fix: healing deadlocks and ordering (#16643)

cmd/admin-heal-ops.go

@@ -407,9 +407,6 @@ type healSequence struct {
 	// bucket, and object on which heal seq. was initiated
 	bucket, object string
 
-	// A channel of entities with heal result
-	respCh chan healResult
-
 	// Report healing progress
 	reportProgress bool
 
@@ -472,7 +469,6 @@ func newHealSequence(ctx context.Context, bucket, objPrefix, clientAddr string,
 	clientToken := mustGetUUID()
 
 	return &healSequence{
-		respCh:         make(chan healResult),
 		bucket:         bucket,
 		object:         objPrefix,
 		reportProgress: true,
@@ -719,15 +715,18 @@ func (h *healSequence) queueHealTask(source healSource, healType madmin.HealItem
 			if serverDebugLog {
 				logger.Info("Task in the queue: %#v", task)
 			}
-		case <-h.ctx.Done():
-			return nil
 		default:
 			// task queue is full, no more workers, we shall move on and heal later.
 			return nil
 		}
-	} else {
-		// respCh must be set for guaranteed result
-		task.respCh = h.respCh
+		// Don't wait for result
+		return nil
+	}
+
+	// respCh must be set to wait for result.
+	// We make it size 1, so a result can always be written
+	// even if we aren't listening.
+	task.respCh = make(chan healResult, 1)
 	select {
 	case globalBackgroundHealRoutine.tasks <- task:
 		if serverDebugLog {
@@ -736,11 +735,10 @@ func (h *healSequence) queueHealTask(source healSource, healType madmin.HealItem
 	case <-h.ctx.Done():
 		return nil
 	}
-	}
 
 	// task queued, now wait for the response.
 	select {
-	case res := <-h.respCh:
+	case res := <-task.respCh:
 		if !h.reportProgress {
 			if errors.Is(res.err, errSkipFile) { // this is only sent usually by nopHeal
 				return nil

cmd/global-heal.go

@@ -48,7 +48,6 @@ func newBgHealSequence() *healSequence {
 	}
 
 	return &healSequence{
-		respCh:      make(chan healResult),
 		startTime:   UTCNow(),
 		clientToken: bgHealingUUID,
 		// run-background heal with reserved bucket