Github user tzulitai commented on the issue: https://github.com/apache/flink/pull/4935 Hi @Aitozi, sorry for the long delay in relaying back to this PR. I'm still not convinced that this is a sane solution. For example, what is a "good" setting for the `KEY_REGISTER_TIMES` property? Isn't 1 enough, since you mentioned that the missing metric is registered in Kafka after the first poll. Making this configurable seems unnecessary to me. I wonder if we can try the following two approaches: 1) Manually register the metrics that we know would be missing before the first poll, or 2) Poll once first outside the loop just to make sure that all Kafka metrics are existent, perform the metrics registration, and then start the regular fetch loop. What do you think?
---