Skip to content

integration-cli: retry Windows npipe connect EOF in TestRunTwoConcurrentContainers - #53395

Open
VedantMadane wants to merge 1 commit into
moby:masterfrom
VedantMadane:fix/53343-flaky-TestRunTwoConcurrentContainers
Open

VedantMadane wants to merge 1 commit into
moby:masterfrom
VedantMadane:fix/53343-flaky-TestRunTwoConcurrentContainers

Conversation

@VedantMadane

Copy link
Copy Markdown
Contributor

Summary

Hardens flaky TestDockerCLIRunSuite/TestRunTwoConcurrentContainers on Windows.

Failure

On windows-2022 CI, one of two concurrent docker run calls intermittently fails with:

docker: error during connect: Post "http://%2F%2F.%2Fpipe%2Fdocker_engine/v1.44/containers/create": EOF.

Exit code 125 after ~60s. Same class of Windows named-pipe transport flake as #49348 / #53269 (concurrent engine API traffic over //./pipe/docker_engine under CI load).

Fix

Keep the concurrent launches (that is what this test covers). Retry each docker run up to 3 times only when the error matches the transient connect EOF pattern (error during connect + EOF). All other failures still fail the test immediately.

Residual risk

  • This does not fix a possible daemon/snapshotter root cause (e.g. slow concurrent create / VHD+boltdb contention on WCOW that can idle the npipe until EOF).
  • If every attempt hangs ~60s before EOF, the test can still be slow or fail after 3 attempts.
  • A create that actually succeeded server-side while the client saw EOF could leave an extra anonymous container; suite teardown cleans those up.

Fixes #53343

Test plan

  • Localized change to existing integration-cli test only
  • Windows CI: TestDockerCLIRunSuite/TestRunTwoConcurrentContainers
  • Confirm Linux path unchanged in behavior (retries only on connect EOF, which is Windows-specific in practice)

TestRunTwoConcurrentContainers flakes on Windows when concurrent docker
CLI processes dial //./pipe/docker_engine and one create hits a transient
connect-time EOF under CI load.

Retry only that specific error-during-connect + EOF failure so the test
still exercises concurrent container starts without failing on the known
named-pipe transport flake.

Signed-off-by: VedantMadane <[email protected]>
@VedantMadane
VedantMadane force-pushed the fix/53343-flaky-TestRunTwoConcurrentContainers branch from 435c91d to 35af8ed Compare August 18, 2026 08:56
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Projects

None yet

Development

Successfully merging this pull request may close these issues.

Flaky test: TestRunTwoConcurrentContainers (Windows named-pipe EOF)

1 participant