Skip to content

RELOPS-2375: tear down SP3 GPU perf dedicated host - #338

Merged
jwmossmoz merged 1 commit into
masterfrom
relops-2375-teardown-sp3-perf
Sep 14, 2026
Merged

RELOPS-2375: tear down SP3 GPU perf dedicated host#338
jwmossmoz merged 1 commit into
masterfrom
relops-2375-teardown-sp3-perf

Conversation

@jwmossmoz

@jwmossmoz jwmossmoz commented Sep 14, 2026

Copy link
Copy Markdown
Contributor

Summary

Removes the gecko-t/win11-64-25h2-gpu-perf-experiment dedicated host (host group + host) that was added as a temporary Speedometer 3 GPU placement experiment.

Why

The resource is tagged lifecycle=temporary-test, ttl=destroy-after-experiment, purpose=speedometer3-dedicated-host-placement-test, but it was never torn down after the experiment concluded. This removes the leftover dedicated host (single NVadsA10v5_Type1 host in RG-WEST-US-3-GECKO-T), which continues to run 24/7 by reservation regardless of task load.

Changes

  • Delete terraform/azure_fxci/gecko-t/dedicated-host-sp3-perf.tf (host group, host, and outputs). No other references to these resources/outputs exist.
  • The reusable taskcluster-arm-template-dedicated-host template spec in template_specs.tf is intentionally left intact.

Landing order

Land the companion fxci-config change (removing the gpu-perf-experiment worker pool) first so worker-manager stops placing VMs on the host, then terraform apply here to destroy the host.

The gecko-t/win11-64-25h2-gpu-perf-experiment dedicated host was added as a
temporary Speedometer 3 GPU placement experiment (tagged
lifecycle=temporary-test, ttl=destroy-after-experiment). It has been billing
~$172/day (~$5.2k/mo) continuously since May, including weekends, because a
dedicated host bills 24/7 by reservation regardless of task load.

Remove the host group and host to stop the spend. The reusable
taskcluster-arm-template-dedicated-host template spec is left intact.
@jwmossmoz
jwmossmoz merged commit 8d019b0 into master Sep 14, 2026
1 check passed
@jwmossmoz
jwmossmoz deleted the relops-2375-teardown-sp3-perf branch September 14, 2026 16:09
jwmossmoz added a commit to mozilla-releng/fxci-config that referenced this pull request Sep 14, 2026
…1211)

The SP3 GPU perf experiment dedicated host has been torn down in
relops_infra_as_code (mozilla-platform-ops/relops_infra_as_code#338).
Remove the corresponding worker pool so worker-manager stops attempting to
provision onto the now-deleted dedicated host.
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.

1 participant