[skylar-simoncelli]

Overview

Start compiling Local Environment Tests alongside the node-image build, then reuse the compiled binary through the existing shared Earthly cache when the images are ready. Previously this compilation happened after the image dependency completed.

The new prepare-local-e2e job builds +local-env-ci-prepare without an image dependency. +local-env-ci inherits the same target; a cache miss rebuilds normally. The execution layer includes the workflow run ID and attempt so tests actually run. Trusted PRs use the existing self-hosted pool; external forks retain inline compilation on hosted runners.

This PR is scoped to local-env compilation scheduling. It contains no bridge test changes, thread-count or faucet changes, RPC configuration changes, Toolkit code or Toolkit workflow changes. Oscar's concurrency work remains in #2408 and Toolkit improvements in #2412. No dependency on either PR is required.

Financial impact

No GHA cash saving or Karpenter cost increase is claimed. No runner infrastructure is added or moved, and existing self-hosted rental cost is unchanged. The benefit is removing local E2E compilation from the post-image-build path. Shared-host queueing and contention can still determine overall PR latency.

🗹 TODO before merging

• Restrict the diff to local-env preparation and its documentation.
• Check compatibility with Oscar's current PR heads.
• Validate compilation reuse and the full local suite on this narrowed head.
• Ready

📌 Submission Checklist

• All commits are signed off (git commit -s) for the DCO
• Changes are backward-compatible (or flagged if breaking)
• Pull request description explains why the change is needed
• Self-reviewed the diff
• I have included a change file, or skipped for this reason: CI scheduling only; no product changes.
• If the changes introduce a new feature, I have bumped the node minor version
• Update documentation (if relevant)
• Updated AGENTS.md if build commands, architecture, or workflows changed
• No new todos introduced

🧪 Testing Evidence

Narrowed head 610721154: CI passed on retry attempt 2. Actual logs confirm:

• Local Environment Tests: 34 passed, 2 ignored, followed by a successful post-suite liveness check. The job took 25m55s; the unchanged serial test suite took 18m14s.
• The preparation compilation layer was reused (cached), and the runtime Cargo check finished in 2.32s. The first preparation compiled in 5m54s (6m37s job).
• Test Toolkit: 186 passed, 1 skipped. No Toolkit code or workflow is changed by this PR.
• Relevant checks are complete with no failures or pending checks; cargo-deny reports neutral.

This demonstrates compilation reuse, not a measured overall PR-time reduction. Shared-runner queueing prevented overlap in the first attempt. The 25m55s local-job duration includes the existing serial tests; Oscar's separate concurrency work is not included here.

The first attempt was interrupted when an unrelated branch restarted shared BuildKit using different cache settings. Retrying the interrupted jobs succeeded without changing test code or that branch.

• Successful retry: https://github.com/midnightntwrk/midnight-node/actions/runs/42855343252/attempts/2

Actionlint, YAML parsing, conflict-marker checks and private-key detection passed locally; no local builds were run. The repository has no pre-commit configuration, so targeted checks used a temporary configuration.

Read-only merge checks show this head combines cleanly with #2408. #2412 has existing conflicts with main in its Toolkit workflow/target; combining this PR adds no new conflict locations. The local-env and Toolkit changes are independently scoped.

Historical evidence from combined head 2ad2d0481: preparation compiled in 6m12s, and the later Cargo check took 2.84s. Local Environment Tests passed all 34 enabled tests plus liveness in a 13m07s job. That head also included Oscar's concurrent tests; its runtime is not a benchmark of this narrowed PR. Earlier 32% and two-chain results likewise do not measure this final scope.

Validate actual compilation reuse, test counts, liveness and latest-head required checks. Record queueing and the complete required-check path separately from individual job runtime. No guaranteed overall PR-time reduction is claimed.

• Additional tests are provided (if possible)

🔱 Fork Strategy

• Node Runtime Update
• Node Client Update
• Other: CI-only preparation on the existing trusted self-hosted pool; fork PRs compile inline.
• N/A

Links

• Independent local-env concurrency: test(e2e): run c2m bridge tests concurrently and raise local e2e threads #2408
• Independent Toolkit improvements: test(toolkit): compile toolkit tests at t=0 and share one node across toolkit_e2e #2412
• Historical combined validation: https://github.com/midnightntwrk/midnight-node/actions/runs/42839327535
• Validation procedure: docs/ci/pr-critical-path.md

[ozgb]

Taking this PR as inspiration, I managed to bump the concurrency of the local-environment tests at a lower-level of granularity - tests now complete in approx. 5 mins: #2408

Suggest we revert the sharding changes here