-
Type: Task
-
Resolution: Unresolved
-
Priority: Major - P3
-
None
-
Affects Version/s: None
-
Component/s: None
-
None
-
Cluster Scalability
-
Cluster Scalability 2023-12-25, Cluster Scalability 2024-1-8, Cluster Scalability 2024-1-22, Cluster Scalability 2024-2-5, Cluster Scalability 2024-2-19, Cluster Scalability 2024-3-4, Cluster Scalability 2024-3-18, Cluster Scalability 2024-4-1, Cluster Scalability 2024-4-15, Cluster Scalability 2024-4-29, Cluster Scalability 2024-5-13, Cluster Scalability 2024-5-27, Cluster Scalability 2024-6-10, Cluster Scalability 06/24/24, Cluster Scalability 2024-07-08
The db hash and validation check for a ReplSetTest involves running listCollections and collStats commands against each mongod while it is fsync locked. On a mongod with replica set endpoint enabled, those collStats commands are handled using the router code paths which may trigger a catatlog cache refresh. The router service on mongod currently uses the ShardServerCatalogCacheLoader (SERVER-82588). So when a refresh occurs, it requires doing a noop write which will then deadlock since the mongod is fsync locked. This deadlock should go away after SERVER-82588. For now, SERVER-81968 has temporarily changed the number of nodes in the replica sets in the noPassthrough/rs_endpoint/ tests with user collections to 1 to avoid the db hash check.
The CheckReplHash hook in replica set endpoint suites appeared to suffer from the same hang issue so it has been disabled from all those suites.
- depends on
-
SERVER-84243 Dedicate a catalog cache and loader to the shard role
- In Progress
-
SERVER-82588 Move CatalogCacheLoader to be a decoration on Service
- Closed