Skip to content

ci: exercise branch checks on gke runners - #5867

Draft
TaylorMutch wants to merge 6 commits into
mainfrom
ci/5866-gke-branch-checks/TaylorMutch
Draft

TaylorMutch wants to merge 6 commits into
mainfrom
ci/5866-gke-branch-checks/TaylorMutch

Conversation

@TaylorMutch

@TaylorMutch TaylorMutch commented Nov 20, 2026 •

Copy link
Copy Markdown
Contributor

Summary

Exercise Linux amd64 Branch Checks on GKE ARC runners. Trusted same-repository jobs use ARC; fork PRs, ARM jobs, and macOS jobs retain hosted runners. Trial isolated persistent Cargo caches for x86 lint and tests.

Related Issue

Closes #5866

Changes

  • Route trusted Linux amd64 jobs to ARC and configure shared memcached sccache.
  • Reuse the runner image's configured Nix cache.
  • Route x86 lint/tests to openshell-ci-amd64-pd-lint and openshell-ci-amd64-pd-test, each with one worker and a dedicated exclusive 200 GiB SSD disk.
  • Select a repository/Nix-shell-specific persistent target directory shared by each job's Cargo commands. Skip GitHub Cargo archive restore/save for these workers.
  • Keep checkouts and credentials ephemeral; document cache and runner diagnostics.

Release Notes

Testing

Job Empty disk Warm disk Execution improvement
x86 Rust lint 14m48s 8m04s 45.5%
x86 Rust tests 17m30s 10m40s 39.0%

Times include setup and cleanup but exclude queue time. Lint compile/check log counts fall from 475/1160 to 12/102. Test compile/check counts fall from 1525/206 to 974/112. The replacement pods start with 4 GiB and 31 GiB target directories respectively, and neither restores nor saves a GitHub Cargo archive. Test cache reuse succeeds after reattachment on a different node. Warm test queue time is 4m05s due to build-pool scale-up/image startup, versus 2m45s on the empty pass; including queue time, test duration improves from 20m15s to 14m45s. One pair of runs is observational and does not isolate CPU contention or sccache warming. An earlier run was canceled to correct pod-environment detection and is excluded from measurements.

The persistent disk trial also consolidates each job's target directories, so comparison with the older archive experiment does not isolate storage alone. Empty/warm runs on the same commit measure reuse across runner replacement. Fresh checkout timestamps and different build configurations can still cause recompilation. Trial lane concurrency is one job per lane; the ordinary ARC pool remains 5–20. Nix sandbox and trust settings are unchanged.

Checklist

  • Follows Conventional Commits
  • Commits are signed off (DCO)
  • CI guidance and monitoring skill are updated

Signed-off-by: Taylor Mutch <taylormutch@gmail.com>
@github-actions github-actions Bot added the state:accepted A maintainer decided OpenShell should pursue this issue label Nov 20, 2026
Signed-off-by: Taylor Mutch <taylormutch@gmail.com>
Signed-off-by: Taylor Mutch <taylormutch@gmail.com>
Signed-off-by: Taylor Mutch <taylormutch@gmail.com>
Signed-off-by: Taylor Mutch <taylormutch@gmail.com>
Signed-off-by: Taylor Mutch <taylormutch@gmail.com>
Comment on lines +53 to +70
run: |
set -euo pipefail
# Pod environment is visible to the shell, not the workflow env context.
if [[ -z "${OPENSHELL_PERSISTENT_CACHE:-}" ]]; then
exit 0
fi
echo "enabled=true" >> "$GITHUB_OUTPUT"
test "$GITHUB_REPOSITORY" = ai-openshell/openshell
test "$OPENSHELL_PERSISTENT_CACHE" = /mnt/ci-cache
mountpoint -q "$OPENSHELL_PERSISTENT_CACHE"
cache_root="$OPENSHELL_PERSISTENT_CACHE/ai-openshell-openshell/$NIX_SHELL_HASH"
mkdir -p "$cache_root/target"
echo "CARGO_TARGET_DIR=$cache_root/target" >> "$GITHUB_ENV"
echo "Persistent cache lane: $OPENSHELL_CACHE_LANE"
echo "Runner pod: $RUNNER_NAME"
echo "Target directory: $cache_root/target"
du -sh "$cache_root/target"
df -h "$cache_root/target"
@github-actions

Copy link
Copy Markdown

Compatibility review (advisory)

New or updated breaking changes; see the full report for all findings.

1. rust-sdk — No explicit risk acceptance found

The exhaustive openshell_sdk::raw::proto::ProviderCredentialRefreshStrategy enum adds GithubAppInstallation to its seven baseline variants.

  • Impact: Existing SDK consumers matching all seven variants without a wildcard no longer compile, although existing gRPC client calls remain wire-compatible.
  • Action: Preserve the baseline enum in v0.1.10; defer its breaking expansion to a minor train with notice and wildcard-match migration guidance.
  • PR: ci: exercise branch checks on gke runners #5867
Review details
  • Execution: complete
  • Candidate: PR #5867 (5133d10c3d56c2fae5bba2d18b411a4b0818390f)
  • Baseline: a1c1e42640fc8c368e4a85e584ab324a33d0f91b (a1c1e42640fc8c368e4a85e584ab324a33d0f91b)
  • Reviewed: rust-sdk, python-sdk, api-behavior, cli, configuration, policy, helm, state

Advisory review; publication gates are unchanged.

Branch check run

This branch has not been deployed

No deployments
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

state:accepted A maintainer decided OpenShell should pursue this issue

Projects

None yet

Development

Successfully merging this pull request may close these issues.

ci: exercise gke runners with branch checks

2 participants