ipng/vpp-maglev
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

checker: fix ResumeBackend leaking goroutine on non-paused backend; v1.0.2

Browse Source 
Calling ResumeBackend on a backend that wasn't actually paused (state
!= StatePaused) would overwrite w.cancel and spawn a fresh probe
goroutine without cancelling the old one, leaving two probe loops
running for the same backend until process exit. The guard now mirrors
EnableBackend's early-return on a non-target state.
This commit is contained in:
Pim van Pelt
2026-04-19 20:39:07 +02:00
parent a8f02b913d
commit 9a3c5c5dc0
3 changed files with 51 additions and 1 deletions
Download Patch File Download Diff File Expand all files Collapse all files

8
internal/checker/checker.go
View File

@@ -416,6 +416,11 @@ func (c *Checker) PauseBackend(name string) (BackendSnapshot, error) {
// goroutine is started and the backend re-enters StateUnknown. The existing
// transition history is preserved.
// Returns an error if the backend is not found or is disabled.
//
// No-ops when the backend isn't paused: PauseBackend is what cancels the
// probe goroutine, so if we started a fresh one here unconditionally we'd
// overwrite w.cancel with a new cancel func and orphan the old goroutine —
// two probe loops would then run for the same backend until process exit.
func (c *Checker) ResumeBackend(name string) (BackendSnapshot, error) {
c.mu.Lock()
defer c.mu.Unlock()
@@ -426,6 +431,9 @@ func (c *Checker) ResumeBackend(name string) (BackendSnapshot, error) {
if !w.entry.Enabled {
return BackendSnapshot{}, fmt.Errorf("backend %q is disabled; enable it first", name)
}
if w.backend.State != health.StatePaused {
return BackendSnapshot{Health: w.backend, Config: w.entry}, nil
}
maxHistory := c.cfg.HealthChecker.TransitionHistory
if w.backend.Resume(maxHistory) {
t := w.backend.Transitions[0]