epoll_ctl: File exists, errno=17

[solsson]

I run fluent/fluent-bit-kubernetes-logging#16 with modest load and very tight memory limits, 10MB. This triggers occasional restarts of the daemonset pods. Despite this it all works very well, but now I have one pod that repeatedly logs the following error:

[2018/01/22 09:08:15] [  Error] epoll_ctl: File exists, errno=17 at /tmp/src/lib/monkey/mk_core/mk_event_epoll.c:133

Before I kill the pod, is there anything I should investigate?

[edsiper]

@solsson is the HTTP Server enabled ?, if so, if you disable it do you see the same issue ?

[solsson]

What HTTP Server are you referring to?

[StevenACoffman]

Prometheus endpoint requires the HTTP server (monkey is the one in fluent-bit). The config that enables the HTTP_Server is here in the fluent-bit-configmap

        HTTP_Server   On
        HTTP_Listen   0.0.0.0
        HTTP_Port     2020

[solsson]

Thanks. Yes, I run the exact config from fluent/fluent-bit-kubernetes-logging#16, bar memory limit changes for fluent/fluent-bit-kubernetes-logging#18. The pod is gone now, so I'll have to see if I run into this again, and next time I will check if I get any prometheus metrics for the pod.

[edsiper]

are you seeing the issue with the latest fluent/fluent-bit-0.13-dev:0.6 image ?

[solsson]

Nope, not seen. It might have been a rare error though, and the new tag hasn't been running for long. Ok with me if you want to close the issue. I can post an update if I see it again.

[edsiper]

thanks, I will close it for now.