github celestiaorg/celestia-core v0.43.0-rc1

pre-release43 minutes ago

Second release candidate of the v0.43.x line, for celestia-app v12 (celestia-app v11 stays on v0.42.x). Since rc0 it keeps the consensus lock across restarts, makes block recovery faster, and hardens consensus, propagation, the CAT mempool, evidence, p2p, state sync and privval. Requires Go 1.26.11.

Breaking changes since rc0

  • App-version gates move from 11 to 12. go-square's BlobTx rules (#3694) and the canonical proposal block encoding voting rule (#3595, #3685) apply from app version 12. Below it a node behaves as on v0.42.x.
  • New consensus WAL record LockInfo (#3529), used to restore the consensus lock after a restart. WALs without it replay as before.

Consensus

  • Restore the consensus lock after a restart, refetch the locked block, and serve the replayed proposal (#3530, #3531, #3533, #3538, #3663, #3665, #3667, #3668).
  • Journal our prevote before acting on an earlier prevote quorum (#3680), act on prevotes collected before the prevote step (#3518), stay in the commit step when a later round gets a quorum (#3738), and re-propose ValidBlock with the parts that got the polka (#3631).
  • Handle our own messages before peer messages (#3672) and skip vote catchup for peers ahead of us (#3703).

Propagation

  • Faster recovery. A node missing original parts rebuilds only those, hands them to consensus first and builds parity in the background (#3755, #3756): at 32 MB, originals reach consensus in 21.3 ms instead of 40.1 ms. Recovery from the mempool is 69 % faster (52.1 → 15.9 ms geomean) and allocates 68 % less (#3713, #3714, #3715, #3716).
  • The proposer pushes up to 3 parts no mempool can rebuild (header, evidence, last commit) right after the compact block. A node accepts unrequested parts only from a peer that sent it the compact block, up to that cap; any other unrequested part disconnects the sender, or is dropped during catchup (#3474).
  • Bound WantParts height, round and per-peer state (#3611); serve committed blocks and wants from the store (#3537, #3661); rely on request expiry for catch-up re-requests (#3662); follow Block.MaxBytes changes without a restart (#3677).

Mempool (CAT)

  • SeenTx claims are kept per peer and a delivery is judged only by its sender's claim (#3708, #3709, #3710, #3711). Each peer gets a byte budget in the received-tx buffer (#3676).
  • The pool height is read without the pool mutex, so Receive no longer waits out recheck (#3693).
  • New metric mempool_recheck_duration_seconds and one info line per recheck (#3736).

Evidence

  • Duplicate vote evidence: one pending entry per validator and height, committed in one write, with leftovers dropped on restart (#3705, #3706, #3707).

p2p, state sync, blocksync, store, privval, light client and RPC

  • p2p: keep accepting connections after temporary accept errors (#3684) and reply without blocking in reactor Receive (#3678). PEX uses the socket address as the address book source and bounds the banned list (#3744, #3743, #3761).
  • State sync tracks the next chunk to allocate and keeps the state sync height until the first block is stored (#3739, #3760). Blocksync counts block responses only against requests made to that peer (#3608).
  • The seen commit is synced to disk (#3750).
  • The privval signer server reconnects after a read error and keeps dialing the node (#3741, #3759).
  • Light client RPC stops forwarding events after unsubscribe or disconnect (#3745); abci_query no longer forwards inclusion proof paths (#3712).
  • Block max bytes are validated against the protocol limit (#3740), and the reserved event type key is ignored in app events (#3746).

Dependencies

  • golang.org/x/crypto v0.59.0, pyroscope-go v1.5.0 and aws-sdk-go-v2 updates; every Dockerfile pinned to Go 1.26.11 (#3730, #3733, #3734, #3735, #3752).

What's Changed

  • ci: [high] let auto-approve cover changes under .github/ by @walldiss in #3671
  • fix(consensus): [high] handle own messages before peer messages by @rootulp in #3672
  • fix(propagation): [high] serve the committed block to nodes still waiting for its parts by @rach-id in #3537
  • fix(consensus): [high] act on prevotes collected before the prevote step by @rach-id in #3518
  • feat(consensus): [high] add the LockInfo WAL record by @rach-id in #3529
  • fix(propagation): [high] bound WantParts height/round and per-peer want state by @rootulp in #3611
  • ci: [low] run generated e2e testnets and deadlock-tagged tests nightly by @rootulp in #3675
  • fix(consensus): [high] restore the consensus lock after a restart by @rach-id in #3530
  • fix(consensus): [high] refetch the locked block and persist the current proposal after a restart by @rach-id in #3531
  • fix(consensus): [high] serve the replayed proposal after a restart by @rach-id in #3533
  • fix(consensus): [high] keep a copy of the locked block while a height is stalled by @rach-id in #3538
  • fix(propagation): [high] rely on request expiry for catch-up re-requests by @rach-id in #3662
  • test(consensus): [low] deflake TestLatePrecommitIncludedAfterSlowExecution by @rach-id in #3679
  • fix(propagation): [high] serve committed wants from the store when the cached entry is incomplete by @rach-id in #3661
  • fix(consensus/propagation): [high] follow Block.MaxBytes changes without restart by @rootulp in #3677
  • fix(consensus): [high] refetch a restored lock once, after replay by @rach-id in #3663
  • fix(consensus): [high] locked-block copy review fixes and tests by @rach-id in #3665
  • test(consensus): [low] cover the lock hand-over guards by @rach-id in #3666
  • fix(consensus): [high] require canonical proposal block encoding by @rootulp in #3595
  • ci: [low] cancel superseded actionlint runs on PRs by @rootulp in #3683
  • fix(consensus): [high] gate the canonical proposal encoding rule on app version 12 by @rootulp in #3685
  • test(consensus): [low] deflake TestCorrelatedRestartDuringPrecommitDelay by @rach-id in #3681
  • test(consensus): [low] deflake TestReactorValidatorSetChanges by @rach-id in #3687
  • test(p2p/pex): [low] deflake TestPEXReactorWhenAddressBookIsSmallerThanMaxDials by @rootulp in #3689
  • fix(blocksync): [high] count block responses only against requests made to the peer by @rootulp in #3608
  • test(consensus): [low] buffer new-block events in reactor tests by @rootulp in #3696
  • perf(mempool): [low] read the CAT pool height without the pool mutex by @rach-id in #3693
  • fix(consensus): [high] small review fixes for the WAL-lock stack by @rach-id in #3667
  • fix(consensus): [high] keep an unlocked proposer's own block across restart by @rach-id in #3668
  • fix(consensus): [medium] keep another block's entry inside the propagator for our own proposals by @rach-id in #3686
  • fix!: [high] gate blob tx validation on app version 12 by @rootulp in #3694
  • fix(consensus): [high] journal our prevote before acting on an earlier prevote quorum by @rach-id in #3680
  • test(consensus): [low] make pacing replay tests independent of TimeoutCommit by @rootulp in #3717
  • fix(rpc): [medium] do not forward inclusion proof paths from abci_query by @rootulp in #3712
  • fix(p2p): [high] keep accepting connections after temporary accept errors by @rootulp in #3684
  • fix(statesync,p2p): [high] use non-blocking sends for replies in reactor Receive by @rootulp in #3678
  • test(blocksync): [low] remove stale fuzz exclusion by @ninabarbakadze in #3702
  • fix(consensus): [high] skip vote catchup for peers ahead of us by @ninabarbakadze in #3703
  • refactor(mempool): [medium] queue SeenTx claims as their own objects by @rootulp in #3708
  • fix(mempool): [high] record which peers announced each SeenTx claim by @rootulp in #3709
  • fix(mempool): [high] judge a delivery only by its sender's SeenTx claim by @rootulp in #3710
  • test(e2e): [low] inject duplicate vote evidence for distinct validator and height pairs by @rootulp in #3704
  • fix(evidence): [high] commit evidence in one write and drop committed leftovers on restart by @rootulp in #3705
  • fix(evidence): [high] keep one pending duplicate vote evidence per validator and height by @rootulp in #3706
  • fix(evidence): [high] record committed duplicate vote evidence per validator and height by @rootulp in #3707
  • fix(consensus): [high] re-propose ValidBlock with the parts that got the polka by @rootulp in #3631
  • fix(mempool/cat): [high] give each peer a byte budget in the received-tx buffer by @rootulp in #3676
  • fix(mempool): [high] keep each peer's SeenTx claim separate by @rootulp in #3711
  • fix(statesync): [high] track the next chunk to allocate and return by @ninabarbakadze in #3739
  • fix(consensus): [high] stay in the commit step when a later round gets a quorum by @ninabarbakadze in #3738
  • fix(types): [high] validate block max bytes against the protocol limit by @ninabarbakadze in #3740
  • perf(crypto/merkle): [high] hash leaves without copying them by @rach-id in #3713
  • perf(propagation): [high] verify locally recovered parts concurrently by @rach-id in #3714
  • perf(propagation): [high] copy recovered transactions into parts once by @rach-id in #3715
  • perf(propagation): [high] build recovered parts concurrently by @rach-id in #3716
  • build(docker): [low] pin every Dockerfile to Go 1.26.11 by @rootulp in #3752
  • build(deps): [high] bump github.com/aws/aws-sdk-go-v2/service/s3 from 1.114.6 to 1.114.10 by @dependabot[bot] in #3735
  • build(deps): [medium] bump amannn/action-semantic-pull-request from 6.1.1 to 6.1.2 by @dependabot[bot] in #3729
  • build(deps): [high] bump golang.org/x/crypto from 0.58.0 to 0.59.0 by @dependabot[bot] in #3730
  • build(deps): [high] bump github.com/grafana/pyroscope-go from 1.4.3 to 1.5.0 by @dependabot[bot] in #3733
  • build(deps): [high] bump github.com/aws/aws-sdk-go-v2/config from 1.33.13 to 1.33.18 by @dependabot[bot] in #3734
  • fix(types): [high] ignore the reserved event type key in app events by @ninabarbakadze in #3746
  • fix(p2p/pex): [high] use the socket address as the address book source by @ninabarbakadze in #3744
  • fix(privval): [high] reconnect the signer server after a read error by @ninabarbakadze in #3741
  • perf(propagation): [high] push the parts no mempool can rebuild, from the proposer only, bounded by @ninabarbakadze in #3474
  • fix(store): [high] sync the seen commit to disk by @ninabarbakadze in #3750
  • fix(p2p/pex): [high] bound the banned peer list by @ninabarbakadze in #3743
  • fix(light/rpc): [high] stop forwarding events after unsubscribe or disconnect by @ninabarbakadze in #3745
  • fix(statesync): [high] keep the state sync height until the first block is stored by @ninabarbakadze in #3760
  • docs(blocksync): [low] shorten the late-block comment in addBlock by @rootulp in #3751
  • perf(p2p/pex): [high] make room for a ban in one pass over the list by @ninabarbakadze in #3761
  • fix(privval): [high] keep dialing the node from the signer server by @ninabarbakadze in #3759
  • perf(propagation): [high] reconstruct only the missing original parts by @rach-id in #3755
  • perf(propagation): [high] hand originals to consensus first, build parity in the background by @rach-id in #3756
  • feat(mempool): [low] report recheck duration by @rach-id in #3736

Full Changelog: v0.43.0-rc0...v0.43.0-rc1

Don't miss a new celestia-core release

NewReleases is sending notifications on new releases.