X-Git-Url: https://git.arvados.org/arvados.git/blobdiff_plain/71fd4da18b22100682ae7e2079aadfd66360d310..3af38eca348413c6f11f6526b2ee2ca7cb53e348:/lib/dispatchcloud/container/queue.go

diff --git a/lib/dispatchcloud/container/queue.go b/lib/dispatchcloud/container/queue.go
index 847fe9e27c..bbe47625a8 100644
--- a/lib/dispatchcloud/container/queue.go
+++ b/lib/dispatchcloud/container/queue.go
@@ -131,7 +131,7 @@ func (cq *Queue) Forget(uuid string) {
 	defer cq.mtx.Unlock()
 	ctr := cq.current[uuid].Container
 	if ctr.State == arvados.ContainerStateComplete || ctr.State == arvados.ContainerStateCancelled {
-		delete(cq.current, uuid)
+		cq.delEnt(uuid, ctr.State)
 	}
 }
 
@@ -196,7 +196,7 @@ func (cq *Queue) Update() error {
 			cq.current[uuid] = cur
 		}
 	}
-	for uuid := range cq.current {
+	for uuid, ent := range cq.current {
 		if _, dontupdate := cq.dontupdate[uuid]; dontupdate {
 			// Don't expunge an entry that was
 			// added/updated locally after we started
@@ -207,7 +207,7 @@ func (cq *Queue) Update() error {
 			// the poll response (evidently it's
 			// cancelled, completed, deleted, or taken by
 			// a different dispatcher).
-			delete(cq.current, uuid)
+			cq.delEnt(uuid, ent.Container.State)
 		}
 	}
 	cq.dontupdate = nil
@@ -216,6 +216,15 @@ func (cq *Queue) Update() error {
 	return nil
 }
 
+// Caller must have lock.
+func (cq *Queue) delEnt(uuid string, state arvados.ContainerState) {
+	cq.logger.WithFields(logrus.Fields{
+		"ContainerUUID": uuid,
+		"State":         state,
+	}).Info("dropping container from queue")
+	delete(cq.current, uuid)
+}
+
 func (cq *Queue) addEnt(uuid string, ctr arvados.Container) {
 	it, err := cq.chooseType(&ctr)
 	if err != nil && (ctr.State == arvados.ContainerStateQueued || ctr.State == arvados.ContainerStateLocked) {
@@ -223,8 +232,25 @@ func (cq *Queue) addEnt(uuid string, ctr arvados.Container) {
 		// error: it wouldn't help to try again, or to leave
 		// it for a different dispatcher process to attempt.
 		errorString := err.Error()
-		cq.logger.WithField("ContainerUUID", ctr.UUID).Warn("cancel container with no suitable instance type")
+		logger := cq.logger.WithField("ContainerUUID", ctr.UUID)
+		logger.WithError(err).Warn("cancel container with no suitable instance type")
 		go func() {
+			if ctr.State == arvados.ContainerStateQueued {
+				// Can't set runtime error without
+				// locking first. If Lock() is
+				// successful, it will call addEnt()
+				// again itself, and we'll fall
+				// through to the
+				// setRuntimeError/Cancel code below.
+				err := cq.Lock(ctr.UUID)
+				if err != nil {
+					logger.WithError(err).Warn("lock failed")
+					// ...and try again on the
+					// next Update, if the problem
+					// still exists.
+				}
+				return
+			}
 			var err error
 			defer func() {
 				if err == nil {
@@ -239,14 +265,8 @@ func (cq *Queue) addEnt(uuid string, ctr arvados.Container) {
 				if latest.State == arvados.ContainerStateCancelled {
 					return
 				}
-				cq.logger.WithField("ContainerUUID", ctr.UUID).WithError(err).Warn("error while trying to cancel unsatisfiable container")
+				logger.WithError(err).Warn("error while trying to cancel unsatisfiable container")
 			}()
-			if ctr.State == arvados.ContainerStateQueued {
-				err = cq.Lock(ctr.UUID)
-				if err != nil {
-					return
-				}
-			}
 			err = cq.setRuntimeError(ctr.UUID, errorString)
 			if err != nil {
 				return
@@ -258,6 +278,12 @@ func (cq *Queue) addEnt(uuid string, ctr arvados.Container) {
 		}()
 		return
 	}
+	cq.logger.WithFields(logrus.Fields{
+		"ContainerUUID": ctr.UUID,
+		"State":         ctr.State,
+		"Priority":      ctr.Priority,
+		"InstanceType":  it.Name,
+	}).Info("adding container to queue")
 	cq.current[uuid] = QueueEnt{Container: ctr, InstanceType: it}
 }