bugfix: NackRedeliveryDelay may cause unexpected retry

delayqueue.go

@@ -512,11 +512,29 @@ func (q *DelayQueue) ack(idStr string) error {
 	return nil
 }
 
+// updateZSetScoreScript update score of a zset member if it exists
+// KEYS[1]: zset
+// ARGV[1]: score
+// ARGV[2]: member
+const updateZSetScoreScript = `
+if redis.call('zrank', KEYS[1], ARGV[2]) ~= nil then
+    return redis.call('zadd', KEYS[1], ARGV[1], ARGV[2])
+else
+    return 0
+end
+`
+
+func (q *DelayQueue) updateZSetScore(key string, score float64, member string) error {
+	scoreStr := strconv.FormatFloat(score, 'f', -1, 64)
+	_, err := q.eval(updateZSetScoreScript, []string{key}, []interface{}{scoreStr, member})
+	return err
+}
+
 func (q *DelayQueue) nack(idStr string) error {
 	atomic.AddInt32(&q.fetchCount, -1)
-	err := q.redisCli.ZAdd(q.unAckKey, map[string]float64{
+	retryTime := float64(time.Now().Add(q.nackRedeliveryDelay).Unix())
-		idStr: float64(time.Now().Add(q.nackRedeliveryDelay).Unix()),
+	// if message consumption has not reach deadlin (still in unAckKey), then update its retry time
-	})
+	err := q.updateZSetScore(q.unAckKey, retryTime, idStr)
 	if err != nil {
 		return fmt.Errorf("negative ack failed: %v", err)
 	}