This Month in Forest - June 2026

[LesnyRumcajs]

This Month in Forest 🌲

Welcome to the June edition of This Month in Forest! We've been mostly busy with performance and stability work, and we have fantastic charts to show for it. Let's dive in.

Highlights

New releases

A couple of releases in the past month, make sure to upgrade to latest:

• v0.33.7 "Shimmergloom"
• v0.33.6 "Ebb"

(v0.33.5 "Drift" was yanked — don't use it.)

RPC Performance improvements

In the past few months we focused heavily on making Forest's Ethereum-compatible JSON-RPC faster up to its limits (and beyond). The results are in. The following charts show the latency of RPC calls from Chain.Love, running Forest in production alongside Lotus.

Forest delivers faster average response times and a substantial reduction in worst-case latency (P95 and P99). The slow outliers that most affect reliability are now far less common, giving users a more consistent and predictable experience. Median response P50 is slightly better.

And resource usage didn't go up to pay for it:

Want to see Forest in action? Check out the RPC performance Grafana dashboard.

Is our job over here? Not at all. There are still some methods that need our love, and we are working on them.

Snapshot export comparison

As part of our performance work, we compared snapshot exports between Forest and Lotus. On a regular machine, Forest finishes the export substantially faster.

ProPGF Round 3

Do you use Forest? Endorse us here! We've applied for the ProPGF Round 3 grant, fingers crossed!

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

The bulk of June's engineering effort was on the Ethereum-compatible JSON-RPC. On the performance side: trace_block now caches results; JWT authentication is paid once per connection rather than per request; hot chain-state reads use lock-free ArcSwap (more) to cut p50 contention; and heavy FVM execution is dispatched to a dedicated blocking thread pool, pulling down p95 and p99.

On the correctness side: eth_call and eth_estimateGas are no longer blocked on the epoch after an expensive migration — matching Lotus, no more unnecessary recomputation; and Filecoin.StateCirculatingSupply now accounts for the datacap actor and stops erroring on mainnet.

New RPC methods and parity: Filecoin.MpoolGetConfig is now implemented (with a Percent type matching Lotus), and Filecoin.StateGetNetworkParams includes the NV29 placeholder.

State-manager and message-pool restructuring

Continuing from May's restructuring, three concrete mempool fixes landed: a confirmed transaction no longer holds up the sender's next one; the next-block selection step only simulates the head change instead of removing real pending transactions; and Forest now accepts the same per-sender pending-transaction limit as Lotus.

Chain-store and storage

EC finality is now integrated into the chain follower, the follower cleans up dangling forks, the validated block cache is bounded, and leftover temporary chain data is cleared on startup.

Stability and diagnostics

The big one: a critical memory leak in the quick_cache crate Forest depends on was tracked down and fixed — the fix contributed back upstream. Other wins: internal P2P-layer maps shrink properly on entry removal, cache size metrics are now accurate, and a jemalloc-profiling feature makes heap diagnostics easier in production. New size metrics for the chain follower, mempool, and zstd frame cache give operators much better visibility into node memory usage. And forest-wallet --encrypt false finally does what it says, even when an encrypted keystore already exists.

Operator-facing improvements

The forest-explorer faucet now offers an RPC provider selector — Glif, Ankr, or Filfox — so users can switch if one is slow or unavailable.

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 — a few methods still need our love.
• More mempool and state-manager improvements.
• Broader automated test coverage motivated by the issues that surfaced this month.

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.

quick_cache

A critical memory leak Forest tracked down landed upstream, now benefiting every project using the crate.

---

Individual reports

@akaladarshi

Highlights

• A confirmed transaction no longer holds up the sender's next one. Fixed how Forest tracks waiting transactions so that once one is included in a block, the sender's next transaction can move forward instead of getting stuck (#7180).
• Working out the next block's transactions no longer disturbs the ones still waiting. Fixed the selection step so it only simulates the result instead of removing real waiting transactions from the queue (#7182).
• Forest now accepts the same number of waiting transactions from unverified senders as Lotus. Raised the per-sender limit to match Lotus (#7127).
• A possible gap between Forest/Lotus and the official Ethereum standard is now flagged for checking. Filed a scoped issue asking whether one log request returns results the way the Ethereum standard expects. Forest currently behaves like Lotus here, but both may differ from the standard (#7193).
• Forest's work to make its responses faster is now backed by real usage data. Built charts from real request traffic (analytics#4).

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

@hanabi1224

Highlights

• Fixed a critical memory leak caused by a bug in the popular quick_cache caching library (#7143), with the fix contributed back upstream.
• Replaced shared locks on frequently accessed chain state with lock-free reads using ArcSwap (#7191, #7213), improving p50 RPC latency by reducing contention across concurrent requests.
• Ensured heavy VM execution calls (FVM) are dispatched to a dedicated blocking thread pool (#7184), improving p95 and p99 RPC latency by preventing them from stalling other async work.

Bug fixes

• Made the validated block cache bounded (#7135) and shrunk internal P2P layer maps on entry removal (#7136), preventing unbounded memory growth in long-running nodes.
• Corrected inaccurate cache size metrics (#7129), which were causing misleading readings in monitoring dashboards.
• Fixed RPC cache prefilling for the Ethereum pending block tag (#7165), improving correctness for Ethereum-compatible tooling.
• Improved chain follower robustness: integrating EC finality (#7171) and cleaning up dangling forks (#7179) to make the node's view of the chain more reliable.
• Fixed Filecoin.StateCirculatingSupply returning an error on mainnet by correctly accounting for the datacap actor (#7217).
• Cleared leftover temporary chain data on startup (#7158), avoiding stale state from previous runs.

Performance improvements

• Reduced unnecessary data copying in the RPC server startup path (#7181) and reduced locking overhead in the snapshot export status handler (#7173).
• Tuned cache sizes for better RPC hit rates (#7152, #7167).

Tooling & observability

• Added size metrics for chain follower internal hash maps (#7131), message pool internal maps (#7133), and the zstd frame cache (#7134), giving operators better visibility into node memory usage.
• Added a jemalloc-profiling feature and a convenience task for heap profiling (#7146, #7153), making it easier to diagnose memory issues in production.
• Improved server-side logging for snapshot exports (#7160).
• Kept Go and npm dependencies up to date (#7149, #7150).

Full list of authored pull requests, June 2026.

@LesnyRumcajs

Highlights

• Forest's JSON-RPC got noticeably faster this month. Led the team-wide effort: most of June's engineering went into the Ethereum-style API for the providers that depend on it. Spent a significant amount of time analysing production RPC traffic to pinpoint Forest's actual bottlenecks, then decomposed the work so the team could attack them in parallel. The month's improvements shipped to operators in Forest v0.33.7 "Shimmergloom" and previous releases. Forest went from usually slower to consistently faster. See the charts above.

• Direct contributions on the performance side included trace_block caching, JWT authentication once per connection, an opt-out for RPC metrics with reduced lock contention, and an in-flight RPC metric for live diagnostics.

• Built out the performance measurement platform, in collaboration with ChainLove. Continued work on the RPC traffic analytics tooling started last month, so Forest's performance decisions are grounded in real production call patterns rather than guesses. Worked directly with ChainLove on the measurement approach and validation of the results — the partnership has been central to deciding what to optimise.

• Analysed security reports and cross-implementation discrepancies. Alongside the performance push, spent meaningful time working through security reports and tracking down behavioural differences between Forest and Lotus — converting findings into actionable work for the team.

Coordination

• Forest team. Set the month's performance priorities, decomposed them into trackable work, and kept the team unblocked through reviews and direction so the improvements landed quickly and in the right order.
• ProPGF application. Prepared Forest's application for Protocol Public Goods Funding (a Filecoin ecosystem grant program) to keep the project funded sustainably.
• RPC providers. Continued the direct feedback loop with providers running Forest, pointing the performance work at the methods that matter most to them.

Full list of authored pull requests, June 2026.

@sudo-shashank

Highlights

• Filecoin.MpoolGetConfig is now supported. Added the Filecoin.MpoolGetConfig RPC method, letting clients read the node's current message-pool settings, with a parity test confirming the response matches Lotus. As part of this, the replace-by-fee ratio is now represented using a Lotus-compatible Percent type (125 = 1.25×), so the value is reported in the same format Lotus uses (#7141).
• Network parameters stay aligned with Lotus. Filecoin.StateGetNetworkParams now includes an UpgradeXxHeight placeholder for the upcoming NV29 upgrade, matching Lotus (#7194, ports lotus#13640, #7192).
• API calls stay correct across network upgrades. Fixed how Forest treats expensive network-upgrade transitions: eth_call and eth_estimateGas are no longer blocked on the epoch immediately after a migration, only on the migration epoch itself — matching Lotus and avoiding unnecessary state recomputation (#7128, ports lotus#13644, #7192).
• RPC networking aligned with Lotus. Forest now enables TCP_NODELAY on incoming RPC connections, matching Go-based nodes like Lotus — so small request/response payloads are sent immediately instead of being buffered (#7187).
• Wallet encryption choice now works as intended. Fixed forest-wallet so --encrypt false correctly uses the unencrypted keystore even when an encrypted keystore already exists (#7211).
• RPC snapshot tests updated. Removed the FOREST_ETH_RPC_COMPUTE_STATE_ON_INDEX_MISS test workaround, expanded EthGetBlockReceipts coverage to Safe and Finalized block tags, and refreshed the affected ETH and Filecoin RPC snapshots (#7154).
• RPC performance charts improved. The analytics toolkit now includes side-by-side Lotus/Forest latency charts, each labelled with its name and version, plus a --reweight option for fairer comparisons when two captures saw different RPC traffic mixes (analytics#1, #3, #8).
• Faucet pages now offer RPC provider choice. Reported inconsistent duplicated CORS headers from the mainnet RPC endpoint (api.node.glif.io), which the Glif maintainer fixed. As a follow-up, added an RPC provider selector across all faucet pages — Glif, Ankr, and Filfox — so users can switch providers if one is slow or unavailable (forest-explorer#460).
• Dependency upkeep. Resolved a cargo deny advisories failure in Forest (#7225) and kept forest-explorer's dependencies patched via triaged Dependabot patch-version and worker-group updates (forest-explorer#456, forest-explorer#457, forest-explorer#458).

Full list of authored pull requests, June 2026.