Skip to content

feat(blockchain): pre-build proposer block one interval early#445

Draft
MegaRedHand wants to merge 15 commits into
mainfrom
feat/proposer-prebuild-aggregation
Draft

feat(blockchain): pre-build proposer block one interval early#445
MegaRedHand wants to merge 15 commits into
mainfrom
feat/proposer-prebuild-aggregation

Conversation

@MegaRedHand

Copy link
Copy Markdown
Collaborator

What

Proposers do the heavy leanVM block-building work (attestation compaction + the Type-1 → Type-2 merge, ~1–2s) synchronously at interval 0 today, squeezed into the slot's first interval. This builds the next slot's block one interval early — at the previous slot's interval 4, synchronously on the blockchain actor — and publishes it at interval 0 after a revalidation. On any mismatch or failure it falls back to the existing synchronous build, so there is no regression.

How

  • Interval 4 (of slot S-1): if one of our validators proposes slot S, prebuild_block builds + signs + merges the full block against the current (read-only) head and stashes it as a PreparedBlock.
  • Interval 0 (of slot S): propose_block publishes the prepared block iff it is still valid (prebuilt_block_is_usable: same slot/proposer, parent still the canonical head, and the built justified slot not behind the store's). Otherwise it rebuilds synchronously.
  • Shared assemble_signed_block / process_and_publish_block helpers back both the pre-build and the fallback path.
  • Signing stays on the actor throughout: XMSS keys are stateful (sign_block_root takes &mut self), so the build cannot be moved off-thread without risking OTS-index reuse.

Finalization fix (included)

Capturing the pre-build head via get_proposal_head — which is not read-only, it ticks the store clock — one interval early advanced the clock prematurely and stalled finalization. Fixed by capturing the head read-only via store.head().

Validation (local devnet: 4 ethlambda nodes, 1 aggregator)

  • Finalization healthy (finalized lag ~3 slots).
  • Near-100% pre-build hit rate; zero stale discards, zero skipped ticks, zero errors.

Tradeoff

Building synchronously blocks the actor for the build duration at interval 4. This is acceptable: between interval 4 and the next slot the actor has no other consensus-critical duty, and the devnet run showed zero skipped ticks.

Tests

make fmt + cargo clippy -p ethlambda-blockchain --all-targets -- -D warnings clean; lib tests pass, including the new block_builder::prebuild_tests covering the usability predicate.

MegaRedHand added a commit that referenced this pull request Jun 22, 2026
Split out of #445 (proposer pre-build).

`gate_proposer` was a one-line wrapper over `duties_allowed()` with a
single call site. This inlines it as `scheduled_proposer.filter(|_|
self.sync_status.duties_allowed())` and removes the method, so the sync
gate reads the same way everywhere.

**Behavior-preserving** — `proposer_validator_id` is computed
identically; every downstream use (the skip-while-syncing log, the
`store::on_tick` proposal flag, the proposal call) is untouched.

The three `gate_proposer`-only unit tests collapse onto the
`duties_allowed()` assertions they already made.

Stacked under the on_tick-regrouping PR; #445 will drop these hunks once
both land.

Co-authored-by: Pablo Deymonnaz <pdeymon@fi.uba.ar>
on_tick ran its per-interval blocks out of order (4-snapshot, tick, 2, 0,
1). Regroup them into ascending interval order behind `==== interval N ====`
markers so the slot timeline reads top to bottom and matches the duty
schedule.

Pure reorder, behavior unchanged. The interval-4 new_payloads snapshot is
the one block that stays ahead of store::on_tick: the interval-4 tick
promotes new_payloads out, so it cannot move into a post-tick group. A
comment now pins that constraint.
…ick flag

Compute scheduled_proposer just before store::on_tick and pass the raw
is_proposer (= scheduled_proposer.is_some()) to it; gate the actual
proposal on duties_allowed() at the call site instead. While syncing and
scheduled to propose, on_tick now accepts attestations early at interval 0
(it did not before); the proposal itself is still skipped.
@MegaRedHand MegaRedHand force-pushed the feat/proposer-prebuild-aggregation branch from d4a251e to 4955800 Compare June 22, 2026 16:34
@MegaRedHand MegaRedHand changed the base branch from main to refactor/group-on-tick-by-interval June 22, 2026 16:34
At interval 4, if one of our validators proposes the next slot, build and
sign its block synchronously on the actor so the heavy leanVM aggregation
(~1-2s) is done before interval 0. The proposer then only has to publish.

Shares one block-build core (produce_block_on_head) across the prebuild and
normal proposal paths; a publish-time usability gate falls back to a fresh
build if the prebuilt block is stale.
@MegaRedHand MegaRedHand force-pushed the feat/proposer-prebuild-aggregation branch from 4955800 to 2c6eb1c Compare June 22, 2026 16:42
MegaRedHand added a commit that referenced this pull request Jun 22, 2026
Split out of #445 (proposer pre-build). **Stacked on #449**
(`gate_proposer` removal) — review/merge that first; this PR's base
retargets to `main` once #449 lands.

`on_tick` ran its per-interval blocks out of order (interval-4 snapshot,
then tick, then 2, 0, 1). This regroups them into ascending interval
order behind `==== interval N ====` markers so the slot timeline reads
top-to-bottom and lines up with the duty schedule.

**Pure reorder, behavior unchanged.**

### One block intentionally does *not* move
The interval-4 `new_payloads` snapshot stays **ahead** of
`store::on_tick`. At interval 4 the tick runs `accept_new_attestations →
promote_new_aggregated_payloads()`, which drains `new_payloads`;
snapshotting after the tick would capture an already-drained set and
silently break the post-block coverage report's "timely" cohort. A
comment now pins that ordering constraint.

> Note: commit `7474c15` on #445 currently moves this snapshot *into*
the post-tick interval-4 group, which hits exactly that regression
(observability-only, so devnet doesn't surface it). #445 should adopt
this PR's placement when it rebases.
Base automatically changed from refactor/group-on-tick-by-interval to main June 22, 2026 16:50
@MegaRedHand MegaRedHand marked this pull request as ready for review June 22, 2026 16:51
@greptile-apps

greptile-apps Bot commented Jun 22, 2026

Copy link
Copy Markdown
Contributor

Greptile Summary

This PR pre-builds the next slot's block at interval 4 of the previous slot, stashing a PreparedBlock that the proposer can publish at interval 0 after a lightweight revalidation, with a full synchronous rebuild as fallback. It also fixes a clock-skew regression where get_proposal_head advanced the store prematurely when called one interval early.

  • PreparedBlock / prebuilt_block_is_usable in block_builder.rs encapsulate the pre-built state and the usability predicate, well-tested by new prebuild_tests.
  • propose_block in lib.rs is refactored into three shared helpers (build_signed_block, assemble_signed_block, process_and_publish_block) used by both the fast and fallback paths; the interval-4 pre-build uses store.head() read-only while the interval-0 proposal still ticks the clock via get_proposal_head.
  • store.rs extracts produce_block_on_head so the pre-build can produce a block against a known head without side effects on the store clock.

Confidence Score: 3/5

The fallback path can call sign_block_root for the same slot twice — once at interval 4 and again at interval 0 — and the has_proposal flag change alters store attestation-promotion for syncing nodes; both need to be resolved before merging.

The core block-builder refactor and the PreparedBlock predicate are clean and well-tested. However, ValidatorSecretKey::sign is non-mutating (&self), using the slot number as the deterministic OTS leaf index with no guard against reuse. When the fallback path triggers — which happens on any head change between interval 4 and interval 0 — the same XMSS OTS leaf is used to sign two distinct block roots, violating the single-use guarantee the rest of the codebase is built around. Additionally, the refactor accidentally dropped the sync_status.duties_allowed() filter from the has_proposal flag passed to store::on_tick, causing syncing nodes to eagerly promote attestations at interval 0 when they will not actually propose.

crates/blockchain/src/lib.rs — the fallback signing path and the has_proposal flag change both need attention

Important Files Changed

Filename Overview
crates/blockchain/src/lib.rs Core actor refactor: introduces pre-build at interval 4, fallback at interval 0, and three shared helpers. The is_proposer flag passed to store::on_tick no longer filters by duties_allowed() (behavioral regression), and the shared build_signed_block helper signs for slot S at interval 4 and potentially again at interval 0 (XMSS OTS reuse risk in the fallback path).
crates/blockchain/src/block_builder.rs Adds PreparedBlock struct and prebuilt_block_is_usable predicate with thorough unit tests. Logic is clean and well-exercised; no issues found.
crates/blockchain/src/store.rs Splits produce_block_with_signatures into a new produce_block_on_head variant that accepts an already-resolved head root without ticking the store clock. get_proposal_head is made pub(crate) with clear documentation. Changes are correct and well-scoped.

Flowchart

%%{init: {'theme': 'neutral'}}%%
flowchart TD
    A[Slot S-1 Interval 4] --> B[prebuild_block next_slot=S]
    B --> C[produce_block_on_head READ-ONLY store.head]
    C --> D[sign_block_root vid slot=S root_A OTS index S consumed]
    D --> E[stash PreparedBlock slot=S]
    E --> F[advance_keys_to S]
    F --> G[Slot S Interval 0]
    G --> H[get_proposal_head ticks store clock]
    H --> I{prebuilt_block_is_usable?}
    I -->|yes| J[process_block prepared]
    J -->|ok| K[publish pre-built block FAST PATH]
    J -->|fail| L[fallback build_signed_block]
    I -->|no - head moved| L
    L --> M[sign_block_root vid slot=S root_B WARNING same OTS index S reused]
    M --> N[publish fallback block]
Loading
%%{init: {'theme': 'base', 'themeVariables': {"darkMode": true, "background": "#0d1117", "primaryColor": "#21262d", "primaryTextColor": "#e6edf3", "primaryBorderColor": "#8b949e", "lineColor": "#8b949e", "textColor": "#e6edf3", "edgeLabelBackground": "#161b22", "actorBkg": "#21262d", "actorBorder": "#8b949e", "actorTextColor": "#e6edf3", "actorLineColor": "#8b949e", "signalColor": "#8b949e", "signalTextColor": "#e6edf3", "noteBkgColor": "#373320", "noteBorderColor": "#d4a72c", "noteTextColor": "#f0e6c0", "labelBoxBkgColor": "#21262d", "labelBoxBorderColor": "#8b949e", "labelTextColor": "#e6edf3", "loopTextColor": "#e6edf3", "activationBkgColor": "#30363d", "activationBorderColor": "#8b949e"}}}%%
flowchart TD
    A[Slot S-1 Interval 4] --> B[prebuild_block next_slot=S]
    B --> C[produce_block_on_head READ-ONLY store.head]
    C --> D[sign_block_root vid slot=S root_A OTS index S consumed]
    D --> E[stash PreparedBlock slot=S]
    E --> F[advance_keys_to S]
    F --> G[Slot S Interval 0]
    G --> H[get_proposal_head ticks store clock]
    H --> I{prebuilt_block_is_usable?}
    I -->|yes| J[process_block prepared]
    J -->|ok| K[publish pre-built block FAST PATH]
    J -->|fail| L[fallback build_signed_block]
    I -->|no - head moved| L
    L --> M[sign_block_root vid slot=S root_B WARNING same OTS index S reused]
    M --> N[publish fallback block]
Loading
Prompt To Fix All With AI
Fix the following 3 code review issues. Work through them one at a time, proposing concise fixes.

---

### Issue 1 of 3
crates/blockchain/src/lib.rs:560-562
**XMSS OTS key reuse in fallback path**

`prebuild_block` at interval 4 calls `sign_block_root(validator_id, slot=S, &block_root_A)`, then if `prebuilt_block_is_usable` returns false at interval 0, the fallback calls `build_signed_block``sign_block_root(validator_id, slot=S, &block_root_B)` — same `slot` argument, different message. `ValidatorSecretKey::sign` takes `&self` (non-mutating) and calls `LeanSignatureScheme::sign(&self.inner, slot, …)` deterministically from the OTS leaf at index `slot`, with no internal "already used" guard. Two distinct messages signed under the same OTS leaf violates the W-OTS+/XMSS single-use guarantee; if both signatures are ever visible to an attacker (e.g., via in-memory recovery or a future change that inadvertently publishes the pre-build), signing keys are compromisable. The PR's own description flags OTS-index reuse as a non-negotiable constraint, yet this fallback path triggers it whenever the head changes between interval 4 and interval 0.

### Issue 2 of 3
crates/blockchain/src/lib.rs:259-265
**`has_proposal` flag now ignores sync status**

The old code filtered `scheduled_proposer` through `sync_status.duties_allowed()` before passing the flag to `store::on_tick`. The new code passes `is_proposer = scheduled_proposer.is_some()` without that filter, so a syncing node that happens to be the scheduled proposer now passes `has_proposal = true`, causing `store::on_tick` to call `accept_new_attestations` at interval 0 even though no block will be built. The old behavior deliberately withheld this early promotion while syncing.

```suggestion
        let scheduled_proposer = (interval == 0 && slot > 0)
            .then(|| self.get_our_proposer(slot))
            .flatten();
        let is_proposer = scheduled_proposer.is_some()
            && self.sync_status.duties_allowed();

        // Tick the store first - this accepts attestations at interval 0 if we have a proposal
        store::on_tick(&mut self.store, timestamp_ms, is_proposer);
```

### Issue 3 of 3
crates/blockchain/src/lib.rs:460-464
**Double `coverage::emit_proposal_coverage` emission**

`build_signed_block` calls `coverage::emit_proposal_coverage` unconditionally. This function is invoked at interval 4 (pre-build) and potentially again at interval 0 (fallback). If the pre-built block passes `prebuilt_block_is_usable` but `process_and_publish_block` returns false (local import failure), the fallback calls `build_signed_block` a second time, emitting proposal-coverage metrics twice for the same slot. Consider moving the coverage emission into `process_and_publish_block` (after a successful local import) so it fires exactly once per published block.

Reviews (1): Last reviewed commit: "feat(blockchain): pre-build proposer blo..." | Re-trigger Greptile

Comment thread crates/blockchain/src/lib.rs Outdated
Comment thread crates/blockchain/src/lib.rs Outdated
Comment thread crates/blockchain/src/lib.rs
#450 (the on_tick interval regrouping) landed in main via squash, so both
sides added the same interval markers/blocks relative to the pre-#450
merge-base. Git kept both with no textual conflict, duplicating the
interval 2/3/4 sections (start_aggregation_session would have run twice).
Removed the duplicate set, keeping a single interval 0-4 sequence with the
prebuild trigger in interval 4.
@MegaRedHand MegaRedHand force-pushed the feat/proposer-prebuild-aggregation branch from c5812bb to d940222 Compare June 22, 2026 17:15
@MegaRedHand MegaRedHand marked this pull request as draft June 22, 2026 21:20
…t start

The proposer pre-built its block at the previous slot's interval 4 and
stashed it for the interval-0 tick to publish. But the build takes longer
than one 0.8s interval, so it overruns the slot boundary; handle_tick's
overrun catch-up (#413) then re-ticks at the current interval and skips
the missed interval-0. The stashed block was never published and block
production halted (head frozen, every proposer blocked by its own prebuild).

Consolidate the proposal into a single path that runs entirely at
interval 4: build on the current head, align publication to the slot
boundary (sleep out the remainder if the build finished early, publish
immediately if it overran), revalidate the head and rebuild if it moved,
then publish. This never depends on the interval-0 tick firing.

Remove the now-dead dual-path machinery: the prepared_block stash,
PreparedBlock/prebuilt_block_is_usable, the interval-0 propose_block, and
the orphaned produce_block_with_signatures/get_proposal_head clock-ticking
build path.
Review follow-up to the interval-4 proposal consolidation; no behavior change:
- Restore the interval-0 section marker in on_tick with a note that the
  block is built and published at interval 4 of the previous slot.
- Re-comment the head-revalidation guard in propose_block: under the
  single-threaded actor it always passes today (no message interleaves
  mid-build, even across the align-sleep), and is kept as a safety net for
  if the build is ever moved off the actor thread.
- Fix stale references to the removed prebuild_block / synchronous proposal
  path / produce_block_with_signatures in doc comments, the CLAUDE.md
  tick-duties table, and the architecture infographic.
With block proposal moved to interval 4 of the previous slot, the
interval-0 accept_new_attestations fired after the block had already been
built and published: too late to be included in it, and it diverged the
proposer's fork-choice view from the one its block closed over. Accepting
should happen immediately before the build, which the unconditional
interval-4 accept already does (it runs at the top of the same tick, just
before propose_block).

Comment out the interval-0 accept and its gate. has_proposal is now unused
but kept in the signature (discarded via `let _`) so re-enabling needs no
call-site change.
Follow-up to disabling the proposer's interval-0 accept_new_attestations:
attestation promotion now happens only at interval 4, so the tick-duties
table and the attestation-pipeline note are updated to match.
The block is published at interval 0 (the slot boundary). The build+publish
code path is merged into the previous slot's interval 4 and aligned to
publish at the boundary, but conceptually publication belongs to interval 0.
Also reflect that the proposer's interval-0 accept_new_attestations was
dropped, so promotion now happens only at interval 4.
Move the proposer's interval-0 deviation out of store::on_tick and into the
actor. Commenting out store::on_tick's interval-0 attestation accept (f34e81c)
changed a function the conformance tests drive directly; revert store::on_tick
to its spec-faithful form.

Instead, propose_block (which runs at the previous slot's interval 4) now
advances the store to the target slot's interval 0 via store::on_tick(.., true)
— accepting attestations exactly as the real interval-0 tick would — before
building, so the block closes over the same interval-0 state a non-prebuilding
proposer would see. Publication is still aligned to the slot boundary, and the
real interval-0 tick is skipped by the idempotency guard since the store clock
is already advanced.
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.

1 participant