CollectD - 5.11 : Memory leak with libnss about collectd HOT 3 CLOSED

Run COllectD 5.11 on Centos7 and valgrind
collectd-wr5.log

All leakage in that Valgrind log is from Bash, not collectd?

from collectd.

thank you for checking @eero-t . Yes, to me it looks like the leaks were from bash.

from collectd.

While it's possible that collectd process had increased memory usage when POD running collectd gets killed due to OOM, it's impossible to say either way, as the attached file was about Bash.

There was also no log of the actual OOM event; was it e.g. whole system OOM, or just result of collectd POD spec specifying too low memory limits for such a long running process. Kernel OOM info does not state whether something leaks, it just states how much memory processes used, and which one of them was killed as result of the OOM situation. Unless processes are prioritized differently, killed process is the one taking most memory in given cgroup.

And even if collectd process memory usage would increase over time, it's not necessarily a leak, or from collectd itself. It could be from a system library, and/or due to data it handles just increasing over time for some valid reason.

PS. Small leaks are very hard to catch, unless they happen right from the beginning (so that something like Valgrind can be used to catch them). Most memory tools have such large overheads that you cannot run them in production, you need special tools to track them [1]. It may be easier just to increase memory limits a bit and/or restart processes at suitable intervals.

[1] Used in Maemo/MeeGo: https://github.com/maemo-tools-old/functracer

(Nowadays similar functionality could probably be written on top of kernel uprobes.)

from collectd.