Skip to content

Remove inf1/inf2/trn1 from RunsOn default spot instance families - #275

Merged
Erik Osterman (Cloud Posse) (osterman) merged 1 commit into
mainfrom
osterman/runs-on-remove-inference-families
Aug 18, 2026
Merged

Remove inf1/inf2/trn1 from RunsOn default spot instance families#275
Erik Osterman (Cloud Posse) (osterman) merged 1 commit into
mainfrom
osterman/runs-on-remove-inference-families

Conversation

@osterman

Copy link
Copy Markdown
Member

Summary

  • Removes inf1.*, inf2.*, and trn1.* from the family: allowlist in the default RunsOn runner profile
  • Cloud Posse does not run inference or training workloads, so these Inferentia/Trainium accelerator families have no business being in the general CI runner pool — they were swept in by the aws ec2 describe-instance-types generation query despite the file's own comment stating the intent to limit to standard (A, C, D, H, I, M, R, T, Z) instance classes
  • Root cause of a MaxSpotInstanceCountExceeded error: the Inf spot quota (L-B5D1601B) is only 32 vCPUs (vs. 1230 vCPUs for the standard family quota, L-34B43A08) and was found fully saturated, forcing CI jobs to fall back to expensive on-demand pricing for hardware they never needed

Test plan

  • Confirm CI jobs across the org continue to schedule successfully on the default/common/small/medium/terraform/large/xlarge runner profiles after merge
  • Confirm MaxSpotInstanceCountExceeded no longer recurs for inf1/inf2/trn1

Cloud Posse does not run inference or training workloads, so these
Inferentia/Trainium accelerator families should not be in the default
CI runner pool. They were swept in by the aws ec2 describe-instance-types
filter used to generate this list, despite the comment above stating the
intent to limit to the standard (A, C, D, H, I, M, R, T, Z) instance
classes.

Their AWS spot quota (L-B5D1601B, 32 vCPUs) is far smaller than the
standard families' quota (L-34B43A08, 1230 vCPUs) and was found fully
saturated, triggering MaxSpotInstanceCountExceeded and an expensive
fallback to on-demand pricing for CI jobs that never needed accelerator
hardware in the first place.

Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
@mergify mergify Bot added the triage Needs triage label Aug 18, 2026
@mergify

mergify Bot commented Aug 18, 2026

Copy link
Copy Markdown

Important

Cloud Posse Engineering Team Review Required

This pull request modifies files that require Cloud Posse's review. Please be patient, and a core maintainer will review your changes.

To expedite this process, reach out to us on Slack in the #pr-reviews channel.

@mergify mergify Bot added the needs-cloudposse Needs Cloud Posse assistance label Aug 18, 2026
@mergify mergify Bot removed the triage Needs triage label Aug 18, 2026
@osterman
Erik Osterman (Cloud Posse) (osterman) merged commit 3911c66 into main Aug 18, 2026
3 checks passed
@osterman
Erik Osterman (Cloud Posse) (osterman) deleted the osterman/runs-on-remove-inference-families branch August 18, 2026 03:09
@mergify mergify Bot removed the needs-cloudposse Needs Cloud Posse assistance label Aug 18, 2026
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants