Merge pull request #1794 from dongluochen/refreshBackoffOnFailure

Implement engine refresh backoff strategy for failing nodes

cluster/engine.go

@@ -66,11 +66,12 @@ func newDelayer(rangeMin, rangeMax time.Duration) *delayer {
 	}
 }
 
-func (d *delayer) Wait() <-chan time.Time {
+// Wait returns timeout event after fixed + randomized time duration
+func (d *delayer) Wait(backoffFactor int) <-chan time.Time {
 	d.l.Lock()
 	defer d.l.Unlock()
 
-	waitPeriod := int64(d.rangeMin)
+	waitPeriod := int64(d.rangeMin) * int64(1+backoffFactor)
 	if delta := int64(d.rangeMax) - int64(d.rangeMin); delta > 0 {
 		// Int63n panics if the parameter is 0
 		waitPeriod += d.r.Int63n(delta)
@@ -637,13 +638,21 @@ func (e *Engine) updateContainer(c dockerclient.Container, containers map[string
 
 // refreshLoop periodically triggers engine refresh.
 func (e *Engine) refreshLoop() {
-
+	const maxBackoffFactor int = 1000
 	for {
 		var err error
 
+		// Engines keep failing should backoff
+		// e.failureCount and e.opts.FailureRetry are type of int
+		backoffFactor := e.failureCount - e.opts.FailureRetry
+		if backoffFactor < 0 {
+			backoffFactor = 0
+		} else if backoffFactor > maxBackoffFactor {
+			backoffFactor = maxBackoffFactor
+		}
 		// Wait for the delayer or quit if we get stopped.
 		select {
-		case <-e.refreshDelayer.Wait():
+		case <-e.refreshDelayer.Wait(backoffFactor):
 		case <-e.stopCh:
 			return
 		}

test/integration/nodemanagement/state.bats

@@ -41,7 +41,7 @@ function setup_discovery_file() {
 	# Restart node
 	docker_host start ${DOCKER_CONTAINERS[0]}
 	# Wait for swarm to detect node recovery
-	retry 5 1 eval "docker_swarm info | grep -q -i 'Status: Healthy'"
+	retry 15 1 eval "docker_swarm info | grep -q -i 'Status: Healthy'"
 }
 
 @test "node pending and recovery" {

test/integration/rescheduling.bats

@@ -155,7 +155,8 @@ function containerRunning() {
 	# Restart node-0
 	docker_host start ${DOCKER_CONTAINERS[0]}
 	# Wait for node-0 to be healthy
-	retry 5 1 eval "test \"$(docker_swarm info | grep \"Status: Unhealthy\" | wc -l)\" = '0'"
+	# Failing node refresh interval increases over time. Provide enough retry here.
+	retry 30 1 eval "test \"$(docker_swarm info | grep \"Status: Unhealthy\" | wc -l)\" = '0'"
 
 	# Stop node-1
 	docker_host stop ${DOCKER_CONTAINERS[1]}