weiqingy commented on issue #893: URL: https://github.com/apache/flink-agents/issues/893#issuecomment-4977432260
Both flaky failure modes tracked here are now root-caused and fixed on `main`: - Failure A (the `llama-server` segfault) was a CPU-microarch build issue: on the mixed runner pool, AVX-512/AMX hosts load an advanced `libggml-cpu` build that crashes ([ollama/ollama#17006](https://github.com/ollama/ollama/issues/17006)), while plain hosts fall back to AVX2 and pass. #904 (merged in `55ccb13`) deletes the advanced builds after install so the whole pool shares the safe AVX2 fallback. Evidence: a 40-runner CPU census, 3 fork runs (33 IT jobs, 0 segfaults), and Apache CI green including a job that landed on an AMX host. - Failure B (the py4j gateway corruption in the MCP e2e test) was fixed in #896. Closing as resolved — I'll reopen if the segfault resurfaces on a new runner variant. Thanks @wenjin272 for the pointer to the Sapphire Rapids/AMX backend issue; it matched what the census showed. -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: [email protected] For queries about this service, please contact Infrastructure at: [email protected]
