How to debugging memory resource usage?

[da young Shin]

Hi Fluentd members,
I'm now moving my daemonset Fluentd to sidecar structure, because the whole cluster was moved from eks to fargate-eks.

In daemonset, Fluentd pod had 100m/512Mi of cpu/memory request, limit amount,
while current sidecar Fluented container has 100m/200Mi for each resources.

My sidecar is sometimes( 1~2 times every 3 days roughly) killed by OOMKilled error, and it ruins whole Nginx pods for few minutes.

I'd like to check if this is because unchecked memory leak in the image I use,
or simply because the memory limit/request is low.

So my question is..

1.  Is there a good fluentd memory tracing/profile tool that can be used in kubernetes? I looked at this and that, but it seems that it cannot be applied to our system because it is obsolete or requires python.
2. Would  you tell me how much your fluentd spend cpu/memory ? Of course, I'll adjust it through a load test later, but it will be helpful if I know the average cpu/memory utilization of yours 

[Kentaro Hayashi]

Hi,

Even though it is not bundled by default, and I'm not sure to match your needs,

you may be able to use fluent-rubyprof.

https://github.com/sonots/fluent-rubyprof

It requires @type debug_agent in .conf.

Then, executing start/stop to tracing/profile.

$ fluent-rubyprof -m MEMORY start

$ fluent-rubyprof stop

Regards,

2020年11月23日月曜日 23:01:14 UTC+9 shin...@gmail.com: