The GitHub Actions job "Flink Kubernetes Operator CI" on 
flink-kubernetes-operator.git/release-1.16 has failed.
Run started by GitHub user gyfora (triggered by gyfora).

Head commit for run:
f4f45824dbbe15f551add54adc059a61c2b3cf39 / Purushottam Sinha 
<[email protected]>
[FLINK-40402] Harden job artifact fetching against slow-host reconcile-thread 
exhaustion (#1186)

* [FLINK-40402] Harden job artifact fetching against slow-host reconcile-thread 
exhaustion.

Add connect/read/total timeouts and a size cap to HttpArtifactFetcher so a
slow or unresponsive FlinkSessionJob jarURI host can no longer pin a
reconcile thread indefinitely.

Generated-by: Claude Code

* [FLINK-40402] Make artifact-fetch failure messages consistent and traceable.

The size-cap and total-timeout errors thrown while streaming the body did not
name the offending jarURI or the configured value, unlike the declared-size
and redirect-timeout errors thrown earlier in the fetch. Pass the URI and the
total timeout into copyBounded so every fetch failure names the artifact, and
unify the wording: both size errors now read "exceeds the configured limit"
and both timeout errors read "Timed out (> <duration>) while fetching artifact
from '<uri>'".

Generated-by: Claude Code

* [FLINK-40402] Add regression tests for the connect and read fetch timeouts.

The existing coverage only exercised the overall (total) fetch timeout via a
slow trickle. Add two tests that isolate the per-connection socket timeouts,
each with a much longer total timeout so it is unambiguously the socket
timeout that bounds them:
  - connect timeout: fetching from a blackholed address (TEST-NET-1) whose
    SYNs are dropped, so the TCP connect never completes;
  - read timeout: a server that sends response headers then stalls without
    sending the body, so a client read blocks.

Generated-by: Claude Code

Report URL: 
https://github.com/apache/flink-kubernetes-operator/actions/runs/32498190077

With regards,
GitHub Actions via GitBox

Reply via email to