Erigon 3.7 turns on parallel state roots and keeps receipts by default
Serguey Asael Shinder2 min readErigon 3.7.0, released on 25 September, is the Ethereum client's version of the same deadline geth announced two days earlier: the Glamsterdam upgrade - Amsterdam on the execution layer and Gloas in Erigon's built-in consensus client, Caplin - is scheduled on the Sepolia testnet, and Sepolia nodes must upgrade before 6 October 2026 at 13:53:36 UTC.
The release is headlined by performance changes. Parallel commitment is now on by default: the state root is computed on multiple cores, with work shared across the account and storage tries, and can be switched off with --experimental.parallel-commitment=false. RPC calls including eth_getLogs, eth_feeHistory and eth_getProof are faster, with lower CPU and memory costs for large responses. And new data directories now persist receipts by default, so receipt and log queries within the retention window no longer require re-executing blocks - at the cost of more disk space.
Several changes can break existing setups. The receipt default applies only to fresh data directories, and changing it on an existing one requires a new one. Release binaries and Docker images now require x86-64-v2 CPUs, with SSE4.2 and POPCNT. Polygon chain names and flags are no longer accepted. A null in a required JSON-RPC argument now returns error -32602 instead of silently becoming zero, matching geth. erigon_getLogs renames timestamp to blockTimestamp. Nodes that run Caplin with checkpoint sync disabled must re-sync it from genesis unless a finalized state is saved locally. Delegated senders under EIP-7702 are limited to one in-flight transaction in the pool.
What it means
For operators, the Sepolia date is the fixed point, and it now appears in the release notes of at least two execution clients. Client diversity only protects a network if every client ships the fork on time, and testnets are where that is rehearsed.
The receipts change is the one with a bill attached. Faster log queries without re-execution are what indexers and RPC providers want, but archive operators are told to budget for the extra history. Because the setting cannot be flipped on an existing directory, it is a decision to make when a node is provisioned, not later.