[ML] Fix test failure updating model deployment with stale cluster state. #128667
+16
−65
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
davidkyle
added
>test
|
Pinging @elastic/ml-core (Team:ML) |
davidkyle
commented
May 30, 2025
| @@ -80,9 +78,6 @@ public class TrainedModelAssignmentClusterService implements ClusterStateListene | |||
|
|
|||
| private static final Logger logger = LogManager.getLogger(TrainedModelAssignmentClusterService.class); | |||
|
|
|||
| private static final TransportVersion RENAME_ALLOCATION_TO_ASSIGNMENT_TRANSPORT_VERSION = TransportVersions.V_8_3_0; | |||
There was a problem hiding this comment.
These version checks are redundant in 9.0 and 9.1. The 8.x backports will need to keep them however.
| ActionListener<ClusterState> updatedStateListener = ActionListener.wrap( | ||
| updatedState -> submitUnbatchedTask("update model deployment", new ClusterStateUpdateTask() { | ||
| ActionListener<TrainedModelAssignmentMetadata.Builder> updatedAssignmentListener = ActionListener.wrap( | ||
| updatedAssignment -> submitUnbatchedTask("update model deployment", new ClusterStateUpdateTask() { |
There was a problem hiding this comment.
This is the fix, here the new assignment state is passed rather than the updated cluster state.
| @@ -375,20 +375,17 @@ public void clusterChanged(ClusterChangedEvent event) { | |||
| final boolean isResetMode = MlMetadata.getMlMetadata(event.state()).isResetMode(); | |||
| TrainedModelAssignmentMetadata modelAssignmentMetadata = TrainedModelAssignmentMetadata.fromState(event.state()); | |||
| final String currentNode = event.state().nodes().getLocalNodeId(); | |||
| final boolean isNewAllocationSupported = event.state() | |||
There was a problem hiding this comment.
Another version change that is irrelevant for 9
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
When updating a model deployment - changing the number of allocations for example- calculating the new state is an expensive operation so it is done outside of the ClusterStateUpdateTask. However, if there was another clusterstate update while computing the update then the submitted update fails due to an out of data cluster state.
The fix is quite easy, compute the model deployment update outside of the ClusterStateUpdateTask then merge it with the latest state when executing the task. The code already has a check that the deployment update is compatible with the new state (
areClusterStatesCompatibleForRebalance(...)) making it safe to merge the new state.Closes #121165