test: deflake e2e_offchain_payment reorg reprocessing race

[spalladino]

Problem

e2e_offchain_payment.test.ts › "reprocesses an offchain-delivered payment after an L1 reorg" flakes (CI run 7d8ad07c73ce84e2) with:

expect(received).toBe(expected) // Object.is equality
  Expected: 100n
  Received: 60n
  > 191 |     expect(aliceAfterRollback).toBe(mintAmount);

60n == 100 - 40: the reverted transfer's effect was still visible to the PXE when the post-reorg balance was read, instead of the rolled-back full mint of 100n.

Root cause

A race between the AutomineSequencer's automatic re-mine of the restored transfer tx and the test's post-reorg assertions:

• forceReorg → revertToCheckpoint(6) prunes block 7 and restores the un-mined transfer to the pending pool (p2pClient.sync() in runRevert, automine_sequencer.ts:629), then returns. It does not itself re-mine.
• The sequencer runs a ~50ms mempool poller (automine_sequencer.ts:175, default pollIntervalMs 50ms) whose buildIfPending autonomously re-mines block 7 from the restored pool.
• The PXE has autoSync: true, so get_balance(...).simulate() re-syncs before every read. Whether Alice's read at line 191 returns 100n (PXE at pruned tip, block 6) or 60n (PXE re-synced to the re-mined block 7) depends purely on this timing.

The failed log shows Updated pxe last block to 7 landing between Bob's read (passes) and Alice's read (fails); the passed log keeps the PXE pinned at block 6 across both reads. Both runs show Building automine checkpoint {checkpointNumber:7, txCount:1}, confirming the poller drives the re-mine.

Fix (test only)

yarn-project/end-to-end/src/e2e_offchain_payment.test.ts:

• forceReorg: automine.pause() before revertToCheckpoint. pause() gates only the mempool poller; explicit ops (revertToCheckpoint, buildEmptyBlock) still run. Pausing before the transfer is restored to the pool means the poller can never enqueue an auto re-mine, so the post-reorg assertions observe a stable pruned state.
• Stay paused through the explicit forceEmptyBlock(). buildEmptyBlock (allowEmpty:true) permits an empty block but still drains the pending pool, so it deterministically re-mines the restored transfer with no poller racing it. The existing retryUntil still waits for the PXE re-sync, so the reprocessing behavior under test is preserved.
• Add an afterEach that calls getAutomineSequencer()?.resume() (idempotent), so a pause is never left dangling.

This removes the nondeterministic auto re-mine that raced the assertions rather than relaxing the aliceAfterRollback == 100 assertion — which is behavioral (it verifies the reorg rolled Alice's note state back) and is kept. Not a skip, not a .test_patterns.yml entry.

Verification

• yarn build: exit 0, no type errors.
• Local e2e red/green not run: the race is nondeterministic (a single local pass would not prove the fix) and the test is infra-heavy. The fix is grounded in the log timeline (PXE re-sync to block 7 lands between Bob's and Alice's reads only in the failed run) and the source mechanism (poller-driven re-mine gated by paused; buildEmptyBlock still drains pending), and was cross-checked to confirm it does not break the later retryUntil.

[AztecBot]

Flakey Tests

🤖 says: This CI run detected 2 tests that failed, but were tolerated due to a .test_patterns.yml entry.

\033FLAKED\033 (8;;http://ci.aztec-labs.com/5333ebc71b758d48�5333ebc71b758d488;;�): yarn-project/kv-store/scripts/run_test.sh src/bench/sqlite-opfs-encrypted/map_bench.test.ts (1s) (code: 0)
\033FLAKED\033 (8;;http://ci.aztec-labs.com/379edb201d919f0f�379edb201d919f0f8;;�): yarn-project/kv-store/scripts/run_test.sh src/sqlite-opfs/internal/ordered-binary-browser.test.ts (2s) (code: 0)