What happened?
The Milvus RAG integration tests fail consistently on the beam_PreCommit_Python_ML workflow (observed on Python 3.13, on both ubuntu-latest and self-hosted runners). This does not look like an intermittent flake: the last 4 completed runs on master all failed.
Failing suites:
apache_beam.ml.rag.enrichment.milvus_search_it_test.TestMilvusSearchEnrichment (e.g. test_hybrid_search, test_empty_input_chunks, test_filtered_search_with_cosine_similarity_and_batching, test_invalid_query_on_non_existent_collection, ...)apache_beam.ml.rag.ingestion.milvus_search_it_test.TestMilvusVectorWriterConfig (e.g. test_idempotent_write, test_write_with_batching, ...)
Several of the failures are 10-minute timeouts, which suggests the Milvus testcontainer fails to come up (or is unreachable) rather than an assertion-level regression.
Recent failing runs on master:
Also reproduced on unrelated PR runs, e.g. https://github.com/apache/beam/actions/runs/27417091271 (PR #38493, which touches no ML code).
Related: #31285 tracks general flakiness of this job; filing this separately since the Milvus suites currently fail on every run.
Issue Priority
Priority: 2 (default / most failing tests should be filed as P2)
Issue Components
What happened?
The Milvus RAG integration tests fail consistently on the
beam_PreCommit_Python_MLworkflow (observed on Python 3.13, on bothubuntu-latestand self-hosted runners). This does not look like an intermittent flake: the last 4 completed runs onmasterall failed.Failing suites:
apache_beam.ml.rag.enrichment.milvus_search_it_test.TestMilvusSearchEnrichment(e.g.test_hybrid_search,test_empty_input_chunks,test_filtered_search_with_cosine_similarity_and_batching,test_invalid_query_on_non_existent_collection, ...)apache_beam.ml.rag.ingestion.milvus_search_it_test.TestMilvusVectorWriterConfig(e.g.test_idempotent_write,test_write_with_batching, ...)Several of the failures are 10-minute timeouts, which suggests the Milvus testcontainer fails to come up (or is unreachable) rather than an assertion-level regression.
Recent failing runs on
master:Also reproduced on unrelated PR runs, e.g. https://github.com/apache/beam/actions/runs/27417091271 (PR #38493, which touches no ML code).
Related: #31285 tracks general flakiness of this job; filing this separately since the Milvus suites currently fail on every run.
Issue Priority
Priority: 2 (default / most failing tests should be filed as P2)
Issue Components