test(integration): validate a real state-sync round-trip + SDK state-sync node support#472
test(integration): validate a real state-sync round-trip + SDK state-sync node support#472bdchatham wants to merge 2 commits into
Conversation
…lowStateSync
The nightly TestWorkflowStateSync ran the state-sync workflow against a
genesis-replayed follower, so "caught up again" could not distinguish a real
re-sync from a no-op. Point it at a genuinely state-sync-bootstrapped node and
assert the wipe + re-sync via earliest_block_height.
SDK support this needs (both additive; map to existing CRD/RPC surface; nil or
absent leaves behavior unchanged):
- NodeSpec.StateSync (*NodeStateSync{RpcServers}) -> fullNode.snapshot.stateSync
+ rpcServers, wired in renderNode; validateNodeSpec enforces the >=2-witness
floor mirroring the CRD MinItems=2.
- EarliestHeight readiness reader (+ earliest_block_height on syncInfo), a mirror
of LatestHeight. earliest>1 proves a node state-synced (not genesis-replayed);
earliest advancing across the workflow proves a genuine wipe + fresh re-sync.
The snapshot-interval wait is derived from the config so retuning the interval
can't silently under-wait and make the assertion flaky.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
PR SummaryMedium Risk Overview SDK: Integration: Reviewed by Cursor Bugbot for commit 8819fe4. Bugbot is set up for automated code reviews on this repo. Configure here. |
![cursor[bot]](https://avatars.githubusercontent.com/in/1210556?s=60&v=4){kind=small}
There was a problem hiding this comment.
Cursor Bugbot has reviewed your changes using default effort and found 3 potential issues.
❌ Bugbot Autofix is OFF. To automatically fix reported issues with cloud agents, enable autofix in the Cursor dashboard.
Reviewed by Cursor Bugbot for commit 7831156. Configure here.
| if err := sei.WaitEVMServing(ctx, hc, node.EVMRPC()); err != nil { | ||
| t.Errorf("follower %s EVM not serving after resync: %v", node.Name(), err) | ||
| if postEarliest <= preEarliest { | ||
| t.Fatalf("follower %s earliest height = %d after resync, want > %d (a genuine wipe + re-sync lands on a newer snapshot)", ssNode.Name(), postEarliest, preEarliest) |
There was a problem hiding this comment.
Earliest advance asserted without new snapshot
Medium Severity
The post-workflow check requires earliest to rise past the pre-wipe floor, but nothing waits for another snapshot_interval of blocks after the first bootstrap. If the chain has not crossed the next snapshot boundary before re-sync, the follower restores the same snapshot and earliest stays unchanged, failing a genuine wipe+resync.
Reviewed by Cursor Bugbot for commit 7831156. Configure here.
…itness set, base-pin Addresses the blinded /xreview slate (systems + kubernetes + idiom): - Witnesses are two DISTINCT light-client sources (genesis validator-0 + the provisioned rpc follower) instead of the aggregate RPC counted twice, which the node CRD (listType=set/MinItems=2) rejects/collapses below the >=2 floor. Keeps a single validator (avoids the 2-validator genesis blocksync stall + 2/3-liveness fragility) and revives the otherwise-dead provisioned follower. - The workflow recipe now inherits the node's resolved DISTINCT syncers (RpcServers: nil) instead of re-syncing from the aggregate-twice unsound set. - validateNodeSpec requires >=2 DISTINCT witnesses (dedup-then-count, mirroring the controller's slices.Sorted+Compact) and reports the distinct count. - Pin the follower's block-store base explicitly (chain.min_retain_blocks=0) so the earliest-height discontinuity assertion rests on a stated invariant, not an image default; corrected the comment (earliest is the CometBFT block-store base governed by min-retain-blocks, not storage.pruning's state-store subsystem). - witnessA/witnessB test consts to clear the goconst golangci-lint gate. golangci-lint run ./sdk/... = 0 issues; go build/vet, go test ./sdk/..., gofmt green. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
1 participant
Strengthens the nightly
TestWorkflowStateSyncso it validates the full state-sync workflow procedure end-to-end (the manual procedure just proven live), plus the minimal SDK surface it needs.Why
The existing nightly ran the state-sync workflow against a genesis-replayed follower, so its "caught up again" check couldn't distinguish a genuine re-sync from a no-op. This points the workflow at a truly state-sync-bootstrapped node and asserts the wipe + re-sync via
earliest_block_height.SDK (additive; nil/absent = unchanged)
NodeSpec.StateSync(*NodeStateSync{ RpcServers []string }) →fullNode.snapshot.stateSync+rpcServersinrenderNode;validateNodeSpecenforces the ≥2-witness floor (mirrors CRDMinItems=2). Doc-comments the bare-host:portinvariant + theurl.Parse().Hostcaller obligation.EarliestHeightreadiness reader (+earliest_block_heightonsyncInfo), a mirror ofLatestHeight.Test
Create a state-sync follower (witnesses = aggregate validator RPC as bare host:port), assert
earliest > 1(proves state-synced, not replayed); run theStateSyncworkflow on it; assertComplete+earliestadvanced (proves a genuine wipe + fresh re-sync). Existing proven network/snapshot config preserved verbatim; snapshot-interval wait derived from the config.Review
Idiom lens: RATIFY (nested-pointer shape matches
WorkflowSpec.StateSync;RpcServersnaming consistent with core; docs conforming). Build/vet/go test ./sdk/...(incl. newTestRenderNode_StateSync+ validate-floor cases) green.Note: consciously duplicates seictl's workflow client pending the future
sdk/seiconsolidation.🤖 Generated with Claude Code