Skip to content

ci: restore the act and deny checks on op-v2.7.0 - #34

Merged
op-will merged 9 commits into
op-v2.7.0from
af/act-stub-secure-runner
Nov 11, 2026
Merged

op-will merged 9 commits into
op-v2.7.0from
af/act-stub-secure-runner

Conversation

@aakoshh

@aakoshh aakoshh commented Nov 5, 2026 •

Copy link
Copy Markdown

What

Restores op-v2.7.0's checks with two commits:

  • Stubs tempoxyz's secure-runner under act, so the act jobs run again, and adds scan-github-actions.yml to the config header's list of excluded workflows.
  • Ignores RUSTSEC-2026-0441 in deny.toml. That advisory marks pin-utils unmaintained, with no safe upgrade. RustSec assigned the ID on 2026-11-05, after the branch was pushed, and lint-deny started failing.

Why

v2.7.0 adds tempoxyz/gh-actions/actions/secure-runner as the first step of 43 jobs; the v2.6.0 workflows had none. It starts StepSecurity's Harden Runner and Socket's Aegis, and its pre step exchanges a GitHub OIDC token for their policy credentials. act never receives that token, so every act job on op-v2.7.0 fails within seconds, before any check runs (act-lint-fmt):

Error: ACTIONS_ID_TOKEN_REQUEST_TOKEN is missing: the job must grant `id-token: write` so Harden Runner can fetch its StepSecurity policy-store credential (event: push)

How

  • act's --local-repository maps a whole repository at a ref. tempoxyz/gh-actions@ee2960a6 also serves the vendored taiki-e/install-action (10 uses) and dtolnay/rust-toolchain (2 uses), which the checks need, so mapping it to the no-op would break them. Instead, a new step fetches that one commit, replaces only actions/secure-runner/action.yml with the existing no-op, and act maps the repository to that copy.
  • A guard fails the step when the workflows pin tempoxyz/gh-actions at any other sha. A later replay then gets a message naming the mismatch, instead of the OIDC error above.
  • scan-github-actions.yml, new in v2.7.0, runs on pull requests through a tempoxyz reusable workflow that needs OIDC. It was never in the act matrix; the header now records the exclusion.
  • pin-utils is reachable only on wasm targets, through alloy-provider → wasmtimer: cargo tree -i pin-utils --all-features prints nothing for the host target.

Caveats

  • With the stub, act-e2e.yml and act-integration.yml now run and fail on two tests that race on CircleCI's runners. The test code is the same in v2.7.0. test(e2e): wait for pruner in RocksDB pruning tests (#27601), harden Osaka blob conversion (#27687) #35, stacked on this PR, cherry-picks upstream's test-only fixes. Neither PR can pass reth-act on its own; test(e2e): wait for pruner in RocksDB pruning tests (#27601), harden Osaka blob conversion (#27687) #35 describes the two ways to merge them.
  • The act jobs run without Harden Runner, which monitors outbound connections on GitHub runners: it audits them by default and can block them under a policy stored in tempoxyz's StepSecurity account. They also run without Aegis's package-download checks. op-v2.6.0's act jobs never had either, and both depend on tempoxyz's credential services, so enabling GitHub Actions on the fork would not restore them. Whether the fork's CircleCI jobs need network monitoring of their own is a question for cloud-security.
  • The tempoxyz sha now appears both in the workflows and in this config. The guard keeps the two in step.
  • op-v2.6.0 has the same pin-utils dependency, so its lint-deny now fails too. The ignore can be cherry-picked there if another PR targets that branch.

Alternatives

  • Pass disable-enforcement: true to each secure-runner step: 43 workflow edits that would conflict on every replay, with flags that look unexplained in upstream's files.
  • Rewrite the 12 vendored uses: lines to their originals and map all of tempoxyz/gh-actions to the no-op. Each .vendored.json shows the copies are exact upstream commits, so this would run the same code. But it adds 12 conflicting edits, and any vendored action upstream adds later would silently become a no-op instead of running.

@aakoshh
aakoshh requested a review from a team as a code owner November 5, 2026 10:03
RUSTSEC-2026-0441 marks pin-utils unmaintained with no safe upgrade.
It is reachable only on wasm targets, through alloy-provider's
wasmtimer.
@aakoshh aakoshh changed the title ci: stub tempoxyz's secure-runner under act ci: restore the act and deny checks on op-v2.7.0 Nov 5, 2026
@aakoshh
aakoshh force-pushed the af/act-stub-secure-runner branch from f602b81 to f13963a Compare November 5, 2026 12:16
@aakoshh
aakoshh added this pull request to stack #36 November 5, 2026 12:27
@einar-oplabs

Copy link
Copy Markdown

Fable 5.1 (xhigh): Heads-up from a local evaluation of this branch (op-v2.7.0 at 2158d3b) against the monorepo's proofs-history ExEx coverage, since the branch has no tag yet.

Setup. Monorepo develop (34625bbcb6) pinned to this branch by rev, with Josh's v2.7.0 adaptations from optimism#23156 carried over, plus the 24 draft test PRs under optimism#23385 and optimism#23386 rebased onto it. Rust suites of the four proofs crates, the four proofs_history_* acceptance packages, the proofs/core e2e suite and the seven non-ExEx acceptance packages, each with op-node and with kona-node as L2 CL.

Result: no 2.7.0 regression. Every failure traced to something already known on 2.6 or to develop drift unrelated to reth (op-geth rc.2's Karst kill switch, develop's build-no-tests skipping the MockSP1Verifier artifact).

Three things for this branch before it is tagged:

  1. reth#28264 is missing. op-v2.6.0-2 carries it via fix(trie): tighten iterator seek boundary #32 (merged 2026-11-05); this branch's newest backport is fix(engine): report payload job start errors paradigmxyz/reth#28015, so the trie iterator seek-boundary fix would regress on the way to 2.7.0.
  2. Pruned-history RPC errors changed wording with upstream fix(rpc): classify pruned changeset errors as unavailable history paradigmxyz/reth#27367 / refactor(rpc): use EthRpcErrorCode::PrunedHistory paradigmxyz/reth#27521: EthApiError::PrunedHistoryUnavailable now reads pruned history unavailable: requested N, earliest available M. Anything matching the 2.6 texts (insufficient changesets ..., state at block #N is pruned) stops matching; the e2e suite had one such matcher.
  3. alloy-trie must stay at 0.9.7 at this rev. Cargo floats to 0.9.9, whose HashBuilder::all_branch_nodes breaks reth-trie-common here; I held it with cargo update alloy-trie --precise 0.9.7. A real bump should pin it or pick up the upstream fix.

Numbers, for the record: Rust 825/828 (the 3 are stack-side: a CLI prune test asserting pre-reconciliation behavior and the two known engine-model crash cases), proofs_history acceptance 4/4 ok on both CLs, e2e 52/54 (the 2 are the known debug_executionWitness above reth's history prune checkpoint, where the EIP-2935 blockhash call reads pruned history; same on 2.6), non-ExEx 4/7 ok with the 3 failures being the op-geth follower lanes refusing Karst.

aakoshh added a commit to ethereum-optimism/optimism that referenced this pull request Nov 10, 2026
Pin the reth crates to the head of ethereum-optimism/reth#35, which
stacks on ethereum-optimism/reth#34 on top of op-v2.7.0. Both are
unmerged, so the pin is a rev until op-v2.7.0-0 is tagged. Sync the
shared dependencies to the versions reth v2.7.0 requires: alloy 2.5.0,
alloy-hardforks 0.4.9, reth-core 0.8.1, revm 43 and revm-inspectors
0.44.

Adapt to the upstream API changes:

- StateProviderDatabase now takes an EvmStateProvider
  (paradigmxyz/reth#27109). Wrap each provider with
  into_evm_state_provider, on a reference where it is used again.
- CancelOnDrop moved from reth-revm to reth-tasks
  (paradigmxyz/reth#27466).
- StateProviderFactory gained a Primitives type
  (paradigmxyz/reth#26923). Where a bound also names
  NodePrimitivesProvider, the debug witness API now says which one it
  means.
- BlockAndReceipts holds SharedReceipts (paradigmxyz/reth#27427).
- The e2e setup functions folded into E2ETestSetupBuilder
  (paradigmxyz/reth#27432).
- MemoryOverlayStateProvider was removed in favour of the overlay
  manager (paradigmxyz/reth#26923). The test engine records every
  validated block in the manager it shares with its provider, so it now
  appends a block to its real parent with state_with_block_appended.
  The flashblock pending state does the same, but on the canonical
  anchor: a speculative block's parent is the pending block before it,
  which the overlay does not know. The next commit shows that this
  breaks for speculative builds.
@aakoshh
aakoshh requested a review from sebastianst November 10, 2026 15:26
@aakoshh

aakoshh commented Nov 11, 2026

Copy link
Copy Markdown
Author

Thanks for the run. Here is where each of the three points ended up:

  1. fix(trie): tighten iterator seek boundary paradigmxyz/reth#28264 is carried as fix(trie): tighten iterator seek boundary (#28264) #40, stacked on test(e2e): wait for pruner in RocksDB pruning tests (#27601), harden Osaka blob conversion (#27687) #35, so op-v2.7.0-0 will include it.
  2. Pruned-history errors: the code changed too, not only the wording. Since fix(rpc): classify pruned changeset errors as unavailable history paradigmxyz/reth#27367, a block below the earliest available changeset fails with 4444 instead of -32603. Nothing merged in the monorepo matches either the old texts or -32603, so the matcher you hit must be in the draft test PRs under op-reth: proofs ExEx test coverage to merge before the reth v2.6.0 bump optimism#23385 and op-reth: test coverage for the reth v2.6.0 bump outside the proofs ExEx optimism#23386, which need the new text when they're rebased onto v2.7.0. rust!: Update reth to v2.7.0 optimism#24024 lists the change for external clients.
  3. alloy-trie stays at 0.9.7 in both monorepo lockfiles. rust!: Update reth to v2.7.0 optimism#24024 notes that the declared range admits 0.9.9, so the lockfiles shouldn't be refreshed with a broad cargo update.

ethereum-optimism/optimism#24024 is merged and pins this stack by rev (0447ceb, the head of #41) until it's tagged.

@geoknee
geoknee enabled stacked auto-merge (rebase) November 11, 2026 17:39
stacked auto-merge was automatically disabled November 11, 2026 17:39

Auto-merge was disabled for another pull request in the stack

@geoknee
geoknee removed this pull request from stack #36 November 11, 2026 17:39
mattsse and others added 6 commits November 11, 2026 17:39
Co-authored-by: Matthias Seitz <19890894+mattsse@users.noreply.github.com>
(cherry picked from commit f269bcc)
Co-authored-by: Matthias Seitz <19890894+mattsse@users.noreply.github.com>
(cherry picked from commit 0fc98b4)
Co-authored-by: Matthias Seitz <19890894+mattsse@users.noreply.github.com>
(cherry picked from commit e29c343)
Co-authored-by: Matthias Seitz <19890894+mattsse@users.noreply.github.com>
(cherry picked from commit cdc4e09)
Co-authored-by: Matthias Seitz <19890894+mattsse@users.noreply.github.com>
(cherry picked from commit 569c770)
@geoknee
geoknee added this pull request to stack #43 November 11, 2026 17:40
@sebastianst

Copy link
Copy Markdown
Member

One op-v2.6.0 CI fix hasn't reached this branch: #38 (806fa3b) pinned Playwright in .github/workflows/book.yml:

-          npx playwright install --with-deps chromium
+          npx playwright@1.64.1 install --with-deps chromium

At the stack head 0447ceb, book.yml#L47 still runs npx playwright install unpinned, and this PR doesn't add the pin. The other half of #38, the RUSTSEC-2026-0441 ignore, is already here (f13963a). The v2.7.0 docs lockfile has no Playwright dependency, so the pin applies cleanly. Please add it to this PR, so the op-v2.7.0 CI squash covers every op-v2.6.0 CI fix.

🤖 Co-created with Claude Opus 5.5

The docs job ran an unpinned `npx playwright install`, which resolves the
newest `playwright` at run time and broke while npm had 1.65.0 of the CLI
but not of `playwright-core`. Pin it to 1.64.1, as op-v2.6.0 does since
#38. Nothing in the docs lockfile depends on a
Playwright version.

(cherry picked from the book.yml half of commit 806fa3b)
@aakoshh

aakoshh commented Nov 11, 2026

Copy link
Copy Markdown
Author

Thanks, added in bc2744c: book.yml now runs npx playwright@1.64.1 install, as op-v2.6.0 does since #38. With the RUSTSEC-2026-0441 ignore (f13963a), every op-v2.6.0 CI fix is now on this branch.

@op-will
op-will enabled stacked auto-merge (rebase) November 11, 2026 19:30
stacked auto-merge was automatically disabled November 11, 2026 19:31

Auto-merge was disabled for another pull request in the stack

@op-will
op-will merged commit 0dcf86d into op-v2.7.0 Nov 11, 2026
30 checks passed
@op-will
op-will deleted the af/act-stub-secure-runner branch November 11, 2026 19:35
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.

5 participants