workers start to consume 100% CPU and stop working
tisba opened this issue · comments
Hey guys,
I'd could use some help/input on a problem I'm currently trying to debug. I'm using cluster to implement quite a simple [1] HTTP server and I'm not quite sure yet if the problem I'm seeing is my fault, or a cluster/node issue.
In production, even under quite normal load (~1k req/s) workers start to saturate a cpu core and not doing anything (strace told me so). If I let it continue to run without restarting, at some point no working workers are left :-/ Even if the request rate gets down over night the misbehaving workers keep using 100% cpu each.
Unfortunately I haven't much success at reproducing the problem on my local machine. The current "solution" is to restart cluster regularly.
[1] The server is just looking at the requests params, JSON-decodes a cookie, updates it and write a few bytes response.
hmm i definitely have not seen this behaviour
I'd be interested in ideas how to debug this. Since this is a comercial project I'm not allowed to share the code, but I hope that I can reproduce the issue maybe with code that I can share.
maybe try some of the v8 profiler / heap inspection tools
Yeah, I can try to use the profiler once I mange to reproduce this issue. Profiling the production system is maybe a bad idea :) The memory usage is pretty low (around 30-40M) and constant.