[ML] Fix Non-Deterministic Training Set Selection in RegressionIT testTwoJobsWithSameRandomizeSeedUseSameTrainingSet #138063
+3
−3
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
valeriy42
commented
Nov 14, 2025
|
|
||
| public void testTwoJobsWithSameRandomizeSeedUseSameTrainingSet() throws Exception { | ||
| String sourceIndex = "regression_two_jobs_with_same_randomize_seed_source"; | ||
| indexData(sourceIndex, 100, 0); |
Contributor
Author
There was a problem hiding this comment.
indexData is calling directly client().admin().indices().prepareCreate() instead of prepareCreate() of the test framework. This ensures that the index has always 1 shard.
However, it still can have multiple segments which then leads to non-deterministics order in which the reservoir sampling might get the documents. Hence, we need to fix both the shards and the segments to 1.
Collaborator
|
Pinging @elastic/ml-core (Team:ML) |
weizijun
added a commit
to weizijun/elasticsearch
that referenced
this pull request
Nov 16, 2025
* main: (135 commits)
Mute org.elasticsearch.upgrades.IndexSortUpgradeIT testIndexSortForNumericTypes {upgradedNodes=1} elastic#138130
Mute org.elasticsearch.upgrades.IndexSortUpgradeIT testIndexSortForNumericTypes {upgradedNodes=2} elastic#138129
Mute org.elasticsearch.search.basic.SearchWithRandomDisconnectsIT testSearchWithRandomDisconnects elastic#138128
[DiskBBQ] avoid EsAcceptDocs bug by calling cost before building iterator (elastic#138127)
Log NOT_PREFERRED shard movements (elastic#138069)
Improve bulk loading of binary doc values (elastic#137995)
Add internal action for getting inference fields and inference results for those fields (elastic#137680)
Address issue with DateFieldMapper#isFieldWithinQuery(...) (elastic#138032)
WriteLoadConstraintDecider: Have separate rate limiting for canRemain and canAllocate decisions (elastic#138067)
Adding NodeContext to TransportBroadcastByNodeAction (elastic#138057)
Mute org.elasticsearch.simdvec.ESVectorUtilTests testSoarDistanceBulk elastic#138117
Mute org.elasticsearch.xpack.esql.qa.single_node.GenerativeIT test elastic#137909
Backport batched_response_might_include_reduction_failure version to 8.19 (elastic#138046)
Add summary metrics for tdigest fields (elastic#137982)
Add gp-llm-v2 model ID and inference endpoint (elastic#138045)
Various tracing fixes (elastic#137908)
[ML] Fixing KDE evaluate() to return correct ValueAndMagnitude object (elastic#128602)
Mute org.elasticsearch.xpack.shutdown.NodeShutdownIT testStalledShardMigrationProperlyDetected elastic#115697
[ML] Fix Flaky Audit Message Assertion in testWithDatastream for RegressionIT and ClassificationIT (elastic#138065)
[ML] Fix Non-Deterministic Training Set Selection in RegressionIT testTwoJobsWithSameRandomizeSeedUseSameTrainingSet (elastic#138063)
...
# Conflicts:
# rest-api-spec/src/yamlRestTest/resources/rest-api-spec/test/search.vectors/200_dense_vector_docvalue_fields.yml
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
3 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 test
testTwoJobsWithSameRandomizeSeedUseSameTrainingSetfails intermittently because documents may be processed in different orders during reindexing. Since we use an online reservoir sampling algorithm, this order actually matters. To ensure deterministic reindexing of the document sequence, both the number of shards and the number of segments must be 1.This PR fixes the test by creating the source index with only 1 segment. This ensures deterministic document order during reindexing, resulting in consistent ID assignments and training set selection when using the same seed.
Fixes #117805