feat(katana): l1 gas sampling

[augustin-v]

Description

Gas oracle on katana related to issue #2558

Designed a 60 capacity FIFO buffer with a worker that runs every 60 seconds following the starknet fee calculation.

Problems encountered:

• Wasn't able to implement the actual logic of when to decide whether to use hardcoded values or values sampled by the oracle, as well as the worker initiation.

[kariy]

Wasn't able to implement the actual logic of when to decide whether to use hardcoded values or values sampled by the oracle, as well as the worker initiation.

So the idea is, by default it should always use the proper sampling mechanism. Which means the user has to specify the L1 provider url. Soon this will be a required argument because it is necessary for other stuff to function correctly as well. But right now, the default would always be hardcoded, unless an L1 provider url is supplied.

So we should add a new cli arg, --l1.provider, and use this to determine the sampling logic.

lets turn the L1GasOracle into an enum and implements the current_gas_prices() and current_data_gas_prices() for both variants.

pub enum L1GasOracle {
    Fixed(FixedL1GasOracle),
    Sampled(SampledL1GasOracle),
}

The component should be built inside of

dojo/crates/katana/node/src/lib.rs

Lines 157 to 164 in 082a0e0

	/// Build the node components from the given [`Config`].
	///
	/// This returns a [`Node`] instance which can be launched with the all the necessary components
	/// configured.
	pub async fn build(mut config: Config) -> Result<Node> {
	if config.metrics.is_some() {
	// Metrics recorder must be initialized before calling any of the metrics macros, in order
	// for it to be registered.

But the worker task should only be executed in

dojo/crates/katana/node/src/lib.rs

Lines 101 to 108 in 082a0e0

	/// Start the node.
	///
	/// This method will start all the node process, running them until the node is stopped.
	pub async fn launch(mut self) -> Result<LaunchedNode> {
	let chain = self.backend.chain_spec.id;
	info!(%chain, "Starting node.");
	// TODO: maybe move this to the build stage

So what we shouldn't do is start the worker task automatically after creating the component and instead have a way to start the task manually on the call to .launch().

Let's create a concrete type for the worker task, that implements a Future. Which we can then spawn onto the task manager.

We should also add this new struct into the Node struct

dojo/crates/katana/node/src/lib.rs

Lines 87 to 99 in 082a0e0

	pub struct Node {
	pub pool: TxPool,
	pub db: Option<DbEnv>,
	pub task_manager: TaskManager,
	pub backend: Arc<Backend<BlockifierFactory>>,
	pub block_producer: BlockProducer<BlockifierFactory>,
	pub rpc_config: RpcConfig,
	pub metrics_config: Option<MetricsConfig>,
	pub sequencing_config: SequencingConfig,
	pub messaging_config: Option<MessagingConfig>,
	forked_client: Option<ForkedClient>,
	}

[kariy]

We don't need to handle cancellation manually here, the task manager already handles that for us. You can simply just spawn the task on it.

Example:

dojo/crates/katana/node/src/lib.rs

Lines 117 to 120 in 082a0e0

	let server = MetricsServer::new(exporter).with_process_metrics().with_reports(reports);
	self.task_manager.task_spawner().build_task().spawn(server.start(cfg.addr));
	info!(addr = %cfg.addr, "Metrics server started.");

[augustin-v]

Thank you so much for the detailed review @kariy. This is incredibly helpful I’ll start making the suggested changes right away and will let you know when the updates are ready for another look. Thanks again