I'm running at 5.6.3 and have two coordinator nodes running 5.5GB HEAP in 8GB of RAM. I've started to see them being oom-killed when elasticsearch is the only service running on these VMs.
Do coordinator nodes have much in the way of off-heap memory requirements? I'm trying to understand what is causing the service to be killed.
depending on your system configuration that does not leave a whole lot of headroom. The OS reserves a certain amount of memory for itself and also the JVM has some overhead on top of what's reserved for the Java heap. You can see some of this overhead with native memory tracking but be aware that this has some performance overhead so you should probably not turn this on in production without prior testing.
Finally, Elasticsearch uses off-heap memory for various purposes. The node stats API provides data on on-heap and off-heap memory usage.
This topic was automatically closed 28 days after the last reply. New replies are no longer allowed.