fix: retry docker volume remove (!399)

Closes coop-cloud/organising#509

Reviewed-on: #399
Reviewed-by: decentral1se <decentral1se@noreply.git.coopcloud.tech>
Co-authored-by: p4u1 <p4u1_f4u1@riseup.net>
Co-committed-by: p4u1 <p4u1_f4u1@riseup.net>

cli/app/remove.go

@@ -3,7 +3,9 @@ package app
 import (
 	"context"
 	"fmt"
+	"log"
 	"os"
+	"time"
 
 	"coopcloud.tech/abra/cli/internal"
 	"coopcloud.tech/abra/pkg/autocomplete"
@@ -124,9 +126,11 @@ flag.
 
 		if len(vols) > 0 {
 			for _, vol := range vols {
-				err := cl.VolumeRemove(context.Background(), vol, internal.Force) // last argument is for force removing
+				err = retryFunc(5, func() error {
+					return cl.VolumeRemove(context.Background(), vol, internal.Force) // last argument is for force removing
+				})
 				if err != nil {
-					logrus.Fatal(err)
+					log.Fatalf("removing volumes failed: %s", err)
 				}
 				logrus.Info(fmt.Sprintf("volume %s removed", vol))
 			}
@@ -143,3 +147,21 @@ flag.
 		return nil
 	},
 }
+
+// retryFunc retries the given function for the given retries. After the nth
+// retry it waits (n + 1)^2 seconds before the next retry (starting with n=0).
+// It returns an error if the function still failed after the last retry.
+func retryFunc(retries int, fn func() error) error {
+	for i := 0; i < retries; i++ {
+		err := fn()
+		if err == nil {
+			return nil
+		}
+		if i+1 < retries {
+			sleep := time.Duration(i+1) * time.Duration(i+1)
+			logrus.Infof("%s: waiting %d seconds before next retry", err, sleep)
+			time.Sleep(sleep * time.Second)
+		}
+	}
+	return fmt.Errorf("%d retries failed", retries)
+}

cli/app/remove_test.go

@@ -0,0 +1,26 @@
+package app
+
+import (
+	"fmt"
+	"testing"
+)
+
+func TestRetryFunc(t *testing.T) {
+	err := retryFunc(1, func() error { return nil })
+	if err != nil {
+		t.Errorf("should not return an error: %s", err)
+	}
+
+	i := 0
+	fn := func() error {
+		i++
+		return fmt.Errorf("oh no, something went wrong!")
+	}
+	err = retryFunc(2, fn)
+	if err == nil {
+		t.Error("should return an error")
+	}
+	if i != 2 {
+		t.Errorf("The function should have been called 1 times, got %d", i)
+	}
+}

tests/integration/app_remove.bats

@@ -104,10 +104,7 @@ teardown(){
 
   _undeploy_app
 
-  # NOTE(d1): to let the stack come down before nuking volumes
-  sleep 5
-
-  run $ABRA app volume rm "$TEST_APP_DOMAIN" --force
+  run $ABRA app volume rm "$TEST_APP_DOMAIN"
   assert_success
 
   run $ABRA app volume ls "$TEST_APP_DOMAIN"
@@ -132,9 +129,6 @@ teardown(){
 
   _undeploy_app
 
-  # NOTE(d1): to let the stack come down before nuking volumes
-  sleep 5
-
   run $ABRA app rm "$TEST_APP_DOMAIN" --no-input
   assert_success
   assert_output --partial 'test-volume'