Elastic Heap size calculation not correct with K8S 1.26


we run several elastic clusters(all with 3 nodes). And until now without any issues until we've upgraded vom k8s version 1.24.8.

We don't have any JVM options set and the calculation from elastic was always fine.
The host has 64GB MEM but elastic has a memory limit at 16GB.

Currently when the instances starts, one of them will try to use much more memory than the other two.
The Heap at start is absurdly high and the pod get OOM killed. So we raised the limit to 32gb MEM so that the pods are able to start and the heap looks like this:

But after a while the Heap get's to the level before we've upgraded k8s:


That is the same cluster and the only thing we've changed is the k8s version.
We also have one cluster left with the old version and there were able to restart everything and elastic is not having such behaviour.

Does anyone encounter the same issue? And more important, could someone try to help me understand what is going on with elastic on startup that it needs so much MEM at startup and then levelling everything after a while?

What is the output from the _cluster/stats?pretty&human API?

This topic was automatically closed 28 days after the last reply. New replies are no longer allowed.