From 0d86ef89181599ced5613fe194d576954e3c7f18 Mon Sep 17 00:00:00 2001 From: Valentin Rothberg Date: Mon, 21 Jun 2021 14:31:26 +0200 Subject: [PATCH] logs: k8s-file: restore poll sleep Commit 84b55eec2796 attempted to fix a race waiting for the container died event. Previously, Podman slept for duration of the polling frequence which I considerred to be a mistake. As it turns out, I was mistaken since the file logger will, in fact, NOT read until EOF and then stop logging but stop logging immediately _after_ it woke up. [NO TESTS NEEDED] as the race condition cannot be hit reliably. Fixes: #10675 Signed-off-by: Valentin Rothberg --- libpod/container_log.go | 14 +++++++++++--- 1 file changed, 11 insertions(+), 3 deletions(-) diff --git a/libpod/container_log.go b/libpod/container_log.go index 8bbb7ebe8d..43b3f77366 100644 --- a/libpod/container_log.go +++ b/libpod/container_log.go @@ -4,10 +4,12 @@ import ( "context" "fmt" "os" + "time" "github.com/containers/podman/v3/libpod/define" "github.com/containers/podman/v3/libpod/events" "github.com/containers/podman/v3/libpod/logs" + "github.com/hpcloud/tail/watch" "github.com/pkg/errors" "github.com/sirupsen/logrus" ) @@ -93,11 +95,14 @@ func (c *Container) readFromLogFile(ctx context.Context, options *logs.LogOption }() // Check if container is still running or paused if options.Follow { + // If the container isn't running or if we encountered an error + // getting its state, instruct the logger to read the file + // until EOF. state, err := c.State() if err != nil || state != define.ContainerStateRunning { - // If the container isn't running or if we encountered - // an error getting its state, instruct the logger to - // read the file until EOF. + // Make sure to wait at least for the poll duration + // before stopping the file logger (see #10675). + time.Sleep(watch.POLL_DURATION) tailError := t.StopAtEOF() if tailError != nil && fmt.Sprintf("%v", tailError) != "tail: stop at eof" { logrus.Errorf("Error stopping logger: %v", tailError) @@ -124,6 +129,9 @@ func (c *Container) readFromLogFile(ctx context.Context, options *logs.LogOption // Now wait for the died event and signal to finish // reading the log until EOF. <-eventChannel + // Make sure to wait at least for the poll duration + // before stopping the file logger (see #10675). + time.Sleep(watch.POLL_DURATION) tailError := t.StopAtEOF() if tailError != nil && fmt.Sprintf("%v", tailError) != "tail: stop at eof" { logrus.Errorf("Error stopping logger: %v", tailError)