weiqingy commented on issue #889: URL: https://github.com/apache/flink-agents/issues/889#issuecomment-4939686165
Status map for anyone picking this up, since the work has spread across a few PRs and a sibling issue. This issue produced two independent PRs: - **#891** — retry the Ollama install and propagate download failures (fixes the transient `.tgz` 404). - **#892** — bound the integration-test jobs with `timeout-minutes` so a hung job fails fast instead of running to the 6-hour ceiling. Both currently show red CI, but the red is **not** these changes. The Ollama install succeeds in every failing run; only the Ollama-backed integration tests crash, and the failing jobs roam from run to run while the changes stay the same. That intermittent flake is tracked separately in **#893**. Root cause (detail in #893): a subset of GitHub `ubuntu-latest` runner VMs trip an upstream `llama.cpp` segmentation fault while serving `qwen3:1.7b` on CPU. A given VM is effectively deterministic, so re-running the *whole* matrix keeps hitting some cursed VM. The dated onset lines up with a GitHub runner-image rollout — and notably the installed Ollama version has been constant across the green→red boundary, so pinning the Ollama version does not apply here. The fix is **#895**: a `workflow_run` companion that re-runs only the failed integration-test legs on a *fresh* runner (which re-rolls the VM), capped at two attempts, with unit/lint/build failures never re-run. It was validated end-to-end on a fork. One caveat noted on that PR: `workflow_run` companions only run from the default branch, so #895 cannot heal its own CI and takes effect only once merged. Suggested order, entirely at the maintainers' discretion: **#895 first** — once merged it heals the integration-test flake for every subsequent run, including #891 and #892 — then #891/#892, which are otherwise green apart from the flake. They can also be reviewed independently. I'll leave the sequencing and the decisions to the maintainers. -- 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]
