reth_node_builder/launch/

common.rs

//! Helper types that can be used by launchers.
//!
//! ## Launch Context Type System
//!
//! The node launch process uses a type-state pattern to ensure correct initialization
//! order at compile time. Methods are only available when their prerequisites are met.
//!
//! ### Core Types
//!
//! - [`LaunchContext`]: Base context with executor and data directory
//! - [`LaunchContextWith<T>`]: Context with an attached value of type `T`
//! - [`Attached<L, R>`]: Pairs values, preserving both previous (L) and new (R) state
//!
//! ### Helper Attachments
//!
//! - [`WithConfigs`]: Node config + TOML config
//! - [`WithMeteredProvider`]: Provider factory with metrics
//! - [`WithMeteredProviders`]: Provider factory + blockchain provider
//! - [`WithComponents`]: Final form with all components
//!
//! ### Method Availability
//!
//! Methods are implemented on specific type combinations:
//! - `impl<T> LaunchContextWith<T>`: Generic methods available for any attachment
//! - `impl LaunchContextWith<WithConfigs>`: Config-specific methods
//! - `impl LaunchContextWith<Attached<WithConfigs, DB>>`: Database operations
//! - `impl LaunchContextWith<Attached<WithConfigs, ProviderFactory>>`: Provider operations
//! - etc.
//!
//! This ensures correct initialization order without runtime checks.

use crate::{
    components::{NodeComponents, NodeComponentsBuilder},
    hooks::OnComponentInitializedHook,
    BuilderContext, ExExLauncher, NodeAdapter, PrimitivesTy,
};
use alloy_eips::eip2124::Head;
use alloy_primitives::{BlockNumber, B256};
use eyre::Context;
use rayon::ThreadPoolBuilder;
use reth_chainspec::{Chain, EthChainSpec, EthereumHardforks};
use reth_config::{config::EtlConfig, PruneConfig};
use reth_consensus::noop::NoopConsensus;
use reth_db_api::{
    database::Database, database_metrics::DatabaseMetrics, models::PartialStateTrieUnwindMarker,
};
use reth_db_common::init::{
    init_genesis_with_settings, init_genesis_with_settings_and_validate, InitStorageError,
};
use reth_downloaders::{bodies::noop::NoopBodiesDownloader, headers::noop::NoopHeaderDownloader};
use reth_engine_local::MiningMode;
use reth_evm::{noop::NoopEvmConfig, ConfigureEvm};
use reth_exex::ExExManagerHandle;
use reth_fs_util as fs;
use reth_network_p2p::headers::client::HeadersClient;
use reth_node_api::{FullNodeTypes, NodeTypes, NodeTypesWithDB, NodeTypesWithDBAdapter};
use reth_node_core::{
    args::{DefaultEraHost, PruneConfigKind},
    dirs::{ChainPath, DataDirPath},
    node_config::NodeConfig,
    primitives::BlockHeader,
    version::version_metadata,
};
use reth_node_metrics::{
    chain::ChainSpecInfo,
    hooks::Hooks,
    recorder::install_prometheus_recorder,
    server::{MetricServer, MetricServerConfig},
    storage::StorageSettingsInfo,
    version::VersionInfo,
};
use reth_provider::{
    providers::{NodeTypesForProvider, ProviderNodeTypes, RocksDBProvider, StaticFileProvider},
    BalStoreHandle, BlockHashReader, BlockNumReader, DBProvider, DatabaseProviderFactory,
    MetadataProvider, MetadataWriter, ProviderError, ProviderFactory, ProviderResult,
    RocksDBBalStore, RocksDBProviderFactory, StageCheckpointReader, StaticFileProviderBuilder,
    StaticFileProviderFactory, StorageSettingsCache,
};
use reth_prune::{PruneMode, PruneModes, PrunerBuilder};
use reth_rpc_builder::config::RethRpcServerConfig;
use reth_rpc_layer::JwtSecret;
use reth_stages::{
    sets::DefaultStages,
    stages::{EraImportSource, MerkleStage},
    MetricEvent, PipelineBuilder, PipelineTarget, StageId, StageSet,
};
use reth_static_file::{blocks_per_file_for_prune_distance, StaticFileProducer, StaticFileSegment};
use reth_storage_overlay::OverlayManager;
use reth_tasks::TaskExecutor;
use reth_tracing::{
    throttle,
    tracing::{debug, error, info, warn},
};
use reth_transaction_pool::TransactionPool;
use std::{num::NonZeroUsize, sync::Arc, thread::available_parallelism, time::Duration};
use tokio::sync::{
    mpsc::{unbounded_channel, UnboundedSender},
    oneshot, watch,
};

use futures::{future::Either, stream, Stream, StreamExt};
use reth_node_ethstats::EthStatsService;
use reth_node_events::{cl::ConsensusLayerHealthEvents, node::NodeEvent};

/// Reusable setup for launching a node.
///
/// This is the entry point for the node launch process. It implements a builder
/// pattern using type-state programming to enforce correct initialization order.
///
/// ## Type Evolution
///
/// Starting from `LaunchContext`, each method transforms the type to reflect
/// accumulated state:
///
/// ```text
/// LaunchContext
///   └─> LaunchContextWith<WithConfigs>
///       └─> LaunchContextWith<Attached<WithConfigs, DB>>
///           └─> LaunchContextWith<Attached<WithConfigs, ProviderFactory>>
///               └─> LaunchContextWith<Attached<WithConfigs, WithMeteredProviders>>
///                   └─> LaunchContextWith<Attached<WithConfigs, WithComponents>>
/// ```
#[derive(Debug, Clone)]
pub struct LaunchContext {
    /// The task executor for the node.
    pub task_executor: TaskExecutor,
    /// The data directory for the node.
    pub data_dir: ChainPath<DataDirPath>,
}

impl LaunchContext {
    /// Create a new instance of the default node launcher.
    pub const fn new(task_executor: TaskExecutor, data_dir: ChainPath<DataDirPath>) -> Self {
        Self { task_executor, data_dir }
    }

    /// Create launch context with attachment.
    pub const fn with<T>(self, attachment: T) -> LaunchContextWith<T> {
        LaunchContextWith { inner: self, attachment }
    }

    /// Loads the reth config with the configured `data_dir` and overrides settings according to the
    /// `config`.
    ///
    /// Attaches both the `NodeConfig` and the loaded `reth.toml` config to the launch context.
    pub fn with_loaded_toml_config<ChainSpec>(
        self,
        config: NodeConfig<ChainSpec>,
    ) -> eyre::Result<LaunchContextWith<WithConfigs<ChainSpec>>>
    where
        ChainSpec: EthChainSpec + reth_chainspec::EthereumHardforks,
    {
        let toml_config = self.load_toml_config(&config)?;
        Ok(self.with(WithConfigs { config, toml_config }))
    }

    /// Loads the reth config with the configured `data_dir` and overrides settings according to the
    /// `config`.
    ///
    /// This is async because the trusted peers may have to be resolved.
    pub fn load_toml_config<ChainSpec>(
        &self,
        config: &NodeConfig<ChainSpec>,
    ) -> eyre::Result<reth_config::Config>
    where
        ChainSpec: EthChainSpec + reth_chainspec::EthereumHardforks,
    {
        let config_path = config.config.clone().unwrap_or_else(|| self.data_dir.config());

        let mut toml_config = reth_config::Config::from_path(&config_path)
            .wrap_err_with(|| format!("Could not load config file {config_path:?}"))?;

        Self::save_pruning_config(&mut toml_config, config, &config_path)?;

        info!(target: "reth::cli", path = ?config_path, "Configuration loaded");

        // Update the config with the command line arguments. Only override when the CLI flag is
        // set, so the TOML value is preserved when the flag is not passed.
        toml_config.peers.trusted_nodes_only |= config.network.trusted_only;

        // Merge static file CLI arguments with config file, giving priority to CLI
        toml_config.static_files =
            config.static_files.merge_with_config(toml_config.static_files, config.pruning.minimal);

        Ok(toml_config)
    }

    /// Save prune config to the toml file if node is a full node or has custom pruning CLI
    /// arguments. Also migrates deprecated prune config values to new defaults.
    fn save_pruning_config<ChainSpec>(
        reth_config: &mut reth_config::Config,
        config: &NodeConfig<ChainSpec>,
        config_path: impl AsRef<std::path::Path>,
    ) -> eyre::Result<()>
    where
        ChainSpec: EthChainSpec + reth_chainspec::EthereumHardforks,
    {
        let mut should_save = reth_config.prune.segments.migrate();

        if let Some(prune_config) = config.prune_config() {
            if reth_config.prune != prune_config {
                reth_config.set_prune_config(prune_config);
                should_save = true;
            }
        } else if !reth_config.prune.is_default() {
            info!(target: "reth::cli", "Pruning configuration is present in the config file, but no CLI arguments are provided. Using config from file.");
        }

        if should_save {
            info!(target: "reth::cli", "Saving prune config to toml file");
            reth_config.save(config_path.as_ref())?;
        }

        Ok(())
    }

    /// Convenience function to [`Self::configure_globals`]
    pub fn with_configured_globals(self, reserved_cpu_cores: usize) -> Self {
        self.configure_globals(reserved_cpu_cores);
        self
    }

    /// Configure global settings this includes:
    ///
    /// - Raising the file descriptor limit
    /// - Configuring the global rayon thread pool for implicit `par_iter` usage
    pub fn configure_globals(&self, reserved_cpu_cores: usize) {
        // Raise the fd limit of the process.
        // Does not do anything on windows.
        match fdlimit::raise_fd_limit() {
            Ok(fdlimit::Outcome::LimitRaised { from, to }) => {
                debug!(from, to, "Raised file descriptor limit");
            }
            Ok(fdlimit::Outcome::Unsupported) => {}
            Err(err) => warn!(%err, "Failed to raise file descriptor limit"),
        }

        // Configure the implicit global rayon pool for `par_iter` usage.
        // TODO: reserved_cpu_cores is currently ignored because subtracting from thread pool
        // sizes doesn't actually reserve CPU cores for other processes.
        let _ = reserved_cpu_cores;
        let num_threads = available_parallelism().map_or(1, NonZeroUsize::get);
        if let Err(err) = ThreadPoolBuilder::new()
            .num_threads(num_threads)
            .thread_name(|i| format!("rayon-{i:02}"))
            .build_global()
        {
            warn!(%err, "Failed to build global thread pool")
        }
    }
}

/// A [`LaunchContext`] along with an additional value.
///
/// The type parameter `T` represents the current state of the launch process.
/// Methods are conditionally implemented based on `T`, ensuring operations
/// are only available when their prerequisites are met.
///
/// For example:
/// - Config methods when `T = WithConfigs<ChainSpec>`
/// - Database operations when `T = Attached<WithConfigs<ChainSpec>, DB>`
/// - Provider operations when `T = Attached<WithConfigs<ChainSpec>, ProviderFactory<N>>`
#[derive(Debug, Clone)]
pub struct LaunchContextWith<T> {
    /// The wrapped launch context.
    pub inner: LaunchContext,
    /// The additional attached value.
    pub attachment: T,
}

impl<T> LaunchContextWith<T> {
    /// Configure global settings this includes:
    ///
    /// - Raising the file descriptor limit
    /// - Configuring the global rayon thread pool
    pub fn configure_globals(&self, reserved_cpu_cores: u64) {
        self.inner.configure_globals(reserved_cpu_cores.try_into().unwrap());
    }

    /// Returns the data directory.
    pub const fn data_dir(&self) -> &ChainPath<DataDirPath> {
        &self.inner.data_dir
    }

    /// Returns the task executor.
    pub const fn task_executor(&self) -> &TaskExecutor {
        &self.inner.task_executor
    }

    /// Attaches another value to the launch context.
    pub fn attach<A>(self, attachment: A) -> LaunchContextWith<Attached<T, A>> {
        LaunchContextWith {
            inner: self.inner,
            attachment: Attached::new(self.attachment, attachment),
        }
    }

    /// Consumes the type and calls a function with a reference to the context.
    // Returns the context again
    pub fn inspect<F>(self, f: F) -> Self
    where
        F: FnOnce(&Self),
    {
        f(&self);
        self
    }
}

impl<ChainSpec> LaunchContextWith<WithConfigs<ChainSpec>> {
    /// Resolves the trusted peers and adds them to the toml config.
    pub fn with_resolved_peers(mut self) -> eyre::Result<Self> {
        if !self.attachment.config.network.trusted_peers.is_empty() {
            info!(target: "reth::cli", "Adding trusted nodes");

            self.attachment
                .toml_config
                .peers
                .trusted_nodes
                .extend(self.attachment.config.network.trusted_peers.clone());
        }
        Ok(self)
    }
}

impl<L, R> LaunchContextWith<Attached<L, R>> {
    /// Get a reference to the left value.
    pub const fn left(&self) -> &L {
        &self.attachment.left
    }

    /// Get a reference to the right value.
    pub const fn right(&self) -> &R {
        &self.attachment.right
    }

    /// Get a mutable reference to the left value.
    pub const fn left_mut(&mut self) -> &mut L {
        &mut self.attachment.left
    }

    /// Get a mutable reference to the right value.
    pub const fn right_mut(&mut self) -> &mut R {
        &mut self.attachment.right
    }
}
impl<R, ChainSpec: EthChainSpec> LaunchContextWith<Attached<WithConfigs<ChainSpec>, R>> {
    /// Adjust certain settings in the config to make sure they are set correctly
    ///
    /// This includes:
    /// - Making sure the ETL dir is set to the datadir
    /// - RPC settings are adjusted to the correct port
    pub fn with_adjusted_configs(self) -> Self {
        self.ensure_etl_datadir().with_adjusted_instance_ports()
    }

    /// Make sure ETL doesn't default to /tmp/, but to whatever datadir is set to
    pub fn ensure_etl_datadir(mut self) -> Self {
        if self.toml_config_mut().stages.etl.dir.is_none() {
            let etl_path = EtlConfig::from_datadir(self.data_dir().data_dir());
            if etl_path.exists() {
                // Remove etl-path files on launch
                if let Err(err) = fs::remove_dir_all(&etl_path) {
                    warn!(target: "reth::cli", ?etl_path, %err, "Failed to remove ETL path on launch");
                }
            }
            self.toml_config_mut().stages.etl.dir = Some(etl_path);
        }

        self
    }

    /// Change rpc port numbers based on the instance number.
    pub fn with_adjusted_instance_ports(mut self) -> Self {
        self.node_config_mut().adjust_instance_ports();
        self
    }

    /// Returns the container for all config types
    pub const fn configs(&self) -> &WithConfigs<ChainSpec> {
        self.attachment.left()
    }

    /// Returns the attached [`NodeConfig`].
    pub const fn node_config(&self) -> &NodeConfig<ChainSpec> {
        &self.left().config
    }

    /// Returns the attached [`NodeConfig`].
    pub const fn node_config_mut(&mut self) -> &mut NodeConfig<ChainSpec> {
        &mut self.left_mut().config
    }

    /// Returns the attached toml config [`reth_config::Config`].
    pub const fn toml_config(&self) -> &reth_config::Config {
        &self.left().toml_config
    }

    /// Returns the attached toml config [`reth_config::Config`].
    pub const fn toml_config_mut(&mut self) -> &mut reth_config::Config {
        &mut self.left_mut().toml_config
    }

    /// Returns the configured chain spec.
    pub fn chain_spec(&self) -> Arc<ChainSpec> {
        self.node_config().chain.clone()
    }

    /// Get the hash of the genesis block.
    pub fn genesis_hash(&self) -> B256 {
        self.node_config().chain.genesis_hash()
    }

    /// Returns the chain identifier of the node.
    pub fn chain_id(&self) -> Chain {
        self.node_config().chain.chain()
    }

    /// Returns true if the node is configured as --dev
    pub const fn is_dev(&self) -> bool {
        self.node_config().dev.dev
    }

    /// Returns the configured [`PruneConfig`]
    ///
    /// Any configuration set in CLI will take precedence over those set in toml
    pub fn prune_config(&self) -> PruneConfig
    where
        ChainSpec: reth_chainspec::EthereumHardforks,
    {
        let Some(mut node_prune_config) = self.node_config().prune_config() else {
            // No CLI config is set, use the toml config.
            return self.toml_config().prune.clone();
        };

        // Otherwise, use the CLI configuration and merge with toml config.
        node_prune_config.merge(self.toml_config().prune.clone());
        node_prune_config
    }

    /// Returns the configured [`PruneModes`], returning the default if no config was available.
    pub fn prune_modes(&self) -> PruneModes
    where
        ChainSpec: reth_chainspec::EthereumHardforks,
    {
        self.prune_config().segments
    }

    /// Returns an initialized [`PrunerBuilder`] based on the configured [`PruneConfig`]
    pub fn pruner_builder(&self) -> PrunerBuilder
    where
        ChainSpec: reth_chainspec::EthereumHardforks,
    {
        PrunerBuilder::new(self.prune_config())
    }

    /// Loads the JWT secret for the engine API
    pub fn auth_jwt_secret(&self) -> eyre::Result<JwtSecret> {
        let default_jwt_path = self.data_dir().jwt();
        let secret = self.node_config().rpc.auth_jwt_secret(default_jwt_path)?;
        Ok(secret)
    }

    /// Returns the [`MiningMode`] intended for --dev mode.
    pub fn dev_mining_mode<Pool>(&self, pool: Pool) -> MiningMode<Pool>
    where
        Pool: TransactionPool + Unpin,
    {
        self.node_config().dev_mining_mode(pool)
    }
}

impl<DB, ChainSpec> LaunchContextWith<Attached<WithConfigs<ChainSpec>, DB>>
where
    DB: Database + Clone + 'static,
    ChainSpec: EthChainSpec + EthereumHardforks + 'static,
{
    /// Returns the [`ProviderFactory`] for the attached storage after executing a consistent check
    /// between the database and static files. **It may execute a pipeline unwind if it fails this
    /// check.**
    pub async fn create_provider_factory<N, Evm>(
        &self,
        overlay_manager: OverlayManager<N::Primitives>,
        rocksdb_provider: Option<RocksDBProvider>,
        disabled_stages: &[StageId],
    ) -> eyre::Result<ProviderFactory<N>>
    where
        N: ProviderNodeTypes<DB = DB, ChainSpec = ChainSpec>,
        Evm: ConfigureEvm<Primitives = N::Primitives> + 'static,
    {
        // Validate static files configuration
        let static_files_config = &self.toml_config().static_files;
        static_files_config.validate()?;

        let prune_config = self.prune_config();

        let mut blocks_per_file = static_files_config.as_blocks_per_file_map();
        // Receipts in static files are pruned by deleting whole files, so with the default file
        // size a distance-based prune target is only reached every 500k blocks. Unless a file size
        // is explicitly configured, derive one from the prune distance so retention tracks the
        // configured distance.
        if blocks_per_file.get(StaticFileSegment::Receipts).is_none() &&
            let Some(PruneMode::Distance(distance)) = prune_config.segments.receipts
        {
            blocks_per_file
                .insert(StaticFileSegment::Receipts, blocks_per_file_for_prune_distance(distance));
        }

        // Apply per-segment blocks_per_file configuration
        let static_file_provider =
            StaticFileProviderBuilder::read_write(self.data_dir().static_files())
                .with_metrics()
                .with_blocks_per_file_for_segments(&blocks_per_file)
                .with_genesis_block_number(self.chain_spec().genesis().number.unwrap_or_default())
                .build()?;

        // Use the provided RocksDB provider or create a new one
        let rocksdb_provider = if let Some(provider) = rocksdb_provider {
            provider
        } else {
            RocksDBProvider::builder(self.data_dir().rocksdb())
                .with_default_tables()
                .with_metrics()
                .with_statistics()
                .build()?
        };

        let bal_store = self
            .node_config()
            .db
            .balstore_cache_size
            .map(|distance| {
                RocksDBBalStore::with_buffer_retention_distance(rocksdb_provider.clone(), distance)
            })
            .unwrap_or_else(|| RocksDBBalStore::new(rocksdb_provider.clone()));
        let bal_store = BalStoreHandle::new(bal_store);
        let factory = ProviderFactory::new(
            self.right().clone(),
            self.chain_spec(),
            static_file_provider,
            rocksdb_provider,
            self.task_executor().clone(),
        )?
        .with_prune_modes(prune_config.segments)
        .with_minimum_pruning_distance(prune_config.minimum_pruning_distance)
        .with_overlay_manager(overlay_manager)
        .with_bal_store(bal_store);

        // Check consistency between the database and static files, returning
        // the unwind targets for each storage layer if inconsistencies are
        // found.
        let (rocksdb_unwind, static_file_unwind) = factory.check_consistency()?;
        let provider_ro = factory.database_provider_ro()?;
        // Finish is committed before Merkle during unwind, so this marker is authoritative when
        // resuming an interrupted partial trie unwind.
        let (partial_trie_unwind, has_persisted_partial_trie_unwind) =
            get_partial_trie_unwind_marker(&provider_ro)?;
        drop(provider_ro);
        let persist_partial_trie_unwind =
            !has_persisted_partial_trie_unwind && partial_trie_unwind.is_some();
        let partial_trie_unwind_target =
            partial_trie_unwind.map(|marker| marker.partial_state_trie);
        // Recover the partial state trie first. Its unwind enables
        // `walk_all_changed_branch_children`, which is more expensive than a normal unwind, so
        // it only runs to the partial trie target. A lower storage-layer target is then unwound
        // normally.
        let storage_unwind = [rocksdb_unwind, static_file_unwind].into_iter().flatten().min();
        let storage_unwind = storage_unwind.filter(|unwind_block| {
            partial_trie_unwind_target.is_none_or(|partial_trie| *unwind_block < partial_trie)
        });

        if partial_trie_unwind_target.is_some() || storage_unwind.is_some() {
            let build_unwind_pipeline = |walk_all_changed_branch_children| {
                let (_tip_tx, tip_rx) = watch::channel(B256::ZERO);
                let mut stages = DefaultStages::new(
                    factory.clone(),
                    tip_rx,
                    Arc::new(NoopConsensus::default()),
                    NoopHeaderDownloader::default(),
                    NoopBodiesDownloader::default(),
                    NoopEvmConfig::<Evm>::default(),
                    self.toml_config().stages.clone(),
                    self.prune_modes(),
                    None,
                )
                .builder()
                .disable_all(disabled_stages);

                if walk_all_changed_branch_children {
                    // Partial trie recovery is not complete until Merkle has unwound.
                    stages =
                        stages.set(MerkleStage::new_unwind(true)).enable(StageId::MerkleUnwind);
                }

                PipelineBuilder::default().add_stages(stages).build(
                    factory.clone(),
                    StaticFileProducer::new(factory.clone(), self.prune_modes()),
                )
            };
            let mut unwinds = Vec::with_capacity(2);

            if let Some(unwind_block) = partial_trie_unwind_target {
                unwinds.push((
                    PipelineTarget::Unwind(unwind_block),
                    "partial state trie".to_owned(),
                    build_unwind_pipeline(true),
                    true,
                ));
            }

            if let Some(unwind_block) = storage_unwind {
                // Highly unlikely to happen, and given its destructive nature, it's better to
                // panic instead. Unwinding to 0 would leave MDBX with a huge free list size.
                let inconsistency_source = match (rocksdb_unwind, static_file_unwind) {
                    (Some(_), Some(_)) => "RocksDB and static file",
                    (Some(_), None) => "RocksDB",
                    (None, Some(_)) => "static file",
                    (None, None) => unreachable!(),
                };
                assert_ne!(
                    unwind_block, 0,
                    "A {inconsistency_source} inconsistency was found that would trigger an unwind to block 0"
                );
                unwinds.push((
                    PipelineTarget::Unwind(unwind_block),
                    inconsistency_source.to_owned(),
                    build_unwind_pipeline(false),
                    false,
                ));
            }

            if persist_partial_trie_unwind {
                // The marker must be durable before any unwind stage can commit.
                let provider_rw = factory.database_provider_rw()?;
                write_partial_trie_unwind_marker(
                    &provider_rw,
                    partial_trie_unwind.expect("partial trie unwind marker must exist"),
                )?;
                provider_rw.commit()?;
            }

            let (tx, rx) = oneshot::channel();
            let factory = factory.clone();

            // Pipeline should be run as blocking and panic if it fails.
            self.task_executor().spawn_critical_blocking_task("pipeline task", async move {
                let result: Result<(), reth_stages::PipelineError> = async {
                    for (unwind_target, inconsistency_source, pipeline, clear_partial_trie_unwind) in
                        unwinds
                    {
                        info!(target: "reth::cli", %unwind_target, %inconsistency_source, "Executing unwind after consistency check.");
                        let (_, result) = pipeline.run_as_fut(Some(unwind_target)).await;
                        result.inspect_err(|err| {
                            error!(target: "reth::cli", %unwind_target, %inconsistency_source, %err, "failed to run unwind");
                        })?;

                        if clear_partial_trie_unwind {
                            let provider_rw = factory.database_provider_rw()?;
                            delete_partial_trie_unwind_marker(&provider_rw)?;
                            provider_rw.commit()?;
                        }
                    }
                    Ok(())
                }
                .await;
                let _ = tx.send(result);
            });
            rx.await??;
        }

        Ok(factory)
    }

    /// Creates a new [`ProviderFactory`] and attaches it to the launch context.
    pub async fn with_provider_factory<N, Evm>(
        self,
        overlay_manager: OverlayManager<N::Primitives>,
        rocksdb_provider: Option<RocksDBProvider>,
        disabled_stages: &[StageId],
    ) -> eyre::Result<LaunchContextWith<Attached<WithConfigs<ChainSpec>, ProviderFactory<N>>>>
    where
        N: ProviderNodeTypes<DB = DB, ChainSpec = ChainSpec>,
        Evm: ConfigureEvm<Primitives = N::Primitives> + 'static,
    {
        let factory = self
            .create_provider_factory::<N, Evm>(overlay_manager, rocksdb_provider, disabled_stages)
            .await?;
        let ctx = LaunchContextWith {
            inner: self.inner,
            attachment: self.attachment.map_right(|_| factory),
        };

        Ok(ctx)
    }
}

impl<T> LaunchContextWith<Attached<WithConfigs<T::ChainSpec>, ProviderFactory<T>>>
where
    T: ProviderNodeTypes,
{
    /// Returns access to the underlying database.
    pub const fn database(&self) -> &T::DB {
        self.right().db_ref()
    }

    /// Returns the configured `ProviderFactory`.
    pub const fn provider_factory(&self) -> &ProviderFactory<T> {
        self.right()
    }

    /// Returns the static file provider to interact with the static files.
    pub fn static_file_provider(&self) -> StaticFileProvider<T::Primitives> {
        self.right().static_file_provider()
    }

    /// This launches the prometheus endpoint.
    ///
    /// Convenience function to [`Self::start_prometheus_endpoint`]
    pub async fn with_prometheus_server(self) -> eyre::Result<Self>
    where
        T::ChainSpec: EthereumHardforks,
    {
        self.start_prometheus_endpoint().await?;
        Ok(self)
    }

    /// Starts the prometheus endpoint.
    pub async fn start_prometheus_endpoint(&self) -> eyre::Result<()>
    where
        T::ChainSpec: EthereumHardforks,
    {
        // ensure recorder runs upkeep periodically
        install_prometheus_recorder().spawn_upkeep();

        let listen_addr = self.node_config().metrics.prometheus;
        if let Some(addr) = listen_addr {
            let prune_config = self.prune_config();
            let pruning_mode =
                PruneConfigKind::from_config(&prune_config, self.chain_spec().as_ref()).as_str();
            // On existing databases, stored settings are authoritative and already cached by the
            // provider factory. Fresh databases do not have storage metadata until genesis is
            // initialized, so report the configured setting during this pre-genesis startup window.
            let storage_settings =
                if self.provider_factory().get_stage_checkpoint(StageId::Headers)?.is_some() {
                    self.provider_factory().cached_storage_settings()
                } else {
                    self.node_config().storage_settings()
                };
            let config = MetricServerConfig::new(
                addr,
                VersionInfo {
                    version: version_metadata().cargo_pkg_version.as_ref(),
                    build_timestamp: version_metadata().vergen_build_timestamp.as_ref(),
                    cargo_features: version_metadata().vergen_cargo_features.as_ref(),
                    git_sha: version_metadata().vergen_git_sha.as_ref(),
                    target_triple: version_metadata().vergen_cargo_target_triple.as_ref(),
                    build_profile: version_metadata().build_profile_name.as_ref(),
                },
                ChainSpecInfo { name: self.chain_id().to_string() },
                self.task_executor().clone(),
                metrics_hooks(self.provider_factory()),
                self.data_dir().pprof_dumps(),
            )
            .with_storage_settings_info(StorageSettingsInfo {
                storage_v2: storage_settings.storage_v2,
                pruning_mode,
                prune_config: serde_json::to_string(&prune_config)
                    .expect("serializing PruneConfig should not fail"),
            })
            .with_push_gateway(
                self.node_config().metrics.push_gateway_url.clone(),
                self.node_config().metrics.push_gateway_interval,
            );

            MetricServer::new(config).serve().await?;
        }

        Ok(())
    }

    /// Convenience function to [`Self::init_genesis`]
    pub fn with_genesis(self) -> Result<Self, InitStorageError> {
        init_genesis_with_settings_and_validate(
            self.provider_factory(),
            self.node_config().storage_settings(),
            !self.node_config().debug.skip_genesis_validation,
        )?;
        Ok(self)
    }

    /// Write the genesis block and state if it has not already been written
    pub fn init_genesis(&self) -> Result<B256, InitStorageError> {
        init_genesis_with_settings(self.provider_factory(), self.node_config().storage_settings())
    }

    /// Creates a new `WithMeteredProvider` container and attaches it to the
    /// launch context.
    ///
    /// This spawns a metrics task that listens for metrics related events and updates metrics for
    /// prometheus.
    pub fn with_metrics_task(
        self,
    ) -> LaunchContextWith<Attached<WithConfigs<T::ChainSpec>, WithMeteredProvider<T>>> {
        let (metrics_sender, metrics_receiver) = unbounded_channel();

        let with_metrics =
            WithMeteredProvider { provider_factory: self.right().clone(), metrics_sender };

        debug!(target: "reth::cli", "Spawning stages metrics listener task");
        let sync_metrics_listener = reth_stages::MetricsListener::new(metrics_receiver);
        self.task_executor()
            .spawn_critical_task("stages metrics listener task", sync_metrics_listener);

        LaunchContextWith {
            inner: self.inner,
            attachment: self.attachment.map_right(|_| with_metrics),
        }
    }
}

impl<N, DB>
    LaunchContextWith<
        Attached<WithConfigs<N::ChainSpec>, WithMeteredProvider<NodeTypesWithDBAdapter<N, DB>>>,
    >
where
    N: NodeTypes,
    DB: Database + DatabaseMetrics + Clone + Unpin + 'static,
{
    /// Returns the configured `ProviderFactory`.
    const fn provider_factory(&self) -> &ProviderFactory<NodeTypesWithDBAdapter<N, DB>> {
        &self.right().provider_factory
    }

    /// Returns the metrics sender.
    fn sync_metrics_tx(&self) -> UnboundedSender<MetricEvent> {
        self.right().metrics_sender.clone()
    }

    /// Creates a `BlockchainProvider` and attaches it to the launch context.
    #[expect(clippy::complexity)]
    pub fn with_blockchain_db<T, F>(
        self,
        create_blockchain_provider: F,
    ) -> eyre::Result<LaunchContextWith<Attached<WithConfigs<N::ChainSpec>, WithMeteredProviders<T>>>>
    where
        T: FullNodeTypes<Types = N, DB = DB>,
        F: FnOnce(ProviderFactory<NodeTypesWithDBAdapter<N, DB>>) -> eyre::Result<T::Provider>,
    {
        let blockchain_db = create_blockchain_provider(self.provider_factory().clone())?;

        let metered_providers = WithMeteredProviders {
            db_provider_container: WithMeteredProvider {
                provider_factory: self.provider_factory().clone(),
                metrics_sender: self.sync_metrics_tx(),
            },
            blockchain_db,
        };

        let ctx = LaunchContextWith {
            inner: self.inner,
            attachment: self.attachment.map_right(|_| metered_providers),
        };

        Ok(ctx)
    }
}

impl<T>
    LaunchContextWith<
        Attached<WithConfigs<<T::Types as NodeTypes>::ChainSpec>, WithMeteredProviders<T>>,
    >
where
    T: FullNodeTypes<Types: NodeTypesForProvider>,
{
    /// Returns access to the underlying database.
    pub const fn database(&self) -> &T::DB {
        self.provider_factory().db_ref()
    }

    /// Returns the configured `ProviderFactory`.
    pub const fn provider_factory(
        &self,
    ) -> &ProviderFactory<NodeTypesWithDBAdapter<T::Types, T::DB>> {
        &self.right().db_provider_container.provider_factory
    }

    /// Fetches the head block from the database.
    ///
    /// If the database is empty, returns the genesis block.
    pub fn lookup_head(&self) -> eyre::Result<Head> {
        self.node_config()
            .lookup_head(self.provider_factory())
            .wrap_err("the head block is missing")
    }

    /// Returns the metrics sender.
    pub fn sync_metrics_tx(&self) -> UnboundedSender<MetricEvent> {
        self.right().db_provider_container.metrics_sender.clone()
    }

    /// Returns a reference to the blockchain provider.
    pub const fn blockchain_db(&self) -> &T::Provider {
        &self.right().blockchain_db
    }

    /// Creates a `NodeAdapter` and attaches it to the launch context.
    pub async fn with_components<CB>(
        self,
        components_builder: CB,
        on_component_initialized: Box<
            dyn OnComponentInitializedHook<NodeAdapter<T, CB::Components>>,
        >,
    ) -> eyre::Result<
        LaunchContextWith<
            Attached<WithConfigs<<T::Types as NodeTypes>::ChainSpec>, WithComponents<T, CB>>,
        >,
    >
    where
        CB: NodeComponentsBuilder<T>,
    {
        // fetch the head block from the database
        let head = self.lookup_head()?;

        let builder_ctx = BuilderContext::new(
            head,
            self.blockchain_db().clone(),
            self.task_executor().clone(),
            self.configs().clone(),
        );

        debug!(target: "reth::cli", "creating components");
        let components = components_builder.build_components(&builder_ctx).await?;

        let blockchain_db = self.blockchain_db().clone();

        let node_adapter = NodeAdapter {
            components,
            task_executor: self.task_executor().clone(),
            provider: blockchain_db,
        };

        debug!(target: "reth::cli", "calling on_component_initialized hook");
        on_component_initialized.on_event(node_adapter.clone())?;

        let components_container = WithComponents {
            db_provider_container: WithMeteredProvider {
                provider_factory: self.provider_factory().clone(),
                metrics_sender: self.sync_metrics_tx(),
            },
            node_adapter,
            head,
        };

        let ctx = LaunchContextWith {
            inner: self.inner,
            attachment: self.attachment.map_right(|_| components_container),
        };

        Ok(ctx)
    }
}

impl<T, CB>
    LaunchContextWith<
        Attached<WithConfigs<<T::Types as NodeTypes>::ChainSpec>, WithComponents<T, CB>>,
    >
where
    T: FullNodeTypes<Types: NodeTypesForProvider>,
    CB: NodeComponentsBuilder<T>,
{
    /// Returns the configured `ProviderFactory`.
    pub const fn provider_factory(
        &self,
    ) -> &ProviderFactory<NodeTypesWithDBAdapter<T::Types, T::DB>> {
        &self.right().db_provider_container.provider_factory
    }

    /// Returns the max block that the node should run to, looking it up from the network if
    /// necessary
    pub async fn max_block<C>(&self, client: C) -> eyre::Result<Option<BlockNumber>>
    where
        C: HeadersClient<Header: BlockHeader>,
    {
        self.node_config().max_block(client, self.provider_factory().clone()).await
    }

    /// Returns the static file provider to interact with the static files.
    pub fn static_file_provider(&self) -> StaticFileProvider<<T::Types as NodeTypes>::Primitives> {
        self.provider_factory().static_file_provider()
    }

    /// Creates a new [`StaticFileProducer`] with the attached database.
    pub fn static_file_producer(
        &self,
    ) -> StaticFileProducer<ProviderFactory<NodeTypesWithDBAdapter<T::Types, T::DB>>> {
        StaticFileProducer::new(self.provider_factory().clone(), self.prune_modes())
    }

    /// Returns the current head block.
    pub const fn head(&self) -> Head {
        self.right().head
    }

    /// Returns the configured `NodeAdapter`.
    pub const fn node_adapter(&self) -> &NodeAdapter<T, CB::Components> {
        &self.right().node_adapter
    }

    /// Returns mutable reference to the configured `NodeAdapter`.
    pub const fn node_adapter_mut(&mut self) -> &mut NodeAdapter<T, CB::Components> {
        &mut self.right_mut().node_adapter
    }

    /// Returns a reference to the blockchain provider.
    pub const fn blockchain_db(&self) -> &T::Provider {
        &self.node_adapter().provider
    }

    /// Returns the initial backfill to sync to at launch.
    ///
    /// This returns the configured `debug.tip` if set, otherwise it will check if backfill was
    /// previously interrupted and returns the block hash of the last checkpoint, see also
    /// [`Self::check_pipeline_consistency`]
    pub fn initial_backfill_target(
        &self,
        disabled_stages: &[StageId],
    ) -> ProviderResult<Option<B256>> {
        let mut initial_target = self.node_config().debug.tip;

        if initial_target.is_none() {
            initial_target = self.check_pipeline_consistency(disabled_stages)?;
        }

        Ok(initial_target)
    }

    /// Returns true if the node should terminate after the initial backfill run.
    ///
    /// This is the case if any of these configs are set:
    ///  `--debug.max-block`
    ///  `--debug.terminate`
    pub const fn terminate_after_initial_backfill(&self) -> bool {
        self.node_config().debug.terminate || self.node_config().debug.max_block.is_some()
    }

    /// Ensures that the database matches chain-specific requirements.
    ///
    /// This checks for OP-Mainnet and ensures we have all the necessary data to progress (past
    /// bedrock height)
    fn ensure_chain_specific_db_checks(&self) -> ProviderResult<()> {
        if self.chain_spec().is_optimism() &&
            !self.is_dev() &&
            self.chain_id() == Chain::optimism_mainnet()
        {
            let latest = self.blockchain_db().last_block_number()?;
            // bedrock height
            if latest < 105235063 {
                error!(
                    "Op-mainnet has been launched without importing the pre-Bedrock state. The chain can't progress without this. See also https://reth.rs/run/sync-op-mainnet.html?minimal-bootstrap-recommended"
                );
                return Err(ProviderError::BestBlockNotFound);
            }
        }

        Ok(())
    }

    /// Check if the pipeline is consistent (all stages have the checkpoint block numbers no less
    /// than the checkpoint of the first stage).
    ///
    /// This will return the pipeline target if:
    ///  * the pipeline was interrupted during its previous run
    ///  * a new stage was added
    ///  * stage data was dropped manually through `reth stage drop ...`
    ///
    /// # Returns
    ///
    /// A target block hash if the pipeline is inconsistent, otherwise `None`.
    pub fn check_pipeline_consistency(
        &self,
        disabled_stages: &[StageId],
    ) -> ProviderResult<Option<B256>> {
        // We skip the era stage if it's not enabled
        let era_enabled = self.era_import_source().is_some();
        let mut all_stages = StageId::ALL
            .into_iter()
            .filter(|id| (era_enabled || id != &StageId::Era) && !disabled_stages.contains(id));

        // Get the expected first stage based on config.
        let first_stage = all_stages.next().expect("there must be at least one stage");

        // If no target was provided, check if the stages are congruent - check if the
        // checkpoint of the last stage matches the checkpoint of the first.
        let first_stage_checkpoint = self
            .blockchain_db()
            .get_stage_checkpoint(first_stage)?
            .unwrap_or_default()
            .block_number;

        // Compare all other stages against the first
        for stage_id in all_stages {
            let stage_checkpoint = self
                .blockchain_db()
                .get_stage_checkpoint(stage_id)?
                .unwrap_or_default()
                .block_number;

            // If the checkpoint of any stage is less than the checkpoint of the first stage,
            // retrieve and return the block hash of the latest header and use it as the target.
            debug!(
                target: "consensus::engine",
                first_stage_id = %first_stage,
                first_stage_checkpoint,
                stage_id = %stage_id,
                stage_checkpoint = stage_checkpoint,
                "Checking stage against first stage",
            );
            if stage_checkpoint < first_stage_checkpoint {
                debug!(
                    target: "consensus::engine",
                    first_stage_id = %first_stage,
                    first_stage_checkpoint,
                    inconsistent_stage_id = %stage_id,
                    inconsistent_stage_checkpoint = stage_checkpoint,
                    "Pipeline sync progress is inconsistent"
                );
                return self.blockchain_db().block_hash(first_stage_checkpoint);
            }
        }

        self.ensure_chain_specific_db_checks()?;

        Ok(None)
    }

    /// Returns the metrics sender.
    pub fn sync_metrics_tx(&self) -> UnboundedSender<MetricEvent> {
        self.right().db_provider_container.metrics_sender.clone()
    }

    /// Returns the node adapter components.
    pub const fn components(&self) -> &CB::Components {
        &self.node_adapter().components
    }

    /// Launches ExEx (Execution Extensions) and returns the ExEx manager handle.
    #[expect(clippy::type_complexity)]
    pub async fn launch_exex(
        &self,
        installed_exex: Vec<(
            String,
            Box<dyn crate::exex::BoxedLaunchExEx<NodeAdapter<T, CB::Components>>>,
        )>,
    ) -> eyre::Result<Option<ExExManagerHandle<PrimitivesTy<T::Types>>>> {
        self.exex_launcher(installed_exex).launch().await
    }

    /// Creates an [`ExExLauncher`] for the installed ExExes.
    ///
    /// This returns the launcher before calling `.launch()`, allowing custom configuration
    /// such as setting the WAL blocks warning threshold for L2 chains with faster block times:
    ///
    /// ```ignore
    /// ctx.exex_launcher(exexes)
    ///     .with_wal_blocks_warning(768)  // For 2-second block times
    ///     .launch()
    ///     .await
    /// ```
    #[expect(clippy::type_complexity)]
    pub fn exex_launcher(
        &self,
        installed_exex: Vec<(
            String,
            Box<dyn crate::exex::BoxedLaunchExEx<NodeAdapter<T, CB::Components>>>,
        )>,
    ) -> ExExLauncher<NodeAdapter<T, CB::Components>> {
        ExExLauncher::new(
            self.head(),
            self.node_adapter().clone(),
            installed_exex,
            self.configs().clone(),
        )
    }

    /// Creates the ERA import source based on node configuration.
    ///
    /// Returns `Some(EraImportSource)` if ERA is enabled in the node config, otherwise `None`.
    pub fn era_import_source(&self) -> Option<EraImportSource> {
        let node_config = self.node_config();
        if !node_config.era.enabled {
            return None;
        }

        EraImportSource::maybe_new(
            node_config.era.source.path.clone(),
            node_config.era.source.url.clone(),
            || node_config.chain.chain().kind().default_era_host(),
            || node_config.datadir().data_dir().join("era").into(),
        )
    }

    /// Creates consensus layer health events stream based on node configuration.
    ///
    /// Returns a stream that monitors consensus layer health if:
    /// - No debug tip is configured
    /// - Not running in dev mode
    ///
    /// Otherwise returns an empty stream.
    pub fn consensus_layer_events(
        &self,
    ) -> impl Stream<Item = NodeEvent<PrimitivesTy<T::Types>>> + 'static
    where
        T::Provider: reth_provider::CanonChainTracker,
    {
        if self.node_config().debug.tip.is_none() && !self.is_dev() {
            Either::Left(
                ConsensusLayerHealthEvents::new(Box::new(self.blockchain_db().clone()))
                    .map(Into::into),
            )
        } else {
            Either::Right(stream::empty())
        }
    }

    /// Spawns the [`EthStatsService`] service if configured.
    pub async fn spawn_ethstats<St>(&self, mut engine_events: St) -> eyre::Result<()>
    where
        St: Stream<Item = reth_engine_primitives::ConsensusEngineEvent<PrimitivesTy<T::Types>>>
            + Send
            + Unpin
            + 'static,
    {
        let Some(url) = self.node_config().debug.ethstats.as_ref() else { return Ok(()) };

        let network = self.components().network().clone();
        let pool = self.components().pool().clone();
        let provider = self.node_adapter().provider.clone();

        info!(target: "reth::cli", "Starting EthStats service at {}", url);

        let ethstats = EthStatsService::new(url, network, provider, pool).await?;

        // If engine events are provided, spawn listener for new payload reporting
        let ethstats_for_events = ethstats.clone();
        let task_executor = self.task_executor().clone();
        task_executor.spawn_task(async move {
            while let Some(event) = engine_events.next().await {
                use reth_engine_primitives::ConsensusEngineEvent;
                match event {
                    ConsensusEngineEvent::ForkBlockAdded(executed, duration) |
                    ConsensusEngineEvent::CanonicalBlockAdded(executed, duration) => {
                        let block_hash = executed.recovered_block.num_hash().hash;
                        let block_number = executed.recovered_block.num_hash().number;
                        if let Err(e) = ethstats_for_events
                            .report_new_payload(block_hash, block_number, duration)
                            .await
                        {
                            debug!(
                                target: "ethstats",
                                "Failed to report new payload: {}", e
                            );
                        }
                    }
                    _ => {
                        // Ignore other event types for ethstats reporting
                    }
                }
            }
        });

        // Spawn main ethstats service
        task_executor.spawn_task(async move { ethstats.run().await });

        Ok(())
    }
}

/// Joins two attachments together, preserving access to both values.
///
/// This type enables the launch process to accumulate state while maintaining
/// access to all previously attached components. The `left` field holds the
/// previous state, while `right` holds the newly attached component.
#[derive(Clone, Copy, Debug)]
pub struct Attached<L, R> {
    left: L,
    right: R,
}

impl<L, R> Attached<L, R> {
    /// Creates a new `Attached` with the given values.
    pub const fn new(left: L, right: R) -> Self {
        Self { left, right }
    }

    /// Maps the left value to a new value.
    pub fn map_left<F, T>(self, f: F) -> Attached<T, R>
    where
        F: FnOnce(L) -> T,
    {
        Attached::new(f(self.left), self.right)
    }

    /// Maps the right value to a new value.
    pub fn map_right<F, T>(self, f: F) -> Attached<L, T>
    where
        F: FnOnce(R) -> T,
    {
        Attached::new(self.left, f(self.right))
    }

    /// Get a reference to the left value.
    pub const fn left(&self) -> &L {
        &self.left
    }

    /// Get a reference to the right value.
    pub const fn right(&self) -> &R {
        &self.right
    }

    /// Get a mutable reference to the left value.
    pub const fn left_mut(&mut self) -> &mut L {
        &mut self.left
    }

    /// Get a mutable reference to the right value.
    pub const fn right_mut(&mut self) -> &mut R {
        &mut self.right
    }
}

/// Helper container type to bundle the initial [`NodeConfig`] and the loaded settings from the
/// reth.toml config
#[derive(Debug)]
pub struct WithConfigs<ChainSpec> {
    /// The configured, usually derived from the CLI.
    pub config: NodeConfig<ChainSpec>,
    /// The loaded reth.toml config.
    pub toml_config: reth_config::Config,
}

impl<ChainSpec> Clone for WithConfigs<ChainSpec> {
    fn clone(&self) -> Self {
        Self { config: self.config.clone(), toml_config: self.toml_config.clone() }
    }
}

/// Helper container type to bundle the [`ProviderFactory`] and the metrics
/// sender.
#[derive(Debug, Clone)]
pub struct WithMeteredProvider<N: NodeTypesWithDB> {
    provider_factory: ProviderFactory<N>,
    metrics_sender: UnboundedSender<MetricEvent>,
}

/// Helper container to bundle the [`ProviderFactory`], [`FullNodeTypes::Provider`]
/// and a metrics sender.
#[expect(missing_debug_implementations)]
pub struct WithMeteredProviders<T>
where
    T: FullNodeTypes,
{
    db_provider_container: WithMeteredProvider<NodeTypesWithDBAdapter<T::Types, T::DB>>,
    blockchain_db: T::Provider,
}

/// Helper container to bundle the metered providers container and [`NodeAdapter`].
#[expect(missing_debug_implementations)]
pub struct WithComponents<T, CB>
where
    T: FullNodeTypes,
    CB: NodeComponentsBuilder<T>,
{
    db_provider_container: WithMeteredProvider<NodeTypesWithDBAdapter<T::Types, T::DB>>,
    node_adapter: NodeAdapter<T, CB::Components>,
    head: Head,
}

/// Returns the metrics hooks for the node.
pub fn metrics_hooks<N: NodeTypesWithDB>(provider_factory: &ProviderFactory<N>) -> Hooks {
    Hooks::builder()
        .with_hook({
            let db = provider_factory.db_ref().clone();
            move || throttle!(Duration::from_secs(5 * 60), || db.report_metrics())
        })
        .with_hook({
            let sfp = provider_factory.static_file_provider();
            move || {
                throttle!(Duration::from_secs(5 * 60), || {
                    if let Err(error) = sfp.report_metrics() {
                        error!(%error, "Failed to report metrics from static file provider");
                    }
                })
            }
        })
        .with_hook({
            let rocksdb = provider_factory.rocksdb_provider();
            move || throttle!(Duration::from_secs(5 * 60), || rocksdb.report_metrics())
        })
        .build()
}

fn get_partial_trie_unwind_marker(
    provider: &(impl MetadataProvider + StageCheckpointReader),
) -> ProviderResult<(Option<PartialStateTrieUnwindMarker>, bool)> {
    if let Some(marker) = provider.get_metadata(PARTIAL_STATE_TRIE_UNWIND_METADATA_KEY)? {
        let marker = serde_json::from_slice::<PartialStateTrieUnwindMarker>(&marker)
            .map_err(ProviderError::other)?;
        if marker.partial_state_trie >= marker.finish_block_number {
            return Err(ProviderError::other(std::io::Error::other(format!(
                "partial state trie unwind target #{} is not below original Finish #{}",
                marker.partial_state_trie, marker.finish_block_number,
            ))))
        }
        return Ok((Some(marker), true))
    }

    let Some(finish_checkpoint) = provider.get_stage_checkpoint(StageId::Finish)? else {
        return Ok((None, false))
    };
    let Some(partial_state_trie) =
        finish_checkpoint.finish_stage_checkpoint().and_then(|finish| finish.partial_state_trie())
    else {
        return Ok((None, false))
    };

    if partial_state_trie > finish_checkpoint.block_number {
        return Err(ProviderError::other(std::io::Error::other(format!(
            "partial state trie frontier #{partial_state_trie} is ahead of Finish #{}",
            finish_checkpoint.block_number,
        ))))
    }

    Ok((
        (partial_state_trie < finish_checkpoint.block_number).then_some(
            PartialStateTrieUnwindMarker {
                finish_block_number: finish_checkpoint.block_number,
                partial_state_trie,
            },
        ),
        false,
    ))
}

/// Metadata key for a partial state trie unwind that has not completed yet.
const PARTIAL_STATE_TRIE_UNWIND_METADATA_KEY: &str = "partial_state_trie_unwind";

fn write_partial_trie_unwind_marker(
    provider: &impl MetadataWriter,
    marker: PartialStateTrieUnwindMarker,
) -> ProviderResult<()> {
    provider.write_metadata(
        PARTIAL_STATE_TRIE_UNWIND_METADATA_KEY,
        serde_json::to_vec(&marker).map_err(ProviderError::other)?,
    )
}

fn delete_partial_trie_unwind_marker(provider: &impl MetadataWriter) -> ProviderResult<()> {
    provider.delete_metadata(PARTIAL_STATE_TRIE_UNWIND_METADATA_KEY)
}

#[cfg(test)]
mod tests {
    use super::{get_partial_trie_unwind_marker, LaunchContext, NodeConfig};
    use reth_config::Config;
    use reth_db_api::models::PartialStateTrieUnwindMarker;
    use reth_node_core::args::PruningArgs;
    use reth_provider::{MetadataProvider, ProviderResult, StageCheckpointReader};
    use reth_stages::{FinishCheckpoint, StageCheckpoint, StageId};

    const EXTENSION: &str = "toml";

    struct MockProvider(Option<Vec<u8>>, Option<StageCheckpoint>);

    impl MetadataProvider for MockProvider {
        fn get_metadata(&self, _: &str) -> ProviderResult<Option<Vec<u8>>> {
            Ok(self.0.clone())
        }
    }

    impl StageCheckpointReader for MockProvider {
        fn get_stage_checkpoint(&self, id: StageId) -> ProviderResult<Option<StageCheckpoint>> {
            assert_eq!(id, StageId::Finish);
            Ok(self.1)
        }

        fn get_stage_checkpoint_progress(&self, _: StageId) -> ProviderResult<Option<Vec<u8>>> {
            Ok(None)
        }

        fn get_all_checkpoints(&self) -> ProviderResult<Vec<(String, StageCheckpoint)>> {
            Ok(Vec::new())
        }
    }

    fn with_tempdir(filename: &str, proc: fn(&std::path::Path)) {
        let temp_dir = tempfile::tempdir().unwrap();
        let config_path = temp_dir.path().join(filename).with_extension(EXTENSION);
        proc(&config_path);
        temp_dir.close().unwrap()
    }

    #[test]
    fn test_save_prune_config() {
        with_tempdir("prune-store-test", |config_path| {
            let mut reth_config = Config::default();
            let node_config = NodeConfig {
                pruning: PruningArgs {
                    full: true,
                    minimal: false,
                    block_interval: None,
                    sender_recovery_full: false,
                    sender_recovery_distance: None,
                    sender_recovery_before: None,
                    transaction_lookup_full: false,
                    transaction_lookup_distance: None,
                    transaction_lookup_before: None,
                    receipts_full: false,
                    receipts_pre_merge: false,
                    receipts_distance: None,
                    receipts_before: None,
                    account_history_full: false,
                    account_history_distance: None,
                    account_history_before: None,
                    storage_history_full: false,
                    storage_history_distance: None,
                    storage_history_before: None,
                    bodies_pre_merge: false,
                    bodies_distance: None,
                    receipts_log_filter: None,
                    bodies_before: None,
                    minimum_distance: None,
                },
                ..NodeConfig::test()
            };
            LaunchContext::save_pruning_config(&mut reth_config, &node_config, config_path)
                .unwrap();

            let loaded_config = Config::from_path(config_path).unwrap();

            assert_eq!(reth_config, loaded_config);
        })
    }

    #[test]
    fn get_partial_trie_unwind_marker_uses_partial_finish_checkpoint() {
        let finish_checkpoint = StageCheckpoint::new(42)
            .with_finish_stage_checkpoint(FinishCheckpoint { partial_state_trie: Some(21) });
        let expected =
            finish_checkpoint.finish_stage_checkpoint().unwrap().partial_state_trie().map(
                |partial_state_trie| PartialStateTrieUnwindMarker {
                    finish_block_number: finish_checkpoint.block_number,
                    partial_state_trie,
                },
            );

        assert_eq!(
            get_partial_trie_unwind_marker(&MockProvider(None, Some(finish_checkpoint))).unwrap(),
            (expected, false)
        );

        let genesis_checkpoint = StageCheckpoint::new(42)
            .with_finish_stage_checkpoint(FinishCheckpoint { partial_state_trie: Some(0) });
        let expected =
            genesis_checkpoint.finish_stage_checkpoint().unwrap().partial_state_trie().map(
                |partial_state_trie| PartialStateTrieUnwindMarker {
                    finish_block_number: genesis_checkpoint.block_number,
                    partial_state_trie,
                },
            );

        assert_eq!(
            get_partial_trie_unwind_marker(&MockProvider(None, Some(genesis_checkpoint))).unwrap(),
            (expected, false)
        );
    }

    #[test]
    fn get_partial_trie_unwind_marker_resumes_persisted_unwind() {
        let marker =
            PartialStateTrieUnwindMarker { finish_block_number: 42, partial_state_trie: 21 };

        assert_eq!(
            get_partial_trie_unwind_marker(&MockProvider(
                Some(serde_json::to_vec(&marker).unwrap()),
                Some(StageCheckpoint::new(21)),
            ),)
            .unwrap(),
            (Some(marker), true)
        );
        assert_eq!(
            get_partial_trie_unwind_marker(&MockProvider(
                Some(serde_json::to_vec(&marker).unwrap()),
                None
            ),)
            .unwrap(),
            (Some(marker), true)
        );
    }

    #[test]
    fn get_partial_trie_unwind_marker_ignores_non_lagging_or_missing_partial_checkpoint() {
        let matching_finish_checkpoint = StageCheckpoint::new(42)
            .with_finish_stage_checkpoint(FinishCheckpoint { partial_state_trie: Some(42) });
        let ahead_finish_checkpoint = StageCheckpoint::new(42)
            .with_finish_stage_checkpoint(FinishCheckpoint { partial_state_trie: Some(43) });
        let missing_partial_finish_checkpoint = StageCheckpoint::new(42)
            .with_finish_stage_checkpoint(FinishCheckpoint { partial_state_trie: None });

        assert_eq!(
            get_partial_trie_unwind_marker(&MockProvider(None, Some(matching_finish_checkpoint)),)
                .unwrap(),
            (None, false)
        );
        assert_eq!(
            get_partial_trie_unwind_marker(&MockProvider(
                None,
                Some(missing_partial_finish_checkpoint)
            ),)
            .unwrap(),
            (None, false)
        );
        assert_eq!(
            get_partial_trie_unwind_marker(&MockProvider(None, None)).unwrap(),
            (None, false)
        );

        let partial_frontier = ahead_finish_checkpoint
            .finish_stage_checkpoint()
            .and_then(|finish| finish.partial_state_trie());
        let result =
            get_partial_trie_unwind_marker(&MockProvider(None, Some(ahead_finish_checkpoint)));
        if partial_frontier.is_some() {
            let error = result.unwrap_err();
            assert!(error.to_string().contains("ahead of Finish"), "unexpected error: {error}");
        } else {
            assert_eq!(result.unwrap(), (None, false));
        }
    }

    #[test]
    fn get_partial_trie_unwind_marker_rejects_invalid_persisted_marker() {
        let marker =
            PartialStateTrieUnwindMarker { finish_block_number: 42, partial_state_trie: 42 };
        let error = get_partial_trie_unwind_marker(&MockProvider(
            Some(serde_json::to_vec(&marker).unwrap()),
            None,
        ))
        .unwrap_err();

        assert!(error.to_string().contains("is not below original Finish"));
    }

    #[test]
    fn get_partial_trie_unwind_marker_rejects_malformed_metadata() {
        assert!(get_partial_trie_unwind_marker(&MockProvider(Some(vec![0xff]), None)).is_err());
    }
}