erigon.eth

Support for @binance's BNB Smart Chain is coming to Erigon. Chapel testnet first, then mainnet, enabled by a flag in regular releases. Compact storage and fast OtterSync bootstrapping for a high-throughput chain with a large history. erigon.tech/blog/bnb-smart-c…

Ask an AI for an Erigon command line and watch where the flags come from: GitHub issues. Which are, by definition, the configurations that failed. A browsing model mines the bug tracker for other people's pasted commands, some from the Erigon 2 era. A local model with no web access just guesses. Either way, it is learning the client from its error cases. So we publish the entire documentation as one file. llms-full.txt is every doc page concatenated: about 500 KB, roughly 68,000 words, a single fetch. Point a model at it, online or off, and it answers from the real reference. It is not an export you take on trust. The file is generated from the docs source and verified in CI, so the build fails if it drifts from the published site. The bytes you read and the bytes the model reads are identical. Small-context models can load llms.txt instead, a 12 KB index, and pull pages on demand. We did not invent this, it follows the llmstxt.org convention. But most docs still make a model work for the answer. This one hands over the whole thing. docs.erigon.tech/llms-full.t… #Erigon #Ethereum

A mainnet archive node is 2.25 TB. You do not need a 2.25 TB NVMe to run one. Erigon splits the datadir across disks, so only the hot data needs a fast device: ? chaindata and snapshots/domain on the NVMe ? history, idx and accessor on a cheap HDD The fast disk stays small while the bulk sits on slow, cheap storage. A full archive on consumer hardware. docs.erigon.tech/fundamental… #Erigon #Ethereum

🦦 Erigon v3.8.1 "Watery Whiskers" is out! Postpones Glamsterdam on Hoodi, fixes block validation races, improves validator reliability & block production. ⚠️ Hoodi 3.8.0 nodes: upgrade before Oct 26, 17:42 UTC. (longer): github.com/erigontech/erigon…

New Zilkworm R&D: MFBD, a flat witness format for zkEVMs. The guest reads the witness in place: structs cast from bytes, perfect-hash maps built on the host, every byte checked against the parent state root. No decoding, no allocation. erigon.tech/blog/introducing…

A zkEVM guest has no disk or network, and every memory read costs proving time. New Zilkworm R&D: use the witness bytes directly as the state. Flat layouts, perfect-hash maps built on the host, and MPT plus pre-state checks in one pass. erigon.tech/blog/re-thinking…

🚀 Erigon v3.7.1 “Velvet Vibrissae” is here! It's mandatory for internal CL (Caplin) users ahead of Glamsterdam on Sepolia on 6 October 2026, 13:53:36 UTC and is recommended for everyone else. No re-sync is required from v3.7.0, making this a quick and valuable upgrade. Why you should upgrade: 🛡️ Critical Security Patch Fixes a vulnerability where unvalidated sync-committee messages could reach gossip. Slot window validation and signature checks are now enforced to keep your node secure. ⚡ Performance & Stability - Fixes OOM failures: Resolves out-of-memory crashes under heavy code-access workloads by properly releasing block-builder state. - Better proposal success: External consensus clients now get up to 6 seconds (one slot) to build payloads, reducing missed proposals during busy periods. - Blob fixes: Fixes the "missing blob bundle" issue when validators publish pre-Gloas blocks to multiple beacon nodes. 🔮 Prep for Sepolia Upgrades This release aligns consensus and APIs with `consensus-specs` v1.7.0-beta.2 and prepares Caplin for the upcoming "Glamsterdam" upgrade on Sepolia. ⚠️ Heads up for Operators: - Graffiti: Custom graffiti is now prefixed with client identification by default (leaving 19 bytes for custom text). Use `--beacon.api.preserve-graffiti` to keep full 32 bytes. - Caplin Ports: If using custom discovery ports, ensure your QUIC port configuration doesn't conflict with discovery UDP binds before upgrading. ⏰ Deadline: If you are on Sepolia, please upgrade before 6 October 2026, 13:53:36 UTC. github.com/erigontech/erigon… #Erigon #Ethereum

We don't test Caplin, Erigon's built-in consensus layer, only on a bench. Since August we've run our own validators on it in Lido's Community Staking Module on Mainnet, with the same bond at risk as any other operator. That pays off. Our first proposal was orphaned: built correctly, published 27 ms before the 4 s attestation deadline. Tuning didn't help. Caplin asked the execution layer for a payload only at production time, leaving one second to deliver, sign, publish and gossip. Starting earlier would have failed too: the builder's budget ran from request arrival, so an early payload froze before the slot began. → Two-step fix: build until the payload's own slot, then prime it ahead of time → Sepolia: 2.82 s of margin instead of 27 ms → Shipped in v3.6.0 for every Erigon operator → No Mainnet misses in 20+ days since How we traced it with Claude: erigon.tech/blog/troubleshoo… #Erigon #Ethereum #Lido

Running an Ethereum node used to mean running two of them. An execution client and a consensus client. Two processes, two datadirs, two config files, two upgrade schedules, and a JWT secret between them that fails in a way nobody enjoys debugging. Erigon ships Caplin, a consensus layer embedded in the same binary. It is on by default, so the plain command is the whole setup, and there is no separate disk store for it either. It is not a cut-down version: → --caplin.states-archive for historical beacon states → --caplin.blocks-archive for full block history → --caplin.blobs-archive for historical blobs For anyone running an op-node, --caplin.blobs-immediate-backfill populates the last 18 days of blobs up front instead of waiting for them to trickle in. The design reason is worth more than the feature. The two layers were split across two processes for historical reasons, not because the data wants to live apart. Putting them in one binary removes a category of operational failure rather than optimising it. docs.erigon.tech/fundamental… #Erigon #Ethereum