search

thanos-io / promql-engine

Multi-threaded PromQL engine implementation based on the Volcano paper.
Apache License 2.0
141 stars 54 forks source link

Optimize Equinix usage #448

Closed jeefy closed 4 months ago

jeefy commented 5 months ago

Hello! Friendly neighborhood CNCF person here.

We're looking to reduce Equinix usage where possible. This is all related to https://contribute.cncf.io/resources/newsletters/2024-02-21-projectnewsletter/#equinix-updates and the resource usage policy linked further below. :)

We see you're using both a pretty beefy machine that, based on bandwidth alone, looks to sit mostly idle. Since these are bare-metal machines, could you instead tear down that machine and only spin something like that up when a benchmark actually needs to run? :)

Additionally, is there a reason you're running a dedicated GHA runner instead of using the GitHub provided runners?

Thanks!

cc @vielmetti (Awesome Equinix staffer :sunglasses:)

GiedriusS commented 5 months ago

Thank you for the message and sorry for the problems. I wanted to use on-demand provisioned physical servers but never found the time to implement this. Currently it runs our continuous benchmarking job but it has been broken for a long time - the website doesn't show newest entries, it sometimes times out or even gets OOM killed. Given that it's basically unmaintained, I think we can just destroy the machine/job and get back to it when we have the time to do it properly. @fpetkovski do you have any thoughts?

fpetkovski commented 5 months ago

Decomissioning the machine for now makes sense. I think we should go back to running our light weight benchmarks using shared GH runners and we can reconsider a dedicated machine if we need it again.

GiedriusS commented 4 months ago

https://github.com/thanos-io/promql-engine/pull/449

vielmetti commented 4 months ago

Thank you!