routing+htlcswitch: fix stuck inflight payments

channeldb/payment_control.go

@@ -115,6 +115,10 @@ var (
 	// amount exceed the total amount.
 	ErrSentExceedsTotal = errors.New("total sent exceeds total amount")
 
+	// ErrRegisterAttempt is returned when a new htlc attempt cannot be
+	// registered.
+	ErrRegisterAttempt = errors.New("cannot register htlc attempt")
+
 	// errNoAttemptInfo is returned when no attempt info is stored yet.
 	errNoAttemptInfo = errors.New("unable to find attempt info for " +
 		"inflight payment")
@@ -342,7 +346,8 @@ func (p *PaymentControl) RegisterAttempt(paymentHash lntypes.Hash,
 
 		// Check if registering a new attempt is allowed.
 		if err := payment.Registrable(); err != nil {
-			return err
+			return fmt.Errorf("%w: %v", ErrRegisterAttempt,
+				err.Error())
 		}
 
 		// If the final hop has encrypted data, then we know this is a

channeldb/payment_control_test.go

@@ -841,9 +841,9 @@ func TestPaymentControlMultiShard(t *testing.T) {
 		b.AttemptID = 3
 		_, err = pControl.RegisterAttempt(info.PaymentIdentifier, &b)
 		if test.settleFirst {
-			require.ErrorIs(t, err, ErrPaymentPendingSettled)
+			require.ErrorIs(t, err, ErrRegisterAttempt)
 		} else {
-			require.ErrorIs(t, err, ErrPaymentPendingFailed)
+			require.ErrorIs(t, err, ErrRegisterAttempt)
 		}
 
 		assertPaymentStatus(t, pControl, info.PaymentIdentifier, StatusInFlight)
@@ -897,32 +897,25 @@ func TestPaymentControlMultiShard(t *testing.T) {
 			require.NoError(t, err, "unable to fail")
 		}
 
-		var (
-			finalStatus PaymentStatus
-			registerErr error
-		)
+		var finalStatus PaymentStatus
 
 		switch {
 		// If one of the attempts settled but the other failed with
 		// terminal error, we would still consider the payment is
 		// settled.
 		case test.settleFirst && !test.settleLast:
 			finalStatus = StatusSucceeded
-			registerErr = ErrPaymentAlreadySucceeded
 
 		case !test.settleFirst && test.settleLast:
 			finalStatus = StatusSucceeded
-			registerErr = ErrPaymentAlreadySucceeded
 
 		// If both failed, we end up in a failed status.
 		case !test.settleFirst && !test.settleLast:
 			finalStatus = StatusFailed
-			registerErr = ErrPaymentAlreadyFailed
 
 		// Otherwise, the payment has a succeed status.
 		case test.settleFirst && test.settleLast:
 			finalStatus = StatusSucceeded
-			registerErr = ErrPaymentAlreadySucceeded
 		}
 
 		assertPaymentStatus(
@@ -931,7 +924,7 @@ func TestPaymentControlMultiShard(t *testing.T) {
 
 		// Finally assert we cannot register more attempts.
 		_, err = pControl.RegisterAttempt(info.PaymentIdentifier, &b)
-		require.Equal(t, registerErr, err)
+		require.ErrorIs(t, err, ErrRegisterAttempt)
 	}
 
 	for _, test := range tests {

routing/payment_lifecycle.go

@@ -274,6 +274,16 @@ lifecycle:
 		// We found a route to try, create a new HTLC attempt to try.
 		attempt, err := p.registerAttempt(rt, ps.RemainingAmt)
 		if err != nil {
+			// If the error is due to we cannot register another
+			// HTLC, we will skip this iteration and continue to
+			// the next one in case there are inflight HTLCs.
+			//
+			// TODO(yy): remove this check once we have a finer
+			// control over errors returned from the switch.
+			if errors.Is(err, channeldb.ErrRegisterAttempt) {
+				continue lifecycle
+			}
+
 			return exitWithErr(err)
 		}