Loading...

XML

Word

Printable

JSON

We have to consider multiple types of deletes in sharded clusters:

Two-phase write-without-shard-key (deletes with no shard key, multi:false). Mongos broadcasts a read phase then sends a targeted _clusterWriteWithoutShardKey internal command to one shard. The shard executes the delete.
Scatter-gather deletes (no shard key, multi:true). Mongos fans out to all shards. Each shard runs performDelete().
Route to one shard (shard key). Mongos can extract the shard key value from the filter, compute the exact chunk it belongs to, and route to exactly one shard. This is the fast path.
Direct-to-shard deletes (only with directShardOperations permissions). I don't think we expect users to do this. Clients connecting directly to a shard (bypassing mongos) with directShardOperations privilege.

depends on

SERVER-122064 Attach metrics to delete response on shards for query stats

SERVER-122074 Shapify an delete command on mongos

is depended on by

SERVER-122080 Support query stats for deletes on mongos for legacy batch write exec

is related to

SERVER-122076 Collect query stats for insert in sharded

SERVER-112050 Collect query stats for update in sharded

SERVER-115142 Shapify an update command on mongos

SERVER-118859 Don't include queryShapeHash for mongos for batched updates

related to

SERVER-128278 DeleteStage docsFetched can return more documents than in the collection

(2 is related to, 1 related to)