-
Type:
Bug
-
Resolution: Fixed
-
Priority:
Major - P3
-
Affects Version/s: None
-
Component/s: None
-
Catalog and Routing
-
Fully Compatible
-
ALL
-
CAR Team 2025-01-20, CAR Team 2025-02-03, CAR Team 2025-02-17
The issue is found through aggregation_unsplittable_collections_on_random_shard_passthrough_with_config_transitions.yml suite with createUnshardedCollectionRandomizeDataShard failpoint.
More details can be found in https://jira.mongodb.org/browse/SERVER-98123
When a config shard is migrated to a dedicated server via
transitionFromDedicatedConfigServer admin command, it possible that after the process started, the config shard will be still available in the shard registry and can be received via shardRegistry->getAllShardIds(opCtx).
It's possible to send a createCollection request to this "inProgress" shard and there could be a race, when this collection will be created after it becomes a dedicated server.
In this case the will be a dedicated config server with a customer collection.
Another problem, that this dedicated server cannot be migrated back to the config shard, as there will be a conflict, that the same database exists already on other shards.
Possible areas to investigate:
- ShardRegistry is not updated after the transition process is started
- When a transition finished, any customer command like createCollection has to be not allowed.
- related to
-
SERVER-98123 Enable failpoint in aggregation_unsplittable_collections_on_random_shard_passthrough_with_config_transitions.yml
-
- Closed
-