No need to log unassigned shards on master shutdown #136149
Conversation
When the master is shutting down, the project is likely being deleted. It is expected shards can become unassigned during this process. There is no need to log them. Relates: ES-12842
|
Pinging @elastic/es-distributed-coordination (Team:Distributed Coordination) |
| if (clusterState.metadata().nodeShutdowns().contains(clusterState.nodes().getLocalNodeId())) { | ||
| // Master is shutting down, the project is likely being deleted. Shards can be unassigned and no need to report | ||
| return; | ||
| } |
There was a problem hiding this comment.
Logging on project deletion is one false postive that I noticed when enabling the logging in QA.
The other false positve is during project creation, e.g. a search shard can be unassigned because search node has not joined yet. This is not addressed in this PR. I don't have a great solution other than some time-based delay, i.e. no logging in the first 5min after node start.
I'd like to fix the issue on project deletion first since it is the more noisy case. I also added input index in the log which might be useful as a hint to decide whether to log or not for project creation if we observe some valid pattern in QA.
…nassigned-during-shutdown
|
Added a test in 5c9ac13. Thanks for the suggestion. |
When the master is shutting down, the project is likely being deleted. It is expected shards can become unassigned during this process. There is no need to log them.
Relates: ES-12842