Repository navigation
Description
Goal: remove every SkipOnOpReth skip. Each test should either run on op-reth or be deleted with a stated argument.
op-geth is EOL as of Karst, and CI runs acceptance tests only with l2_el_kind: op-reth (main.yml#L1980-L1983, #L2679, #L2702). So every skipped test is currently dead coverage. Most of the skips came in with #19787, when op-reth was first run against all acceptance tests.
Status
| Test | Resolution |
|---|---|
TestTruncateDatabaseOnELResync, TestPreserveDatabaseOnCLResync, TestELSyncSafeRetractedByOffset |
#23827, which adds the L2ELNode.Wipe() primitive |
TestCLUnsafeNotRewoundOnInvalidDuringELSync |
#24073, stacked on #23827 |
TestVerifierManualSync, TestL2ELP2PCanonicalChainAdvancedByFCU, TestSyncTesterHardforks |
#24092; the second is deleted |
The SkipOnOpReth helper itself goes with the in-process op-geth L2 EL (#21744).
Inventory
The inventory below covers every SkipOnOpReth call in op-acceptance-tests, op-devstack and op-e2e; there is no other EL-kind skip helper. Verdicts were reached by reading the tests, not by running them:
- (a) needs an op-reth capability or devstack primitive
- (b) asserts op-geth-only behaviour
- (c) the skip looks stale
1. EL resync: needs an EL-wipe primitive — (a)
These tests force a resync by stopping the verifier EL. They rely on "Since the sysgo EL uses in-memory storage this also wipes its database", which is true only for op-geth. op-reth's sysgo datadir is a per-node temp dir that persists across Stop/Start (mixed_runtime.go#L382-L389, #L423). On op-reth the restarted node therefore never resyncs, and the "peering issue" reasons are misleading.
TestELSyncSafeRetractedByOffset: "not supported (peering issue)".TestTruncateDatabaseOnELResync: "not supported (peering issue)". The quoted error isPeerWithtiming out after the restart.TestPreserveDatabaseOnCLResync: same reason and samePeerWithtimeout.TestCLUnsafeNotRewoundOnInvalidDuringELSync: "not supported". It wipes the EL "to start from genesis", and the quoted failure is the EL advancing when it should not.
Needed:
- An op-devstack primitive that restarts an op-reth L2 EL from a fresh datadir, or an explicit
L2ELNodewipe. PeerWithmust work reliably against a restarted op-reth. It gets a new port (--port=0) but keeps its p2p key; that this causes the timeout is inferred, not verified.
These matter more now: the EL-sync action tests in op-e2e/actions/sync are being removed as part of #20257 (the op-reth-test-engine stack has no devp2p). That leaves these sysgo tests as the end-to-end coverage of op-node EL resync. The completion-head logic (offset, safedb) is unit-tested in op-node/rollup/engine.
2. Asserts that non-canonical blocks are readable by hash — (b) for the assertion, the rest likely passes
Both tests fail on op-reth with "failed to determine block-hash … could not get payload: not found". They expect a block that engine_newPayload executed but no forkchoice update canonicalized to be served by eth_getBlockByHash. op-geth does this because it persists side-chain blocks. reth keeps executed side-chain blocks in the engine tree, which RPC doesn't serve. The Engine API does not require either behaviour.
TestVerifierManualSync: drop the by-hash check before FCU (L63-L64) and unskip. The rest (newPayload VALID, not canonical by number until FCU, then canonical) is EL-agnostic.TestL2ELP2PCanonicalChainAdvancedByFCU: beyond the by-hash check (L122), it documents op-geth skeleton-sync internals: whether SYNCING payloads are retained, and expected geth log lines. It tests the EL, not op-node, and largely overlapsTestELP2PFCUUnavailableHashandTestSafeDoesNotAdvanceWhenUnsafeIsSyncing_NoELP2Pin the same file, which already run on op-reth. Candidate for deletion, or for cutting down to spec-mandated semantics (FCU with missing ancestors returns SYNCING, then VALID once peered).
3. Hardfork-transition sync — (a), cause unidentified
TestSyncTesterHardforks: "not supported". The chain activates Bedrock through Jovian sequentially after genesis, and the quoted failure is that unsafe never advances. This is not op-geth-only behaviour: it covers running a chain through in-chain fork activations, which op-reth must handle. It needs an investigation into whether the cause is the op-reth sequencer at an activation or the sync-tester read path against op-reth.
🤖 Co-created with Claude Opus 5.5
Activity
[-]op-acceptance-tests: EL-resync sysgo tests are SkipOnOpReth and never run in CI[/-][+]op-acceptance-tests: remove every SkipOnOpReth skip[/+]sebastianst commentedon Oct 12, 2026
🤖 Fable says: Cross-reference from the op-reth-test-engine stack: #21786 adds the action test
RestartEngineFreshState(op-e2e/actions/derivation/reorg_test.go), which replacesRestartOpGethand checks op-node's recovery when its EL comes back with all state lost (reset on the next forkchoice update, re-derive the safe chain from L1, sequencer rebuilds unbatched unsafe blocks). It exists only because the sysgo EL-wipe tests in group 1 here areSkipOnOpRethand therefore never run in CI.Direction: this issue unblocks deleting
RestartEngineFreshState, not the other way round. Once the EL-wipe primitive exists and the group-1 tests run on op-reth, the system-level tests are the better home (real op-node against production op-reth), and the action test should go. Before deleting it, check that the un-skipped sysgo tests also cover the sequencer side (unsafe blocks lost with the EL being re-built), which the group-1 verifier-side tests may not assert today.op-devstack,op-acceptance-tests: address review on EL-resync tests
op-devstack,op-acceptance-tests: address review on EL-resync tests
sebastianst commentedon Nov 13, 2026
🤖 Fable says: #24092 merged, so no test calls
SkipOnOpRethany more. The helper itself is removed with the sysgo EL-kind seam in #21744.