we are using a Java process with jemalloc on Kubernetes, and occasionally the pod gets OOMKilled. Just before the kill, there is a sharp spike in memory usage and the pod gets killed. I have saved the jemalloc logs at - HERE .
My questions are:
- How to interpret the jemalloc log to find out the amount of fragmentation and available space in various arenas?
- How to limit the number of arenas and ask to unmap the mapping faster to help reduce the peak memory usage?
Thanks!