Skip to content

Releases: ethereum/go-ethereum

Repair Actuals (v1.16.1)

02 Jul 08:13
@fjl fjl
v1.16.1
12b4131
Compare
Choose a tag to compare
Repair Actuals (v1.16.1) Latest
Latest

This is a patch release, fixing some regressions with v1.16.0.

  • abigen v2 now creates TryPack methods that return an error for invalid parameter values. (#31692)
  • geth --vmtrace, enabling the live tracer, did not work due to a bug. (#32107)
  • The blockTimestamp as returned by eth_getLogs is now hex-encoded. (#32129)
  • eth_getLogs and related endpoints now limit the number of queried addresses to 1000. (#31876)
  • eth_getTransactionReceipt should be a bit faster. (#32021)
  • A buggy interaction of the new archive node and snap sync is resolved. (#32104)
  • A rare crash in geth --dev mode related to debug_setHead is resolved. (#31871)
  • Release archives now contain statically-linked binaries. (#32118)

For a full rundown of the changes please consult the Geth 1.16.1 release milestone.

As with all our previous releases, you can find the:

Assets 2
Loading

Terran Rivets (v1.16.0)

26 Jun 18:59
@fjl fjl
v1.16.0
57f8971
Compare
Choose a tag to compare
Terran Rivets (v1.16.0)

We are proud to present the v1.16.0 release, introducing path-based archive node and several other improvements to the state database implementation. There is also the usual amount of bug fixes and small improvements.

Archive Node

This release includes a new, experimental implementation of the archive node (--gcmode=archive). As an archive node, Geth stores all historical states, and can thus retrieve account balances, nonces and storage values at old blocks.

In previous versions of Geth, running as an archive node was only possible using the legacy 'hash-based' state storage scheme (--state.scheme=hash). When syncing mainnet the resulting database would reach sizes larger than 20TB, and it all had to be stored on a high-speed SSD.

The new implementation provides the state archive in combination with 'path-based' storage (--state.scheme=path). Comparing to hash-based storage, there are three basic advantages:

  • Geth can be configured to store historical states on a dedicated disk, and it doesn't have to be an SSD (it will be ~3x faster to access when stored on SSD though).
  • For the full state history of mainnet, the state database will have a size of ~1.9TB (this is without the tx-by-hash index and log index).
  • The amount of historical state kept by the node is configurable as well. By default, Geth will keep historical state for the last 90k blocks, and older state is pruned from disk automatically as the chain advances. You can set the number of states to keep with the --history.state flag.

Notably, there is also one drawback with the new archive node: eth_getProof is not supported for historical blocks. With --state.scheme=path, Geth can only serve state proofs for the latest 128 blocks. We believe this is an acceptable tradeoff. If you need older proofs, you can still a hash-based archive node as before, but we'd also really like to hear about your use case. Please contact us via email at [email protected] so we can learn more, and possibly find a solution that can be implemented within Geth.

How it works

With 'path-based' state storage, Geth keeps exactly one full state in the database. Specifically, the persistent state is 128 blocks in the past. For newer blocks up to the head, forward diffs are kept in memory. In order to support rolling back to blocks older than head-128, Geth also keeps 'state history' in the form of reverse key-value diffs. When the chain is reset to an old block, these diffs are applied to the persistent state, going backwards one diff at a time until the target block is reached.

A reverse state diff contains the previous values of all modified accounts and storage locations for a specific block. There is a reverse diff for each historical block. This makes reverse diffs suitable for storage into the 'freezer', i.e. they do not need to live within Pebble/LevelDB.

The new archive mode is built on the realization that reverse state diffs contain all necessary data to serve historical state queries. For example, in order to get the historical balance of an account X at block B, we can search forward through diffs starting at block B until we find a block where account X is modified. This diff will contain the balance of the account, since it stores the all modified pre-values.

To accelerate the search for a suitable diff, Geth creates a database index storing the block numbers in which an account was modified. This index is small compared to the overall state history, but it is crucial for correct operation of the archive node. The state index is stored in PebbleDB and will be generated automatically while geth is syncing the chain. It takes ~30h to build the archive state index for mainnet, and historical state will only be available when the index is fully built. Geth will report a syncing status through eth_syncing while the indexing happens.

Other state implementation changes

In v1.16.0, the implementation of the 'state snapshot' has been rewritten for improved integration with path-based state storage. As you may recall, Geth stores the state in two different ways:

  • The 'snapshot' contains the flat keys and values of the state, in a layer-based struture.
  • The trie database contains state trie nodes, necessary for the state root computation.

When the implementation of the snapshot and its layer tree was introduced in Geth v1.10.0, it was built to be entirely separate from the trie database. We added path-based state storage in v1.13.0, and that system contains a layer tree as well, with slight differences in semantics and implementation.

As of this release, the layer trees of the state snapshot and trie database have been merged into a single implementation (#30661). In path-based mode, the trie database now holds a reference to the corresponding snapshot layer for each tree state. The new layer tree has also been enhanced with an additional index of layers in which an account was modified (#30971). Note this is similar in spirit to the historical state index. The new index will enable us to increase the number of in-memory state layers in the future.

Full list of changes in this release

Geth

  • The default block gas limit is now 45M. (#32087)
  • Geth can now use era1 files to access history when running with pruned history. (#31604)
  • There is a new geth download-era subcommand for fetching era1 files. (#31823, #31951, #32081)
  • When Geth is interrupted while it is syncing, pending block imports will now be canceled, leading to a quicker shutdown. (#32062)
  • Ready for fork activation messages are now printed based on the current clock time. (#31800)
  • Geth no longer allows setting a different network ID when running a network preset. (#31630)
  • The beacon-chain light client, blsync, has been fixed to work correctly after Pectra. (#31775)
  • geth --dev now shows a listing of pre-funded dev chain accounts and their keys. (#31898)
  • geth --dev now always enables preimage recording. (#31821)

Fusaka

We have started merging some EIP implementations for the upcoming Fusaka fork:

Core

  • There is a new archive-node implementation using path-based storage. (#31156, #31161, #31725, #31781, #32091, #32099)
  • The pathdb layer tree has been rewritten for tight integration with the state snapshot, and has an improved lookup structure that helps it scale to more than 128 layers. (#30661, #32086, #32097)
  • The state prefetcher has been improved to run for all inserted blocks, not just during full sync. (#31557, #32075, #32071)
  • Metrics for state cache utilization and the prefetcher have been added. (#31998)
  • We have also added metrics for gas usage and mgas/s (#31753, #32004)
  • The Pebble database is now synced to disk at explicit safepoints. In the beginning of the v1.15.x release cycle, we added a change that disabled fsync on the database level, then spent a bunch more time throughout the cycle triaging issues related to that. We eventually had to row back and enable fsync again in v1.15.8, causing a drop in performance. The new approach of syncing explicitly gives us the ability to disable fsync for most writes, improving performance. (#31703)
  • Pebble parameters have been tweaked to ensure better compaction behavior. Specifically, some changes were made to smooth out compaction-related CPU and I/O spikes. (#31970, #31988, #31782, #30464)

RPC

  • Logs returned by eth_getLogs and similar requests will now return the blockTimestamp. (#31887)
  • eth_getBlockReceipts will no...
Read more
Loading

Armored Compartments (v1.15.11)

05 May 14:21
@fjl fjl
v1.15.11
36b2371
Compare
Choose a tag to compare
Armored Compartments (v1.15.11)

This is a maintenance release, correcting issues with log and transaction indexing. Upgrading to this release is not required to follow the Pectra fork on mainnet, you can keep using v1.15.9 or v1.15.10, but we recommend you upgrade at your convenience to fix RPC issues.

All changes

  • A stall condition in eth_getTransactionByHash is resolved. (#31752)
  • Two bugs in the new log indexer are resolved. Note: upon upgrading the log index will be rebuilt, expect high CPU load after startup. (#31750, #31734)
  • Log indexing performance has been improved. (#31716)
  • eth_simulateV1 now correctly returns the transaction sender (from). (#31480)
  • A corner-case in eth_estimateGas related to floorDataGas is resolved. (#31735)
  • ethclient's BlockByNumber can now retrieve the pending block. (#31504)
  • geth init will once again exit in error when trying to re-initialize a database with an incompatible config. (#31743)
  • Transaction pool reorgs are slightly faster. (#31715)
  • A transaction pool data race in geth --dev mode is resolved. (#31758)

For a full rundown of the changes please consult the Geth 1.15.11 release milestone

As with all our previous releases, you can find the:

Loading

Omni-Capacitors (v1.15.10)

25 Apr 11:27
@fjl fjl
v1.15.10
2bf8a78
Compare
Choose a tag to compare
Omni-Capacitors (v1.15.10)

This is a bug-fix release that corrects an issue with the new log indexer and configures the beacon chain light client for the Electra fork on mainnet (we forgot about this in v1.15.9).

This release is also suitable for following the Pectra fork on Mainnet.

All changes

  • A deadlock condition in the log indexer is resolved. This only affected deployments of Geth in archive mode, and old databases still using LevelDB. (#31708)
  • The beacon chain light client is now configured for the Pectra fork on mainnet. (#31706)
  • The default block gas limit has been increased to 36M. (#31705)
  • ethclient now allows passing an EIP-7702 authorizationList to calls. (#31198)
  • A new RPC endpoint debug_setMemoryLimit has been added for tweaking Go garbage collector behavior. We do not recommended using this, it's just a facility for debugging Geth. (#31441)

For a full rundown of the changes please consult the Geth 1.15.10 release milestone

As with all our previous releases, you can find the:

Loading

Multicapacitor (v1.15.9)

21 Apr 12:56
@fjl fjl
v1.15.9
74165a8
Compare
Choose a tag to compare
Multicapacitor (v1.15.9)

This release enables the Prague execution-layer fork on mainnet.

Prague

As of this release, the Prague fork is scheduled to occur on mainnet at block timestamp 1746612311 (Wed May 07 10:05:11 2025 UTC). As a reminder, the fork contains the following EIPs:

  • EIP-2537: Precompile for BLS12-381 curve operations
  • EIP-2935: Save historical block hashes in state
  • EIP-6110: Supply validator deposits on chain
  • EIP-7002: Execution layer triggerable exits
  • EIP-7251: Increase the MAX_EFFECTIVE_BALANCE
  • EIP-7549: Move committee index outside Attestation
  • EIP-7623: Increase calldata cost
  • EIP-7685: General purpose execution layer requests
  • EIP-7691: Blob throughput increase
  • EIP-7702: Set EOA account code
  • EIP-7840: Add blob schedule to EL config files

All changes

  • The Prague fork timestamp was added for mainnet. (#31535)
  • Transaction-sending RPCs will now add txs to the 'locals' tracker only when they have any chance of inclusion. This is a bit of a revert from the behavior we added in v1.15.4, where APIs such as eth_sendRawTransaction would always return a txhash, even if the transaction wasn't includable on chain. (#31618)
  • If an EVM system call fails during block execution, the block is considered invalid. (#31639)
  • An corner-case crash in eth_feeHistory related to blob fees is resolved. (#31663)
  • Several correctness bugs in the new log indexer have been fixed. (#31590, #31680, #31671, #31668, #31642)
  • The history pruning implementation was further improved. (#31638, #31636, #31656)
  • Geth will now print periodic logs when a non-activated fork is configured. (#31340)
  • Geth will now occasionally drop peers at random after being fully synced. (#31476)
  • CPU usage of tx propagation has been optimized. (#31657)
  • Peer disconnect metrics are improved. (#31629, #31621)

For a full rundown of the changes please consult the Geth 1.15.9 release milestone

As with all our previous releases, you can find the:

Loading

Combatives Upgrade (v1.15.8)

11 Apr 09:32
@fjl fjl
v1.15.8
f8ff24e
Compare
Choose a tag to compare
Combatives Upgrade (v1.15.8)

This is a bug-fix release with some performance improvements.

Geth

  • geth import now applies database and cache flags correctly. (#31577, #31534)
  • The new log indexer now exports metrics about its operation. (#31511)
  • The beacon chain light client, blsync now has a feature to export checkpoint files. (#31469)

Core library

  • Database writes have been made fully synchronous again. We disabled the use of fsync a while ago to improve performance on slow filesystems, but it has lead to reports of instability. The performance hit from enabling data sync is negligible. (#31519)
  • The transaction pool now takes pending blob transactions into account when limiting pending EIP-7702 authorizations for an account. (#31526)
  • A logic race in EIP-7702 transaction validation is resolved. (#31373)
  • The blob transaction pool performs less disk reads when sending transaction announcements. (#31433)
  • The EVM now has a special fast-path for PUSH2, which is the most common instruction. (#31267)
  • The Trezor hardware wallet implementation now supports 32-bit chain IDs. (#17439)
  • Geth can now stop syncing history at the PoS merge point. This behavior is not enabled yet. (#31414)

RPC

  • When trying to access pruned history, all RPC APIs now return error code 4444. (#31361)

P2P networking

  • UPnP support has been improved and some bugs got fixed. (#30265, #31486, #31566)
  • The discv5 'talk request' API has been changed to pass *enode.Node to handlers. This is a breaking change, but the only known user of this API is the shisui portal network client. (#31075)
  • A flaw in the recently added discv5 challenge resend logic was fixed. (#31543)
  • The eth protocol now properly handles very large skip values when processing GetBlockHeaders messages from peers. This is not a security fix, despite looking like one, it's more about correctness. (#31522)

Build

  • This release is built with Go 1.24.2 (#31538)
  • Note: due to issues with our build environment, we can no longer provide binary builds for macOS. These may be restored at a later date, hopefully soon. For now, you'll have to install from Homebrew.
  • The previous release's git tag, v1.15.7, was published twice, leading to an issue with the Go module cache.

For a full rundown of the changes please consult the Geth 1.15.8 release milestone

As with all our previous releases, you can find the:

Loading

Juggernaut Shield (v1.15.7)

31 Mar 16:31
@fjl fjl
v1.15.7
827d3fc
Compare
Choose a tag to compare
Juggernaut Shield (v1.15.7)

This is a bug fix release. We are putting is out specifically to address a critical issue that could break archive node databases.

  • Fixed an issue for --state.scheme=hash where the log indexer would accidentally delete trie nodes. (#31525)
  • Fixed an issue with tx submission, where the local pool didn't track pending nonces correctly. (#31496)
  • The log indexer will now disable itself when hitting missing receipts in the database. (#31500)
  • Another txpool issue reorg issue in ethclient/simulated.Backend is fixed in this release. (#31228)
  • Memory allocation for trie operations has been reduced significantly. (#30932)
  • eth_createAccessList now supports state overrides like eth_call. (#31497)
  • eth_createAccessList will now exclude 7702 authorities from the result. (#31336)
  • The abigen library now correctly forwards access lists to eth_estimateGas. (#31394)

For a full rundown of the changes please consult the Geth 1.15.7 release milestone

As with all our previous releases, you can find the:

Loading

Mental Focuser (v1.15.6)

25 Mar 11:25
@fjl fjl
v1.15.6
19d2b4c
Compare
Choose a tag to compare
Mental Focuser (v1.15.6)

🚧 Note: we are investigating an issue with this release that affects archive nodes (--gcmode=archive).
If you are running Geth in this mode, please hold off upgrading. 🚧

This is a feature release, with two exciting upgrades:

  • Log filtering in Geth receives a huge performance upgrade with the introduction of our new 'filtermaps' index. Unlike the previous 'bloombits' index, query performance no longer suffers as the density of logs in a block increases. The new index design is also a step towards a future where filtering results can be proven by the server. See the PR and associated design documents for more information.

    In practical terms, the new index is a bit larger than the old one. As before, you can choose the amount of historical blocks to be indexed using the --history.logs command-line flag. We have adjusted the default value of this flag to cover one year of history, and the resulting index has a size of ~10GB for Ethereum mainnnet. Indexing of the entire chain with --history.logs=0 will take up ~61GB.

    Once the index is built, searches will be fast, but note that querying outside of the indexed block range will fall back to a very slow unindexed search. We will continue optimizing log searches in future releases, and welcome your feedback and bug reports in this area. (#31065, #31079, #31080, #31081, #31419, #31429, #31450, #31463, #31455)

  • abigen v2 is finally here. abigen is a tool for creating Go bindings for Solidity contracts. In v1, the generated bindings presented an API for sending transactions, filtering logs, and performing read-only calls as Go methods on the contract object. In the new version, we have updated the interface of the generated code to focus purely on encoding and decoding ABI payloads. Generic helper functions are provided in a library package to enable the same interactions as before, but you can also use your own custom method of signing & sending transactions. Generated bindings are also significantly smaller. (#31379)

Other changes in this release:

RPC

  • A regression in eth_sendRawTransaction - where transactions with too-low nonce would be accepted by the API - has been fixed. (#31473)
  • eth_call/estimateGas RPC methods will now always return error code 3 for reverts. It previously only returned this code when the EVM produced revert data. (#31456)
  • eth_simulateV1 now returns a correct logs bloom value in the simulated block (#31411)
  • eth_simulateV1 supports block overrides for the beacon root and withdrawals (#31304)
  • debug_traceCall: the movePrecompileTo override feature should now work correctly (#31348)
  • debug_traceCall and other related RPC methods now hex-encode the EVM return value. This is a breaking change. (#31216, #31445)
  • ethclient has a new method EstimateGasAtBlock (#27508)

Geth

  • Support for the Hoodi testnet has been added. (#31406)
  • The beacon chain light client has been updated to support the Electra fork. (#31243, #31470)
  • geth import now properly handles Ctrl-C interrupts (#31360)
  • geth --dev mode will now pre-fund all precompile accounts supported by the Pectra upgrade (#31342)
  • geth --dev now respects the --miner.pending.feeRecipient flag (#31316)

Core library

  • The performance of transaction signature validation has improved significantly (#31242, #31258, #31434)
  • The transaction pool has seen some correctness and performance improvements. (#31430, #31307, #31332)
  • We have added a lot of preliminary changes for History Expiry (EIP-4444). There shouldn't be any user-facing consequences of these just yet, just reporting it here since this area was a big focus this cycle. (#31424, #31355, #31362, #31383, #31365, #31117, #31356, #31384, #31393)

For a full rundown of the changes please consult the Geth 1.15.6 release milestone

As with all our previous releases, you can find the:

Loading

Stronghold Package (v1.15.5)

05 Mar 10:56
@fjl fjl
v1.15.5
4263936
Compare
Choose a tag to compare
Stronghold Package (v1.15.5)

Alright 🥲 This is a hotfix release for an issue on the Sepolia testnet. Shortly after the Pectra fork activation, an issue in deposit contract logs parsing was discovered. Sepolia has a custom version of the deposit contract that also implements ERC-20, and thus unexpected log events may be created by transactions to it. This issue affects all Ethereum execution clients.

The Sepolia testnet may take a short while to recover while the update is adopted by nodes. An incident report will be published once the network has fully recovered. ❤️‍🩹

Other changes in this release:

  • In output of debug_traceTransaction, the "memory" and "storage" fields will be omitted if empty. (#31289)
  • This release is built with Go 1.24.1 (#31313)

For a full rundown of the changes please consult the Geth 1.15.5 release milestone

As with all our previous releases, you can find the:

Loading

Commando Package (v1.15.4)

01 Mar 20:38
@fjl fjl
v1.15.4
8ccca24
Compare
Choose a tag to compare
Commando Package (v1.15.4)

This is a bug fix release.

Note: you need to upgrade to v1.15.3 or this release to be compatible with the Pectra fork on the Sepolia network (activates Wed, Mar 5 at 07:29:36 UTC).

  • Fixed a v1.15.0 regression in eth_feeHistory that caused incorrect blobGasRatio return values. (#31246, #31270)
  • A v1.15.0 regression in RPC transaction submission has been fixed: if a transaction did not pass txpool verification (e.g. low fees), an error was returned by RPC, but the transaction would be added to the local pool anyway. This is now fixed and no error will be returned by the API in this case. (#31202)
  • Txpool logic was reworked to avoid an error log flood about EIP-7702 authorities. (#31249)
  • Certain invalid blob transactions no longer cause disconnect issues in the p2p layer. (#31219)
  • ethclient now provides a BlobBaseFee method to request the current blob basefee. (#31290)
  • The PPA package build was fixed after being broken in v1.15.3 by the upgrade to Go 1.24. (#31282, #31283)

For a full rundown of the changes please consult the Geth 1.15.4 release milestone

As with all our previous releases, you can find the:

Loading