feat(state-prune): support tail replay on existing output#3991
Merged
Conversation
|
The latest updates on your projects. Learn more about Vercel for GitHub.
1 Skipped Deployment
|
Dependency Review✅ No vulnerabilities or license issues or OpenSSF Scorecard issues found.Scanned Files
|
Contributor
There was a problem hiding this comment.
Pull request overview
Adds operational tooling and plumbing for mainnet repair/recovery flows by enabling (1) tail replay against an existing replay output directory, (2) a finalize-only path for partially completed outputs, (3) optional skipping of final state-node compaction, and (4) a helper to backfill missing indexed L2 transaction history into a destination DB without initializing the destination indexer.
Changes:
- Extend
IncrementalReplayerwithtail_replay_existing_output(...)andfinalize_existing_output(...), plus askip_final_compactreplay option. - Wire new
db state-prune tail-replay/finalize-replay-outputCLI flows and adddb import-indexed-transactions. - Add design docs describing tail replay and a selective-CF-copy replay redesign.
Reviewed changes
Copilot reviewed 11 out of 12 changed files in this pull request and generated 5 comments.
Show a summary per file
| File | Description |
|---|---|
| docs/dev-guide/mainnet_tail_replay_design_20260408.md | Tail replay design doc and proposed CLI/steps. |
| docs/dev-guide/mainnet_replay_cf_copy_redesign_20260407.md | Design doc for reducing replay output size via selective CF copy. |
| crates/rooch/src/commands/db/mod.rs | Registers the new ImportIndexedTransactions DB subcommand. |
| crates/rooch/src/commands/db/commands/state_prune/replay.rs | Adds TailReplayCommand and FinalizeReplayOutputCommand; exposes skip_final_compact. |
| crates/rooch/src/commands/db/commands/state_prune/command.rs | Adds new state-prune subcommands to the CLI enum dispatch. |
| crates/rooch/src/commands/db/commands/mod.rs | Exposes the new import_indexed_transactions command module. |
| crates/rooch/src/commands/db/commands/import_indexed_transactions.rs | Implements import/backfill of missing L2 tx + execution info referenced by the target indexer. |
| crates/rooch/Cargo.toml | Adds diesel dependency needed by the new import command. |
| crates/rooch-pruner/src/state_prune/incremental_replayer.rs | Implements tail replay + finalize flows, accumulator tail rebuild, and skip-final-compact support. |
| crates/rooch-pruner/Cargo.toml | Adds accumulator dependency for tail accumulator rebuild. |
| crates/rooch-config/src/state_prune.rs | Adds skip_final_compact to ReplayConfig. |
| Cargo.lock | Locks new dependencies (diesel, accumulator) into the build. |
Comment on lines
+27
to
+29
| - [replay.rs](/Users/jolestar/opensource/src/github.com/rooch-network/rooch/crates/rooch/src/commands/db/commands/state_prune/replay.rs) | ||
| - [incremental_replayer.rs](/Users/jolestar/opensource/src/github.com/rooch-network/rooch/crates/rooch-pruner/src/state_prune/incremental_replayer.rs) | ||
|
|
There was a problem hiding this comment.
These markdown links point to an absolute local filesystem path (/Users/...). That won’t work for other developers or in GitHub; please switch to repo-relative links (e.g., crates/rooch/src/... or a GitHub permalink).
Comment on lines
+276
to
+303
| let mut metadata = StatePruneMetadata::new( | ||
| crate::state_prune::OperationType::Replay { | ||
| snapshot_path: PathBuf::new(), | ||
| from_order: from_order.unwrap_or(0), | ||
| to_order, | ||
| output_dir: output_dir.to_path_buf(), | ||
| }, | ||
| serde_json::json!({ | ||
| "mode": "tail_replay_existing_output", | ||
| "from_order": from_order, | ||
| "to_order": to_order, | ||
| "output_dir": output_dir, | ||
| "config": self.config | ||
| }), | ||
| ); | ||
|
|
||
| let source_moveos_store = self.source_moveos_store()?; | ||
| let (output_store, output_rooch_store) = self.load_output_stores(output_dir)?; | ||
| let output_startup_info = output_store | ||
| .get_config_store() | ||
| .get_startup_info()? | ||
| .ok_or_else(|| anyhow::anyhow!("No startup info found in existing replay output"))?; | ||
| let output_sequencer_info = output_rooch_store | ||
| .get_meta_store() | ||
| .get_sequencer_info()? | ||
| .ok_or_else(|| anyhow::anyhow!("No sequencer info found in existing replay output"))?; | ||
|
|
||
| let resolved_from_order = from_order.unwrap_or(output_sequencer_info.last_order + 1); |
There was a problem hiding this comment.
StatePruneMetadata::new is initialized with from_order: from_order.unwrap_or(0), but the actual start order is resolved later from the existing output’s sequencer_info. This makes the recorded operation type/metadata inaccurate (and can produce misleading filenames/diagnostics). Consider resolving from_order first (or updating metadata.operation_type) so it records the effective resolved_from_order.
Suggested change
| let mut metadata = StatePruneMetadata::new( | |
| crate::state_prune::OperationType::Replay { | |
| snapshot_path: PathBuf::new(), | |
| from_order: from_order.unwrap_or(0), | |
| to_order, | |
| output_dir: output_dir.to_path_buf(), | |
| }, | |
| serde_json::json!({ | |
| "mode": "tail_replay_existing_output", | |
| "from_order": from_order, | |
| "to_order": to_order, | |
| "output_dir": output_dir, | |
| "config": self.config | |
| }), | |
| ); | |
| let source_moveos_store = self.source_moveos_store()?; | |
| let (output_store, output_rooch_store) = self.load_output_stores(output_dir)?; | |
| let output_startup_info = output_store | |
| .get_config_store() | |
| .get_startup_info()? | |
| .ok_or_else(|| anyhow::anyhow!("No startup info found in existing replay output"))?; | |
| let output_sequencer_info = output_rooch_store | |
| .get_meta_store() | |
| .get_sequencer_info()? | |
| .ok_or_else(|| anyhow::anyhow!("No sequencer info found in existing replay output"))?; | |
| let resolved_from_order = from_order.unwrap_or(output_sequencer_info.last_order + 1); | |
| let source_moveos_store = self.source_moveos_store()?; | |
| let (output_store, output_rooch_store) = self.load_output_stores(output_dir)?; | |
| let output_startup_info = output_store | |
| .get_config_store() | |
| .get_startup_info()? | |
| .ok_or_else(|| anyhow::anyhow!("No startup info found in existing replay output"))?; | |
| let output_sequencer_info = output_rooch_store | |
| .get_meta_store() | |
| .get_sequencer_info()? | |
| .ok_or_else(|| anyhow::anyhow!("No sequencer info found in existing replay output"))?; | |
| let resolved_from_order = from_order.unwrap_or(output_sequencer_info.last_order + 1); | |
| let mut metadata = StatePruneMetadata::new( | |
| crate::state_prune::OperationType::Replay { | |
| snapshot_path: PathBuf::new(), | |
| from_order: resolved_from_order, | |
| to_order, | |
| output_dir: output_dir.to_path_buf(), | |
| }, | |
| serde_json::json!({ | |
| "mode": "tail_replay_existing_output", | |
| "from_order": from_order, | |
| "resolved_from_order": resolved_from_order, | |
| "to_order": to_order, | |
| "output_dir": output_dir, | |
| "config": self.config | |
| }), | |
| ); |
Comment on lines
+322
to
+350
| metadata.mark_in_progress( | ||
| format!( | ||
| "Loading canonical tail entries [{}..={}]", | ||
| resolved_from_order, to_order | ||
| ), | ||
| 20.0, | ||
| ); | ||
| let tail_entries = self.load_tail_entries( | ||
| &source_moveos_store, | ||
| resolved_from_order, | ||
| to_order, | ||
| &mut report, | ||
| )?; | ||
| self.progress_tracker.set_total(tail_entries.len() as u64); | ||
|
|
||
| metadata.mark_in_progress("Applying tail changesets".to_string(), 40.0); | ||
| let (actual_state_root, expected_state_root, expected_global_size) = self | ||
| .tail_replay_entries_batched( | ||
| tail_entries, | ||
| &output_store, | ||
| &output_rooch_store, | ||
| output_startup_info.state_root, | ||
| output_startup_info.size, | ||
| output_sequencer_info.last_accumulator_info.clone(), | ||
| &mut report, | ||
| &mut metadata, | ||
| ) | ||
| .await?; | ||
|
|
There was a problem hiding this comment.
load_tail_entries materializes the entire [resolved_from_order..=to_order] range into memory before any apply happens. For large tail ranges, this can cause high peak memory usage and longer time-to-first-progress. Consider processing in streaming/chunked fashion (load a batch of changesets + tx metadata, apply, then drop) instead of building a single Vec<TailReplayEntry>.
Suggested change
| metadata.mark_in_progress( | |
| format!( | |
| "Loading canonical tail entries [{}..={}]", | |
| resolved_from_order, to_order | |
| ), | |
| 20.0, | |
| ); | |
| let tail_entries = self.load_tail_entries( | |
| &source_moveos_store, | |
| resolved_from_order, | |
| to_order, | |
| &mut report, | |
| )?; | |
| self.progress_tracker.set_total(tail_entries.len() as u64); | |
| metadata.mark_in_progress("Applying tail changesets".to_string(), 40.0); | |
| let (actual_state_root, expected_state_root, expected_global_size) = self | |
| .tail_replay_entries_batched( | |
| tail_entries, | |
| &output_store, | |
| &output_rooch_store, | |
| output_startup_info.state_root, | |
| output_startup_info.size, | |
| output_sequencer_info.last_accumulator_info.clone(), | |
| &mut report, | |
| &mut metadata, | |
| ) | |
| .await?; | |
| let total_tail_entries = to_order - resolved_from_order + 1; | |
| let tail_chunk_size: u64 = 1_000; | |
| self.progress_tracker.set_total(total_tail_entries); | |
| metadata.mark_in_progress("Applying tail changesets".to_string(), 40.0); | |
| let mut current_from_order = resolved_from_order; | |
| let mut current_state_root = output_startup_info.state_root; | |
| let mut current_global_size = output_startup_info.size; | |
| let mut current_accumulator_info = output_sequencer_info.last_accumulator_info.clone(); | |
| let mut actual_state_root = current_state_root; | |
| let mut expected_state_root = current_state_root; | |
| let mut expected_global_size = current_global_size; | |
| while current_from_order <= to_order { | |
| let current_to_order = | |
| std::cmp::min(current_from_order + tail_chunk_size - 1, to_order); | |
| metadata.mark_in_progress( | |
| format!( | |
| "Loading and applying canonical tail entries [{}..={}]", | |
| current_from_order, current_to_order | |
| ), | |
| 40.0, | |
| ); | |
| let tail_entries = self.load_tail_entries( | |
| &source_moveos_store, | |
| current_from_order, | |
| current_to_order, | |
| &mut report, | |
| )?; | |
| (actual_state_root, expected_state_root, expected_global_size) = self | |
| .tail_replay_entries_batched( | |
| tail_entries, | |
| &output_store, | |
| &output_rooch_store, | |
| current_state_root, | |
| current_global_size, | |
| current_accumulator_info.clone(), | |
| &mut report, | |
| &mut metadata, | |
| ) | |
| .await?; | |
| current_state_root = actual_state_root; | |
| current_global_size = expected_global_size; | |
| current_accumulator_info = output_rooch_store | |
| .get_meta_store() | |
| .get_sequencer_info()? | |
| .ok_or_else(|| { | |
| anyhow::anyhow!("No sequencer info found after tail replay chunk") | |
| })? | |
| .last_accumulator_info; | |
| current_from_order = current_to_order + 1; | |
| } |
Comment on lines
+500
to
+508
| let mut entries = Vec::with_capacity(changesets.len()); | ||
|
|
||
| for (tx_order, changeset_ext) in changesets { | ||
| let tx_hash = self | ||
| .rooch_store | ||
| .transaction_store | ||
| .get_tx_hashes(vec![tx_order])? | ||
| .pop() | ||
| .flatten() |
There was a problem hiding this comment.
Inside the loop, get_tx_hashes(vec![tx_order]) issues a separate DB query per order (N+1) and allocates a Vec each time. Since you already know the full order range, consider fetching tx hashes in bulk (e.g., one get_tx_hashes call for the whole range or per-batch) and then multi-get transactions / execution infos to reduce IO overhead.
Suggested change
| let mut entries = Vec::with_capacity(changesets.len()); | |
| for (tx_order, changeset_ext) in changesets { | |
| let tx_hash = self | |
| .rooch_store | |
| .transaction_store | |
| .get_tx_hashes(vec![tx_order])? | |
| .pop() | |
| .flatten() | |
| let tx_orders: Vec<u64> = changesets | |
| .iter() | |
| .map(|(tx_order, _)| *tx_order) | |
| .collect(); | |
| let tx_hashes = self.rooch_store.transaction_store.get_tx_hashes(tx_orders)?; | |
| if tx_hashes.len() != changesets.len() { | |
| return Err(anyhow::anyhow!( | |
| "Mismatched tx hash count for range {}..={}: expected {}, got {}", | |
| from_order, | |
| to_order, | |
| changesets.len(), | |
| tx_hashes.len() | |
| )); | |
| } | |
| let mut entries = Vec::with_capacity(changesets.len()); | |
| for ((tx_order, changeset_ext), tx_hash_opt) in changesets.into_iter().zip(tx_hashes.into_iter()) | |
| { | |
| let tx_hash = tx_hash_opt |
Comment on lines
+108
to
+112
| let tx_db_path = target_indexer_dir.join("transactions"); | ||
| let tx_db_url = tx_db_path | ||
| .to_str() | ||
| .ok_or_else(|| anyhow::anyhow!("Invalid target transactions indexer path"))?; | ||
| let mut conn = SqliteConnection::establish(tx_db_url) |
There was a problem hiding this comment.
load_indexed_transaction_batch opens a new SqliteConnection on every call, and the import loop calls this once per batch. Reusing a single connection (or using Diesel’s existing connection pool helper like the indexer does) would avoid repeated connection setup and improve import throughput for large backfills.
jolestar
force-pushed
the
feat/state-prune-tail-replay
branch
from
June 16, 2026 09:32
177eb00 to
b2e94bd
Compare
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
finalize-replay-outputflowIncluded changes
IncrementalReplayercan now finalize an existing output store without reopening the same RocksDB handle in-placedb state-pruneadds the existing-output tail replay/finalize pathdb import-indexed-transactionsimports missing indexed L2 transactions and avoids initializing the target indexer on the destination DBTesting