Skip to content

stagedsync: replace TxLookup prune full-table scan with block-based deletion #19975

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Closed
domiwei wants to merge 2 commits into from
Closed

stagedsync: replace TxLookup prune full-table scan with block-based deletion #19975

Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
1bcb935
stagedsync: replace TxLookup prune full-table scan with precise block…
domiwei Mar 18, 2026
84efdb7
stagedsync: cap TxLookup prune to 100 blocks per cycle at chain tip
domiwei Mar 18, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
80 changes: 18 additions & 62 deletions execution/stagedsync/stage_txlookup.go
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -20,20 +20,17 @@ import (
"context"
"encoding/binary"
"fmt"
"time"

"github.com/erigontech/erigon/common"
"github.com/erigontech/erigon/common/dbg"
"github.com/erigontech/erigon/common/hexutil"
"github.com/erigontech/erigon/common/log/v3"
"github.com/erigontech/erigon/db/etl"
"github.com/erigontech/erigon/db/kv"
mdbx2 "github.com/erigontech/erigon/db/kv/mdbx"
"github.com/erigontech/erigon/db/kv/prune"
"github.com/erigontech/erigon/db/kv/rawdbv3"
"github.com/erigontech/erigon/db/rawdb"
"github.com/erigontech/erigon/db/services"
"github.com/erigontech/erigon/db/state"
"github.com/erigontech/erigon/execution/stagedsync/stages"
"github.com/erigontech/erigon/execution/types"
)
Expand Down Expand Up @@ -227,73 +224,32 @@ func PruneTxLookup(s *PruneState, tx kv.RwTx, cfg TxLookupCfg, ctx context.Conte
blockTo = cfg.blockReader.CanPruneTo(s.ForwardProgress)
}

if blockFrom >= blockTo {
return nil
}

txNumReader := cfg.blockReader.TxnumReader()
txFrom, err := txNumReader.Min(ctx, tx, blockFrom)
if err != nil {
return fmt.Errorf("txNumReader.Min(%d): %w", blockFrom, err)
}
txTo, err := txNumReader.Max(ctx, tx, blockTo)
if err != nil {
return fmt.Errorf("txNumReader.Max(%d): %w", blockTo, err)
}
if txFrom >= txTo {
return nil
// At chain tip, cap the number of blocks pruned per cycle to bound the
// time spent inside a single DB transaction. Each block adds ~150 random
// key deletions to TxLookup, so 100 blocks ≈ 15 000 deletes — a few
// tens of milliseconds. During initial sync there is no slot budget
// pressure, so we let it run unrestricted.
if !s.CurrentSyncCycle.IsInitialCycle {
const maxBlocksPerCycle = 100
blockTo = min(blockTo, blockFrom+maxBlocksPerCycle)
}

prevStat, err := state.GetPruneValProgress(tx, []byte(kv.TxLookup))
if err != nil {
return err
}
if prevStat != nil && prevStat.TxFrom == txFrom && prevStat.TxTo == txTo && prevStat.ValueProgress == prune.Done {
if blockFrom >= blockTo {
return nil
}
if prevStat == nil {
prevStat = &prune.Stat{}
}

valsRwCursor, err := tx.RwCursor(kv.TxLookup)
if err != nil {
return fmt.Errorf("create TxLookup cursor: %w", err)
}
defer valsRwCursor.Close()
var valsCursor kv.PseudoDupSortRwCursor
switch c := valsRwCursor.(type) {
case *mdbx2.MdbxCursor:
valsCursor = &mdbx2.MdbxCursorPseudoDupSort{MdbxCursor: c}
default:
return fmt.Errorf("unexpected cursor type %T for table %s", valsRwCursor, kv.TxLookup)
}

logEvery := time.NewTicker(logInterval)
defer logEvery.Stop()

pruneTimeout := 2 * time.Second
if s.CurrentSyncCycle.IsInitialCycle {
pruneTimeout = time.Hour
}
pruneCtx, pruneCancel := context.WithTimeout(ctx, pruneTimeout)
defer pruneCancel()

pruneStat, err := prune.TableScanningPrune(pruneCtx, logPrefix, "txlookup", txFrom, txTo, 0, 1,
logEvery, logger, nil, valsCursor, false, prevStat, prune.ValueOffset8StorageMode)
if err != nil {
// Use precise block-based deletion instead of full table scan.
// TxLookup keys are tx hashes (random), so scanning the entire table to find
// entries within a txNum range is O(table_size). Instead, we iterate
// HeaderCanonical for the block range, look up each block's transactions,
// and delete their TxLookup entries directly — O(txs_to_delete).
// etl.Transform uses ExtractEndKey as exclusive bound, therefore blockTo + 1
if err := deleteTxLookupRange(tx, logPrefix, blockFrom, blockTo+1, ctx, cfg, logger); err != nil {
return fmt.Errorf("prune TxLookup: %w", err)
}
defer func() {
pruneStat.TxFrom, pruneStat.TxTo = txFrom, txTo
if e := state.SavePruneValProgress(tx, kv.TxLookup, pruneStat); e != nil {
logger.Error("[snapshots] save prune val progress", "name", "txlookup", "err", e)
}
}()

if pruneStat.ValueProgress == prune.Done {
if err = s.DoneAt(tx, blockTo); err != nil {
return err
}
if err = s.DoneAt(tx, blockTo); err != nil {
return err
}

return nil
Expand Down
Loading