Bug Report : Memory leak when readiness probe Curls to check the cluster health when security is enabled

[Cyril_James]

@DavidTurner

consumer despite only having 1.5MB resident

Thanks David. Then we may have to look more into the part where we do the curl request and also verify any bug with kernel causing this memory issue.

[Marc_Wolff]

Hi Cyril,
have you found a solution for that problem?

We have the same issue with https.
Our settings are close:
CentOS 7.6
K8s: 1.14
Docker 18.09.7

I think the RSS from the java process is not the problem.
But i don´t know why the node mem limit is reached.

memory.stats:
rss 1451929600

memory.usage_in_bytes:
2137989120 and growing

[Marc_Wolff]

My workaround for this problem is change readinessprobe

httpGet:
httpHeaders:
- name: Authorization
value: Basic [base64 encoded user:pw]
path: /_cluster/health?local=true
port: 9200
scheme: HTTPS

[Cyril_James]

This seem to be an issue caused by Centos kernel bug.