netann: fix disable issue for channels

We make sure we do not disable a channel if the channel is active
in the switch and the enable timer hasn't fired yet during
reconnection.

Author: ziggie1984

netann/chan_status_manager.go

@@ -568,6 +568,23 @@ func (m *ChanStatusManager) disableInactiveChannels() {
 			continue
 		}
 
+		// Re-verify the channel is still inactive before disabling.
+		// This prevents the race condition where a channel becomes
+		// active again after being marked as pending disabled but
+		// before the disable timeout expires. Otherwise there is a
+		// race condition where the channel is disabled even though it
+		// is active.
+		chanID := lnwire.NewChanIDFromOutPoint(outpoint)
+		if m.cfg.IsChannelActive(chanID) {
+			// Channel became active again, cancel the pending
+			// disable.
+			log.Debugf("Channel(%v) became active, canceling "+
+				"scheduled disable", outpoint)
+
+			m.chanStates.markEnabled(outpoint)
+			continue
+		}
+
 		log.Infof("Announcing channel(%v) disabled "+
 			"[detected]", outpoint)
 

netann/chan_status_manager_test.go

@@ -909,6 +909,58 @@ var stateMachineTests = []stateMachineTest{
 			)
 		},
 	},
+	{
+		name:         "channel reconnects before disable",
+		startActive:  true,
+		startEnabled: true,
+		fn: func(h testHarness) {
+			// This test demonstrates the race condition fix
+			// where a channel becomes inactive, gets marked as
+			// pending disabled, then becomes active again before
+			// the disable timeout expires.
+			// The channel should NOT be disabled in this case.
+
+			// Step 1: Simulate disconnection - channel becomes
+			// inactive.
+			h.markInactive(h.graph.chans())
+
+			// Step 2: Wait for the channel to be marked as
+			// pending disabled. Sample interval of the manager
+			// is 50ms, so we wait for 50ms + 50ms (buffer).
+			time.Sleep(50*time.Millisecond + 50*time.Millisecond)
+
+			// Step 3: Simulate reconnection - channel becomes
+			// active again.
+			//
+			// NOTE: This does not reflect the actual behavior of
+			// LND because as soon as the channel becomes active it
+			// will start an enable timer and send an enable update.
+			// However we want to avoid testing these timings
+			// here. In general it is important that the channel
+			// does not get disabled in case it reconnects before
+			// the disable timeout expires.
+			h.markActive(h.graph.chans())
+
+			// Step 4: Wait for the disable timeout to expire.
+			// The disable timeout (1 second) expires, but our fix
+			// should prevent the disable because the channel is
+			// active again.
+			time.Sleep(time.Second + 200*time.Millisecond)
+
+			// Step 5: Verify that NO disable update was sent.
+			// The channel should remain enabled because it became
+			// active again before the disable timeout expired,
+			// and our fix re-checked its status before disabling.
+			h.assertNoUpdates(500 * time.Millisecond)
+
+			// Step 6: Verify that the channel is still enabled by
+			// checking that we can still request enable without
+			// sending an update. This means the channel is still
+			// enabled.
+			h.assertEnables(h.graph.chans(), nil, false)
+			h.assertNoUpdates(500 * time.Millisecond)
+		},
+	},
 }
 
 // TestChanStatusManagerStateMachine tests the possible state transitions that