Camunda 8.5 - Zeebe broker is not reclaiming buffer/cache memory when system is idle

Camunda 8 Topics
,
1

Hi All,

We are running load test on Camunda 8, in that test we have observed pod memory is gradually increasing during the test, however it is not being reclaimed once test is completed and system is idle.

Please find the below configuration.
No of zeebe instance :3
No of partitions : 6

resources:
requests:
cpu: 800m
memory: 2000Mi
limits:
cpu: 1600m
memory: 4000Mi
Below is the current state of broker which still hold 21150.8 in cache
MiB Mem : 32089.8 total,
1317.9 free,
10084.8 used,
21150.8 buff/cache

Please share you thought if anyone faced this challenge and configuration change to fix the issue.

2

Hi @Koksingh - I believe this is the expected behavior of the JVM. I’m not an expert in this area, but IIRC the garbage collector is quite greedy. Because the application just needed that many resources, the assumption is that it will likely be needed again and so it holds on to it. I would suggest looking into JVM tuning.