This Month in Forest - May 2026

[LesnyRumcajs]

This Month in Forest 🌲

Welcome to the May edition of This Month in Forest! Busy month all around — NV28 finally activated on mainnet, a couple of releases went out, and a healthy chunk of work landed across the mempool, wallet, and Ethereum API. Let's get into it.

Highlights

NV28 on mainnet

Following the April calibration upgrade, NV28 activated on mainnet at epoch 6052800 on May 27.

New releases

Two Forest releases this month:

• v0.33.3 "Dawn" — early-May stability and Ethereum-API fixes.
• v0.33.4 "Stray" — NV28 mainnet support bundled with performance and stability improvements.

Forest adoption

Bridge validators and DEXes now using Forest. There are discussions with Block Explorers, and more ecosystem teams moving to Forest including users of Forest-exclusive trace API calls trace_call and debug_traceTransaction.

Filecoin Security & Antithesis ❤️ Forest

It's worth highlighting that several reports coming from Filecoin Security & Antithesis actually helped resolve some edge cases around Forest. Those included, e.g., duplicate beacon entries or recently Forest getting stuck.

Development

Random improvements and fixes

fix n. & v. trans. [Latin figere "to attack with reproaches, to render immovable."] 1 n. A palliative shot in the system's arm, becoming less effective with each application 2 v. trans. To remove [a bug] by redefining the program specification in order to take advantage of an unexpected feature.
The Computer Contradictionary, Stan Kelly-Bootle

Ethereum RPC: correctness and performance

A focused push to make Forest's Ethereum-compatible JSON-RPC faster and more correct for the providers that rely on it:

• scoping receipt logs by message and capping event filters, fixing block-trace indexing, and parallelising block-trace execution.
• parallel processing of JSON-RPC batches (#7101, #7093), larger connection backlog, disabling default RPC compression, and a faster caching library.
• prefilled RPC caches for newly-validated tipsets and for collect_events, plus an optimised tipset-to-Ethereum-block conversion.
• stricter API parity comparisons to catch response-format drift earlier.

Alongside the targeted fixes, we also started building tooling to turn RPC traffic into structured datasets we can query — so we can spot the slowest and most-used methods empirically rather than guessing. Results? We'll share more detailed ones later, but here's a sneak peek

Wallet usability

• forest-wallet set-default no longer fails when the keystore has no existing default — friendlier first run.
• The Calibnet wallet test flow was rewritten from shell scripts into Rust (and follow-up) for easier maintenance.

Mempool tooling for stuck messages

If you've ever had a wallet wedged behind a stuck pending message, this one's for you. forest-tool now ships nonce-fix and replace mempool helpers (#7084) to recover from nonce gaps and reprice stuck pending messages. The replace-by-fee behaviour is now aligned with Lotus as well.

State-manager and message-pool restructuring

A multi-PR (#6864, #6965, #7006, #7033) reorganisation of two of Forest's largest modules into smaller, focused files. Less spectacular than a new feature, but it makes bugs easier to find and the code easier to compare against Lotus — and it surfaced a number of latent issues now tracked as scoped follow-ups. A message chain ordering fix during republish came out of this work too.

API correctness around state forks

Network-upgrade epochs needed special handling in the API surface. #7061 implements state-fork handling in call_raw so expensive migrations don't get triggered unnecessarily, and Eth RPC requests around fork epochs return correct results.

Chain-store and storage

A persisted tipset-by-height lookup table makes height-based queries cheaper, and the database layer was refactored from a generic to an enum-based dispatch to reduce maintenance overhead.

Stability and diagnostics

A node crash that could occur under high event-traffic load is fixed, keeping nodes online during peak traffic. Errors that previously vanished silently in the FVM consensus trait and the randomness layer are now inspected and logged. Internal tipset-lookup paths now distinguish a real failure from a "not found yet", reducing hidden errors.

Operator-facing improvements

• RPC hardware guidance for low-traffic nodes.
• A native-CPU build option for operators willing to trade portability for a bit of extra performance.
• A gperftools profiling helper for operators and contributors diagnosing performance issues.
• CI link times sped up with mold.

See the CHANGELOG for the full list of client-facing changes; the commit history has the rest. 🤓

Upcoming work

Our immediate focus:

• continued Eth-RPC performance work along with performance figures,
• more mempool and state-manager improvements, plus broader automated test coverage motivated by the issues that surfaced this month,
• complementary snapshots with receipts, events and tipset index.

And then, of course, we will continue with our regular maintenance and improvements.

Outside of Forest

This is a section where we share our activities outside of the core Forest client development (but still related to the Filecoin ecosystem). Disclaimer: this is for reporting purposes only.

lotus

A cross-client fix discovered while restructuring Forest:

• fix(api): trace transaction api to return correct error filecoin-project/lotus#13614

rust-fil-proofs

Aligning toolchain management with the rest of the Filecoin Rust ecosystem:

• chore: move to rust-toolchain.toml filecoin-project/rust-fil-proofs#1799

serde_ipld_dagcbor

CBOR decoder upgrade we needed in Forest, contributed upstream first:

• chore!: bump cbor4ii to 1.2.2 ipld/serde_ipld_dagcbor#52

ref-fvm

Routine maintenance:

• chore: fix deny advisories filecoin-project/ref-fvm#2300
• chore: bump wasmtime to 36.0.8 filecoin-project/ref-fvm#2301

---

Individual reports

Disclaimer: this is for reporting purposes only.

@akaladarshi

Highlights

• Forest's two most-changed code areas got cheaper to maintain. Restructured the message-pool and state-manager into smaller, focused files so future patches ship with smaller risk surface, faster review, and tighter consistency with the Lotus (#6864, #6965, #7006, #7033).

• Forest reached real-time Ethereum subscription parity with Lotus. Added subscription support for pending transactions, smart-contract events, and new blocks, closing a parity gap with Lotus for wallets and EVM tooling that depend on these streams (#6941).

• Forest nodes stay online under high traffic. Fixed a node crash that could occur under peak load (#7020).

• Five silent bugs in Forest's most-used code paths are now tracked. While working closely with the message-pool and Ethereum-RPC code, filed each bug as a scoped issue so the team can triage and fix them (#7074, #7075, #7076, #7091, #7096).

• Forest's Ethereum-API performance work now has concrete targets. Investigated why Forest's Ethereum read APIs (such as eth_call) run slower than Lotus's in production, identifying where the time is lost. Also scoped automated test designs for the message-pool and state-manager so future regressions get caught at PR time, not in production.

Wider ecosystem

Both Filecoin clients now return consistent Ethereum error codes. Contributed an error-code fix to Lotus's trace API (lotus#13614), shipped the matching fix on the Forest side (#7098), and filed a separate cross-repo bug against Lotus during the same investigation (lotus#13613). Wallets and developers across Filecoin can now diagnose failed Ethereum transactions reliably on both clients.

Full list of authored pull requests and filed issues, May 2026.

@hanabi1224

Summary

This month's work focused on making Forest faster, more reliable, and easier to maintain. Key themes included eliminating cold-start overhead through persistent storage improvements, reducing RPC latency with targeted caching and parallelism, improving consensus-safety observability, adding a state-diff tool for debugging replay mismatches, and a database architecture change that makes storage code statically safe to extend.

Highlights

Faster, more efficient API responses

Pre-warming RPC caches for newly validated tipsets and side products(message receipts and events, etc.), parallelizing batch request processing, and an optimized tipset-to-Ethereum-block conversion — reducing latency on the hot paths RPC providers hit most.
#7093 #7077 #7068 #7062

Improved diagnostics and error handling

Better logging and clearer error reporting make it easier to investigate and resolve issues.
#7083 #7082

Tooling, build, and release preparation

A faster build process, new state-diff tool, and release preparation work.
#7078 #7081 #7021

Refactors and storage cleanup

Internal improvements to database and storage code, reducing complexity and risk of bugs.
#7051 #7038 #7023

Tests and stability fixes

Expanded test coverage and targeted fixes for edge cases, improving reliability.
#7057 #7056 #7019 #7002

Maintenance and dependency updates

Routine upkeep, network resets, and dependency updates.
#7003 #7072 #6999 #6998

Design and coordination

Drafted a design proposal for a global index spanning multiple ForestCAR files to guide future storage improvements.

Full list of authored pull requests: hanabi1224 PRs, May 2026.

@LesnyRumcajs

Highlights

• Forest stayed on the network through NV28. Mainnet activated on May 27; Forest shipped support ahead of time (#7037) and rolled it out in v0.33.4 "Stray" alongside performance and stability work.
• RPC providers got a faster, more correct Ethereum API. Receipt logs and event filters now have bounded responses (#7025), block traces are indexed correctly and run in parallel (#7001, #7107), and batch-request metrics no longer mislead operators (#7113).
• Latency and tail timeouts dropped under production traffic. JSON-RPC batches now process in parallel (#7101), the connection backlog scales with max-connections (#7034), default RPC compression is off where it didn't pay (#7066), and the cache layer is faster (#7073).
• Eth-RPC performance work is now data-driven. Started building tooling that turns RPC traffic into structured datasets, so the team can prioritize the slowest and most-used methods empirically — rather than by guesswork.
• Operators have better tools for their own performance work. Hardware guidance for low-traffic nodes, a native-CPU build option, and a gperftools profiling helper lower the barrier to diagnosing performance issues locally.

Coordination

Coordination work this month (no code, but it kept the rest of the work shippable):

• The Forest team shipped a coordinated release across NV28. Technical direction, prioritization, code review, and unblocking kept the team aligned with the month's release and performance goals.
• ChainSafe production environments kept up with the codebase. Coordinated with internal teams on the infrastructure that runs and tests Forest.
• RPC providers' bottlenecks shaped the month's Eth-RPC fixes. Direct feedback loop with providers running Forest fed straight into the issues prioritized this month.
• Filecoin Security review work landed. Participated in a security review and the resulting follow-up.

Wider ecosystem contributions

• rust-fil-proofs — Toolchain management is now consistent across the Filecoin Rust ecosystem (migrated to rust-toolchain.toml).
• serde_ipld_dagcbor — A CBOR decoder upgrade Forest needed landed upstream and is now in Forest.

Full list of authored pull requests: LesnyRumcajs PRs, May 2026.

@sudo-shashank

Highlights

• New mpool recovery tools. Added the nonce-fix and replace message-pool CLI tools to help recover from nonce gaps and reprice stuck pending messages, making it easier to unblock affected wallet activity (#7085, #7086, #7084).

• Replace-by-fee value now matches Lotus. Lowered the minimum premium bump for a replacement message from 1.25x to 1.10x, so users get consistent repricing behavior. Ports filecoin-project/lotus#10416 (#7084).

• set-default now works on a fresh keystore. forest-wallet set-default previously could fail when the keystore had no default entry; it now works on a fresh keystore and persists across restarts, removing a sharp edge from the new-user experience (#7018, #7087).

• Wallet testing is now maintainable in the long term. Migrated Calibnet wallet checks from brittle shell scripts into Rust, making the test flow easier to maintain, extend, and review (#6850, #6983, #7095).

• Stricter checks keep Forest's API in lockstep with Lotus. Enabled strict API comparison against Lotus, fixed several response-format mismatches, and refreshed test snapshots — closing compatibility gaps and guarding against regressions (#5635, #6970).

• API calls stay correct across network upgrades. Implemented state fork handling and corrected behavior around migration epochs, keeping results correct while avoiding unnecessary migrations by explicitly handling expensive forks (#6446, #5965, #7061).

• Messages get republished in the right order. Fixed an incorrect swap index in the message-republish selection logic, so pending messages are re-broadcast in the correct priority order instead of being silently reordered (#7075, #7090).

• Genuine failures are no longer silently swallowed. Refactored internal tipset lookup paths to distinguish a real failure from a simple "not found yet," reducing masked errors and making node behavior easier to reason about and debug (#6755, #7005).

Security

• Public-facing services got a closer look. Investigated security advisories in forest-explorer covering transaction-signing safeguards, faucet quota accounting, and per-wallet rate limiting — helping keep user-facing infrastructure safer and more resilient.

Full list of authored pull requests: sudo-shashank PRs, May 2026.