paradigmxyz/reth - CodSpeed

reth

Blog Docs Changelog

Performance History

Latest Results

fix(rocksdb): validate storage settings on database load Add validation at ProviderFactory creation to catch invalid configurations when loading an existing database, not just at genesis init. This ensures nodes with pre-existing invalid configurations (RocksDB history without changesets in static files) fail with a clear error message rather than experiencing subtle divergence bugs. Added InvalidStorageSettings error variant to ProviderError.

fix/rocksdb-history-divergence

3 minutes ago

perf(prune): use bulk table clear for PruneMode::Full When PruneMode::Full is configured for segments with min_blocks() == 0 (SenderRecovery, TransactionLookup), the pruner now clears the entire table in a single MDBX operation instead of iterating entry-by-entry. This provides O(1) table clearing vs O(n) row-by-row deletion, which is a significant speedup for tables with millions of entries. Changes: - Add clear_table<T>() method to DbTxPruneExt trait - SenderRecovery: use bulk clear for Full mode - TransactionLookup: use bulk clear for Full mode Closes #21296

alexey/bulk-prune-full-mode

8 minutes ago

perf(prune): use bulk table clear for PruneMode::Full When PruneMode::Full is configured for segments with min_blocks() == 0 (SenderRecovery, TransactionLookup), the pruner now clears the entire table in a single MDBX operation instead of iterating entry-by-entry. This provides O(1) table clearing vs O(n) row-by-row deletion, which is a significant speedup for tables with millions of entries. Changes: - Add clear_table<T>() method to DbTxPruneExt trait - SenderRecovery: use bulk clear for Full mode - TransactionLookup: use bulk clear for Full mode Closes #21296

gakonst:perf-bulk-prune-full

9 minutes ago

docs: expand sharding explanation with key structure

yk/dedup-history-shard-ops

9 minutes ago

fix(rocksdb): enforce history indices require changesets in static files Per Joshie's feedback, the proper fix for RocksDB history divergence is not changing commit order, but enforcing the architectural invariant: RocksDB history indices (AccountsHistory, StoragesHistory) are derived data that index into changesets. They can only be safely stored in RocksDB when changesets are in static files (crash-stable), not MDBX (can be unwound). Changes: - Revert commit order back to: static_file → RocksDB → MDBX - Add StorageSettings::validate() to enforce the constraint - Add validation at genesis init with clear error message - Update invariants.rs documentation - Add tests for validation logic The constraint: account_history_in_rocksdb OR storages_history_in_rocksdb requires account_changesets_in_static_files=true

fix/rocksdb-history-divergence

10 minutes ago

andrewshab3:redundant-clone

10 minutes ago

docs: explain what a shard is

yk/dedup-history-shard-ops

11 minutes ago

docs: add comments to flush_shards_partial

yk/dedup-history-shard-ops

13 minutes ago

Active Branches

perf(prune): use bulk table clear for PruneMode::Full

last run

8 minutes ago

0%

refactor(stages): deduplicate history shard loading with HistoryShardWriter trait

last run

9 minutes ago

0%

fix(rocksdb): correct history index divergence causing nonce errors after unwind

last run

10 minutes ago

0%

© 2026 CodSpeed Technology

Home Terms Privacy Docs