchRow) -> None:
         self._snapshot_branches.insert(branch)
 
     def snapshot_count_branches(
         self, snapshot_id: Sha1Git, branch_name_exclude_prefix: Optional[bytes] = None,
     ) -> Dict[Optional[str], int]:
         """Returns a dictionary from type names to the number of branches
         of that type."""
         counts: Dict[Optional[str], int] = defaultdict(int)
         for branch in self._snapshot_branches.get_from_partition_key((snapshot_id,)):
             if branch_name_exclude_prefix and branch.name.startswith(
                 branch_name_exclude_prefix
             ):
                 continue
             if branch.target_type is None:
                 target_type = None
             else:
                 target_type = branch.target_type
             counts[target_type] += 1
         return counts
 
     def snapshot_branch_get(
         self,
         snapshot_id: Sha1Git,
         from_: bytes,
         limit: int,
         branch_name_exclude_prefix: Optional[bytes] = None,
     ) -> Iterable[SnapshotBranchRow]:
         count = 0
         for branch in self._snapshot_branches.get_from_partition_key((snapshot_id,)):
             prefix = branch_name_exclude_prefix
             if branch.name >= from_ and (
                 prefix is None or not branch.name.startswith(prefix)
             ):
                 count += 1
                 yield branch
             if count >= limit:
                 break
 
     ##########################
     # 'origin' table
     ##########################
 
     def origin_add_one(self, origin: OriginRow) -> None:
         self._origins.insert(origin)
         self.increment_counter("origin", 1)
 
     def origin_get_by_sha1(self, sha1: bytes) -> Iterable[OriginRow]:
         return self._origins.get_from_partition_key((sha1,))
 
     def origin_get_by_url(self, url: str) -> Iterable[OriginRow]:
         return self.origin_get_by_sha1(origin_url_to_sha1(url))
 
     def origin_list(
         self, start_token: int, limit: int
     ) -> Iterable[Tuple[int, OriginRow]]:
         """Returns an iterable of (token, origin)"""
         matches = [
             (token, row)
             for (token, partition) in self._origins.data.items()
             for (clustering_key, row) in partition.items()
             if token >= start_token
         ]
         matches.sort()
         return matches[0:limit]
 
     def origin_iter_all(self) -> Iterable[OriginRow]:
         return (
             row
             for (token, partition) in self._origins.data.items()
             for (clustering_key, row) in partition.items()
         )
 
     def origin_generate_unique_visit_id(self, origin_url: str) -> int:
         origin = list(self.origin_get_by_url(origin_url))[0]
         visit_id = origin.next_visit_id
         origin.next_visit_id += 1
         return visit_id
 
     ##########################
     # 'origin_visit' table
     ##########################
 
     def origin_visit_get(
         self, origin_url: str, last_visit: Optional[int], limit: int, order: ListOrder,
     ) -> Iterable[OriginVisitRow]:
         visits = list(self._origin_visits.get_from_partition_key((origin_url,)))
 
         if last_visit is not None:
             if order == ListOrder.ASC:
                 visits = [v for v in visits if v.visit > last_visit]
             else:
                 visits = [v for v in visits if v.visit < last_visit]
 
         visits.sort(key=lambda v: v.visit, reverse=order == ListOrder.DESC)
 
         visits = visits[0:limit]
 
         return visits
 
     def origin_visit_add_one(self, visit: OriginVisitRow) -> None:
         self._origin_visits.insert(visit)
         self.increment_counter("origin_visit", 1)
 
     def origin_visit_get_one(
         self, origin_url: str, visit_id: int
     ) -> Optional[OriginVisitRow]:
         return self._origin_visits.get_from_primary_key((origin_url, visit_id))
 
     def origin_visit_get_all(self, origin_url: str) -> Iterable[OriginVisitRow]:
         return self._origin_visits.get_from_partition_key((origin_url,))
 
     def origin_visit_iter(self, start_token: int) -> Iterator[OriginVisitRow]:
         """Returns all origin visits in order from this token,
         and wraps around the token space."""
         return (
             row
             for (token, partition) in self._origin_visits.data.items()
             for (clustering_key, row) in partition.items()
         )
 
     ##########################
     # 'origin_visit_status' table
     ##########################
 
     def origin_visit_status_get_range(
         self,
         origin: str,
         visit: int,
         date_from: Optional[datetime.datetime],
         limit: int,
         order: ListOrder,
     ) -> Iterable[OriginVisitStatusRow]:
         statuses = list(self.origin_visit_status_get(origin, visit))
 
         if date_from is not None:
             if order == ListOrder.ASC:
                 statuses = [s for s in statuses if s.date >= date_from]
             else:
                 statuses = [s for s in statuses if s.date <= date_from]
 
         statuses.sort(key=lambda s: s.date, reverse=order == ListOrder.DESC)
 
         return statuses[0:limit]
 
     def origin_visit_status_add_one(self, visit_update: OriginVisitStatusRow) -> None:
         self._origin_visit_statuses.insert(visit_update)
         self.increment_counter("origin_visit_status", 1)
 
     def origin_visit_status_get_latest(
         self, origin: str, visit: int,
     ) -> Optional[OriginVisitStatusRow]:
         """Given an origin visit id, return its latest origin_visit_status
 
          """
         return next(self.origin_visit_status_get(origin, visit), None)
 
     def origin_visit_status_get(
         self, origin: str, visit: int,
     ) -> Iterator[OriginVisitStatusRow]:
         """Return all origin visit statuses for a given visit
 
         """
         statuses = [
             s
             for s in self._origin_visit_statuses.get_from_partition_key((origin,))
             if s.visit == visit
         ]
         statuses.sort(key=lambda s: s.date, reverse=True)
         return iter(statuses)
 
     ##########################
     # 'metadata_authority' table
     ##########################
 
     def metadata_authority_add(self, authority: MetadataAuthorityRow):
         self._metadata_authorities.insert(authority)
         self.increment_counter("metadata_authority", 1)
 
     def metadata_authority_get(self, type, url) -> Optional[MetadataAuthorityRow]:
         return self._metadata_authorities.get_from_primary_key((url, type))
 
     ##########################
     # 'metadata_fetcher' table
     ##########################
 
     def metadata_fetcher_add(self, fetcher: MetadataFetcherRow):
         self._metadata_fetchers.insert(fetcher)
         self.increment_counter("metadata_fetcher", 1)
 
     def metadata_fetcher_get(self, name, version) -> Optional[MetadataAuthorityRow]:
         return self._metadata_fetchers.get_from_primary_key((name, version))
 
     #########################
     # 'raw_extrinsic_metadata' table
     #########################
 
     def raw_extrinsic_metadata_add(self, raw_extrinsic_metadata):
         self._raw_extrinsic_metadata.insert(raw_extrinsic_metadata)
         self.increment_counter("raw_extrinsic_metadata", 1)
 
     def raw_extrinsic_metadata_get_after_date(
         self,
         target: str,
         authority_type: str,
         authority_url: str,
         after: datetime.datetime,
     ) -> Iterable[RawExtrinsicMetadataRow]:
         metadata = self.raw_extrinsic_metadata_get(
             target, authority_type, authority_url
         )
         return (m for m in metadata if m.discovery_date > after)
 
     def raw_extrinsic_metadata_get_after_date_and_id(
         self,
         target: str,
         authority_type: str,
         authority_url: str,
         after_date: datetime.datetime,
         after_id: bytes,
     ) -> Iterable[RawExtrinsicMetadataRow]:
         metadata = self._raw_extrinsic_metadata.get_from_partition_key((target,))
         after_tuple = (after_date, after_id)
         return (
             m
             for m in metadata
             if m.authority_type == authority_type
             and m.authority_url == authority_url
             and (m.discovery_date, m.id) > after_tuple
         )
 
     def raw_extrinsic_metadata_get(
         self, target: str, authority_type: str, authority_url: str
     ) -> Iterable[RawExtrinsicMetadataRow]:
         metadata = self._raw_extrinsic_metadata.get_from_partition_key((target,))
         return (
             m
             for m in metadata
             if m.authority_type == authority_type and m.authority_url == authority_url
         )
 
     #########################
     # 'extid' table
     #########################
     def _extid_add_finalize(self, extid: ExtIDRow) -> None:
         self._extid.insert(extid)
         self.increment_counter("extid", 1)
 
     def extid_add_prepare(self, extid: ExtIDRow):
         finalizer = functools.partial(self._extid_add_finalize, extid)
         return (self._extid.token(self._extid.partition_key(extid)), finalizer)
 
-    def extid_index_add_one(self, extid: ExtIDRow, token: int) -> None:
+    def extid_index_add_one(self, row: ExtIDByTargetRow) -> None:
         pass
 
     def extid_get_from_pk(
         self, extid_type: str, extid: bytes, target: ExtendedSWHID,
     ) -> Optional[ExtIDRow]:
         primary_key = self._extid.primary_key_from_dict(
             dict(
                 extid_type=extid_type,
                 extid=extid,
                 target_type=target.object_type.value,
                 target=target.object_id,
             )
         )
         return self._extid.get_from_primary_key(primary_key)
 
     def extid_get_from_extid(self, extid_type: str, extid: bytes) -> Iterable[ExtIDRow]:
         return (
             row
             for pk, row in self._extid.iter_all()
             if row.extid_type == extid_type and row.extid == extid
         )
 
     def extid_get_from_target(
         self, target_type: str, target: bytes
     ) -> Iterable[ExtIDRow]:
         return (
             row
             for pk, row in self._extid.iter_all()
             if row.target_type == target_type and row.target == target
         )
 
 
 class InMemoryStorage(CassandraStorage):
     _cql_runner: InMemoryCqlRunner  # type: ignore
 
     def __init__(self, journal_writer=None):
         self.reset()
         self.journal_writer = JournalWriter(journal_writer)
 
     def reset(self):
         self._cql_runner = InMemoryCqlRunner()
         self.objstorage = ObjStorage({"cls": "memory"})
 
     def check_config(self, *, check_write: bool) -> bool:
         return True