Skip to content

Bump Base to v1.5.5 - #42

Merged
maoueh merged 10 commits into
release/1.xfrom
bump/base-1.5.5
Nov 18, 2026
Merged

maoueh merged 10 commits into
release/1.xfrom
bump/base-1.5.5

Conversation

@dfuse-bot

@dfuse-bot dfuse-bot commented Nov 13, 2026 •

Copy link
Copy Markdown

Merges upstream Base v1.5.5 into release/1.x. Tag v1.5.5-fh3.1 on the merge commit once merged.

Upstream recommends this release for Base mainnet and Base Sepolia, with no deadline: it fixes the eth_call latency regression of v1.5.4 by returning --engine.memory-block-buffer-target to reth's default of 5 blocks (70 in v1.5.4). It also makes the sequencer seal payloads as soon as they are built.

v1.5.5 keeps upstream's reth pin (base-v2.5.2.11) and alloy-evm 0.38, so streamingfast/reth stays on base-v2.5.2.11-fh3.1 and streamingfast/evm on v0.38.0-sf. No other StreamingFast repository needs a release first. firehose-tracer stays on 5.7.0 and the Firehose block model is unchanged, so the tag keeps fh3.1 (v1.5.4-fh3.1 with the upstream version replaced).

Besides the merge:

  • Conflict in crates/execution/payload/src/builder.rs only. In try_build, upstream sends a payload-ready hint for frozen sequencer builds (!no_tx_pool), and we trace blocks built from the payload attributes alone (no_tx_pool). Both are kept, along with upstream's comment in the closure. The two conditions never hold together.
  • Cargo.toml: every streamingfast/reth crate and [patch] entry, and the alloy-evm patch, is now pinned by rev with the tag in a comment, instead of by tag. base-v2.5.2.11-fh3.1 points to the merge commit of Bump Base reth to base-v2.5.2.11 reth#89 (24849d77d), while Cargo.lock held that PR's head (9f9f5efa1). The two commits have the same tree, so no code changes. Cargo.lock changes only those sources and the version of our Firehose crates.
  • ACTIVATIONS.md: statuses refreshed against v1.5.5; no new upgrade and no schedule change.
  • CHANGELOG entry below.

Dependency table check: the fork pins streamingfast/reth (release/base-2.x) and streamingfast/evm (release/0.x), as the release tracker lists. release/base-2.x is still at the tagged commit. No public StreamingFast repository pins this fork: a code search for v1.5.4-fh3.1 only finds our deployment configuration.

CHANGELOG

## v1.5.5-fh3.1

### Changed

* Rebased on upstream Base `v1.5.5`, recommended for Base mainnet and Base Sepolia: it fixes the
  `eth_call` latency regression of `v1.5.4`. Uses `streamingfast/reth` `base-v2.5.2.11-fh3.1`;
  `firehose-tracer` stays on 5.7.0 and the Firehose block model is unchanged.
* Upstream returns `--engine.memory-block-buffer-target` of `base-reth-node` to 5 blocks (70 in
  `v1.5.4`). After an unclean shutdown the node still executes again and emits again the FIRE BLOCK
  lines of up to about 70 blocks, now fewer on average.

Test Execution Summary

Branch bump/base-1.5.5: release/1.x at v1.5.4-fh3.1 + git merge v1.5.5, built against streamingfast/reth base-v2.5.2.11-fh3.1 (24849d77d) and streamingfast/evm v0.38.0-sf (0ac393245), both fetched from GitHub.

Build Status

  • cargo build --release --package base-reth-node --bin base-reth-node --package base-consensus --bin base-consensus (same command as Dockerfile.sf): PASSED (4 min on a clean target directory, after the dependencies were fetched)
  • cargo check --workspace --all-targets: PASSED, after forge soldeer install && forge build in crates/utilities/test-utils/contracts (without the compiled contracts, base-test-utils fails on the missing contracts/out/*.json)
  • cargo clippy -p base-reth-node -p base-builder-multiplex -p base-execution-payload-builder --tests: PASSED. The warnings are in crates/firehose-flashblocks/src/processor.rs, crates/execution/firehose and crates/execution/flashblocks/src/traits.rs, which this merge does not change (same set as the v1.5.4-fh3.1 bump). rustfmt --check passes on the resolved builder.rs.

Unit Tests

Package Status Notes
cargo test -p base-firehose-tests PASSED 14 tests (prestate replay, EIP-8130 tracing)
cargo test -p base-firehose-flashblocks PASSED 44 tests
cargo test -p base-reth-node PASSED 2 tests
cargo test -p base-execution-firehose PASSED no unit tests in the crate
cargo test -p base-execution-chainspec PASSED 49 passed, 1 ignored (upstream)
cargo test -p base-common-genesis PASSED 103 tests
cargo test -p base-common-chains PASSED 52 passed, 1 ignored (upstream)
cargo test -p base-execution-payload-builder PASSED 148 passed, 1 ignored; covers the conflict resolved in builder.rs
cargo test -p base-reth-cli PASSED 16 passed, 1 ignored; asserts the memory block buffer target is back to 5
cargo test -p base-builder-multiplex PASSED 16 tests, including upstream's new payload_ready tests and in_process_smoke, which failed on the v1.5.4 bump and passes now
cargo test -p base-node-runner FAILED (not merge-related) 8 unit and integration tests pass. The BaseNode::provider_factory_builder doctest (crates/execution/runner/src/node.rs) does not compile: it sets ReadOnlyConfig { watch_static_files }, a field that neither upstream base/reth base-v2.5.2.11 nor our reth fork has (it is watch). The doc example is the same in upstream v1.5.4 and v1.5.5, so it is broken upstream
cargo test -p reth-firehose -p reth-firehose-tests (in streamingfast/reth at 24849d77d, base-v2.5.2.11-fh3.1) PASSED 35 + 2 tests
cargo test -p reth-engine-tree --test firehose_live_tracing --test firehose_live_tracing_bal (same checkout) PASSED 2 tests

Integration Tests

  • Status: PASSED
  • RUSTFLAGS="-C debug-assertions=off" RUST_MIN_STACK=33554432 cargo test -p base-system-tests --test firehose_denim --test firehose_b20 --test firehose_cobalt_b20, Docker-backed (after forge soldeer install && forge build in crates/utilities/test-utils/contracts):
Test Status Time
firehose_b20::b20_transfer_is_traced PASSED 25.1 s
firehose_cobalt_b20::cobalt_b20_surfaces_are_traced PASSED 53.7 s
firehose_denim::denim_activation_block_traces_base_time_install PASSED 51.6 s

Battlefield Tests

  • Status: PASSED
  • Pass/Fail: 91 passing, 47 pending, 0 failing
  • streamingfast/battlefield-ethereum at adb1921 (master), base-devnet target, with this branch's release base-reth-node and base-consensus as the Firehose node and the upstream ghcr.io/base/node:v1.5.5 image as the sequencer:
    BASE_REPO=<this checkout> BASE_NODE_IMAGE=ghcr.io/base/node:v1.5.5 ./scripts/base/run_base_devnet.sh
    BASE_RETH_NODE_BINARY=<this checkout>/target/release/base-reth-node \
      BASE_CONSENSUS_BINARY=<this checkout>/target/release/base-consensus \
      ./scripts/run_firehose_base_devnet.sh
    pnpm test:fh3.1:base-devnet
  • The devnet runs Base's own upgrades: Azul, Beryl and Cobalt at blocks 20 to 22, Denim at 25 (200 ms blocks from there on) and Everest at 30. Every Base test passed: the Denim BaseTime predeploy install, the BaseTime deposit and distinct Firehose times for blocks sharing a header timestamp, the B-20 precompile tests, the Everest nonce manager install, and both EIP-8130 transaction tests. Header corresponds to RPC and Compare blocks: Firehose blocks match RPC blocks passed. No snapshot changed, and the node log has no ERROR and no panic (about 900 blocks). The reader node reported FIRE INIT 3.1 reth base-1.5.5.
  • The 47 pending tests are gated on features this chain does not have (Amsterdam/BAL, slot number, Transfer logs, blobs, Arbitrum, and other chains' upgrades), plus "Reverted token creation keeps its writes in the trace", which the Battlefield README lists as a known pending case for this tracer.
  • Zenith is genesis-only, and no devnet schedules it, so the Keystore account changes are covered only by the unit tests in base-firehose-tests.

Firehose bug patterns

  1. Serial order: the release's changes to the payload builder (standard transactions snapshotted at build start, uncompressed block size limit, payload-ready hint), the txpool, the sequencer actor (sealing on freeze notifications), the batcher, the tx-manager and transaction forwarding are on the sequencer and builder side. None of them runs the EVM on the traced executor or hands it transactions out of order. BuiltBlock::trace for blocks built from payload attributes alone is kept. None found.
  2. Emit after accept: no change to engine validation, consensus checks or the reth fork. The memory block buffer target only changes how many blocks stay in memory after a persistence run. Persistence still starts once more than 70 blocks are unpersisted, so after an unclean shutdown the node executes and prints again up to about 70 blocks, as in v1.5.4, and usually fewer. None found.
  3. Activations: v1.5.5 changes nothing in crates/common/chains or crates/utilities/upgrade-signal, and defines no new BaseUpgrade. Denim is active on Base mainnet and Sepolia. Everest is still not scheduled: there is no timestamp in the code, and getSchedule() on the configured L1 protocol_versions_address still reverts on both L1s (checked 2026-11-13). Zenith is genesis-only. ACTIVATIONS.md refreshed.
  4. Fields: no change to the consensus, RPC or receipt types in crates/common that feed the block, and no change to the reth mapper. The new PayloadReady type is an engine notification and does not reach the block. None found.
  5. Hashes: no new transaction without a hash of its own. None found.

github-actions Bot and others added 10 commits November 6, 2026 13:59
Co-authored-by: github-actions[bot] <github-actions[bot]@users.noreply.github.com>
…ackport base#6477 to v1.5.5) (base#6484)

* feat(builder): limit uncompressed block size in the native builder

Add --builder.max-uncompressed-block-size to the native payload builder.
It caps the cumulative EIP-2718 encoded size of a block's transactions.
The Flashblocks builder already has this limit.

Sequencer transactions always count toward the total. Pool transactions
that would push the block past the limit are skipped for that block. The
check uses the encoded length cached on each pooled transaction, so the
build loop does no extra encoding. The limit is unset by default.

(cherry picked from commit 6b14f87)

* test(builder): cover sequencer bytes in the uncompressed block size limit

(cherry picked from commit 20fbb91)

* style: cargo fmt

---------

Co-authored-by: Danyal Prout <git@dany.al>
* fix(tx-manager): compute transaction hash locally in publish_tx (base#5878)

* fix(tx-manager): compute transaction hash locally in publish_tx

Co-authored-by: Toshi <toshi-noreply@coinbase.com>

* fix(tx-manager): address review round 1 findings

Co-authored-by: Toshi <toshi-noreply@coinbase.com>

* docs(tx-manager): document publish_tx local hash rule (review feedback)

Co-authored-by: Toshi <toshi-noreply@coinbase.com>

---------

Co-authored-by: Toshi <toshi-noreply@coinbase.com>
(cherry picked from commit ec65d30)

* feat(batcher): add min tip cap and min basefee flags (base#6427)

The batcher always prices its L1 transactions with the priority fee its L1
node suggests, which is close to zero on Ethereum today. A fee bump doubles
that tip, and doubling a near-zero tip keeps it near zero, so when L1 blocks
are full a batch transaction can wait a long time before it is included.

The tx manager already enforces a minimum tip cap and a minimum base fee,
and the other services expose them through `define_tx_manager_cli!`.
op-batcher has the equivalent `--txmgr.min-tip-cap` and
`--txmgr.min-basefee` settings. Expose them on `base batcher` as
`--min-tip-cap` / `BASE_BATCHER_MIN_TIP_CAP` and `--min-basefee` /
`BASE_BATCHER_MIN_BASEFEE`, in gwei with decimal strings accepted, and pass
them through `BatcherConfig` to the tx manager. Both default to 0, so
nothing changes until an operator sets them.

(cherry picked from commit 132bbc1)

* feat(batcher): send a channel's blobs in one L1 transaction (base#6429)

* feat(batcher): send a channel's blobs in one L1 transaction

A blob transaction now takes the ready blobs up to and including the first one that holds a channel's last frame, at most max_blobs_per_tx. Blobs of a channel still open wait until they fill a transaction instead of each going out alone, since the safe head only moves once the channel's last frame is on L1. A ready calldata frame also ends the run, so blobs before it never wait behind it.

DaEgress now builds new artifacts whenever the ready ones cannot be leased, instead of only when none is ready, so waiting blobs never stop the encoder from framing more output.

* fix(batcher): keep a calldata cut when a blob run is requeued

A calldata artifact now ends the blob run before it in any state, not only
when ready. Once the calldata went out, a requeued run cut by it no longer
waits for more channel output: it is leased again as the same transaction.

Co-authored-by: Toshi <toshi-noreply@coinbase.com>

---------

Co-authored-by: Toshi <toshi-noreply@coinbase.com>
(cherry picked from commit 40a9ebf)

* feat(tx-manager): add tx inclusion latency metric (base#6428)

* feat(tx-manager): add tx inclusion latency metric

Co-authored-by: Toshi <toshi-noreply@coinbase.com>

* fix(tx-manager): record inclusion latency with the send latency

Take the inclusion time from the canonical block that the receipt poller
already reads, and record it in the same place as the confirmed send
latency. Every confirmed send now records both, so the dashboard's
included-to-confirmed series compares the same sends. This drops the
extra block read and the case where it failed.

* fix(tx-manager): export ConfirmedTx from the crate root

Co-authored-by: Toshi <toshi-noreply@coinbase.com>

* test(tx-manager): use mockall for the inclusion latency test

Generate `MockTxMetrics` with `automock` instead of hand-rolling a
`TxMetrics` double, as the repository testing rules require. The test
now expects exactly one confirmed send latency of 0 ms and one inclusion
latency of 19_700 ms.

Co-authored-by: Toshi <toshi-noreply@coinbase.com>

---------

Co-authored-by: Toshi <toshi-noreply@coinbase.com>
(cherry picked from commit 3555800)

---------

Co-authored-by: Jack Chuma <jack.chuma@coinbase.com>
Co-authored-by: Toshi <toshi-noreply@coinbase.com>
…RE (base#6522)

A mempool node forwards every transaction to every configured builder, and
each builder that rejected it made the node journal a forward failure. The
receiving builder already journals the same rejection as
TXPOOL_VALIDATED_INSERT_REJECTED while it is the active sequencer, so the
node-side row either duplicated the leader's verdict or recorded a standby or
canary verdict that never decides inclusion. On mainnet (2026-11-08 09:00 UTC)
these were 20.2M rows/hour, 17.6% of all transaction events, 88% of them from
the two canary sequencers.

Drop the emit and retire the event type; historical rows stay readable.
Forwarding, failure reports to the reader, and the per-builder rejection
metric are unchanged. TXPOOL_BUILDER_FORWARD_ATTEMPT and _DROPPED stay, since
a send that never reaches a builder has no builder-side record.

CHAIN-6721

(cherry picked from commit 34c165f)
… layer (backport base#6557 to v1.5.5) (base#6567)

* feat(sequencer): seal payloads on execution freeze notifications

Backport of base#6557 to releases/v1.5.5.

* test(builder): exercise native payload-ready wiring

* test(sequencer): cover searcher next-block inclusion with early broadcast

* docs: update the docs index for frozen payload notifications

* chore(payload): order the tokio dependency by line length

* feat(sequencer): add early-seal metrics

Count seals by trigger, record when getPayload and gossip happen relative to the slot start, and count frozen-payload hints sent by the builder and dropped by the sequencer.

* docs(runner): note the exported prefix of early-seal metrics

* fix(sequencer): seal early when a hinted rebuild reuses the last sealed payload id

Track the frozen hint on the payload handle so it cannot outlive its build, and treat a hint as having lost to the deadline only when it matches no active build.

* refactor(sequencer): drop the gossip offset metric

It stopped at the handoff to the network actor rather than the publish, so it added little over the getPayload offset.

* feat(sequencer): record when each block build starts in its slot

A Denim build reads the transaction pool once, after the build is requested, so this offset is the deadline for landing in that block.

* fix(sequencer): measure slot start from the parent timestamp on the release branch

The isolated-sequencer schedule delay that block_deadline applies on main is not on releases/v1.5.5.

* test(sequencer): pin when early-seal offsets are taken

Cover build starts recorded outside the ticker, assert the getPayload offset is taken before the request, and describe the build start offset as part of the searcher reaction window.

* docs: restamp the runner README in the docs index

* test(builder): stop deriving multiplex smoke timestamps from wall clock (base#6455)

Denim activated on mainnet, so the smoke test's wall-clock-based payload
attributes became post-Denim attributes, which the engine rejects unless
they carry the BaseTime update deposit at tx[1]. Anchor the build
timestamp to the last legacy block slot before Denim activation instead:
past every scheduled fork and still pre-Denim, so a single legacy block
stays valid regardless of when the test runs.

Amp-Thread-ID: https://ampcode.com/threads/T-09d6a81e-204d-45bb-a227-69764d686e37

Co-authored-by: Amp <amp@ampcode.com>
(cherry picked from commit 407af44)

---------

Co-authored-by: Julian Meyer <julian.meyer@coinbase.com>
Co-authored-by: Amp <amp@ampcode.com>
…ase#6554, base#6555 to v1.5.5) (base#6569)

Backport of base#6554 and base#6555 to releases/v1.5.5 as one change.

The parking pool iterator now calls no_updates() on the reth standard
pool, so every store (standard, sidecar, validity) is snapshotted when
the iterator is created. A transaction that arrives mid-build waits for
the next block instead of joining this one, closing bottom-of-block
sniping by low-fee standard transactions and letting targeted backruns
that arrive with their target compete on fee against resting validity
transactions. Adds an integration-pool test that a late arrival waits
for the next build in all three stores.

(cherry picked from commit b8f6dfa)
(cherry picked from commit c1cbcb4)

Co-authored-by: Toshi <toshi-noreply@coinbase.com>
base-v2.5.2.11-fh3.1 points to the merge commit of the reth PR, whose tree matches
the PR head the lock file held.
@dfuse-bot
dfuse-bot requested a review from maoueh November 13, 2026 22:37
@maoueh
maoueh merged commit 7ac1bde into release/1.x Nov 18, 2026
2 checks passed
@maoueh
maoueh deleted the bump/base-1.5.5 branch November 18, 2026 20:04
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.

7 participants