mirror of
https://github.com/agent-substrate/substrate.git
synced 2026-10-02 03:24:42 +08:00
Benjamin Elder
e4844d0fd2
When an e2e test failed, the evidence was deleted before anyone could read it. The suite deleted every namespace it created on the way out, taking the worker pods with it, and the workflow's post-failure dump only looked at three fixed namespaces — never the suites' randomly-named ones. So a failure inside an actor (#619: a micro-VM resume where the guest died at boot) left nothing behind but the RPC error the test printed. Keep the namespaces when the suite failed, and dump every worker pod in every namespace, so the ateom logs — which carry the guest's console tail — reach the failed run's output. Kept namespaces are nobody's to reclaim, and each holds a WorkerPool's worth of running pods, so they now carry an ate.dev/e2e label and hack/cleanup-e2e.sh deletes them once the logs have served their purpose. CI throws its cluster away, but a development cluster accumulates them run after run. Fixes #<issue_number_goes_here> Built while debugging https://github.com/agent-substrate/substrate/issues/619 > It's a good idea to open an issue first for discussion. - [x] Tests pass - [x] Appropriate changes to documentation are included in the PR