Merge pull request #10455 from ziggie1984/fix-onchain-timeout-terminal

Fix flaky itest multihop testcase

itest/lnd_multi-hop_force_close_test.go

@@ -517,19 +517,25 @@ func runLocalClaimOutgoingHTLC(ht *lntest.HarnessTest,
 	// Now that Bob has claimed his HTLCs, Alice should mark the two
 	// payments as failed.
 	//
-	// Alice will mark this payment as failed with no route as the only
+	// Alice's payment can fail with either NO_ROUTE or TIMEOUT depending
-	// route she has is Alice->Bob->Carol. This won't be the case if she
+	// on timing. There's a race between:
-	// has a second route, as another attempt will be tried.
+	// 1. The channel closure propagating to Alice's graph (-> NO_ROUTE)
-	//
+	// 2. The payment attempt timeout firing (-> TIMEOUT)
-	// TODO(yy): we should instead mark this payment as timed out if she has
+	// Both failure reasons are correct.
-	// a second route to try this payment, which is the timeout set by Alice
+	p := ht.AssertPaymentFailureReasonAny(alice, preimage,
-	// when sending the payment.
+		lnrpc.PaymentFailureReason_FAILURE_REASON_NO_ROUTE,
-	expectedReason := lnrpc.PaymentFailureReason_FAILURE_REASON_NO_ROUTE
+		lnrpc.PaymentFailureReason_FAILURE_REASON_TIMEOUT,
-	p := ht.AssertPaymentFailureReason(alice, preimage, expectedReason)
+	)
+
+	// The HTLC-level failure code should be PERMANENT_CHANNEL_FAILURE
+	// regardless of which payment-level failure reason we got.
 	require.Equal(ht, lnrpc.Failure_PERMANENT_CHANNEL_FAILURE,
 		p.Htlcs[0].Failure.Code)
 
-	p = ht.AssertPaymentFailureReason(alice, preimageDust, expectedReason)
+	p = ht.AssertPaymentFailureReasonAny(alice, preimageDust,
+		lnrpc.PaymentFailureReason_FAILURE_REASON_NO_ROUTE,
+		lnrpc.PaymentFailureReason_FAILURE_REASON_TIMEOUT,
+	)
 	require.Equal(ht, lnrpc.Failure_PERMANENT_CHANNEL_FAILURE,
 		p.Htlcs[0].Failure.Code)
 }

lntest/harness_assertion.go

@@ -1648,6 +1648,40 @@ func (h *HarnessTest) AssertPaymentFailureReason(
 	return payment
 }
 
+// AssertPaymentFailureReasonAny asserts that the given node lists a payment
+// with the given preimage which has one of the expected failure reasons.
+func (h *HarnessTest) AssertPaymentFailureReasonAny(
+	hn *node.HarnessNode, preimage lntypes.Preimage,
+	reasons ...lnrpc.PaymentFailureReason) *lnrpc.Payment {
+
+	var payment *lnrpc.Payment
+
+	payHash := preimage.Hash()
+	err := wait.NoError(func() error {
+		p, err := h.findPayment(hn, payHash.String())
+		if err != nil {
+			return err
+		}
+
+		payment = p
+
+		// Check if the payment failure reason matches any of the
+		// expected reasons.
+		for _, reason := range reasons {
+			if reason == p.FailureReason {
+				return nil
+			}
+		}
+
+		return fmt.Errorf("payment: %v failure reason not match, "+
+			"want one of %v, got %s(%d)", payHash, reasons,
+			p.FailureReason, p.FailureReason)
+	}, DefaultTimeout)
+	require.NoError(h, err, "timeout checking payment failure reason")
+
+	return payment
+}
+
 // AssertActiveNodesSynced asserts all active nodes have synced to the chain.
 func (h *HarnessTest) AssertActiveNodesSynced() {
 	for _, node := range h.manager.activeNodes {