multiprocessing.queue在gunicorn worker超时后无法获取数据

时间:2017-10-10 02:29:44

标签: python gunicorn python-multiprocessing

我有一个FLASK APP并使用gunicorn(同步模式)作为Web服务器。对于异步推送信息,我使用“gunicorn服务器挂钩”启动一个维护进程(multiprocessing.Process())当gunicorn启动时,并使用multiprocessing.Queue()(它实际上是logging.handlers.QueueHandler(Queue)兼容用python日志记录)来发送消息。 但是我发现如果gunicorn工作者在“[CRITICAL] WORKER TIMEOUT”时重新启动,则维护进程不会从gunicorn worker发送的队列消息(queue.qsize()不是0,并根据日志,它将消息放入成功排队,但Queue.get(超时)引发空异常),但可以从gunicorn主进程获取消息。 我的日志:

  34 pid:24831 wechatlog : 2017-10-10 06:32:43,552 wechat_middle.py[line:34] DEBUG recive <LogRecord: wechat, 40, /www_upload/src/api_server.py, 543, "{'tag_list': 1, 'msg': 'company_test sid:8607550100000080 id: 8607550100000080 his: 1', 'lastsend': 'serial_error'}">
  35 pid:23930 wechat    : 2017-10-10 06:38:56,805 api_server.py[line:543] ERROR {'tag_list': 1, 'msg': 'company_test sid:8607550100000080 id: 8607550100000080 his: 1', 'lastsend': 'serial_error'}
  36 pid:24831 wechatlog : 2017-10-10 06:38:56,807 wechat_middle.py[line:34] DEBUG recive <LogRecord: wechat, 40, /www_upload/src/api_server.py, 543, "{'tag_list': 1, 'msg': 'company_test sid:8607550100000080 id: 8607550100000080 his: 1', 'lastsend': 'serial_error'}">
  37 pid:24887 wechat    : 2017-10-10 07:07:50,904 api_server.py[line:543] ERROR {'tag_list': 1, 'msg': 'company_test sid:8607550100000080 id: 8607550100000080 his: 1', 'lastsend': 'serial_error'}
  38 pid:24831 wechatlog : 2017-10-10 07:07:51,810 maintain_task.py[line:274] INFO current qsize: 1, debug_size: 0
  39 pid:24831 wechatlog : 2017-10-10 07:07:55,813 maintain_task.py[line:274] INFO current qsize: 1, debug_size: 1
  40 pid:24831 wechatlog : 2017-10-10 07:07:57,813 wechat_middle.py[line:25] INFO in debug mode, queue id 139972199063056, size 1
  41 pid:24831 wechatlog : 2017-10-10 07:07:59,816 wechat_middle.py[line:31] ERROR in debug mode, queue get nothing.
  42 pid:24831 wechatlog : 2017-10-10 07:07:59,816 maintain_task.py[line:274] INFO current qsize: 1, debug_size: 1
  43 pid:24831 wechatlog : 2017-10-10 07:08:00,817 maintain_task.py[line:281] ERROR queue is empty
  44 pid:24831 wechatlog : 2017-10-10 07:08:00,818 maintain_task.py[line:283] ERROR the message block the queue: None
2017-10-10 06:38:56到2017-10-10 07:07:50之间,gunicorn日志报告此:

 [2017-10-10 06:41:08 +0800] [23906] [CRITICAL] WORKER TIMEOUT (pid:24838)

我的代码:

maintain_task.py
def wechat_push_thread(queue):
    we = wechat_middler_ware(queue=queue)
    wechat_log_logger = configs.make_logger_handler('wechatlog', filename='wechat')
    wechat_log_logger.info(f'queue id: {id(queue)}')
    debug_size = 0
    while True:
        try:
            we.listen(2)
        except Exception as e:
            wechat_log_logger.exception(e)
        # for debug
        if queue.qsize() > 0:
            wechat_log_logger.info(f'current qsize: {queue.qsize()}, debug_size: {debug_size}')
            if debug_size == queue.qsize():
                if we.debug_flag:
                    try:
                        msg = queue.get(timeout=1)
                    except Empty:
                        msg = None
                        wechat_log_logger.error(f'queue is empty')
                    wechat_log_logger.error(f'the message block the queue: {msg}')
                we.debug_flag = True
            debug_size = queue.qsize()
        else:
            we.debug_flag = False
            debug_size = 0
        # endfor debug
        if quit_event.wait(timeout=2):
            break
    logger.info('wechat_push_thread clean env')

wechat_middle.py
class wechat_middler_ware:
    def __init__(self, queue):
        self.q = queue
        self.logger = configs.make_logger_handler('wechatlog', filename='wechat')
        self.push_api = Push_Server(logger=self.logger)
        self.debug_flag = False

    def listen(self, timeout):
        while True:
            if self.debug_flag:
                self.logger.info(f'in debug mode, queue id {id(self.q)}, size {self.q.qsize()}')
            try:
                msg = self.q.get(timeout=timeout)
                self.logger.debug(f'recive {msg}')
            except Empty:
                if self.debug_flag:
                    self.logger.error(f'in debug mode, queue get nothing.')
                break
            else:
                ...

1 个答案:

答案 0 :(得分:0)

根据py doc:

  

警告如果在关联进程使用a时使用此方法   管道或队列然后管道或队列容易被破坏   其他过程可能无法使用。同样,如果过程有   获得锁定或信号量等,然后终止它是负责任的   导致其他进程陷入僵局。

gunicorn master杀死了“超时”工作者,因此队列已被其他进程无法使用。 现在,我使用multiprocessing.manager.queue而不是multiprocessing.queue。即使工人被主人杀死也能很好地工作。