feat(pipeline): introduce pipeline views for building by Maddiaa0 · Pull Request #21026 · AztecProtocol/aztec-packages

Maddiaa0 · 2026-03-03T02:23:19Z

Overview

Epoch cache operations now return two views, the current slot and the pipelined slot / epoch

Testing

The main test which showcases this functionality is epoch_mbps.pipeline, this runs the sequencers with the pipeline mode enabled. At the moment it only expects 1 block per slot as it still waits until the proposal slot to send the checkpoint to l1. For this PR it uses a blocking sleep here, that stops all sequencers for this test. This is addressed in a pr stacked ontop.

Maddiaa0 · 2026-03-03T20:51:21Z

spalladino

Changes look good, but I don't love the new API for the epoch cache. I'd suggest either:

Rolling back the changes in epoch cache, have it just report the now info as it did, and have clients apply the pipelining offset as needed (most likely via a shared helper somewhere).
Keeping the same API as we did in the epoch cache, but add new methods like getEpochAndSlotPipeline which return the pipelined view for their counterparts (getEpochAndSlotNow here). I prefer this option.

Aside from that, I'd review naming: I found the "pipeline slot" confusing. The suggestion I like the most from Claude was current/target instead of now/pipeline.

spalladino · 2026-03-04T14:19:30Z

yarn-project/sequencer-client/src/sequencer/checkpoint_proposal_job.ts

+        submissionSlot: this.pipelineSlot,
+        submissionSlotTimestamp,
+      });
+      await sleepUntil(new Date(Number(submissionSlotTimestamp) * 1000), this.dateProvider.nowAsDate());


I think we can sleep until 12s before the submission slot timestamp, so L1 txs can get included on the first L1 block of the L2 slot

yarn-project/sequencer-client/src/sequencer/sequencer.ts

spalladino · 2026-03-04T14:24:49Z

yarn-project/validator-client/src/validator.ts


-    // Check that I have any address in current committee before attesting
-    const inCommittee = await this.epochCache.filterInCommittee(slotNumber, this.getValidatorAddresses());
+    // TODO(md): we check based on the slot number that is in the proposal - but not that the slot number makes sense?????


IIRC that's handled on the p2p proposal validator, but yeah, still worth looking into

spalladino · 2026-03-04T14:58:47Z

yarn-project/sequencer-client/src/sequencer/sequencer.ts

-      proposer ?? EthAddress.ZERO,
-      invalidateCheckpoint,
-    );
+    const canProposeCheck = await publisher.canProposeAtNextEthBlock(syncedTo.archive, proposer ?? EthAddress.ZERO, {


I think we should update the publisher methods so, instead of setting an offset to now, we just pass the publishing timestamp range for the slot and let the publisher handle it.

spalladino · 2026-03-04T15:05:17Z

yarn-project/ethereum/src/contracts/rollup.ts

@@ -749,10 +749,11 @@ export class RollupContract {
    archive: Buffer,
    account: `0x${string}` | Account,
    slotDuration: number,
-    opts: { forcePendingCheckpointNumber?: CheckpointNumber } = {},
+    opts: { forcePendingCheckpointNumber?: CheckpointNumber; additionalSlotOffset?: number } = {},
  ): Promise<{ slot: SlotNumber; checkpointNumber: CheckpointNumber; timeOfNextL1Slot: bigint }> {


I'd change this to a canProposeAt for robustness

Maddiaa0 · 2026-03-17T16:17:29Z

yarn-project/epoch-cache/src/epoch_cache.ts

    return this.l1constants;
  }

+  public isProposerPipeliningEnabled(): boolean {


Majority of work here is creating these views -> then other components use the version that matters to them.

if pipelining is disabled - this return what it previously did - so backwards compatible

spalladino

Epoch cache looks great now. Left a few comments, but looks good to merge.

yarn-project/p2p/src/msg_validators/clock_tolerance.ts

spalladino · 2026-03-18T13:38:15Z

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

+    // When pipelining, anchor the simulation timestamp to the checkpoint's own slot start time
+    // rather than the current L1 block timestamp, which may overshoot into the next slot if the build ran late.


I think we should always anchor to the checkpoint, rather than relying on current L1 block timestamp? We'll have to fix it in #14766 anyway.

yarn-project/sequencer-client/src/sequencer/checkpoint_proposal_job.ts

spalladino · 2026-03-18T13:47:46Z

yarn-project/sequencer-client/src/sequencer/checkpoint_proposal_job.ts

+      // Sync the pending broadcast block to the archiver before broadcasting the proposal.
+      // Same as non-last blocks in the build loop, this catches consistency errors early.
+      if (blockPendingBroadcast) {
+        await this.syncProposedBlockToArchiver(blockPendingBroadcast.block);
+      }


I think I already did this change in #21554

right on ill remove

spalladino · 2026-03-18T13:52:40Z

yarn-project/sequencer-client/src/sequencer/sequencer.ts

-    this.log.info(`Preparing checkpoint proposal ${checkpointNumber} at slot ${slot}`, { ...logCtx, proposer });
+    this.lastSlotForCheckpointProposalJob = targetSlot;
+
+    await this.p2pClient.prepareForSlot(targetSlot);


Let's make a note to review this. IIRC prepareForSlot(N) unprotects txs from slot N-1. Since we're now calling with the target slot, I'm worried we may be unprotecting txs we still need.

P2PClent also calls prepareForSlot so would need to be aligned.

I think that calling prepareForSlot with the targetSlot is correct. Ill update the p2p client to use it whenever we have a pendingCheckpointProposal (introduced in the next pr) therefore are pipelining

spalladino · 2026-03-18T14:07:57Z

yarn-project/sequencer-client/src/sequencer/sequencer.ts

    }

    // Check all components are synced to latest as seen by the archiver (queries all subsystems)
    const syncedTo = await this.checkSync({ ts, slot });


Maybe it comes on a later PR, but it seems like checkSync is missing something. Let's say we're at slot N, building N+1. This checks that we have synced N-1, not N. Maybe we should wait a bit if we have seen block proposals for N, but are still waiting on or processing the checkpoint proposal? Otherwise we risk start building our slot too early while there's still lingering execution from slot N.

These changes are in the PR above - are these enough?

I don't think so. These are only checking that components are 1) synced between each other, and 2) synced at least to slot N-1. But this should not affect the happy path.

In the pr above it also checks the existence of a pendingCheckpointProposal, we can reconsider this there. I do agree that it may be abit thin

PhilWindle · 2026-03-18T16:49:13Z

yarn-project/sequencer-client/src/sequencer/checkpoint_proposal_job.ts

+    }
+
    // Then send everything to L1
    const l1Response = await this.publisher.sendRequests();


Presumably we will only want to submit the proposal if our parent checkpoint didn't get pruned.

Maddiaa0 · 2026-03-18T21:40:01Z

yarn-project/end-to-end/src/e2e_epochs/epochs_mbps.parallel.test.ts

+    // ensures the deploy tx is in the pool before the call tx regardless of gossip timing.
    const timeout = test.L2_SLOT_DURATION_IN_S * 5;
-    logger.warn(`Sending both txs and waiting for checkpointed receipts`);
+    logger.warn(`Sending deploy tx first, then call tx`);


fix a flake

BEGIN_COMMIT_OVERRIDE fix(p2p): fall back to maxTxsPerCheckpoint for per-block tx validation (#21605) chore: fixing M3 devcontainer builds (#21611) fix: clamp finalized block to oldest available in world-state (#21643) chore: fix proving logs script (#21335) fix: (A-649) tx collector bench test (#21619) fix(validator): process block proposals from own validator keys in HA setups (#21603) fix: add bounds when allocating arrays in deserialization (#21622) fix: skip handleChainFinalized when block is behind oldest available (#21656) chore: demote finalized block skip log to trace (#21661) fix: skip -march auto-detection for cross-compilation presets (#21356) chore: revert "add bounds when allocating arrays in deserialization" (#21622) (#21666) fix: capture txs not available error reason in proposal handler (#21670) fix: estimate gas in bot and make BatchCall.simulate() return SimulationResult (#21676) fix: prevent HA peer proposals from blocking equivocation in duplicate proposal test (#21673) fix(p2p): penalize peers for errors during response reading (#21680) feat(sequencer): add build-ahead config and metrics (#20779) chore: fixing build on mac (#21685) fix: HA deadlock for last block edge case (#21690) fix: process all contract classes in storeBroadcastedIndividualFunctions (A-683) (#21686) chore: add slack success post on nightly scenario (#21701) fix(builder): persist contractsDB across blocks within a checkpoint (#21520) fix: only delete logs from rolled-back blocks, not entire tag (A-686) (#21687) chore(p2p): lower attestation pool per-slot caps to 2 (#21709) chore(p2p): remove unused method (#21678) fix(p2p): penalize peer on tx rejected by pool (#21677) fix(test): workaround slow mock creation (#21708) fix(sequencer): fix checkpoint budget redistribution for multi-block slots (#21692) fix: batch checkpoint unwinding in handleEpochPrune (A-690) (#21668) fix(sequencer): add missing opts arg to checkpoint_builder tests (#21733) fix: race condition in fast tx collection (#21496) fix: increase default postgres disk size from 1Gi to 10Gi (#21741) fix: update batch_tx_requester tests to use RequestTracker (#21734) chore: replace dead BOOTSTRAP_TO env var with bootstrap.sh build arg (#21744) fix(sequencer): extract gas and blob configs from valid requests only (A-677) (#21747) fix: deflake attempt for l1_tx_utils (#21743) fix(test): fix flaky keystore reload test (#21749) fix(test): fix flaky duplicate_attestation_slash test (#21753) feat(pipeline): introduce pipeline views for building (#21026) END_COMMIT_OVERRIDE

Maddiaa0 mentioned this pull request Mar 3, 2026

feat(pipeline): allow syncing blocks ontop of the proposed chain #21025

Open

Maddiaa0 force-pushed the feat/build-ahead-pr1-config-types branch from 9e57ad3 to 82d017c Compare March 3, 2026 20:51

Maddiaa0 force-pushed the md/update-epoch-cache-for-buildahead branch from 3a75fdb to 1ca98d8 Compare March 3, 2026 20:51

Maddiaa0 mentioned this pull request Mar 3, 2026

feat(sequencer): add build-ahead config and metrics #20779

Merged

Maddiaa0 mentioned this pull request Mar 3, 2026

chore(pipeline): timetable adjustments under pipelining #21076

Open

spalladino reviewed Mar 4, 2026

View reviewed changes

Maddiaa0 marked this pull request as ready for review March 6, 2026 13:17

Maddiaa0 force-pushed the md/update-epoch-cache-for-buildahead branch from b1b1b14 to ce0e422 Compare March 9, 2026 12:12

Maddiaa0 mentioned this pull request Mar 9, 2026

feat(pipeline): handle pipeline prunes #21250

Open

Maddiaa0 force-pushed the md/update-epoch-cache-for-buildahead branch from ce0e422 to d8b8d31 Compare March 9, 2026 16:01

Maddiaa0 force-pushed the feat/build-ahead-pr1-config-types branch from 82d017c to 21053ab Compare March 9, 2026 16:01

Maddiaa0 force-pushed the md/update-epoch-cache-for-buildahead branch from d8b8d31 to 6973d24 Compare March 9, 2026 16:41

Maddiaa0 mentioned this pull request Mar 10, 2026

chore(pipeline): spartan config #21285

Open

Maddiaa0 force-pushed the feat/build-ahead-pr1-config-types branch from 21053ab to f3565ef Compare March 12, 2026 15:52

Maddiaa0 force-pushed the md/update-epoch-cache-for-buildahead branch from 6973d24 to 21cdcd5 Compare March 12, 2026 15:52

This was referenced Mar 12, 2026

feat(pipeline): minimize deadzone w cross slot attesting #21435

Draft

chore: Update latency explorer #21590

Draft

chore(pipeline): shorten last block in checkpoint build duration to allow checkpoint construction #21591

Draft

Maddiaa0 force-pushed the feat/build-ahead-pr1-config-types branch from f3565ef to e14a469 Compare March 16, 2026 18:08

Maddiaa0 force-pushed the md/update-epoch-cache-for-buildahead branch from 21cdcd5 to 9543a61 Compare March 16, 2026 18:08

Maddiaa0 changed the base branch from feat/build-ahead-pr1-config-types to graphite-base/21026 March 17, 2026 12:28

Maddiaa0 mentioned this pull request Mar 17, 2026

chore(pipelining): thread last block duration in config #21626

Draft

Maddiaa0 force-pushed the graphite-base/21026 branch from e14a469 to 09904a9 Compare March 17, 2026 12:34

Maddiaa0 force-pushed the md/update-epoch-cache-for-buildahead branch from 9543a61 to 6644382 Compare March 17, 2026 12:34

Maddiaa0 changed the base branch from graphite-base/21026 to feat/build-ahead-pr1-config-types March 17, 2026 12:34

Maddiaa0 force-pushed the md/update-epoch-cache-for-buildahead branch from 6644382 to ed993b1 Compare March 17, 2026 16:15

Maddiaa0 force-pushed the feat/build-ahead-pr1-config-types branch from 09904a9 to 27a398a Compare March 17, 2026 16:15

Maddiaa0 commented Mar 17, 2026

View reviewed changes

Base automatically changed from feat/build-ahead-pr1-config-types to merge-train/spartan March 17, 2026 16:33

Maddiaa0 force-pushed the md/update-epoch-cache-for-buildahead branch 3 times, most recently from d527e2e to c60ca91 Compare March 17, 2026 18:04

spalladino approved these changes Mar 18, 2026

View reviewed changes

Maddiaa0 force-pushed the md/update-epoch-cache-for-buildahead branch 2 times, most recently from 9bba2e5 to 485647b Compare March 18, 2026 15:27

Maddiaa0 enabled auto-merge (squash) March 18, 2026 15:30

PhilWindle reviewed Mar 18, 2026

View reviewed changes

Maddiaa0 force-pushed the md/update-epoch-cache-for-buildahead branch 3 times, most recently from 4d09a7d to 14c625e Compare March 18, 2026 21:35

Maddiaa0 commented Mar 18, 2026

View reviewed changes

feat(pipeline): switch proposer slot at build-ahead boundary

255f12c

Maddiaa0 force-pushed the md/update-epoch-cache-for-buildahead branch from 14c625e to 255f12c Compare March 18, 2026 22:14

Maddiaa0 merged commit 4902133 into merge-train/spartan Mar 18, 2026
12 checks passed

Maddiaa0 deleted the md/update-epoch-cache-for-buildahead branch March 18, 2026 22:31

AztecBot mentioned this pull request Mar 18, 2026

feat: merge-train/spartan #21627

Merged

		// When pipelining, anchor the simulation timestamp to the checkpoint's own slot start time
		// rather than the current L1 block timestamp, which may overshoot into the next slot if the build ran late.

Conversation

Maddiaa0 commented Mar 3, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Overview

Testing

Uh oh!

Maddiaa0 commented Mar 3, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

spalladino left a comment • 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!

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!

spalladino left a comment

Choose a reason for hiding this comment

Uh oh!

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!

3 participants

Maddiaa0 commented Mar 3, 2026 •

edited

Loading

Maddiaa0 commented Mar 3, 2026 •

edited

Loading

spalladino left a comment •

edited

Loading