celery不能檢測到已關閉的經紀人連接，并在一段時間后凍結。 -有解無憂

我有celery==3.1.16和Python 2.7，它有一些任務，其代理是Redis。它在開始的時候性能很好，但過了一會兒就凍結了。我檢查了容器中的TCP連接，我看到了這個：

tcp 1 0 WORKER:34472 REDIS_HOST:6379 CLOSE_WAIT tcp 1 0 WORKER:39884 REDIS_HOST:6379 CLOSE_WAIT tcp 1 0 WORKER:57292 REDIS_HOST:6379 CLOSE_WAIT tcp 0 0 WORKER:60030 REDIS_HOST:6379 ESTABLISHED tcp 1 0 Worker:39906 REDIS_HOST:6379 CLOSE_WAIT tcp 1 0 WORKER:57102 REDIS_HOST:6379 CLOSE_WAIT tcp 1 0 Worker:34508 REDIS_HOST:6379 CLOSE_WAIT tcp 1 0 WORKER:39874 REDIS_HOST:6379 CLOSE_WAIT tcp 1 0 WORKER:57182 REDIS_HOST:6379 CLOSE_WAIT tcp 1 0 WORKER:57106 REDIS_HOST:6379 CLOSE_WAIT tcp 1 0 WORKER:39870 REDIS_HOST:6379 CLOSE_WAIT tcp 1 0 WORKER:57056 REDIS_HOST:6379 CLOSE_WAIT tcp 1 0 WORKER:39902 REDIS_HOST:6379 CLOSE_WAIT tcp 1 0 WORKER:34494 REDIS_HOST:6379 CLOSE_WAIT tcp 0 0 WORKER:40878 REDIS_HOST:6379 ESTABLISHED tcp 1 0 Worker:39878 REDIS_HOST:6379 CLOSE_WAIT tcp 0 0 WORKER:53138 REDIS_HOST:6379 ESTABLISHED tcp 1 0 Worker:43818 REDIS_HOST:6379 CLOSE_WAIT tcp 0 0 WORKER:39876 REDIS_HOST:6379 ESTABLISHED tcp 1 0 Worker:50586 REDIS_HOST:6379 CLOSE_WAIT tcp 1 0 WORKER:59800 REDIS_HOST:6379 CLOSE_WAIT tcp 1 0 WORKER:57128 REDIS_HOST:6379 CLOSE_WAIT tcp 1 0 WORKER:57238 REDIS_HOST:6379 CLOSE_WAIT tcp 1 0 WORKER:57346 REDIS_HOST:6379 CLOSE_WAIT tcp 1 0 Worker:57050 REDIS_HOST:6379 CLOSE_WAIT tcp 1 0 WORKER:39896 REDIS_HOST:6379 CLOSE_WAIT tcp 1 0 Worker:44850 REDIS_HOST:6379 CLOSE_WAIT tcp 1 0 WORKER:57124 REDIS_HOST:6379 CLOSE_WAIT tcp 1 0 WORKER:39904 REDIS_HOST:6379 CLOSE_WAIT tcp 0 0 WORKER:39872 REDIS_HOST:6379 ESTABLISHED tcp 1 0 Worker:57160 REDIS_HOST:6379 CLOSE_WAIT tcp 1 0 WORKER:57190 REDIS_HOST:6379 CLOSE_WAIT tcp 1 0 WORKER:59724 REDIS_HOST:6379 CLOSE_WAIT tcp 1 0 WORKER:57260 REDIS_HOST:6379 CLOSE_WAIT tcp 1 0 WORKER:34492 REDIS_HOST:6379 CLOSE_WAIT tcp 1 0 WORKER:59740 REDIS_HOST:6379 CLOSE_WAIT tcp 1 0 Worker:55426 REDIS_HOST:6379 CLOSE_WAIT

我認為celery worker沒有檢測到關閉的連接來重新建立它們，我必須定期重置worker pod才能再次作業。

有什么想法嗎？

uj5u.com熱心網友回復：

我用solo pool-size模式運行celery，并使用多個Kubernetes pod而不是一個pod，因為solo模式是單執行緒的，所以問題就解決了。

https://docs.celeryproject.org/en/stable/internals/reference/celery.concurrency.solo.html

轉載請註明出處，本文鏈接：https://www.uj5u.com/qiye/309337.html

標籤：

上一篇：根據在資料框架中的位置計算元素的數量

下一篇：為物件分配類別的問題