Metric check for listen queue len seems wrong

[snc]

Based on https://serverfault.com/a/355546 the listen queue len value is decreased on connection. So initially the value might be for example 128. When I run the check via php-fpm-healthcheck --listen-queue-len=10 it fails with 'listen queue len' value '128' is greater than expected '10'. In this case this should only fail if the value of listen queue len is lower than 10.

[renatomefi]

Hey @snc thanks for reporting, after reading the issue you pointed out and trying to remember how I coded it what I understand of the metric is:

• If the queue len increases too much this means your container is not handling the requests
• by setting the --listen-queue-len=10 it'll be seen as unhealthy since your expected length is less than 10

I might have interpreted it wrongly, but would that make sense? Otherwise would you have a suggestion on how we can tweak this metric?

For now you could not use it since it might not be clear, if we find a conclusion we can document it better!

Thanks

[achton]

It seems to me that listen queue len is a static number? Like an upper limit for the listen queue value (which is the real-time status of the connection queue). In which case checking anything against that value doesn't make sense.

Which means that the example in the README - "check if you have more than 10 processes in the queue" - would make better sense if it was based on listen queue or max listen queue fields. Eg:

And you can also check if you have more than 10 processes in the queue:

$ php-fpm-healthcheck --accepted-conn=3000 --max-listen-queue=10
$ echo $?
0

Admittedly, I haven't dug into the php-fpm docs for the meanings of these values, but that is my reading of it.

[renatomefi]

It's very difficult to find information about it, even the code doesn't have much documentation either, one thing that I found says it's the amount of pending connection, check this article: https://easyengine.io/tutorials/php/fpm-status-page/

So as @achton said we could use the listen-queue metric instead in the documentation, since using the listen-queue-len might be useful only on very specific use cases!

If you find more material about it please let me know!

Also I found this very interesting blogpost on how to monitor phpfpm: https://www.malasuk.com/linux/php-fpm-status-page/

[renatomefi]

Hello everyone, have you seen the changes from #25?

It addresses this issue here, and has a nice update to the docs and explanation in the PR!

Thanks a lot @tjespers!