Add feature to TXM to detect and purge stuck transactions

.changeset/red-eagles-cry.md

@@ -0,0 +1,5 @@
+---
+"chainlink": minor
+---
+
+#added Added an auto-purge feature to the EVM TXM that identifies terminally stuck transactions either through a chain specific method or heurisitic then purges them to unblock the nonce. Included 4 new toml configs under Transactions.AutoPurge to configure this new feature: Enabled, Threshold, MinAttempts, and DetectionApiUrl.

common/config/chaintype.go

@@ -20,6 +20,7 @@ const (
 	ChainWeMix           ChainType = "wemix"
 	ChainXDai            ChainType = "xdai" // Deprecated: use ChainGnosis instead
 	ChainXLayer          ChainType = "xlayer"
+	ChainZkEvm           ChainType = "zkevm"
 	ChainZkSync          ChainType = "zksync"
 )
 
@@ -33,13 +34,14 @@ var ErrInvalidChainType = fmt.Errorf("must be one of %s or omitted", strings.Joi
 	string(ChainScroll),
 	string(ChainWeMix),
 	string(ChainXLayer),
+	string(ChainZkEvm),
 	string(ChainZkSync),
 }, ", "))
 
 // IsValid returns true if the ChainType value is known or empty.
 func (c ChainType) IsValid() bool {
 	switch c {
-	case "", ChainArbitrum, ChainCelo, ChainGnosis, ChainKroma, ChainMetis, ChainOptimismBedrock, ChainScroll, ChainWeMix, ChainXDai, ChainXLayer, ChainZkSync:
+	case "", ChainArbitrum, ChainCelo, ChainGnosis, ChainKroma, ChainMetis, ChainOptimismBedrock, ChainScroll, ChainWeMix, ChainXDai, ChainXLayer, ChainZkEvm, ChainZkSync:
 		return true
 	}
 	return false

common/txmgr/confirmer.go

@@ -122,12 +122,13 @@ type Confirmer[
 	lggr    logger.SugaredLogger
 	client  txmgrtypes.TxmClient[CHAIN_ID, ADDR, TX_HASH, BLOCK_HASH, R, SEQ, FEE]
 	txmgrtypes.TxAttemptBuilder[CHAIN_ID, HEAD, ADDR, TX_HASH, BLOCK_HASH, SEQ, FEE]
-	resumeCallback ResumeCallback
-	chainConfig    txmgrtypes.ConfirmerChainConfig
-	feeConfig      txmgrtypes.ConfirmerFeeConfig
-	txConfig       txmgrtypes.ConfirmerTransactionsConfig
-	dbConfig       txmgrtypes.ConfirmerDatabaseConfig
-	chainID        CHAIN_ID
+	stuckTxDetector txmgrtypes.StuckTxDetector[CHAIN_ID, ADDR, TX_HASH, BLOCK_HASH, SEQ, FEE]
+	resumeCallback  ResumeCallback
+	chainConfig     txmgrtypes.ConfirmerChainConfig
+	feeConfig       txmgrtypes.ConfirmerFeeConfig
+	txConfig        txmgrtypes.ConfirmerTransactionsConfig
+	dbConfig        txmgrtypes.ConfirmerDatabaseConfig
+	chainID         CHAIN_ID
 
 	ks               txmgrtypes.KeyStore[ADDR, CHAIN_ID, SEQ]
 	enabledAddresses []ADDR
@@ -163,6 +164,7 @@ func NewConfirmer[
 	txAttemptBuilder txmgrtypes.TxAttemptBuilder[CHAIN_ID, HEAD, ADDR, TX_HASH, BLOCK_HASH, SEQ, FEE],
 	lggr logger.Logger,
 	isReceiptNil func(R) bool,
+	stuckTxDetector txmgrtypes.StuckTxDetector[CHAIN_ID, ADDR, TX_HASH, BLOCK_HASH, SEQ, FEE],
 ) *Confirmer[CHAIN_ID, HEAD, ADDR, TX_HASH, BLOCK_HASH, R, SEQ, FEE] {
 	lggr = logger.Named(lggr, "Confirmer")
 	return &Confirmer[CHAIN_ID, HEAD, ADDR, TX_HASH, BLOCK_HASH, R, SEQ, FEE]{
@@ -179,6 +181,7 @@ func NewConfirmer[
 		ks:               keystore,
 		mb:               mailbox.NewSingle[HEAD](),
 		isReceiptNil:     isReceiptNil,
+		stuckTxDetector:  stuckTxDetector,
 	}
 }
 
@@ -206,6 +209,9 @@ func (ec *Confirmer[CHAIN_ID, HEAD, ADDR, TX_HASH, BLOCK_HASH, R, SEQ, FEE]) sta
 	if err != nil {
 		return fmt.Errorf("Confirmer: failed to load EnabledAddressesForChain: %w", err)
 	}
+	if err = ec.stuckTxDetector.LoadPurgeBlockNumMap(ctx, ec.enabledAddresses); err != nil {
+		ec.lggr.Debugf("Confirmer: failed to load the last purged block num for enabled addresses. Process can continue as normal but purge rate limiting may be affected.")
+	}
 
 	ec.ctx, ec.ctxCancel = context.WithCancel(context.Background())
 	ec.wg = sync.WaitGroup{}
@@ -297,6 +303,13 @@ func (ec *Confirmer[CHAIN_ID, HEAD, ADDR, TX_HASH, BLOCK_HASH, R, SEQ, FEE]) pro
 	ec.lggr.Debugw("Finished CheckForReceipts", "headNum", head.BlockNumber(), "time", time.Since(mark), "id", "confirmer")
 	mark = time.Now()
 
+	if err := ec.ProcessStuckTransactions(ctx, head.BlockNumber()); err != nil {
+		return fmt.Errorf("ProcessStuckTransactions failed: %w", err)
+	}
+
+	ec.lggr.Debugw("Finished ProcessStuckTransactions", "headNum", head.BlockNumber(), "time", time.Since(mark), "id", "confirmer")
+	mark = time.Now()
+
 	if err := ec.RebroadcastWhereNecessary(ctx, head.BlockNumber()); err != nil {
 		return fmt.Errorf("RebroadcastWhereNecessary failed: %w", err)
 	}
@@ -435,6 +448,57 @@ func (ec *Confirmer[CHAIN_ID, HEAD, ADDR, TX_HASH, BLOCK_HASH, R, SEQ, FEE]) Che
 	return nil
 }
 
+// Determines if any of the unconfirmed transactions are terminally stuck for each enabled address
+// If any transaction is found to be terminally stuck, this method sends an empty attempt with bumped gas in an attempt to purge the stuck transaction
+func (ec *Confirmer[CHAIN_ID, HEAD, ADDR, TX_HASH, BLOCK_HASH, R, SEQ, FEE]) ProcessStuckTransactions(ctx context.Context, blockNum int64) error {
+	// Use the detector to find a stuck tx for each enabled address
+	stuckTxs, err := ec.stuckTxDetector.DetectStuckTransactions(ctx, ec.enabledAddresses, blockNum)
+	if err != nil {
+		return fmt.Errorf("failed to detect stuck transactions: %w", err)
+	}
+	if len(stuckTxs) == 0 {
+		return nil
+	}
+
+	var wg sync.WaitGroup
+	wg.Add(len(stuckTxs))
+	errorList := []error{}
+	var errMu sync.Mutex
+	for _, tx := range stuckTxs {
+		// All stuck transactions will have unique from addresses. It is safe to process separate keys concurrently
+		// NOTE: This design will block one key if another takes a really long time to execute
+		go func(tx txmgrtypes.Tx[CHAIN_ID, ADDR, TX_HASH, BLOCK_HASH, SEQ, FEE]) {
+			defer wg.Done()
+			lggr := tx.GetLogger(ec.lggr)
+			// Create an purge attempt for tx
+			purgeAttempt, err := ec.TxAttemptBuilder.NewPurgeTxAttempt(ctx, tx, lggr)
+			if err != nil {
+				errMu.Lock()
+				errorList = append(errorList, fmt.Errorf("failed to create a purge attempt: %w", err))
+				errMu.Unlock()
+				return
+			}
+			// Save purge attempt
+			if err := ec.txStore.SaveInProgressAttempt(ctx, &purgeAttempt); err != nil {
+				errMu.Lock()
+				errorList = append(errorList, fmt.Errorf("failed to save purge attempt: %w", err))
+				errMu.Unlock()
+				return
+			}
+			lggr.Warnw("marked transaction as terminally stuck", "etx", tx)
+			// Send purge attempt
+			if err := ec.handleInProgressAttempt(ctx, lggr, tx, purgeAttempt, blockNum); err != nil {
+				errMu.Lock()
+				errorList = append(errorList, fmt.Errorf("failed to send purge attempt: %w", err))
+				errMu.Unlock()
+				return
+			}
+		}(tx)
+	}
+	wg.Wait()
+	return errors.Join(errorList...)
+}
+
 func (ec *Confirmer[CHAIN_ID, HEAD, ADDR, TX_HASH, BLOCK_HASH, R, SEQ, FEE]) separateLikelyConfirmedAttempts(from ADDR, attempts []txmgrtypes.TxAttempt[CHAIN_ID, ADDR, TX_HASH, BLOCK_HASH, SEQ, FEE], minedSequence SEQ) []txmgrtypes.TxAttempt[CHAIN_ID, ADDR, TX_HASH, BLOCK_HASH, SEQ, FEE] {
 	if len(attempts) == 0 {
 		return attempts
@@ -485,15 +549,21 @@ func (ec *Confirmer[CHAIN_ID, HEAD, ADDR, TX_HASH, BLOCK_HASH, R, SEQ, FEE]) fet
 			j = len(attempts)
 		}
 
-		ec.lggr.Debugw(fmt.Sprintf("Batch fetching receipts at indexes %v until (excluded) %v", i, j), "blockNum", blockNum)
+		ec.lggr.Debugw(fmt.Sprintf("Batch fetching receipts at indexes %d until (excluded) %d", i, j), "blockNum", blockNum)
 
 		batch := attempts[i:j]
 
 		receipts, err := ec.batchFetchReceipts(ctx, batch, blockNum)
 		if err != nil {
 			return fmt.Errorf("batchFetchReceipts failed: %w", err)
 		}
-		if err := ec.txStore.SaveFetchedReceipts(ctx, receipts, ec.chainID); err != nil {
+		validReceipts, purgeReceipts := ec.separateValidAndPurgeAttemptReceipts(receipts, batch)
+		// Saves the receipts and mark the associated transactions as Confirmed
+		if err := ec.txStore.SaveFetchedReceipts(ctx, validReceipts, TxConfirmed, nil, ec.chainID); err != nil {
+			return fmt.Errorf("saveFetchedReceipts failed: %w", err)
+		}
+		// Save the receipts but mark the associated transactions as Fatal Error since the original transaction was purged
+		if err := ec.txStore.SaveFetchedReceipts(ctx, purgeReceipts, TxFatalError, ec.stuckTxDetector.StuckTxFatalError(), ec.chainID); err != nil {
 			return fmt.Errorf("saveFetchedReceipts failed: %w", err)
 		}
 		promNumConfirmedTxs.WithLabelValues(ec.chainID.String()).Add(float64(len(receipts)))
@@ -506,6 +576,25 @@ func (ec *Confirmer[CHAIN_ID, HEAD, ADDR, TX_HASH, BLOCK_HASH, R, SEQ, FEE]) fet
 	return nil
 }
 
+func (ec *Confirmer[CHAIN_ID, HEAD, ADDR, TX_HASH, BLOCK_HASH, R, SEQ, FEE]) separateValidAndPurgeAttemptReceipts(receipts []R, attempts []txmgrtypes.TxAttempt[CHAIN_ID, ADDR, TX_HASH, BLOCK_HASH, SEQ, FEE]) (valid []R, purge []R) {
+	receiptMap := make(map[TX_HASH]R)
+	for _, receipt := range receipts {
+		receiptMap[receipt.GetTxHash()] = receipt
+	}
+	for _, attempt := range attempts {
+		if receipt, ok := receiptMap[attempt.Hash]; ok {
+			if attempt.IsPurgeAttempt {
+				// Setting the purged block num here is ok since we have confirmation the tx has been purged with the receipt
+				ec.stuckTxDetector.SetPurgeBlockNum(attempt.Tx.FromAddress, receipt.GetBlockNumber().Int64())
+				purge = append(purge, receipt)
+			} else {
+				valid = append(valid, receipt)
+			}
+		}
+	}
+	return
+}
+
 func (ec *Confirmer[CHAIN_ID, HEAD, ADDR, TX_HASH, BLOCK_HASH, R, SEQ, FEE]) getMinedSequenceForAddress(ctx context.Context, from ADDR) (SEQ, error) {
 	return ec.client.SequenceAt(ctx, from, nil)
 }
@@ -699,7 +788,6 @@ func (ec *Confirmer[CHAIN_ID, HEAD, ADDR, TX_HASH, BLOCK_HASH, R, SEQ, FEE]) Fin
 		lggr.Infow(fmt.Sprintf("Found %d transactions to be re-sent that were previously rejected due to insufficient native token balance", len(etxInsufficientFunds)), "blockNum", blockNum, "address", address)
 	}
 
-	// TODO: Just pass the Q through everything
 	etxBumps, err := ec.txStore.FindTxsRequiringGasBump(ctx, address, blockNum, gasBumpThreshold, bumpDepth, chainID)
 	if ctx.Err() != nil {
 		return nil, nil

common/txmgr/types/stuck_tx_detector.go

@@ -0,0 +1,26 @@
+package types
+
+import (
+	"context"
+
+	feetypes "github.com/smartcontractkit/chainlink/v2/common/fee/types"
+	"github.com/smartcontractkit/chainlink/v2/common/types"
+)
+
+// StuckTxDetector is used by the Confirmer to determine if any unconfirmed transactions are terminally stuck
+type StuckTxDetector[
+	CHAIN_ID types.ID, // CHAIN_ID - chain id type
+	ADDR types.Hashable, // ADDR - chain address type
+	TX_HASH, BLOCK_HASH types.Hashable, // various chain hash types
+	SEQ types.Sequence, // SEQ - chain sequence type (nonce, utxo, etc)
+	FEE feetypes.Fee, // FEE - chain fee type
+] interface {
+	// Uses either a chain specific API or heuristic to determine if any unconfirmed transactions are terminally stuck. Returns only one transaction per enabled address.
+	DetectStuckTransactions(ctx context.Context, enabledAddresses []ADDR, blockNum int64) ([]Tx[CHAIN_ID, ADDR, TX_HASH, BLOCK_HASH, SEQ, FEE], error)
+	// Loads the internal map that tracks the last block num a transaction was purged at using the DB state
+	LoadPurgeBlockNumMap(ctx context.Context, addresses []ADDR) error
+	// Sets the last purged block num after a transaction has been successfully purged with receipt
+	SetPurgeBlockNum(fromAddress ADDR, blockNum int64)
+	// Returns the error message to set in the transaction error field to mark it as terminally stuck
+	StuckTxFatalError() *string
+}

common/txmgr/types/tx.go

@@ -183,6 +183,7 @@ type TxAttempt[
 	State                   TxAttemptState
 	Receipts                []ChainReceipt[TX_HASH, BLOCK_HASH] `json:"-"`
 	TxType                  int
+	IsPurgeAttempt          bool
 }
 
 func (a *TxAttempt[CHAIN_ID, ADDR, TX_HASH, BLOCK_HASH, SEQ, FEE]) String() string {

common/txmgr/types/tx_attempt_builder.go

@@ -42,4 +42,7 @@ type TxAttemptBuilder[
 
 	// NewEmptyTxAttempt is used in ForceRebroadcast to create a signed tx with zero value sent to the zero address
 	NewEmptyTxAttempt(ctx context.Context, seq SEQ, feeLimit uint64, fee FEE, fromAddress ADDR) (attempt TxAttempt[CHAIN_ID, ADDR, TX_HASH, BLOCK_HASH, SEQ, FEE], err error)
+
+	// NewPurgeTxAttempt is used to create empty transaction attempts with higher gas than the previous attempt to purge stuck transactions
+	NewPurgeTxAttempt(ctx context.Context, etx Tx[CHAIN_ID, ADDR, TX_HASH, BLOCK_HASH, SEQ, FEE], lggr logger.Logger) (attempt TxAttempt[CHAIN_ID, ADDR, TX_HASH, BLOCK_HASH, SEQ, FEE], err error)
 }

common/txmgr/types/tx_store.go

@@ -39,7 +39,7 @@ type TxStore[
 	FindTxesPendingCallback(ctx context.Context, blockNum int64, chainID CHAIN_ID) (receiptsPlus []ReceiptPlus[R], err error)
 	// Update tx to mark that its callback has been signaled
 	UpdateTxCallbackCompleted(ctx context.Context, pipelineTaskRunRid uuid.UUID, chainId CHAIN_ID) error
-	SaveFetchedReceipts(ctx context.Context, receipts []R, chainID CHAIN_ID) (err error)
+	SaveFetchedReceipts(ctx context.Context, r []R, state TxState, errorMsg *string, chainID CHAIN_ID) error
 
 	// additional methods for tx store management
 	CheckTxQueueCapacity(ctx context.Context, fromAddress ADDR, maxQueuedTransactions uint64, chainID CHAIN_ID) (err error)

core/chains/evm/config/chain_scoped_transactions.go

@@ -1,6 +1,7 @@
 package config
 
 import (
+	"net/url"
 	"time"
 
 	"github.com/smartcontractkit/chainlink/v2/core/chains/evm/config/toml"
@@ -33,3 +34,27 @@ func (t *transactionsConfig) MaxInFlight() uint32 {
 func (t *transactionsConfig) MaxQueued() uint64 {
 	return uint64(*t.c.MaxQueued)
 }
+
+func (t *transactionsConfig) AutoPurge() AutoPurgeConfig {
+	return &autoPurgeConfig{c: t.c.AutoPurge}
+}
+
+type autoPurgeConfig struct {
+	c toml.AutoPurgeConfig
+}
+
+func (a *autoPurgeConfig) Enabled() bool {
+	return *a.c.Enabled
+}
+
+func (a *autoPurgeConfig) Threshold() uint32 {
+	return *a.c.Threshold
+}
+
+func (a *autoPurgeConfig) MinAttempts() uint32 {
+	return *a.c.MinAttempts
+}
+
+func (a *autoPurgeConfig) DetectionApiUrl() *url.URL {
+	return a.c.DetectionApiUrl.URL()
+}

core/chains/evm/config/config.go

@@ -2,6 +2,7 @@ package config
 
 import (
 	"math/big"
+	"net/url"
 	"time"
 
 	gethcommon "github.com/ethereum/go-ethereum/common"
@@ -100,6 +101,14 @@ type Transactions interface {
 	ReaperThreshold() time.Duration
 	MaxInFlight() uint32
 	MaxQueued() uint64
+	AutoPurge() AutoPurgeConfig
+}
+
+type AutoPurgeConfig interface {
+	Enabled() bool
+	Threshold() uint32
+	MinAttempts() uint32
+	DetectionApiUrl() *url.URL
 }
 
 //go:generate mockery --quiet --name GasEstimator --output ./mocks/ --case=underscore