[ML] Fix double-counting of inference memory in the assignment rebalancer #133919
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
elasticsearchmachine added needs:triage 
valeriy42 added >bug :ml 
| Hi @valeriy42, I've created a changelog YAML for you. |
| Pinging @elastic/ml-core (Team:ML) |
valeriy42 added v9.1.4 v8.18.7 v8.19.4 auto-backport 
jan-elastic reviewed Sep 1, 2025
| // We subtract native inference memory as the planner expects available memory for | ||
| // native inference including current assignments. | ||
| getNodeFreeMemoryExcludingPerNodeOverheadAndNativeInference(load), | ||
| load.getFreeMemoryExcludingPerNodeOverhead(), |
There was a problem hiding this comment.
I don't understand the method name getFreeMemoryExcludingPerNodeOverhead.
Shouldn't that just be getFreeMemory()?
There was a problem hiding this comment.
It seems that those numbers should be the same most of the time.
Maybe there is a corner case where the ML node is used only for inference in Java code, so that we don't account for the 30MB of the native code overhead, and NodeLoad.assignedNativeCodeOverheadMemory is 0.
This was referenced Sep 3, 2025
Merged
Merged
valeriy42 added a commit to valeriy42/elasticsearch that referenced this pull request Sep 3, 2025
…ncer (elastic#133919) The static method TrainedModelAssignmentRebalancer.getNodeFreeMemoryExcludingPerNodeOverheadAndNativeInference was used to subtract load.getAssignedNativeInferenceMemory() from load.getFreeMemoryExcludingPerNodeOverhead(). However, in NodeLoad.getFreeMemoryExcludingPerNodeOverhead(), native inference memory was already subtracted as part of the getAssignedJobMemoryExcludingPerNodeOverhead() calculation. This led to double-counting of the native inference memory. Avoiding this double-counting allows us to remove the private method getNodeFreeMemoryExcludingPerNodeOverheadAndNativeInference() entirely.
valeriy42 added a commit to valeriy42/elasticsearch that referenced this pull request Sep 3, 2025
…ncer (elastic#133919) The static method TrainedModelAssignmentRebalancer.getNodeFreeMemoryExcludingPerNodeOverheadAndNativeInference was used to subtract load.getAssignedNativeInferenceMemory() from load.getFreeMemoryExcludingPerNodeOverhead(). However, in NodeLoad.getFreeMemoryExcludingPerNodeOverhead(), native inference memory was already subtracted as part of the getAssignedJobMemoryExcludingPerNodeOverhead() calculation. This led to double-counting of the native inference memory. Avoiding this double-counting allows us to remove the private method getNodeFreeMemoryExcludingPerNodeOverheadAndNativeInference() entirely.
valeriy42 added a commit to valeriy42/elasticsearch that referenced this pull request Sep 3, 2025
…ncer (elastic#133919) The static method TrainedModelAssignmentRebalancer.getNodeFreeMemoryExcludingPerNodeOverheadAndNativeInference was used to subtract load.getAssignedNativeInferenceMemory() from load.getFreeMemoryExcludingPerNodeOverhead(). However, in NodeLoad.getFreeMemoryExcludingPerNodeOverhead(), native inference memory was already subtracted as part of the getAssignedJobMemoryExcludingPerNodeOverhead() calculation. This led to double-counting of the native inference memory. Avoiding this double-counting allows us to remove the private method getNodeFreeMemoryExcludingPerNodeOverheadAndNativeInference() entirely.
valeriy42 added a commit to valeriy42/elasticsearch that referenced this pull request Sep 3, 2025
…ncer (elastic#133919) The static method TrainedModelAssignmentRebalancer.getNodeFreeMemoryExcludingPerNodeOverheadAndNativeInference was used to subtract load.getAssignedNativeInferenceMemory() from load.getFreeMemoryExcludingPerNodeOverhead(). However, in NodeLoad.getFreeMemoryExcludingPerNodeOverhead(), native inference memory was already subtracted as part of the getAssignedJobMemoryExcludingPerNodeOverhead() calculation. This led to double-counting of the native inference memory. Avoiding this double-counting allows us to remove the private method getNodeFreeMemoryExcludingPerNodeOverheadAndNativeInference() entirely.
elasticsearchmachine pushed a commit that referenced this pull request Sep 3, 2025
…ncer (#133919) (#134053) The static method TrainedModelAssignmentRebalancer.getNodeFreeMemoryExcludingPerNodeOverheadAndNativeInference was used to subtract load.getAssignedNativeInferenceMemory() from load.getFreeMemoryExcludingPerNodeOverhead(). However, in NodeLoad.getFreeMemoryExcludingPerNodeOverhead(), native inference memory was already subtracted as part of the getAssignedJobMemoryExcludingPerNodeOverhead() calculation. This led to double-counting of the native inference memory. Avoiding this double-counting allows us to remove the private method getNodeFreeMemoryExcludingPerNodeOverheadAndNativeInference() entirely.
elasticsearchmachine pushed a commit that referenced this pull request Sep 3, 2025
…ncer (#133919) (#134052) The static method TrainedModelAssignmentRebalancer.getNodeFreeMemoryExcludingPerNodeOverheadAndNativeInference was used to subtract load.getAssignedNativeInferenceMemory() from load.getFreeMemoryExcludingPerNodeOverhead(). However, in NodeLoad.getFreeMemoryExcludingPerNodeOverhead(), native inference memory was already subtracted as part of the getAssignedJobMemoryExcludingPerNodeOverhead() calculation. This led to double-counting of the native inference memory. Avoiding this double-counting allows us to remove the private method getNodeFreeMemoryExcludingPerNodeOverheadAndNativeInference() entirely.
elasticsearchmachine pushed a commit that referenced this pull request Sep 3, 2025
…ncer (#133919) (#134051) The static method TrainedModelAssignmentRebalancer.getNodeFreeMemoryExcludingPerNodeOverheadAndNativeInference was used to subtract load.getAssignedNativeInferenceMemory() from load.getFreeMemoryExcludingPerNodeOverhead(). However, in NodeLoad.getFreeMemoryExcludingPerNodeOverhead(), native inference memory was already subtracted as part of the getAssignedJobMemoryExcludingPerNodeOverhead() calculation. This led to double-counting of the native inference memory. Avoiding this double-counting allows us to remove the private method getNodeFreeMemoryExcludingPerNodeOverheadAndNativeInference() entirely.
elasticsearchmachine pushed a commit that referenced this pull request Sep 8, 2025
…ncer (#133919) (#134054) The static method TrainedModelAssignmentRebalancer.getNodeFreeMemoryExcludingPerNodeOverheadAndNativeInference was used to subtract load.getAssignedNativeInferenceMemory() from load.getFreeMemoryExcludingPerNodeOverhead(). However, in NodeLoad.getFreeMemoryExcludingPerNodeOverhead(), native inference memory was already subtracted as part of the getAssignedJobMemoryExcludingPerNodeOverhead() calculation. This led to double-counting of the native inference memory. Avoiding this double-counting allows us to remove the private method getNodeFreeMemoryExcludingPerNodeOverheadAndNativeInference() entirely.
sarog pushed a commit to portsbuild/elasticsearch that referenced this pull request Sep 11, 2025
…ncer (elastic#133919) (elastic#134054) The static method TrainedModelAssignmentRebalancer.getNodeFreeMemoryExcludingPerNodeOverheadAndNativeInference was used to subtract load.getAssignedNativeInferenceMemory() from load.getFreeMemoryExcludingPerNodeOverhead(). However, in NodeLoad.getFreeMemoryExcludingPerNodeOverhead(), native inference memory was already subtracted as part of the getAssignedJobMemoryExcludingPerNodeOverhead() calculation. This led to double-counting of the native inference memory. Avoiding this double-counting allows us to remove the private method getNodeFreeMemoryExcludingPerNodeOverheadAndNativeInference() entirely.
sarog pushed a commit to portsbuild/elasticsearch that referenced this pull request Sep 19, 2025
…ncer (elastic#133919) (elastic#134054) The static method TrainedModelAssignmentRebalancer.getNodeFreeMemoryExcludingPerNodeOverheadAndNativeInference was used to subtract load.getAssignedNativeInferenceMemory() from load.getFreeMemoryExcludingPerNodeOverhead(). However, in NodeLoad.getFreeMemoryExcludingPerNodeOverhead(), native inference memory was already subtracted as part of the getAssignedJobMemoryExcludingPerNodeOverhead() calculation. This led to double-counting of the native inference memory. Avoiding this double-counting allows us to remove the private method getNodeFreeMemoryExcludingPerNodeOverheadAndNativeInference() entirely.
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
4 participants
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.
The static method
TrainedModelAssignmentRebalancer.getNodeFreeMemoryExcludingPerNodeOverheadAndNativeInferencewas used to subtractload.getAssignedNativeInferenceMemory()fromload.getFreeMemoryExcludingPerNodeOverhead(). However, inNodeLoad.getFreeMemoryExcludingPerNodeOverhead(), native inference memory was already subtracted as part of thegetAssignedJobMemoryExcludingPerNodeOverhead()calculation.This led to double-counting of the native inference memory. Avoiding this double-counting allows us to remove the private method
getNodeFreeMemoryExcludingPerNodeOverheadAndNativeInference()entirely.