Skip to main content
Kona brings together a powerful suite of no-std and std Rust components, purpose-built for the OP Stack. At the heart of this ecosystem is the kona-node — a modern, modular rollup node (L2 consensus node) that can be used as a drop-in binary or as a foundation for custom services. The kona-node is a fully compliant implementation of the “Rollup Node” Specifications and is released as a binary in the kona repository. Whether you’re running a production network, building new rollup features, or experimenting with the OP Stack, kona-node is built to be both robust and extensible.

​
Background

A rollup node is responsible for deriving the canonical L2 chain from L1 blocks and their receipts. It validates these blocks using the Engine API, passing them to the execution layer for processing. Paired with an execution engine like op-reth or op-geth, kona-node tracks the unsafe, safe, and finalized tips of the L2 OP Stack chain, ensuring the node is always in sync with the latest state. :::note Rollup nodes hold minimal state. Unsafe and safe payloads alike are sent away to the execution engine client which holds the chain’s db. This way, the rollup node holds a view of the tip of the chain - unsafe, safe, and finalized block info. All data otherwise needed is held in memory. ::: There are a few core architectural pieces of the kona-node.
  • Derivation Pipeline: Constructs L2 payload attributes from L1 blocks, forming the backbone of rollup logic.
  • Execution Engine Integration: Executes L2 payload attributes via the Engine API, abstracting away different EL clients.
  • P2P Networking: Enables block gossip and peer discovery.
For an in-depth breakdown of these three pillars and a detailed design of the kona-node, visit the Node Design section. Additionally, an RPC server exposes essential methods, including the L2 Output RPC method.

​
Syncing

The kona-node syncs the L2 chain in two main phases:
  1. Execution Layer (EL) Sync: When starting, the node initially has an empty engine task queue. Once an unsafe block is received from P2P gossip, an InsertUnsafe task is executed. The InsertUnsafe task executes a forkchoice update through the engine api. Since the engine state is lazily initialized (the safe and finalized heads are zero), the forkchoice update kicks off EL sync on the execution layer (EL) client (such as op-reth or op-geth). EL sync instructs the execution client to fetch and sync L2 blocks directly from peers. During this phase, the kona-node effectively waits for the EL client to reach the chain tip, when it returns that it is synced. No L2 blocks are derived from L1 during this period.
  2. Consensus Layer (CL) Sync: Once the EL client is fully synced to the tip, the node transitions to consensus layer (CL) sync. In this phase, kona-node begins deriving new L2 payload attributes from the L1 chain, following the rollup derivation process. OpPayloadAttributesWithParent derived this way are executed as an engine task which submits the payloads to the execution engine for validation and execution.
:::note
  • kona-node does not support historical CL sync or backfilling L2 blocks from L1 for past chain history. It relies on the EL client to perform the initial sync of the L2 chain.
  • Only after the EL is fully synced does the node begin deriving and following new L2 blocks from L1.
:::

​
Extensibility

The kona-node is designed as a modular, actor-based node SDK, making it possible to extend or customize node behavior by adding new actors or swapping out existing ones. This extensibility is currently in beta, but the architecture is intentionally built to support advanced use cases and custom integrations.

​
Actor Model

At the core of kona-node is the concept of actors—independent, async services that communicate via channels. Each actor implements the NodeActor trait, which defines how the actor is built, started, and how it communicates with other actors. Key built-in actors include:
  • EngineActor: Manages the execution layer (EL) Engine API.
  • DerivationActor: Runs the derivation pipeline to produce L2 payloads.
  • NetworkActor: Handles P2P networking and block gossip.
  • RpcActor: Runs the node’s RPC server.
  • SupervisorActor: Integrates with the supervisor RPC API.
  • L1WatcherActor: Watches L1 for new blocks and events.

​
Extending with Custom Actors

You can add your own actors to the node by implementing the NodeActor trait. This allows you to introduce new background services, event processors, or integrations with external systems. Example: Defining a Custom Actor 
use kona_node_service::NodeActor;
use async_trait::async_trait;
use tokio_util::sync::CancellationToken;

struct MyCustomActor;

#[async_trait]
impl NodeActor for MyCustomActor {
    type Error = std::io::Error;
    type InboundData = MyCustomContext;
    type OutboundData = ();
    type State = ();

    fn build(_state: Self::State) -> (Self::OutboundData, Self) {
        ((), MyCustomActor)
    }

    async fn start(self, ctx: Self::InboundData) -> Result<(), Self::Error> {
        // Your actor logic here
        Ok(())
    }
}

struct MyCustomContext {
    cancellation: CancellationToken,
}

impl kona_node_service::CancellableContext for MyCustomContext {
    fn cancelled(&self) -> tokio_util::sync::WaitForCancellationFuture<'_> {
        self.cancellation.cancelled()
    }
}

​
Integrating Custom Actors

To integrate your custom actor, you can create your own implementation of the RollupNodeService trait, which defines the set of actors and pipelines used by the node. You can swap out any of the built-in actors for your own, or add entirely new ones. Example: Custom RollupNodeService 
use kona_node_service::{RollupNodeService, NodeActor, ...};

struct MyNodeService;

#[async_trait]
impl RollupNodeService for MyNodeService {
    // Use built-in actors, or your own custom ones
    type DataAvailabilityWatcher = MyCustomActor;
    type DerivationPipeline = ...;
    type DerivationActor = ...;
    type EngineActor = ...;
    type NetworkActor = ...;
    type SupervisorExt = ...;
    type SupervisorActor = ...;
    type RpcActor = ...;
    type Error = ...;

    // Implement required trait methods...
}
You can then instantiate and run your custom node service in your own binary or integration test.

​
Programmatic Node Construction

The RollupNodeBuilder provides a convenient way to construct a standard node, but for advanced use cases, you can build your own node by composing actors and services directly, or by implementing your own builder pattern.

​
Current Limitations

  • The extensibility API is beta and may change.
  • Most users will want to start by subclassing or wrapping the standard RollupNode and only override specific actors or pipelines as needed.
  • Documentation and examples for advanced extensibility are still evolving—contributions and feedback are welcome!

​
Learn More