> For the complete documentation index, see [llms.txt](https://docs.adi.foundation/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://docs.adi.foundation/references/changelog.md).

# Changelog

Release history for the ADI Server - the node behind ADI's sequencer and external nodes.

### Deployed versions

| Component      | Mainnet version | Testnet version |
| -------------- | --------------- | --------------- |
| Protocol       | `v0.30.2`       | `v0.31.2`       |
| Sequencer      | `v0.21.1-b4`    | `v0.24.2-b7`    |
| External nodes | `v0.24.2-b7`    | `v0.24.2-b7`    |

### Release notes

## [v0.24.2-b7](https://github.com/ADI-Foundation-Labs/ADI-Stack-EN-Setup-script/blob/main/upgrades/v0.20.12_to_v0.24.2.md)

*2026-10-21 - latest*

External node version on testnet and mainnet, superseding `v0.21.1-b4` and `v0.20.12-b1`, the previous default of the EN setup script. The sequencer stays on `v0.21.1-b4` until the main-node upgrade. This entry lists the ADI and upstream changes across the `v0.21` -> `v0.24` line that affect external node operators.

**Breaking changes**

* **upgrade order** - upgrade external nodes **before** the main node. A `v0.24.2` main node writes blocks with `execution_version: 0`, and older nodes panic on these blocks with `Unsupported ZKsync OS execution version`. A `v0.24.2` external node follows the current `v0.21` main node, so you can upgrade early.
  * There is no rollback. After a node stores blocks from a `v0.24.2` main node, it cannot go back to an older version.
  * See the [v0.20.12 -> v0.24.2 migration guide](https://github.com/ADI-Foundation-Labs/ADI-Stack-EN-Setup-script/blob/main/upgrades/v0.20.12_to_v0.24.2.md) for the steps.
* **l1** - a new interop-roots watcher also runs on external nodes. On every start, it reads `eth_getLogs` from L1 block 0 up to the current block, one call per 1,000 blocks, with no pause between calls.
  * The L1 RPC must answer `eth_getLogs` for every 1,000-block range from block 0, so it cannot prune old history. To use smaller ranges, set `l1_watcher_max_blocks_to_process`. Smaller ranges mean more calls.
  * The node retries a rate-limited call 5 times, about 1 second apart. If a call still fails, or the RPC rejects the range or the history, the node stops and the scan starts again on the next start. Many free public endpoints fail these checks.
* **execution** - `execution_version` is deprecated. `zks_getBlockMetadataByNumber` no longer returns the `execution_version` field, and the metric `execution_last_execution_version` is replaced by `execution_last_protocol_version`. The new metric is a gauge with a `version` label: the current version has the value `1`, and the previous version has `0`. Update your dashboard queries.
* **state** - the `Compacted` state backend and the `general_state_backend` key are removed. The node always uses `FullDiffs` and ignores the old key. A node that ran `Compacted` rebuilds its state by a replay of all blocks on upgrade, which is a full resync.

**Features**

* **evm** - support EIP-7702 set-code transactions. They turn on with the next chain upgrade. Restart external nodes after that chain upgrade lands.

**Fixes**

* **l1-watcher** - fix a panic when a node replays the `v0.30.1` -> `v0.30.2` patch upgrade block after governance stage 1 of the `v31` upgrade: `expected patch protocol upgrade 0.30.1->0.30.2 but found minor protocol upgrade 0.31.2`. Stage 1 has not run on testnet or mainnet yet. After it runs, an older node that syncs from zero or resyncs stops with this panic.
* **l1-watcher** - an L1 RPC on reth returns an empty revert without a `data` field. The upgrade-tx watcher now treats this as a missing method and uses the legacy upgrade-cut scan. Before, it crashed on the pre-`v31` Chain Type Manager that both networks use.
* **l1-watcher** - the commit-tx lookup now backs off exponentially from 200 ms to 10 s, about 2.5 minutes in total, and accepts a pending copy of the commit tx. Before, a slow load-balanced L1 RPC could panic the watcher and restart the node.
* **network** - start without waiting for the boot-node DNS name to resolve.
* **network** - key peer session state by connection, not by peer id.
* **mempool** - do not wait for the initial replay upgrade at startup.
* **pipeline** - make graceful shutdown race-free.

**Images**

* `harbor.sde.adifoundation.ai/adi-public/chain/server:v0.24.2-b7`

## [v0.21.1-b4](https://github.com/ADI-Foundation-Labs/ADI-Stack-Server/releases/tag/v0.21.1-b4)

*2026-09-07*

Sequencer and external node version on testnet and mainnet, superseding `v0.13.0-b4`. This covers everything ADI changed across the `v0.13` -> `v0.21` line, plus the one upstream change that every node operator has to act on: block replay moved off HTTP onto devp2p.

**Breaking changes**

* **network** - block replay no longer runs over HTTP. External nodes now sync over [devp2p](https://github.com/ethereum/devp2p) on port `3060` (TCP for RLPx, UDP for peer discovery) instead of pulling replay records from the sequencer's HTTP replay server on port `3053`. An external node left on the HTTP-replay build stops syncing and cannot peer with the network.
  * Removed config: `sequencer_block_replay_download_address`, `sequencer_block_replay_server_enabled`, `sequencer_block_replay_server_address`.
  * New required config on external nodes: `general_node_role: external`.
  * The cluster needs at least one node reachable at a stable address for ENs to peer against, and networking must be enabled on main and external nodes alike.
  * See the [v0.13.0 -> v0.20.12 migration guide](https://github.com/ADI-Foundation-Labs/ADI-Stack-EN-Setup-script/blob/main/upgrades/v0.13.0_to_v0.20.12.md) for the full rollout.

**Features**

* **network** - resolve hostnames in `enode` URLs, so boot nodes can be configured by DNS name instead of a hardcoded IP (#16).
* **network** - raise the P2P active-connection cap from 25 to 100, apply it to inbound connections, and disable peer rotation, so a growing external-node fleet is not churned off the sequencer.
* **l1** - retry L1 requests in the L1 sender and in the commit and execute watchers, instead of failing the operation on the first transient RPC error (#17).
* **rocksdb** - cap open files at 4096 and move SST index and bloom filters into a sized block cache (2 GiB repository, 2 GiB state, 1 GiB preimages) on read-heavy databases. The previous defaults pinned every SST file's index and filter in RAM permanently, so memory grew without bound as the database grew.
* **memory** - use jemalloc as the global allocator. On mainnet, app plus allocator overhead measured \~22 GiB on the glibc build versus \~3 GiB on the jemalloc build under the same database and workload, because jemalloc returns freed memory to the OS.
* **prover** - support circuit-soundness proving versions V8 (protocol v30.2) and V9 (protocol v31.1).

**Fixes**

* **network** - an external node whose replay connection goes quiet now disconnects and re-requests instead of stalling indefinitely. The tolerated silence is configurable via `en_replay_inactivity_timeout` (default 300s); it must comfortably exceed the chain's block cadence, so raise it on chains that can legitimately idle for longer (#1513).
* **sequencer** - apply runtime fee overrides before the mempool reads the base fee, so an override takes effect on the next block instead of one block late.
* **rpc(fees)** - propagate a config-override base fee to every fee-exposing RPC endpoint, derive `native_price` from the base fee when it is not set explicitly, and default `pubdata_price` to `0` unless overridden, so the base fee covers all additional cost.
* **rpc(estimation)** - `eth_estimateGas` caps the gas limit by what the sender can afford, but divided the balance by the block base fee while validation charges the effective gas price. Any request carrying a priority fee therefore produced an unaffordable cap and failed with `LackOfFundForMaxFee` for every account below `fee_cap * block_gas_limit`. The allowance is now divided by the request's fee cap, which is affordable by construction.
* **merkle\_tree** - handle empty batch proofs instead of failing on them (#1535).
* **rpc** - log config-override messages at debug instead of info.

**Images**

* `harbor.sde.adifoundation.ai/adi-public/chain/server:v0.21.1-b4`

## [v0.13.0-b4](https://github.com/ADI-Foundation-Labs/ADI-Stack-Server/releases/tag/v0.13.0-b4)

*2026-06-30*

L1 watcher hardening.

**Fixes**

* **l1-watcher** - leave 2 confirmation blocks unprocessed, matching the upgrade watcher's behavior, to avoid `block range extends beyond current head block` errors.
* **tests** - add the missing private API server config and bind it to a dynamically allocated port.

**Images**

* `harbor.sde.adifoundation.ai/adi-public/chain/external-node@v0.13.0-b4`
* `harbor.sde.adifoundation.ai/adi-public/chain/server@v0.13.0-b4`

## [v0.13.0-b3](https://github.com/ADI-Foundation-Labs/ADI-Stack-Server/releases/tag/v0.13.0-b3)

*2026-06-11*

The `zks` finality RPC and ADI-native CI/CD.

**Features**

* **rpc(finality)** - add `zks_getBlockFinality`, `zks_getTransactionFinality`, `zks_getBatchFinality`, and `zks_getFinalityStatus`, reporting how far a block, transaction, or batch has moved through L1 finality (`pending` -> `committed` -> `executed`), plus a single-call snapshot of every finality frontier. See the [JSON-RPC API](https://github.com/ADI-Foundation-Labs/ADI-Network-Documentation/blob/main/references/json-rpc-api.md#finality).

**CI/CD**

* Switch to ADI-native CI/CD workflows (multi-arch container image build and push to GHCR).

## [v0.13.0-b2](https://github.com/ADI-Foundation-Labs/ADI-Stack-Server/releases/tag/v0.13.0-b2)

*2026-06-03*

L1 block-finality hardening.

**Fixes**

* **l1** - add 2 confirmation blocks to mitigate reorg issues in L1 block finality.

## [v0.13.0-b1](https://github.com/ADI-Foundation-Labs/ADI-Stack-Server/releases/tag/v0.13.0-b1)

*2026-02-04*

Replay-connection resilience.

**Fixes**

* **replay** - retry on a dropped replay connection and raise the initial delay to 500s (#12).

## [v0.13.0-b](https://github.com/ADI-Foundation-Labs/ADI-Stack-Server/releases/tag/v0.13.0-b)

*First ADI beta*

First ADI beta over upstream `v0.13.0` - a private runtime-config RPC plus RPC and mempool fixes.

**Features**

* **rpc(private)** - private RPC API for runtime configuration overrides, with on-disk persistence, `config_removeOverrides`, and `pubdata_price = 0` support.

**Fixes**

* **rpc** - fix gas estimation for accounts with small balances.
* **mempool** - use the pending base fee on canonical-state change (#9); add a transaction fee cap (#8).
* **fees** - fee estimation now honours config overrides (#7).

**Chores**

* Integrate ADI repositories and improvements; refresh dependencies, README, licenses, and authors.
