fix: potential append race

pkg/rolling/rolling.go

@@ -30,6 +30,10 @@ type Rolling struct {
 	state     *state
 	opts      options.RestartOptions
 	restarter restarters.Restarter
+
+	// TODO jorres@: maybe turn this into a local `map`
+	// variable in `processActionGroupStates`
+	completedActions []*Ydb_Maintenance.ActionUid
 }
 
 type state struct {
@@ -231,9 +235,11 @@ func (r *Rolling) processActionGroupStates(actions []*Ydb_Maintenance.ActionGrou
 
 	r.logger.Infof("%d ActionGroupStates moved to PERFORMED, will restart now...", len(performed))
 
-	actionsCompletedThisStep := []*Ydb_Maintenance.ActionUid{}
+	r.completedActions = []*Ydb_Maintenance.ActionUid{}
 	wg := new(sync.WaitGroup)
 
+	mutexForCompletedActions := new(sync.Mutex)
+
 	for _, gs := range performed {
 		var (
 			as   = gs.ActionStates[0]
@@ -242,7 +248,7 @@ func (r *Rolling) processActionGroupStates(actions []*Ydb_Maintenance.ActionGrou
 		)
 
 		if collections.Contains(r.state.unreportedButFinishedActionIds, as.ActionUid.ActionId) {
-			actionsCompletedThisStep = append(actionsCompletedThisStep, as.ActionUid)
+			r.completedActions = append(r.completedActions, as.ActionUid)
 			r.logger.Debugf(
 				"Node id %v already restarted, but CompleteAction failed on last iteration, "+
 					"so CMS does not know it is complete yet.",
@@ -271,22 +277,19 @@ func (r *Rolling) processActionGroupStates(actions []*Ydb_Maintenance.ActionGrou
 				r.state.retriesMadeForNode[node.NodeId]++
 
 				if r.state.retriesMadeForNode[node.NodeId] == r.opts.RestartRetryNumber {
-					// TODO reduce copypaste with literally 5 lines below
-					r.state.unreportedButFinishedActionIds = append(r.state.unreportedButFinishedActionIds, as.ActionUid.ActionId)
-					actionsCompletedThisStep = append(actionsCompletedThisStep, as.ActionUid)
+					r.atomicRememberComplete(mutexForCompletedActions, as.ActionUid)
 					r.logger.Warnf("Failed to retry node %v specified number of times %v", node.NodeId, r.opts.RestartRetryNumber)
 				}
 			} else {
-				r.state.unreportedButFinishedActionIds = append(r.state.unreportedButFinishedActionIds, as.ActionUid.ActionId)
-				actionsCompletedThisStep = append(actionsCompletedThisStep, as.ActionUid)
+				r.atomicRememberComplete(mutexForCompletedActions, as.ActionUid)
 				r.logger.Debugf("Successfully restarted node with id: %d", node.NodeId)
 			}
 		}()
 	}
 
 	wg.Wait()
 
-	result, err := r.cms.CompleteAction(actionsCompletedThisStep)
+	result, err := r.cms.CompleteAction(r.completedActions)
 	if err != nil {
 		r.logger.Warnf("Failed to complete action: %+v", err)
 		return false
@@ -298,6 +301,14 @@ func (r *Rolling) processActionGroupStates(actions []*Ydb_Maintenance.ActionGrou
 	return len(actions) == len(result.ActionStatuses)
 }
 
+func (r *Rolling) atomicRememberComplete(m *sync.Mutex, actionUID *Ydb_Maintenance.ActionUid) {
+	m.Lock()
+	defer m.Unlock()
+
+	r.state.unreportedButFinishedActionIds = append(r.state.unreportedButFinishedActionIds, actionUID.ActionId)
+	r.completedActions = append(r.completedActions, actionUID)
+}
+
 func (r *Rolling) prepareState() (*state, error) {
 	tenants, err := r.cms.Tenants()
 	if err != nil {