那是 window cleanup timer。
Flink timer 需要存储在 state ，所以没法像普通程序那样可以存储丰富的回调上下文。当 timer 触发时，operator
基本都是把这个 timer 派送给所有可能的接受者，接受者需要过滤掉非期望的 timer 回调。在 WindowOperator 中，即使
cleanup timer 不是 trigger 注册的，trigger 也会收到 cleanup timer 的回调，trigger
需要自己过滤掉这个回调。你可以看下 EventTimeTrigger,
Thank you, @David Anderson and @Fuyao Li. This answered my question and
cleared my confusions.
On Fri, Mar 26, 2021 at 11:08 AM David Anderson
wrote:
> Prometheus is a metrics system; you can use Flink's Prometheus metrics
> reporter to send metrics to Prometheus.
>
> Grafana can also be
FYI, we experience a similar error again, lost leadership but not due to
timeout but a disconnect from zookeeper. This time I examined logs for other
errors related to zookeeper and found the kafka cluster that uses the same
zookeeper also was disconnected.
We run on AWS and this seems to be
