REQUEST: Repository maintenance on `opentelemetry-rust`

[scottgerring]

Affected Repository

https://github.com/open-telemetry/opentelemetry-rust

Requested changes

Please grant access of https://github.com/organizations/open-telemetry/settings/actions/runner-groups/8 (Benchmark Bare Metal Runners) to this repository, as was done for opentelemetry-go in #2266. We're running our benchmarks in the regular pools and the results are slow and variable.

Related: open-telemetry/opentelemetry-rust#2706

Purpose

Enable bare metal runners to run benchmarks.

Runner group name: 'Benchmark Bare Metal Runners'

Expected Duration

permanently

Repository Maintainers

* [@open-telemetry/rust-maintainers](https://github.com/orgs/open-telemetry/teams/rust-maintainers)

Adding @cijothomas for visibility

[trask]

done! please check and close the issue once you've confirmed it's working, thanks

[scottgerring]

@trask thanks! I'll test it out in the next couple days and loop back to you.

[scottgerring]

Hey @trask , our job doesn't seem to be picked up by the runners. I'm not sure if there's some other configuration we need - I think I saw something, somewhere, about whitelisting particular workflows?

Here's a blocked run and here's the config - but really I've just lifted the runs-on from opentelemetry-go:

jobs:
  runBenchmark:
    name: run benchmark
    permissions:
      pull-requests: write
    runs-on: self-hosted

[trask]

I think I saw something, somewhere, about whitelisting particular workflows?

oh, you're right!

here's the current allowed workflows:

open-telemetry/opentelemetry-collector-contrib/.github/workflows/load-tests.yml@refs/heads/main,
open-telemetry/opentelemetry-go/.github/workflows/benchmark.yml@refs/heads/main,
open-telemetry/opentelemetry-java/.github/workflows/benchmark.yml@refs/heads/main,
open-telemetry/opentelemetry-java/.github/workflows/benchmark-tags.yml@refs/heads/main,
open-telemetry/opentelemetry-js/.github/workflows/benchmark.yml@refs/heads/main,
open-telemetry/opentelemetry-python/.github/workflows/benchmarks.yml@refs/heads/main

it looks like we're only running these on main in order to conserve the limited bare metal runners, can you do something similar in rust repo?

[scottgerring]

We're new to this in Rust-land, so this might not be the best idea, but we've made it contingent on setting a label on the PR so it runs:

https://github.com/open-telemetry/opentelemetry-rust/blob/18e87185531c723dc76bd0dd404889535b22b10e/.github/workflows/pr_criterion.yaml#L9 - this way we can look at a particular PR and say "what's the impact of this thing we're suspicious of on performance", through the label on, and opt into the benchmarks:

    if: ${{ contains(github.event.pull_request.labels.*.name, 'performance') }}

Do you think we could get away with a wildcard to support this, or are we going a bit off the rails?

[trask]

hm, it looks like the entries in that list "must be pinned to a ref, tag, or full SHA".

so I could add

open-telemetry/opentelemetry-rust/.github/workflows/pr_criterion.yaml@refs/pull/2813/merge

but not something that matches any PR ref branch

I'm guessing the concern with removing this restriction is that any PR could add a workflow that uses the bare metal runner

cc @jpkrohling @open-telemetry/technical-committee as admin for the bare metal runners in case they have some background on this

[scottgerring]

Hey @trask thanks for the info!

I'll leave it using shared workers for our opt-in PR perf regressions then, and setup the job to do HEAD vs HEAD~1 perf diff on main using this pool for a "higher quality comparison".

I figure this way we get the benefit of the bare metal runner for high-fidelity testing on main and follow the same pattern as the other projects.

What do you think?

I'll see if I can structure this within the same workflow file nicely, or not, and ping back on this thread with the full reference to add to the entries list 💪

[scottgerring]

Hey @trask , thanks for the patience!
Can you bless this job ?

The workflow file is open-telemetry/opentelemetry-rust/.github/workflows/benchmark.yml and it runs on pushes to main

https://github.com/open-telemetry/opentelemetry-rust/actions/runs/13948851079/job/39042702985

[trask]

done, give it another try

[scottgerring]

it runs! But, I don't have a good idea what's in the env - we're missing e.g. unzip -> https://github.com/open-telemetry/opentelemetry-rust/actions/runs/13950335436/job/39047893019

I'm not sure what is the easier path forward. If it's easy enough to add things to the env - e.g. unzip - maybe we can start there? But it's hard to say how deep the rabbit hole will go.

Alternatively, i've made a branch that you could explicitly bless, and I can then use that to fix the build, then notify you i'm done, then un-bless it. Hacking on it it in main is pretty undesirable.

https://github.com/scottgerring/opentelemetry-rust/tree/chore/fix-ci-environment-deps

Happy to follow your lead here, and apologies this is fiddlier than I expected!