Slow Supervisor Boot Time

[lswith]

Context: I've added the prometheus_ex library to my phoenix application. When I do this, my mix tasks and overall boot time for the whole application has gone up by a minute. I've started trying to determine where this occurs, and it's definitely happening the moment this application is added to the supervision tree.

Any thoughts on how I can improve runtime boot time?

[deadtrickster]

This might happen because of collectors autodiscovery. To check that you can configure collectors list manually (note default alias) https://github.com/deadtrickster/prometheus.erl#configuration

[lswith]

I love the quick response!

I've actually set the default_collectors to an empty list to test the boot time performance and its still very similar.

[lswith]

So, setting the instrumenters to an empty list as well as the collectors solved the boot time issue. Any ideas what might be the reason for this?

[deadtrickster]

Not sure, what you mean by instrumenters?

[lswith]

you can configure config :prometheus, instrumenters: [].

[lswith]

I think the issue might be in this module: https://github.com/deadtrickster/prometheus.erl/blob/master/src/prometheus_misc.erl

Getting the default instrumenters seems very slow.

[lswith]

I'm just curious what the default instrumenters are? if I wanted to set this configuration to the default, what would it be?

[deadtrickster]

hmm looks like this is a leftover from the past, I was trying to automate instrumenters set up like declaring metrics, but due interactive/embedded mode differences it's not possible to do that reliably. As a temporary fix please add instrumenters: [] to the prometheus config. I'll remove it entirely I think.

[dylan-chong]

Any progress on optimising this? The module_info spam is still causing 9 seconds of startup time for me, a bit annoying for a fast TDD cycle when tests take so long to start