X-Git-Url: https://git.arvados.org/arvados.git/blobdiff_plain/fe33cfe407912df46ae34b5419ebdc84650829f6..cd020c016106fbe844501c5f434c16f4def4e08d:/lib/dispatchcloud/scheduler/fix_stale_locks.go

diff --git a/lib/dispatchcloud/scheduler/fix_stale_locks.go b/lib/dispatchcloud/scheduler/fix_stale_locks.go
index 264f9e4ec6..1f9338f7b8 100644
--- a/lib/dispatchcloud/scheduler/fix_stale_locks.go
+++ b/lib/dispatchcloud/scheduler/fix_stale_locks.go
@@ -19,24 +19,15 @@ import (
 func (sch *Scheduler) fixStaleLocks() {
 	wp := sch.pool.Subscribe()
 	defer sch.pool.Unsubscribe(wp)
+
+	var stale []string
 	timeout := time.NewTimer(sch.staleLockTimeout)
 waiting:
-	for {
-		unlock := false
-		select {
-		case <-wp:
-			// If all workers have been contacted, unlock
-			// containers that aren't claimed by any
-			// worker.
-			unlock = sch.pool.CountWorkers()[worker.StateUnknown] == 0
-		case <-timeout.C:
-			// Give up and unlock the containers, even
-			// though they might be working.
-			unlock = true
-		}
-
+	for sch.pool.CountWorkers()[worker.StateUnknown] > 0 {
 		running := sch.pool.Running()
 		qEntries, _ := sch.queue.Entries()
+
+		stale = nil
 		for uuid, ent := range qEntries {
 			if ent.Container.State != arvados.ContainerStateLocked {
 				continue
@@ -44,14 +35,24 @@ waiting:
 			if _, running := running[uuid]; running {
 				continue
 			}
-			if !unlock {
-				continue waiting
-			}
-			err := sch.queue.Unlock(uuid)
-			if err != nil {
-				sch.logger.Warnf("Unlock %s: %s", uuid, err)
-			}
+			stale = append(stale, uuid)
+		}
+		if len(stale) == 0 {
+			return
+		}
+
+		select {
+		case <-wp:
+		case <-timeout.C:
+			// Give up.
+			break waiting
+		}
+	}
+
+	for _, uuid := range stale {
+		err := sch.queue.Unlock(uuid)
+		if err != nil {
+			sch.logger.Warnf("Unlock %s: %s", uuid, err)
 		}
-		return
 	}
 }