Make sure media hashes are not queried until the index is up (#18302)

changelog.d/18302.feature

@@ -0,0 +1 @@
+Hashes of media files are now tracked by Synapse. Media quarantines will now apply to all files with the same hash.

synapse/storage/databases/main/media_repository.py

@@ -999,6 +999,15 @@ class MediaRepositoryStore(MediaRepositoryBackgroundUpdateStore):
             None if the media_id doesn't exist.
         """
 
+        # If we don't have the index yet, performance tanks, so we return False.
+        # In the background updates, remote_media_cache_sha256_idx is created
+        # after local_media_repository_sha256_idx, which is why we only need to
+        # check for the completion of the former.
+        if not await self.db_pool.updates.has_completed_background_update(
+            "remote_media_cache_sha256_idx"
+        ):
+            return False
+
         def get_matching_media_txn(
             txn: LoggingTransaction, table: str, sha256: str
         ) -> bool:

synapse/storage/schema/main/delta/91/01_media_hash.sql

@@ -16,6 +16,13 @@ ALTER TABLE local_media_repository ADD COLUMN sha256 TEXT;
 ALTER TABLE remote_media_cache ADD COLUMN sha256 TEXT;
 
 -- Add a background updates to handle creating the new index.
-INSERT INTO background_updates (ordering, update_name, progress_json) VALUES
-    (9101, 'local_media_repository_sha256_idx', '{}'),
-    (9101, 'remote_media_cache_sha256_idx', '{}');
+--
+-- Note that the ordering of the update is not following the usual scheme. This
+-- is because when upgrading from Synapse 1.127, this index is fairly important
+-- to have up quickly, so that it doesn't tank performance, which is why it is
+-- scheduled before other background updates in the 1.127 -> 1.128 upgrade
+INSERT INTO
+  background_updates (ordering, update_name, progress_json)
+VALUES
+  (8890, 'local_media_repository_sha256_idx', '{}'),
+  (8891, 'remote_media_cache_sha256_idx', '{}');