Hey all, the ECK operator is doing amazing things for me. The only thing I'm noticing is that on upgrades, my 3 coordinator pods seem to roll one right after the other and I wind up with a non-responsive queries on my ingress.
Just wondering what the operator waits for before terminating the next coordinator pod and if I can change/check it. Started looking through the code to see what was happening, but that looks like it'll take me a while
We are searching the clusters through CCS Proxy Connections -> Load Balancer -> Ingress to pods
We see all the proxy connections drop, so we drop the remote cluster and add it, which fixes that. But I'm wondering if they rolled slower, would the proxy connections stay up.
Have you configured a more aggressive update strategy with
maxUnavailable set to higher value than 1?
We did not, but thanks for that link. We tried just manually rolling the pods one by one and the Proxy connections still drop to zero. We've just implemented a second route through the Load Balancer and sidecar mechanism that flip/flops the proxy connection from one route to the other if the number of proxy connections drop below 18. This has resolved it, and it's the same thing we had to do with Seed connections as well. Elastic seems to latch onto the remote IPs and won't fallback to the load balancer once it has some.
This topic was automatically closed 28 days after the last reply. New replies are no longer allowed.