diff options
Diffstat (limited to 'pkg/sentry/watchdog/watchdog.go')
-rw-r--r-- | pkg/sentry/watchdog/watchdog.go | 14 |
1 files changed, 10 insertions, 4 deletions
diff --git a/pkg/sentry/watchdog/watchdog.go b/pkg/sentry/watchdog/watchdog.go index 101497ed6..748273366 100644 --- a/pkg/sentry/watchdog/watchdog.go +++ b/pkg/sentry/watchdog/watchdog.go @@ -77,7 +77,10 @@ var DefaultOpts = Opts{ // trigger it. const descheduleThreshold = 1 * time.Second -var stuckTasks = metric.MustCreateNewUint64Metric("/watchdog/stuck_tasks_detected", true /* sync */, "Cumulative count of stuck tasks detected") +var ( + stuckStartup = metric.MustCreateNewUint64Metric("/watchdog/stuck_startup_detected", true /* sync */, "Incremented once on startup watchdog timeout") + stuckTasks = metric.MustCreateNewUint64Metric("/watchdog/stuck_tasks_detected", true /* sync */, "Cumulative count of stuck tasks detected") +) // Amount of time to wait before dumping the stack to the log again when the same task(s) remains stuck. var stackDumpSameTaskPeriod = time.Minute @@ -220,6 +223,9 @@ func (w *Watchdog) waitForStart() { // We are fine. return } + + stuckStartup.Increment() + var buf bytes.Buffer buf.WriteString(fmt.Sprintf("Watchdog.Start() not called within %s", w.StartupTimeout)) w.doAction(w.StartupTimeoutAction, false, &buf) @@ -323,13 +329,13 @@ func (w *Watchdog) report(offenders map[*kernel.Task]*offender, newTaskFound boo func (w *Watchdog) reportStuckWatchdog() { var buf bytes.Buffer - buf.WriteString("Watchdog goroutine is stuck:") + buf.WriteString("Watchdog goroutine is stuck") w.doAction(w.TaskTimeoutAction, false, &buf) } // doAction will take the given action. If the action is LogWarning, the stack -// is not always dumpped to the log to prevent log flooding. "forceStack" -// guarantees that the stack will be dumped regarless. +// is not always dumped to the log to prevent log flooding. "forceStack" +// guarantees that the stack will be dumped regardless. func (w *Watchdog) doAction(action Action, forceStack bool, msg *bytes.Buffer) { switch action { case LogWarning: |