feat(aggregator): Aggregator Retry improvements

aggregator/pkg/aggregator.go

@@ -419,7 +419,7 @@ func (agg *Aggregator) InitializeNewTask(batchIndex uint32, taskCreatedBlock uin
 		}
 		return err
 	}
-	return retry.Retry(initializeNewTask_func, retry.MinDelay, retry.RetryFactor, retry.NumRetries, retry.MaxInterval, retry.MaxElapsedTime)
+	return retry.Retry(initializeNewTask_func, retry.DefaultRetryConfig())
 }
 
 // Long-lived goroutine that periodically checks and removes old Tasks from stored Maps

aggregator/pkg/server.go

@@ -53,7 +53,7 @@ func (agg *Aggregator) ProcessOperatorSignedTaskResponseV2(signedTaskResponse *t
 		"operatorId", hex.EncodeToString(signedTaskResponse.OperatorId[:]))
 	taskIndex := uint32(0)
 
-	taskIndex, err := agg.GetTaskIndex(signedTaskResponse.BatchIdentifierHash)
+	taskIndex, err := agg.GetTaskIndexRetryable(signedTaskResponse.BatchIdentifierHash)
 
 	if err != nil {
 		agg.logger.Warn("Task not found in the internal map, operator signature will be lost. Batch may not reach quorum")
@@ -72,7 +72,7 @@ func (agg *Aggregator) ProcessOperatorSignedTaskResponseV2(signedTaskResponse *t
 
 	agg.logger.Info("Starting bls signature process")
 	go func() {
-		err := agg.ProcessNewSignature(
+		err := agg.ProcessNewSignatureRetryable(
 			context.Background(), taskIndex, signedTaskResponse.BatchIdentifierHash,
 			&signedTaskResponse.BlsSignature, signedTaskResponse.OperatorId,
 		)
@@ -123,7 +123,7 @@ func (agg *Aggregator) ServerRunning(_ *struct{}, reply *int64) error {
   - Retry times (3 retries): 12 sec (1 Blocks), 24 sec (2 Blocks), 48 sec (4 Blocks)
   - NOTE: TaskNotFound errors from the BLS Aggregation service are Transient errors as block reorg's may lead to these errors being thrown.
 */
-func (agg *Aggregator) ProcessNewSignature(ctx context.Context, taskIndex uint32, taskResponse interface{}, blsSignature *bls.Signature, operatorId eigentypes.Bytes32) error {
+func (agg *Aggregator) ProcessNewSignatureRetryable(ctx context.Context, taskIndex uint32, taskResponse interface{}, blsSignature *bls.Signature, operatorId eigentypes.Bytes32) error {
 	processNewSignature_func := func() error {
 		err := agg.blsAggregationService.ProcessNewSignature(
 			ctx, taskIndex, taskResponse,
@@ -137,10 +137,15 @@ func (agg *Aggregator) ProcessNewSignature(ctx context.Context, taskIndex uint32
 		return err
 	}
 
-	return retry.Retry(processNewSignature_func, retry.MinDelayChain, retry.RetryFactor, retry.NumRetries, retry.MaxIntervalChain, retry.MaxElapsedTime)
+	return retry.Retry(processNewSignature_func, retry.ChainRetryConfig())
 }
 
-func (agg *Aggregator) GetTaskIndex(batchIdentifierHash [32]byte) (uint32, error) {
+// Checks Internal mapping for Signed Task Response, returns its TaskIndex.
+/*
+- All errors are considered Transient Errors
+- Retry times (3 retries): 1 sec, 2 sec, 4 sec
+*/
+func (agg *Aggregator) GetTaskIndexRetryable(batchIdentifierHash [32]byte) (uint32, error) {
 	getTaskIndex_func := func() (uint32, error) {
 		agg.taskMutex.Lock()
 		agg.AggregatorConfig.BaseConfig.Logger.Info("- Locked Resources: Starting processing of Response")
@@ -153,6 +158,5 @@ func (agg *Aggregator) GetTaskIndex(batchIdentifierHash [32]byte) (uint32, error
 			return taskIndex, nil
 		}
 	}
-
-	return retry.RetryWithData(getTaskIndex_func, retry.MinDelay, retry.RetryFactor, retry.NumRetries, retry.MaxInterval, retry.MaxElapsedTime)
+	return retry.RetryWithData(getTaskIndex_func, retry.DefaultRetryConfig())
 }

core/chainio/avs_subscriber.go

@@ -68,13 +68,13 @@ func (s *AvsSubscriber) SubscribeToNewTasksV2(newTaskCreatedChan chan *servicema
 	// Subscribe to new tasks
 	sub, err := SubscribeToNewTasksV2Retryable(&bind.WatchOpts{}, s.AvsContractBindings.ServiceManager, internalChannel, nil)
 	if err != nil {
-		s.logger.Error("Primary failed to subscribe to new AlignedLayer V2 tasks after %d retries", retry.NumRetries, "err", err)
+		s.logger.Error("Primary failed to subscribe to new AlignedLayer V2 tasks after %d retries", retry.DefaultMaxNumRetries, "err", err)
 		return nil, err
 	}
 
 	subFallback, err := SubscribeToNewTasksV2Retryable(&bind.WatchOpts{}, s.AvsContractBindings.ServiceManagerFallback, internalChannel, nil)
 	if err != nil {
-		s.logger.Error("Fallback failed to subscribe to new AlignedLayer V2 tasks after %d retries", retry.NumRetries, "err", err)
+		s.logger.Error("Fallback failed to subscribe to new AlignedLayer V2 tasks after %d retries", retry.DefaultMaxNumRetries, "err", err)
 		return nil, err
 	}
 	s.logger.Info("Subscribed to new AlignedLayer V2 tasks")

core/chainio/avs_writer.go

@@ -93,6 +93,17 @@ func (w *AvsWriter) SendAggregatedResponse(batchIdentifierHash [32]byte, batchMe
 	txOpts.NoSend = false
 	i := 0
 
+	// Set Retry config for RespondToTaskV2
+	respondToTaskV2Config := retry.DefaultRetryConfig()
+	respondToTaskV2Config.MaxNumRetries = 0
+	respondToTaskV2Config.MaxElapsedTime = 0
+
+	// Set Retry config for WaitForTxRetryable
+	waitForTxConfig := retry.DefaultRetryConfig()
+	waitForTxConfig.MaxInterval = 2 * time.Second
+	waitForTxConfig.MaxNumRetries = 0
+	waitForTxConfig.MaxElapsedTime = timeToWaitBeforeBump
+
 	respondToTaskV2Func := func() (*types.Receipt, error) {
 		gasPrice, err := utils.GetGasPriceRetryable(w.Client, w.ClientFallback)
 		if err != nil {
@@ -126,7 +137,7 @@ func (w *AvsWriter) SendAggregatedResponse(batchIdentifierHash [32]byte, batchMe
 			return nil, err
 		}
 
-		receipt, err := utils.WaitForTransactionReceiptRetryable(w.Client, w.ClientFallback, tx.Hash(), timeToWaitBeforeBump)
+		receipt, err := utils.WaitForTransactionReceiptRetryable(w.Client, w.ClientFallback, tx.Hash(), waitForTxConfig)
 		if receipt != nil {
 			w.checkIfAggregatorHadToPaidForBatcher(tx, batchIdentifierHash)
 			return receipt, nil
@@ -143,7 +154,7 @@ func (w *AvsWriter) SendAggregatedResponse(batchIdentifierHash [32]byte, batchMe
 		return nil, fmt.Errorf("transaction failed")
 	}
 
-	return retry.RetryWithData(respondToTaskV2Func, retry.MinDelay, retry.RetryFactor, 0, retry.MaxInterval, 0)
+	return retry.RetryWithData(respondToTaskV2Func, respondToTaskV2Config)
 }
 
 // Calculates the transaction cost from the receipt and compares it with the batcher respondToTaskFeeLimit