core: agree on no block for payload attestation duty

[KaloyanTanev]

Lets the cluster agree that a slot has no block to attest to, instead of each node acting on its own beacon node's view. When the beacon node answers 204 for payload attestation data, the fetcher proposes an empty UnsignedDataSet like any other value and consensus runs as usual: the round leader's proposal is what the cluster agrees on, so the outcome follows the leader's beacon node view, as for every other duty. The agreed empty set is stored in the dutydb, which resolves AwaitPayloadAttestationData with ErrNoPayloadAttestationData, and the validator API answers the VC with 204 No Content (~0.2s after the fetch, no VC timeouts). The tracker records empty-set consensus/dutydb events explicitly and reports an agreed no-block as a no-op rather than bug_fetch_error, removing the per-slot consensus timeout and Duty failed noise on empty slots.

category: bug
ticket: #4324

[codecov]

Codecov Report

❌ Patch coverage is 87.27273% with 7 lines in your changes missing coverage. Please review.
⚠️ Please upload report for BASE (gloas@ad5b2a6). Learn more about missing BASE report.

Files with missing lines	Patch %	Lines
core/tracing.go	0.00%	3 Missing ⚠️
core/dutydb/memory.go	88.88%	2 Missing ⚠️
core/tracker/tracker.go	90.00%	2 Missing ⚠️

Additional details and impacted files

@@           Coverage Diff            @@
##             gloas    #4732   +/-   ##
========================================
  Coverage         ?   66.35%           
========================================
  Files            ?      247           
  Lines            ?    31947           
  Branches         ?        0           
========================================
  Hits             ?    21197           
  Misses           ?    10749           
  Partials         ?        1           

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[OisinKyne]

I think i'm with opus here in terms of doing option B just because its our weak or far away bns that might let the cluster down even more than they could otherwise if we do a).

If nodes enter the participate game even though their bn saw no timely block, if the other faster nodes did see one on-time, this node can incrementally help in getting the (non slashable!) PTC duty getting signed, (e.g. if someone's good BN has a validator client down or something)

(Approving to unblock because i think as is is perfectly safe, just maybe a small opportunity lost)

For truly empty slots this is right. The spec says not to attest when you haven't seen a block (validator.md:407), and it removes the consensus-timeout noise.

The cost is the case we care about most: nodes disagree about whether they saw the block. I checked the QBFT code. Incoming messages for a duty are only buffered, and the instance only runs on Propose or Participate. So a node that got a 204 now:

1. can't count towards quorum on a peer's value, and
2. has already told its VC 204, so it won't sign even if the cluster agrees on data.

Before this PR, such a node joined consensus, adopted the leader's value and signed. Example: 4 nodes, 2 of them see the block late → consensus needs 3 → it used to succeed, now it fails. That's exactly the "not deterministic across nodes, so agree on it live" case. The impact is bounded: a block still unseen at 6s is rare, and PTC messages aren't slashable, so the downside is a missed PTC vote, never a safety problem. Still, it's a regression in the scenario DV is supposed to handle best.

Options:

• (a) Accept it as-is, and state the trade-off in the PR body and the Participate comment.
• (b) Keep Participate for PTC. Only answer the VC with a 204 once the consensus instance ends without a decision (or at a cutoff), and just suppress the timeout log/tracker noise when the node had no value of its own. This keeps the split-view case working.

I'd push lightly for (b), or at least a comment so the choice is explicit. Also:

• Misleading comment. // Stored data takes precedence over the no-block mark, e.g. for a late block is effectively unreachable: a node that got a 204 never runs the instance, so it never stores a decided value. The test forces the path artificially. Remove the comment or reword it.
• Separate, older question: fetching at 6s gets the final answer for payload_present, but the spec says blob_data_available should only be set false once 9s has passed (validator.md:416-417). A 6s fetch can lock in an early false. It's the usual DV trade-off, since consensus and signing need time before 9s, but worth a ticket.

[Copilot]

Copilot review overview

🟡 Changes recommended

The no-block decision can be overwritten after clients observe it, and existing query-cleanup coverage was removed.

Review effort: Balanced
Findings: 1 · 1

Open (2)

• Preserve zero-marker decisions against later data overwrites · New
• Retain cancellation-pruning regression coverage · New

What changed in this PR

Enables cluster consensus on payload-attestation slots with no block and returns HTTP 204 to validator clients.

Changes:

• Proposes and serializes empty payload-attestation data sets through consensus.
• Stores no-block outcomes and exposes them through the validator API.
• Tracks no-block duties as no-ops and adds related tests.

File	Description
core/​validatorapi/​validatorapi.go	Maps no-block outcomes to the client sentinel.
core/​validatorapi/​validatorapi_test.go	Tests sentinel propagation.
core/​validatorapi/​router.go	Produces HTTP 204 responses.
core/​validatorapi/​router_internal_test.go	Tests the 204 route response.
core/​tracker/​tracker.go	Records empty-set events as no-ops.
core/​tracker/​tracker_internal_test.go	Tests empty-set tracking.
core/​tracing.go	Propagates the updated DutyDB result.
core/​proto.go	Allows empty payload-attestation sets.
core/​proto_test.go	Tests empty-set serialization.
core/​interfaces.go	Updates DutyDB and validator API contracts.
core/​fetcher/​fetcher.go	Proposes empty sets for no-block responses.
core/​fetcher/​fetcher_test.go	Tests no-block fetching.
core/​dutydb/​memory.go	Stores and resolves no-block markers.
core/​dutydb/​memory_test.go	Tests marker behavior.
core/​dutydb/​memory_internal_test.go	Tests marker identification.

---

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.

[sonarqubecloud]

Quality Gate passed

Issues
2 New issues
0 Accepted issues

Measures
0 Security Hotspots
0.0% Coverage on New Code
0.0% Duplication on New Code

See analysis details on SonarQube Cloud