Currently, multi-writes in sharding attach shardVersion: IGNORED for multi-writes, which need to be broadcast to more than one shard. Shards, upon receiving shard version IGNORED, unconditionally treat it as "a match" and allow the operation to proceed without triggering a refresh.
In combination with SERVER-32198, this causes the side effect that if a shard is restarted, multi-writes writes will not trigger a refresh on the shard and the shard will treat the collection as unsharded, thus not writing the shard key to the oplog entries. This in turn impacts any change streams on that collection.
- depends on
-
SERVER-45100 Make the BatchWriteExecutor retry multi-writes only against unsuccessful shards
- Closed