use std::{path::Path, sync::Arc}; use anyhow::{Context as _, Result}; use arc_swap::ArcSwap; use common::block::Block; // TODO: Remove after testnet use futures::StreamExt as _; pub use ingest_error::BlockIngestError; use log::{error, info, warn}; use logos_blockchain_zone_sdk::{ CommonHttpClient, Slot, ZoneMessage, adapter::NodeHttpClient, indexer::ZoneIndexer, }; pub use stall_reason::StallReason; use crate::{ block_store::{AcceptOutcome, IndexerStore}, config::IndexerConfig, status::{IndexerStatus, IndexerSyncStatus}, }; pub mod block_store; pub mod config; pub mod ingest_error; pub mod stall_reason; pub mod status; /// Result of comparing the indexer's stored tip against the channel. enum ChainIdentityOutcome { /// Proceed from the persisted cursor. Either the channel still serves our tip /// (same chain), there is nothing to compare (empty store / parked / no cursor), /// or the check was inconclusive (L1 unreadable, or bedrock's LIB still behind /// our tip slot) — none of which prove a reset. Consistent, /// The channel serves a *different* block at the tip's id — a chain reset. Mismatch { detail: String }, } #[derive(Clone)] pub struct IndexerCore { pub zone_indexer: Arc>, pub config: IndexerConfig, pub store: IndexerStore, /// Live ingestion status; updated by the ingest stream, read by `status`. pub status: Arc>, } impl IndexerCore { pub fn new(config: IndexerConfig, storage_dir: &Path) -> Result { // Namespace the DB by channel so indexers on different channels can // share a storage dir without their RocksDB state colliding. let home = storage_dir.join(format!("rocksdb-{}", config.channel_id)); let basic_auth = config.bedrock_config.auth.clone().map(Into::into); let node = NodeHttpClient::new( CommonHttpClient::new(basic_auth), config.bedrock_config.addr.clone(), ); let zone_indexer = ZoneIndexer::new(config.channel_id, node); Ok(Self { zone_indexer: Arc::new(zone_indexer), config, store: IndexerStore::open_db(&home)?, status: Arc::new(ArcSwap::from_pointee(IndexerSyncStatus::starting())), }) } /// Builds the core, then verifies the stored chain matches the channel's by /// re-reading the channel at the stored tip's position. On mismatch: refuse /// (error) unless `allow_reset`, in which case wipe the store and re-index from /// scratch. Used at service/FFI startup in place of `new`. pub async fn new_with_chain_check( config: IndexerConfig, storage_dir: &Path, allow_reset: bool, ) -> Result { let home = storage_dir.join(format!("rocksdb-{}", config.channel_id)); let core = Self::new(config.clone(), storage_dir)?; match core.chain_identity_outcome().await? { ChainIdentityOutcome::Consistent => Ok(core), ChainIdentityOutcome::Mismatch { detail } if allow_reset => { warn!( "Chain reset detected ({detail}). Wiping indexer store at {} and re-indexing.", home.display() ); drop(core); // sole owner before the ingest task is spawned → closes the DB storage::indexer::RocksDBIO::destroy(&home)?; Self::new(config, storage_dir) } ChainIdentityOutcome::Mismatch { detail } => Err(anyhow::anyhow!( "Indexer store at {} holds a different chain than the channel now serves \ ({detail}). Run `just clean`, point at a fresh storage dir, or set \ `allow_chain_reset` in the indexer config.", home.display() )), } } /// Verifies the channel still serves our chain at the tip's L1 slot (the persisted /// cursor), by comparing the first block it serves there against our stored block /// of the *same id*. /// /// This is mostly a development convenience: it lets the indexer notice that its /// local state belongs to a different chain than the one it is now connected to /// (e.g. a wiped/restarted bedrock) and reset instead of silently diverging. It /// will not trigger during normal live indexing. Reading at the cursor — which is /// recent — keeps this to ~one batch rather than a scan from genesis (L1 slots /// are wall-clock-derived, so genesis is millions of empty slots away). async fn chain_identity_outcome(&self) -> Result { // We deliberately do NOT skip parked stores: a parked store must still be able // to detect a reset, or it stays parked across reboots forever (its persisted // stall would short-circuit the check every boot). A same-chain park is still // safe — the parked block sits at an id we never applied, so the lookup below // misses and we stay Consistent. let Some(cursor) = self.store.get_zone_cursor()? else { return Ok(ChainIdentityOutcome::Consistent); // empty / cold store: nothing to verify }; // Compare the first block the channel serves at/after the cursor against our // stored block of the same id. On the same chain that is our tip and matches; // a reset serves a different block here — crucially including a freshly // restarted, *shorter* chain whose low-id block at this slot differs from ours // (the old "look for our tip id" approach missed this: a short chain has no // block at our tip id). // // Anything inconclusive stays Consistent (proceed, let ingest park if truly // divergent) rather than wiping a valid store: an empty/unreadable read (most // importantly bedrock's LIB still behind our tip slot), or a channel block at // an id we don't hold. Blind spot: a store holding only genesis can't be // distinguished (genesis is deterministic across chains), but that window is // transient. let Some(channel_block) = self.fetch_channel_block_from(cursor).await? else { return Ok(ChainIdentityOutcome::Consistent); }; let Some(ours) = self.store.get_block_at_id(channel_block.header.block_id)? else { return Ok(ChainIdentityOutcome::Consistent); }; Ok(compare_block(&ours, &channel_block)) } /// Reads the channel starting at the tip's L1 slot (the `cursor`) and returns the /// first block it serves there. `next_messages` is exclusive of its argument, so /// `cursor - 1` is passed to include the tip's own slot. /// /// Cheap: the cursor is recent, so this reads roughly one batch. `None` covers the /// inconclusive cases — a slow/unreachable L1 (timeout/error) or bedrock's LIB /// still behind our tip slot (empty stream) — neither of which proves a reset. async fn fetch_channel_block_from(&self, cursor: Slot) -> Result> { const TIP_FETCH_TIMEOUT: std::time::Duration = std::time::Duration::from_secs(30); // A slot-0 cursor is degenerate (real inscriptions live at wall-clock slots); // bail rather than let `next_messages(None)` fall back to a from-genesis scan. let Some(from_slot) = cursor.into_inner().checked_sub(1) else { return Ok(None); }; let fetch = async { let stream = self .zone_indexer .next_messages(Some(Slot::from(from_slot))) .await?; let mut stream = std::pin::pin!(stream); while let Some((msg, _slot)) = stream.next().await { if let ZoneMessage::Block(zone_block) = msg { let block: Block = borsh::from_slice(&zone_block.data) .context("Failed to deserialize channel block")?; return Ok::, anyhow::Error>(Some(block)); } } Ok(None) }; match tokio::time::timeout(TIP_FETCH_TIMEOUT, fetch).await { Ok(Ok(found)) => Ok(found), Ok(Err(err)) => { warn!("Failed to read channel tip for the consistency check; proceeding: {err:#}"); Ok(None) } Err(_elapsed) => { warn!("Timed out reading channel tip for the consistency check; proceeding"); Ok(None) } } } /// Snapshot of the current ingestion status (sync state + indexed tip). /// /// Combines the ingest loop's live status with the L2 tip read fresh from the /// store, so callers (FFI/RPC) can tell "catching up" from "failed". #[must_use] pub fn status(&self) -> IndexerStatus { let sync = IndexerSyncStatus::clone(&self.status.load()); let indexed_block_id = self.store.get_last_block_id().ok().flatten(); let stall_reason = self.store.get_stall_reason().ok().flatten(); IndexerStatus { sync, indexed_block_id, stall_reason, } } /// Atomically publish a new ingestion status for readers of `status`. fn set_status(&self, status: IndexerSyncStatus) { self.status.store(Arc::new(status)); } pub fn subscribe_parse_block_stream(&self) -> impl futures::Stream> + '_ { let poll_interval = self.config.consensus_info_polling_interval; let initial_cursor = self .store .get_zone_cursor() .expect("Failed to load zone-sdk indexer cursor"); async_stream::stream! { let mut cursor = initial_cursor; if cursor.is_some() { info!("Resuming indexer from cursor {cursor:?}"); } else { info!("Starting indexer from beginning of channel"); } loop { let stream = match self.zone_indexer.next_messages(cursor).await { Ok(s) => s, Err(err) => { error!("Failed to start zone-sdk next_messages stream: {err}"); self.set_status(IndexerSyncStatus::error(format!( "cannot reach L1 / read channel: {err}" ))); tokio::time::sleep(poll_interval).await; continue; } }; let mut stream = std::pin::pin!(stream); let mut announced_syncing = false; let mut had_cycle_error = false; while let Some((msg, slot)) = stream.next().await { if !announced_syncing { self.set_status(IndexerSyncStatus::syncing()); announced_syncing = true; } let zone_block = match msg { ZoneMessage::Block(b) => b, ZoneMessage::Deposit(_) | ZoneMessage::Withdraw(_) => continue, }; let l1_slot = serde_json::to_value(slot).unwrap_or(serde_json::Value::Null); let block: Block = match borsh::from_slice(&zone_block.data) { Ok(b) => b, Err(e) => { error!("Failed to deserialize L2 block from zone-sdk: {e}"); if let Err(err) = self.store.record_deserialize_stall(l1_slot, e.to_string()) { warn!("Failed to record stall reason: {err:#}"); } self.set_status(IndexerSyncStatus::stalled(format!( "failed to deserialize L2 block: {e}" ))); // Advance the L1 read cursor past the broken inscription; // the validated tip stays frozen. cursor = Some(slot); if let Err(err) = self.store.set_zone_cursor(&slot) { warn!("Failed to persist indexer cursor: {err:#}"); } continue; } }; match self.store.accept_block(&block, l1_slot).await { Ok(AcceptOutcome::Applied) => { info!("Indexed L2 block {}", block.header.block_id); self.set_status(IndexerSyncStatus::syncing()); cursor = Some(slot); if let Err(err) = self.store.set_zone_cursor(&slot) { warn!("Failed to persist indexer cursor: {err:#}"); } yield Ok(block); } Ok(AcceptOutcome::Parked(ingest_err)) => { error!( "Parked at block {}: {ingest_err}", block.header.block_id ); self.set_status(IndexerSyncStatus::stalled(ingest_err.to_string())); // Advance the L1 read cursor; tip stays frozen, no yield. cursor = Some(slot); if let Err(err) = self.store.set_zone_cursor(&slot) { warn!("Failed to persist indexer cursor: {err:#}"); } } Err(err) => { // Infrastructure error (DB read/write), not a bad block. // Keep the cursor put; re-poll the same position next cycle. error!( "Store error applying block {}: {err:#}", block.header.block_id ); self.set_status(IndexerSyncStatus::error(format!( "store error: {err:#}" ))); had_cycle_error = true; break; } } } if had_cycle_error { tokio::time::sleep(poll_interval).await; continue; } // Stream drained. Stay Stalled if parked; otherwise we are caught up. if self.store.get_stall_reason().ok().flatten().is_none() { self.set_status(IndexerSyncStatus::caught_up()); } tokio::time::sleep(poll_interval).await; } } } } /// Pure comparison of our stored block against the channel's block at the same id: /// a mismatch is differing hashes. Missing/unreadable cases are handled upstream. fn compare_block(ours: &Block, channel: &Block) -> ChainIdentityOutcome { if ours.header.hash == channel.header.hash { ChainIdentityOutcome::Consistent } else { ChainIdentityOutcome::Mismatch { detail: format!( "stored block {} {} != channel block {} {}", ours.header.block_id, ours.header.hash, channel.header.block_id, channel.header.hash ), } } } #[cfg(test)] mod chain_identity_tests { use common::{HashType, block::Block, test_utils::produce_dummy_block}; use super::{ChainIdentityOutcome, compare_block}; fn block_with_prev(prev_seed: u8) -> Block { produce_dummy_block(5, Some(HashType([prev_seed; 32])), vec![]) } #[test] fn matching_block_is_consistent() { let b = block_with_prev(1); assert!(matches!( compare_block(&b, &b), ChainIdentityOutcome::Consistent )); } #[test] fn differing_block_is_mismatch() { let stored = block_with_prev(1); let current = block_with_prev(2); assert!(matches!( compare_block(&stored, ¤t), ChainIdentityOutcome::Mismatch { .. } )); } }