This message was deleted.
# troubleshooting
s
This message was deleted.
s
I think the best person to answer that is @Cory Johannsen. Cory has been working on such a deployment and made improvements to the extension.
y
If I can do without zk, I will 🙂
c
There's an open PR with a fix for a resource leak. We run a locally built version of druid that includes this fix. It's not perfect, but it does work around the underlying issue.
We automatically cycle the coordinator pods every ~12 hours to force a leader change and clean up any leaked http connections
We run this configuration in our staging and production environments and so far it has been stable.
y
Nice! I want to use in an early stage Prod.
Will only start being customer facing in ~ 3months
c
If you use the extension as-is (without the fix in place) you will see the leader coordinator leak an http connection every time an indexer pod is removed.
y
When will the fix be GA?
Which PR are you referring to? We could just build a version for us
s
Additionally, there are also options to avoid ZK for certain functions. In Apache Druid 25.0 most of these are now the defaults. Check out the PR: https://github.com/apache/druid/pull/13092
y
Great. Thanks, I'll check it out
c
I would love to test is using the extension that removes the middlemanager actually works around this issue, I believe it may. I have not had time.
y
If we can help, let me know