Stanislav Kozlovski created KAFKA-8607:
------------------------------------------
Summary: Reduce AdminClient Metadata request rate when invalid
node id is given
Key: KAFKA-8607
URL: https://issues.apache.org/jira/browse/KAFKA-8607
Project: Kafka
Issue Type: Task
Affects Versions: 2.3.0, 2.2.0, 2.1.0, 2.0.0
Reporter: Stanislav Kozlovski
While testing KAFKA-7800 (KIP-412), we were playing around with the config
command CLI and [noticed that it hangs for very
long|https://github.com/apache/kafka/pull/6903#discussion_r297434016] when
given an invalid broker id.
After investigating a bit more, I noticed that we endlessly retry metadata
updates. Locally, my AdminClient issued 78 requests for 10 seconds - averaging
at a rate of 7.8 requests/sec. The call times out after 2 minutes by default -
we end up sending 1149 requests.
This respects the "retry.backoff.ms" config but it may be better to have some
sort of exponential backoff to ease the needless load on the cluster.
It is unlikely for this to be a high-impact change but it sounds worth it to
have the protection. Orchestration systems like Kubernetes make it easier for a
user to mass-deploy a wrong config and inadvertenly DDoS his cluster via
metadata requests
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
