diff options
author | Matthew Heon <matthew.heon@pm.me> | 2021-09-21 14:24:23 -0400 |
---|---|---|
committer | Matthew Heon <matthew.heon@pm.me> | 2021-09-22 16:33:22 -0400 |
commit | d45cbbda3445d204588723c03609ff2641a70781 (patch) | |
tree | 699a864f9ed224d528590768ed3561791c7a4605 | |
parent | 70da2e858bfecf3d3301b5b85ff5d7332911855d (diff) | |
download | podman-d45cbbda3445d204588723c03609ff2641a70781.tar.gz podman-d45cbbda3445d204588723c03609ff2641a70781.tar.bz2 podman-d45cbbda3445d204588723c03609ff2641a70781.zip |
Add a backoff and retries to retrieving exited event
There's a potential race around extremely short-running
containers and events with journald. Events may not be written
for some time (small, but appreciable) after they are received,
and as such we can fail to retrieve it if there is a sufficiently
short time between us writing the event and trying to read it.
Work around this by just retrying, with a 0.25 second delay
between retries, up to 4 times.
[NO TESTS NEEDED] because I have no idea how to reproduce this
race in CI.
Fixes #11633
Signed-off-by: Matthew Heon <matthew.heon@pm.me>
-rw-r--r-- | pkg/domain/infra/abi/containers.go | 55 |
1 files changed, 25 insertions, 30 deletions
diff --git a/pkg/domain/infra/abi/containers.go b/pkg/domain/infra/abi/containers.go index dc5f7a0df..affed64d1 100644 --- a/pkg/domain/infra/abi/containers.go +++ b/pkg/domain/infra/abi/containers.go @@ -830,21 +830,7 @@ func (ic *ContainerEngine) ContainerStart(ctx context.Context, namesOrIds []stri } return reports, errors.Wrapf(err, "unable to start container %s", ctr.ID()) } - - if ecode, err := ctr.Wait(ctx); err != nil { - if errors.Cause(err) == define.ErrNoSuchCtr { - // Check events - event, err := ic.Libpod.GetLastContainerEvent(ctx, ctr.ID(), events.Exited) - if err != nil { - logrus.Errorf("Cannot get exit code: %v", err) - exitCode = define.ExecErrorCodeNotFound - } else { - exitCode = event.ContainerExitCode - } - } - } else { - exitCode = int(ecode) - } + exitCode = ic.GetContainerExitCode(ctx, ctr) reports = append(reports, &entities.ContainerStartReport{ Id: ctr.ID(), RawInput: rawInput, @@ -985,21 +971,7 @@ func (ic *ContainerEngine) ContainerRun(ctx context.Context, opts entities.Conta report.ExitCode = define.ExitCode(err) return &report, err } - - if ecode, err := ctr.Wait(ctx); err != nil { - if errors.Cause(err) == define.ErrNoSuchCtr { - // Check events - event, err := ic.Libpod.GetLastContainerEvent(ctx, ctr.ID(), events.Exited) - if err != nil { - logrus.Errorf("Cannot get exit code: %v", err) - report.ExitCode = define.ExecErrorCodeNotFound - } else { - report.ExitCode = event.ContainerExitCode - } - } - } else { - report.ExitCode = int(ecode) - } + report.ExitCode = ic.GetContainerExitCode(ctx, ctr) if opts.Rm && !ctr.ShouldRestart(ctx) { if err := ic.Libpod.RemoveContainer(ctx, ctr, false, true); err != nil { if errors.Cause(err) == define.ErrNoSuchCtr || @@ -1013,6 +985,29 @@ func (ic *ContainerEngine) ContainerRun(ctx context.Context, opts entities.Conta return &report, nil } +func (ic *ContainerEngine) GetContainerExitCode(ctx context.Context, ctr *libpod.Container) int { + exitCode, err := ctr.Wait(ctx) + if err == nil { + return int(exitCode) + } + if errors.Cause(err) != define.ErrNoSuchCtr { + logrus.Errorf("Could not retrieve exit code: %v", err) + return define.ExecErrorCodeNotFound + } + // Make 4 attempt with 0.25s backoff between each for 1 second total + var event *events.Event + for i := 0; i < 4; i++ { + event, err = ic.Libpod.GetLastContainerEvent(ctx, ctr.ID(), events.Exited) + if err != nil { + time.Sleep(250 * time.Millisecond) + continue + } + return int(event.ContainerExitCode) + } + logrus.Errorf("Could not retrieve exit code from event: %v", err) + return define.ExecErrorCodeNotFound +} + func (ic *ContainerEngine) ContainerLogs(ctx context.Context, containers []string, options entities.ContainerLogsOptions) error { if options.StdoutWriter == nil && options.StderrWriter == nil { return errors.New("no io.Writer set for container logs") |