feat(pipeline): minimize deadzone w cross slot attesting by Maddiaa0 · Pull Request #21435 · AztecProtocol/aztec-packages

Maddiaa0 · 2026-03-12T15:53:16Z

Overview

Allow validators to attest to old proposals slightly into the next slot, this allows validators from the slot before to send their checkpoint proposals
later in their own slot.

Key points

Allows timetable to extend past the current slot
Decouples attestation gathering from the hot path for sequencers - it can move async
Refactor of the timetable model into stdlib so it can be consumed elsewhere

Maddiaa0 · 2026-03-12T15:53:27Z

feat(pipeline): minimize deadzone w cross slot attesting #21435 : 2 dependent PRs (#21590 , #21591 ) 👈 (View in Graphite)
chore(pipeline): spartan config #21285
feat(pipeline): handle pipeline prunes #21250
chore(pipeline): timetable adjustments under pipelining #21076
feat(pipeline): allow syncing blocks ontop of the proposed chain #21025
feat(pipeline): introduce pipeline views for building #21026
feat(sequencer): add build-ahead config and metrics #20779
merge-train/spartan

This stack of pull requests is managed by Graphite. Learn more about stacking.

Maddiaa0 · 2026-03-23T14:10:50Z

aztec-packages/yarn-project/sequencer-client/src/publisher/sequencer-publisher.ts

Line 826 in d88635a

const ts = checkpoint.header.timestamp;

fix here

spalladino · 2026-03-31T15:19:39Z

      initialValidators: validators,
      enableProposerPipelining: true, // <- yehaw
      mockGossipSubNetwork: true,
+      mockGossipSubNetworkLatency: 500, // 200 ms delay in message prop - adverse network conditions


It's amazing how fast comments become outdated

spalladino · 2026-03-31T15:19:44Z

      disableAnvilTestWatcher: true,
      startProverNode: true,
-      perBlockAllocationMultiplier: 1,
+      perBlockAllocationMultiplier: 8,


This will lead to very big blocks at the beginning of the checkpoint, we sure about it?

i had the value so high as it essentially disables it. ceil(remainingBudget / remainingBlocks * multiplier)

spalladino · 2026-03-31T15:20:07Z

+      ethereumSlotDuration: 12,
+      aztecSlotDuration: 72,
      blockDurationMs: 8000,
-      l1PublishingTime: 2,
-      attestationPropagationTime: 0.5,
+      // maxDABlockGas: 786432, // Set max DA block gas to be the same as the checkpoint
+      // l1PublishingTime: 2,
+      // attestationPropagationTime: 1,


Why these changes?

legacy and ill have to make sure if still true - l1PublishingTime + attestationPropagationTime removal was to use the network defaults in this test.

maxDaBlockGas was due dividing the defaults across blocks at one point in time was not enough to get a tx into a block

spalladino · 2026-03-31T15:50:14Z

-        // Check if message is for previous slot and within clock tolerance
-        if (!isWithinClockTolerance(slotNumber, targetSlot, this.epochCache)) {
+        // When pipelining, accept attestations for the current slot (built in the previous slot)
+        // if we're within the first ethereumSlotDuration/2 seconds of the slot.


Comment is wrong? The implementation accepts up until the p2p propagation time.

spalladino · 2026-03-31T15:50:40Z

+ * @param epochCache - EpochCache to get timing and pipelining state
+ * @returns true if pipelining is enabled, the message is for the current slot, and we're within the grace period
+ */
+export function isWithinPipeliningGracePeriod(messageSlot: SlotNumber, epochCache: EpochCacheInterface): boolean {


Shouldn't this function also consider the clock disparity time?

Actually, can't we just merge this function with the one above? They are all used in all the same places after all.

ive done a little refactor here

spalladino · 2026-03-31T18:22:16Z

+        if (this.pipelining) {
+          // L1 submission happens in target slot, after attestations
+          return this.aztecSlotDuration + this.pipeliningAttestationGracePeriod;
+        }


This limit feels too tight. We're saying here that we're only willing to collect attestations until this time, after that, we just give up. I'd update this to this.aztecSlotDuration * 2 - this.l1PublishingTime.

spalladino · 2026-03-31T18:22:44Z

        stateDiff: [
          { slot: toPaddedHex(RollupContract.checkBlobStorageSlot, true), value: toPaddedHex(0n, true) },
+          ...forcePendingArchiveStateDiff,
          ...forcePendingCheckpointNumberStateDiff,


I'm surprised we need the archive override tbh, given the checkpoint number override works for simulating. Maybe we can merge them and keep a single override for "current state of the network"?

I dont think we do either, theres got to be a good reason i didnt migrate it in to the earlier PR. This appears to just be stale

Right, its required if supporting a pipeline depth > 1. When building ontop of a proposed checkpoint that is not confirmed, then this is required else the value of the archive for the checkpoint being built on is 0

merging them is a good idea

spalladino · 2026-03-31T18:52:05Z

-
-    if (checkpoint) {
-      this.metrics.recordCheckpointProposalSuccess();
+    if (!broadcast) {


Maybe it's in a later PR, but if we failed to broadcast (or if there was any error broadcasting) we should clean up our proposed blocks from our archiver, so we can attest in the next slot.

This is true, right now the wider rollback covers this - but we will need to tighten it up

spalladino · 2026-03-31T18:52:51Z

+    // Background the attestation → signing → L1 pipeline so the work loop is unblocked
+    this.pendingL1Submission = this.waitForAttestationsAndEnqueueSubmissionAsync(broadcast, votesPromises);


I'm trying to figure out if there's an issue with this getting overwritten before completion if we are the proposer for two slots in a row. But I don't think so - this is just used for awaiting completion, right?

spalladino · 2026-04-08T18:59:50Z

+  const elapsedMs = Number(nowMs - slotStartMs);
+  const windowMs = windowSeconds * 1000;
+
+  return elapsedMs < windowMs;


We should still account for clock disparity here

spalladino · 2026-04-08T19:10:47Z

+};
+
+/**
+ * Shared checkpoint timing model used by the sequencer timetable and P2P validators.


Good refactor

WDYT about having different impls for pipelining and not pipelining, to avoid all the conditionals in the getters? No need to implement now, for sure.

yeah good idea

spalladino · 2026-04-08T20:07:28Z

+    }
+
+    const timeAvailableForBlocks = this.aztecSlotDuration - this.checkpointInitializationTime - this.timeReservedAtEnd;
+    return Math.max(1, Math.floor(timeAvailableForBlocks / this.blockDuration));


Can this mask errors where we have an invalid config altogether? Should we instead throw if the resulting blocks per slot is below 1?

Yeah good point

spalladino · 2026-04-08T20:13:26Z

+  public get pipeliningAttestationGracePeriod(): number {
+    return (this.blockDuration ?? 0) + this.p2pPropagationTime;
+  }


Can you remind me why this value?

Aside from that, it'd be good to keep comments on what each value represents, and the rationale for it.

BEGIN_COMMIT_OVERRIDE fix(stdlib): use bigint arithmetic in GasFees.mul() for non-integer scalars (#22383) fix(node-lib): reuse existing fileStore in snapshot sync instead of recreating (#22375) fix: gate req/resp data protocols for unauthenticated peers (#22406) fix(p2p): use per-batch ops array in AztecDatastore.batch() (#22357) chore(pipeline): spartan config (#21285) chore: add claude skill to send txs (#22439) feat(pipeline): minimize deadzone w cross slot attesting (#21435) fix(p2p): avoid 32-bit overflow in attestation pool block position key (#22412) fix(prover-client): increment retry count on timeout re-enqueue to prevent infinite loop (#22355) fix: remove redundant p2pClient.start() call (#22438) chore: add kubectl binary to spartan .gitignore (#22454) END_COMMIT_OVERRIDE

## Overview Allow validators to attest to old proposals slightly into the next slot, this allows validators from the slot before to send their checkpoint proposals later in their own slot. ## Key points - Allows timetable to extend past the current slot - Decouples attestation gathering from the hot path for sequencers - it can move async - Refactor of the timetable model into stdlib so it can be consumed elsewhere

Maddiaa0 force-pushed the md/pipeline-overlap branch from 730df54 to 67be055 Compare March 16, 2026 12:13

Maddiaa0 force-pushed the md/pipeline-spartan-config branch from 26acaf0 to f18cc20 Compare March 16, 2026 12:13

This was referenced Mar 16, 2026

chore: Update latency explorer #21590

Closed

chore(pipeline): add metrics for pipeling building timelines #21591

Merged

Maddiaa0 force-pushed the md/pipeline-spartan-config branch from f18cc20 to b7b0c11 Compare March 16, 2026 18:08

Maddiaa0 force-pushed the md/pipeline-overlap branch from 67be055 to 72312e1 Compare March 16, 2026 18:08

Maddiaa0 mentioned this pull request Mar 17, 2026

chore(pipelining): add metrics for pipeline build timing #21626

Closed

Maddiaa0 force-pushed the md/pipeline-overlap branch from 72312e1 to 9f24073 Compare March 18, 2026 11:33

Maddiaa0 force-pushed the md/pipeline-spartan-config branch 2 times, most recently from 0944eb7 to de53468 Compare March 18, 2026 11:59

Maddiaa0 force-pushed the md/pipeline-overlap branch from 9f24073 to c2d1bb4 Compare March 18, 2026 11:59

Maddiaa0 marked this pull request as ready for review March 19, 2026 22:02

Maddiaa0 force-pushed the md/pipeline-overlap branch from c2d1bb4 to ef383be Compare March 19, 2026 22:03

Maddiaa0 force-pushed the md/pipeline-spartan-config branch from de53468 to 82f2dbd Compare March 19, 2026 22:03

Maddiaa0 force-pushed the md/pipeline-overlap branch 2 times, most recently from a399cd5 to c1e72e3 Compare March 20, 2026 13:21

Maddiaa0 force-pushed the md/pipeline-spartan-config branch from 82f2dbd to 4f49f4a Compare March 20, 2026 13:21

spalladino reviewed Mar 31, 2026

View reviewed changes

Maddiaa0 changed the base branch from md/pipeline-spartan-config to graphite-base/21435 April 8, 2026 11:46

Maddiaa0 force-pushed the md/pipeline-overlap branch from c1e72e3 to bfadfa1 Compare April 8, 2026 16:38

Maddiaa0 requested review from a team and jeanmon as code owners April 8, 2026 16:38

Maddiaa0 requested review from IlyasRidhuan, charlielye and fcarreiro as code owners April 8, 2026 16:38

Maddiaa0 changed the base branch from graphite-base/21435 to md/pipeline-spartan-config April 8, 2026 16:39

Maddiaa0 removed request for a team, IlyasRidhuan, charlielye, fcarreiro and jeanmon April 8, 2026 17:11

Maddiaa0 added the ci-full Run all master checks. label Apr 8, 2026

Maddiaa0 changed the base branch from md/pipeline-spartan-config to graphite-base/21435 April 8, 2026 19:51

Maddiaa0 force-pushed the md/pipeline-overlap branch from 59aca30 to 8fd3bbe Compare April 8, 2026 19:52

Maddiaa0 force-pushed the graphite-base/21435 branch from 7c93544 to 0f459dd Compare April 8, 2026 19:52

Maddiaa0 changed the base branch from graphite-base/21435 to md/pipeline-spartan-config April 8, 2026 19:52

Maddiaa0 changed the base branch from md/pipeline-spartan-config to graphite-base/21435 April 8, 2026 20:03

Maddiaa0 force-pushed the graphite-base/21435 branch from 0f459dd to 95557a0 Compare April 8, 2026 20:08

Maddiaa0 force-pushed the md/pipeline-overlap branch from 8fd3bbe to 0c09382 Compare April 8, 2026 20:08

Maddiaa0 changed the base branch from graphite-base/21435 to md/pipeline-spartan-config April 8, 2026 20:08

spalladino approved these changes Apr 8, 2026

View reviewed changes

Base automatically changed from md/pipeline-spartan-config to merge-train/spartan April 9, 2026 11:15

Maddiaa0 force-pushed the md/pipeline-overlap branch 2 times, most recently from 2c391ba to de83910 Compare April 9, 2026 12:24

chore: adjust clock tolerance

da950b2

Maddiaa0 force-pushed the md/pipeline-overlap branch from de83910 to da950b2 Compare April 9, 2026 14:43

Maddiaa0 merged commit 3548b45 into merge-train/spartan Apr 9, 2026
13 checks passed

Maddiaa0 deleted the md/pipeline-overlap branch April 9, 2026 15:19

AztecBot mentioned this pull request Apr 9, 2026

feat: merge-train/spartan #22428

Merged

Maddiaa0 mentioned this pull request Apr 10, 2026

chore(pipelining): update next net #22466

Merged

		// Background the attestation → signing → L1 pipeline so the work loop is unblocked
		this.pendingL1Submission = this.waitForAttestationsAndEnqueueSubmissionAsync(broadcast, votesPromises);

Conversation

Maddiaa0 commented Mar 12, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Overview

Key points

Uh oh!

Maddiaa0 commented Mar 12, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Maddiaa0 commented Mar 23, 2026

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Maddiaa0 Apr 8, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

Maddiaa0 commented Mar 12, 2026 •

edited

Loading

Maddiaa0 commented Mar 12, 2026 •

edited

Loading

Maddiaa0 Apr 8, 2026 •

edited

Loading