Skip to content

ci: run every e2e job on a single kurtosis-cdk commit and allow an optional GCP identity for image pulls - #1977

Open
arnaubennassar wants to merge 4 commits into
developfrom
ci/single-kurtosis-cdk-pin
Open

arnaubennassar wants to merge 4 commits into
developfrom
ci/single-kurtosis-cdk-pin

Conversation

@arnaubennassar

@arnaubennassar arnaubennassar commented Nov 17, 2026 •

Copy link
Copy Markdown
Contributor

Summary

  • One kurtosis-cdk commit for every e2e job. KURTOSIS_CDK_COMMIT now points to 85fcdea (the merge of feat(aggkit): add the AggOracleCommittee v2, with GER validators 0xPolygon/kurtosis-cdk#1044, which has the AggOracle committee v2). The committee v2 job takes it from get-kurtosis-cdk-commit like the other jobs, so KURTOSIS_CDK_COMMITTEE_V2_COMMIT and the committee-v2-kurtosis-ref output are removed.
  • The nightly run tests the pinned commit. get-kurtosis-cdk-commit no longer switches to kurtosis-cdk main on schedule; every event uses KURTOSIS_CDK_COMMIT. Moving to a newer kurtosis-cdk is an explicit bump of that one value.
  • kurtosis CLI aligned. The multi-chain workflow installs kurtosis CLI 1.23.0 (with its published SHA-256), the version kurtosis-cdk installs at the pinned commit (kurtosis-pre-run default and its own CI). The nightly run on kurtosis-cdk main failed the multi-chain jobs with Cannot construct 'PortSpec' because CLI 1.22.2 did not match.
  • agglayer/e2e on a5e723b (main) everywhere, which adds the optional gcp-workload-identity-provider / gcp-service-account inputs of the single-chain workflow. The single-chain jobs pass them from the GCP_WORKLOAD_IDENTITY_PROVIDER and KURTOSIS_GCP_SERVICE_ACCOUNT repository variables. When the variables are not set the inputs are empty and kurtosis-pre-run keeps its default identity, so behaviour here does not change.

Test plan

  • zizmor --min-severity medium .github/workflows (v1.32.0): no findings
  • actionlint on both workflows: clean
  • Bats E2E Tests green on this branch: single-chain (OP pessimistic, op-succinct, AggOracle committee v2) and multi-chain (2 and 3 chains)

Pin KURTOSIS_CDK_COMMIT to the kurtosis-cdk commit that has the AggOracle
committee v2 (85fcdea) and use it for every e2e job, including the
committee v2 job, so the second committee-only pin goes away. The nightly
run uses the same pinned commit instead of following kurtosis-cdk main.

Move every agglayer/e2e reference to a5e723b (main) and pass the optional
gcp-workload-identity-provider / gcp-service-account inputs from the
GCP_WORKLOAD_IDENTITY_PROVIDER and KURTOSIS_GCP_SERVICE_ACCOUNT repository
variables. When they are not set the inputs are empty and kurtosis-pre-run
keeps its default identity.

Install kurtosis CLI 1.23.0 in the multi-chain workflow, the version that
kurtosis-cdk expects at the pinned commit.
@arnaubennassar arnaubennassar self-assigned this Nov 17, 2026
@arnaubennassar

Copy link
Copy Markdown
Contributor Author

@claude

@claude

claude Bot commented Nov 17, 2026 •

Copy link
Copy Markdown

Claude finished @arnaubennassar's task in 28s —— View job


Review of PR #1977

  • Read the diff against origin/develop
  • Checked for leftover references to the removed variables and for consistency across the workflows
  • Posted the review

The change is small and consistent. I found no blocking issues.

Findings

🔴 Required: none.

🟡 Suggested

  1. The nightly run no longer detects kurtosis-cdk drift (test-e2e.yml, get-kurtosis-cdk-commit). This is intentional and explained in the PR description, and I think it's the right call, because the PortSpec failure came from tracking main with a mismatched CLI. The cost is that nobody learns about breakage in newer kurtosis-cdk until someone bumps KURTOSIS_CDK_COMMIT. If you still want an early warning, consider a separate non-blocking scheduled job or workflow_dispatch input that runs against main.
  2. The kurtosis CLI version is coupled to KURTOSIS_CDK_COMMIT by a comment only. test-e2e.yml says the CLI in test-e2e-multi-chains.yml "must match", but nothing enforces it. A bump of one without the other will fail the same way the nightly did. A short comment next to KURTOSIS_VERSION in test-e2e-multi-chains.yml pointing back to KURTOSIS_CDK_COMMIT would make the coupling visible from both sides.
  3. get-kurtosis-cdk-commit is now mostly a pass-through. It checks out a pinned SHA and returns git rev-parse HEAD, which equals the input. It still gives an early failure if the SHA is unreachable, and it keeps the needs wiring and the Slack message unchanged, so keeping it is defensible. You could also drop the job and use env.KURTOSIS_CDK_COMMIT directly, but that is a larger diff and not required.

🟢 Optional

  1. The $GITHUB_OUTPUT quoting fix in get-commit is a good small hardening. Removing the determine-ref step also removes an expression interpolated directly into a shell run: block, which is a plus for zizmor.
  2. The multi-chain workflow still has no committee v2 variant, and test-e2e-exit_cenrtificate_tool.yml still pins kurtosis 1.14.1. Both look out of scope here. I'm only noting that the CLI versions across the e2e workflows are now uneven.

Things I checked

  • Leftover references: no references to KURTOSIS_CDK_COMMITTEE_V2_COMMIT or committee-v2-kurtosis-ref remain. AGGLAYER_CONTRACTS_COMMITTEE_V2_COMMIT is a different variable and is still used.
  • Job dependencies: the committee v2 job now has get-kurtosis-cdk-commit in needs, which it requires in order to read that job's output.
  • agglayer/e2e pins: every agglayer-e2e-ref and every uses: pin is now a5e723b, with no stale 75202d2 left anywhere.
  • Context availability: vars.* is allowed in jobs.<id>.with for reusable-workflow calls. env.* is allowed in the actions/checkout with:. Unset variables evaluate to an empty string, which matches the "keeps default identity" claim. They are passed as repository variables, not secrets, so there is no secret-handling concern. This relies on the a5e723b reusable workflow treating empty inputs as "unset", which I could not check from this repo, but your passing Bats E2E Tests run covers it.
  • Pinning: the new kurtosis CLI download keeps its SHA-256 verification. I couldn't verify the checksum value from the sandbox, but the multi-chain jobs passing in your test plan confirms it.

Summary template


🚀 What's New

  • All e2e jobs (single-chain OP pessimistic, op-succinct, AggOracle committee v2, and multi-chain) now run against one pinned kurtosis-cdk commit, 85fcdea (the merge of feat(aggkit): add the AggOracleCommittee v2, with GER validators 0xPolygon/kurtosis-cdk#1044). The separate committee v2 pin and the committee-v2-kurtosis-ref output are removed.
  • The nightly schedule now tests the pinned commit instead of kurtosis-cdk main.
  • The kurtosis CLI in the multi-chain workflow is bumped from 1.22.2 to 1.23.0 (with its SHA-256) to match the pinned kurtosis-cdk.
  • agglayer/e2e is pinned to a5e723b everywhere. The single-chain jobs can optionally pass a GCP workload identity and service account for image pulls, taken from the GCP_WORKLOAD_IDENTITY_PROVIDER and KURTOSIS_GCP_SERVICE_ACCOUNT repository variables.

🐛 Bug Fixes

  • Fixes the nightly multi-chain failures (Cannot construct 'PortSpec') caused by a kurtosis CLI/kurtosis-cdk version mismatch.

📋 Config Updates

  • Optional repository variables GCP_WORKLOAD_IDENTITY_PROVIDER and KURTOSIS_GCP_SERVICE_ACCOUNT. Default: unset, which leaves behaviour unchanged.
  • KURTOSIS_CDK_COMMIT is now 85fcdea024297e62328f1af7aa42ff2807988c5e. KURTOSIS_CDK_COMMITTEE_V2_COMMIT is removed.

⚠️ Breaking Changes

  • None for the code. CI behaviour changes: the nightly run no longer tracks kurtosis-cdk main, so moving to a newer kurtosis-cdk now requires bumping KURTOSIS_CDK_COMMIT (and the CLI version if it changed).

Verdict: looks good to merge from my side. Suggestion 2 is the one I'd consider adding.
| Branch: ci/single-kurtosis-cdk-pin

@arnaubennassar

Copy link
Copy Markdown
Contributor Author

Thanks for the review. Summary of what was done with it:

  • Suggestion 2 (CLI version coupled to KURTOSIS_CDK_COMMIT by a comment only): added a comment next to KURTOSIS_VERSION in test-e2e-multi-chains.yml that points back to KURTOSIS_CDK_COMMIT. No behaviour change.
  • Suggestion 1 (early warning for kurtosis-cdk drift): not part of this PR. Tracking main on the nightly run is what broke it, because the pinned CLI did not match the moving kurtosis-cdk. A separate non-blocking job against main is a fair idea, but it is new CI behaviour and better reviewed on its own.
  • Suggestion 3 (drop the pass-through get-kurtosis-cdk-commit job): left as is. It still fails early when the pinned SHA is unreachable and keeps the needs wiring and the Slack message unchanged, and removing it would make this diff larger without being required.
  • The optional notes (no committee v2 multi-chain variant, test-e2e-exit_cenrtificate_tool.yml still on kurtosis 1.14.1) are out of scope here.

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

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants