Use Large Runners in Build Jobs #142
Conversation
|
Seems like 8 or 16 cores probably is the sweet spot. Billing is per-minute, so currently 16 cores isn't as bad as it looks, as it finishes in under 2 minutes, while 8 cores will be billed for the full 3 minutes. Of course if build times increase this could flip. |
|
@pdewilde I created a new runner group (Beefy) with two machine configurations (4-core and 8-core). |
…-experiment # Conflicts: # .github/workflows/ci.yml
…-experiment # Conflicts: # .github/workflows/ci.yml
pdewilde
changed the title
Signed-off-by: Parker DeWilde <[email protected]>
| @@ -79,7 +79,8 @@ jobs: | |||
|
|
|||
| # Build github-metrics-aggregator image and push to artifact registry | |||
| build: | |||
| runs-on: 'ubuntu-latest' | |||
| runs-on: | |||
| group: 'Beefy' # group with a 4-core and 8-core runner for faster builds | |||
There was a problem hiding this comment.
@pdewilde - will it randomly choose one? I created the group because that's the only way to restrict the runners to a specific repo. I was thinking you would use either the 4-core or 8-core machine label directly.
There was a problem hiding this comment.
Aah, I assumed you intended for me to use the group for some reason (maybe so a slow job on the 8-core machine doesn't block all CI-jobs in the beefy group). I'm not sure what the logic is. I did one test and saw it corresponded to the 8-core machine and was hoping it would use the more powerful machine first.
I'll have to take a look and see if it seems to be consistently using the 8-core machine.
There was a problem hiding this comment.
Looks like there is a long-existing ask but not movement on such a prioritization feature in a year. (That is for Self-Hosted rather than Large runners, but I would imagine the behavior is similar).
Probably best to just slap on the 8-core machine label directly. Probably no need to have a 4-core for now in the Beefy group as most of the other jobs run in pretty similar times.
The existing CI build job took over 5 minutes to run. This will use a new runner group locked down to the github-metrics-aggregator repo called "beefy" which has a 4-core and a 8-core runner to help speed up ci jobs significantly.