fix(executor): Replace default retry in executor with an increased va…

…lue retryer (#3891)
argoproj · Sep 2, 2020 · 4b30fa4 · 4b30fa4
1 parent ae537cd
commit 4b30fa4
Showing 1 changed file with 5 additions and 4 deletions.
diff --git a/workflow/executor/executor.go b/workflow/executor/executor.go
@@ -39,7 +39,8 @@ import (
 	os_specific "github.com/argoproj/argo/workflow/executor/os-specific"
 )
 
-var MainContainerStartRetry = wait.Backoff{
+// ExecutorRetry is a retry backoff settings for WorkflowExecutor
+var ExecutorRetry = wait.Backoff{
 	Steps:    8,
 	Duration: 1 * time.Second,
 	Factor:   1.0,
@@ -621,7 +622,7 @@ func (we *WorkflowExecutor) getPod() (*apiv1.Pod, error) {
 	podsIf := we.ClientSet.CoreV1().Pods(we.Namespace)
 	var pod *apiv1.Pod
 	var err error
-	_ = wait.ExponentialBackoff(retry.DefaultRetry, func() (bool, error) {
+	_ = wait.ExponentialBackoff(ExecutorRetry, func() (bool, error) {
 		pod, err = podsIf.Get(we.PodName, metav1.GetOptions{})
 		if err != nil {
 			log.Warnf("Failed to get pod '%s': %v", we.PodName, err)
@@ -922,7 +923,7 @@ func (we *WorkflowExecutor) Wait() error {
 	annotationUpdatesCh := we.monitorAnnotations(ctx)
 	go we.monitorDeadline(ctx, annotationUpdatesCh)
 
-	_ = wait.ExponentialBackoff(retry.DefaultRetry, func() (bool, error) {
+	_ = wait.ExponentialBackoff(ExecutorRetry, func() (bool, error) {
 		err = we.RuntimeExecutor.Wait(mainContainerID)
 		if err != nil {
 			log.Warnf("Failed to wait for container id '%s': %v", mainContainerID, err)
@@ -949,7 +950,7 @@ func (we *WorkflowExecutor) waitMainContainerStart() (string, error) {
 		var err error
 		var watchIf watch.Interface
 
-		err = wait.ExponentialBackoff(MainContainerStartRetry, func() (bool, error) {
+		err = wait.ExponentialBackoff(ExecutorRetry, func() (bool, error) {
 			watchIf, err = podsIf.Watch(opts)
 			if err != nil {
 				log.Debugf("Failed to establish watch, retrying: %v", err)