From b77c1eb03fa75e4fed4c8d25902e1127f9884067 Mon Sep 17 00:00:00 2001
From: Ciprian Hacman <ciprian@hakman.dev>
Date: Sat, 11 Jul 2026 10:21:43 +0300
Subject: [PATCH] fix(logwatchers/kmsg): don't block Stop() when the log
 channel is full

The log monitor stops draining logCh before calling watcher.Stop(), so
with a full channel (e.g. a kmsg burst at shutdown) watchLoop blocked
on the send forever, never called tomb.Done(), and Stop() hung.
Select on tomb.Stopping() alongside the send.
---
 .../logwatchers/kmsg/log_watcher_linux.go     |  9 ++++-
 .../kmsg/log_watcher_linux_test.go            | 39 +++++++++++++++++++
 2 files changed, 47 insertions(+), 1 deletion(-)

diff --git a/pkg/systemlogmonitor/logwatchers/kmsg/log_watcher_linux.go b/pkg/systemlogmonitor/logwatchers/kmsg/log_watcher_linux.go
index 83818e8e..fa6d8cf2 100644
--- a/pkg/systemlogmonitor/logwatchers/kmsg/log_watcher_linux.go
+++ b/pkg/systemlogmonitor/logwatchers/kmsg/log_watcher_linux.go
@@ -135,9 +135,16 @@ func (k *kernelLogWatcher) watchLoop() {
 				continue
 			}
 
-			k.logCh <- &logtypes.Log{
+			// The consumer stops draining logCh before calling Stop(), so a
+			// plain send on a full channel could block forever and deadlock Stop().
+			select {
+			case k.logCh <- &logtypes.Log{
 				Message:   strings.TrimSpace(msg.Message),
 				Timestamp: msg.Timestamp,
+			}:
+			case <-k.tomb.Stopping():
+				klog.Infof("Stop watching kernel log")
+				return
 			}
 		}
 	}
diff --git a/pkg/systemlogmonitor/logwatchers/kmsg/log_watcher_linux_test.go b/pkg/systemlogmonitor/logwatchers/kmsg/log_watcher_linux_test.go
index bc4db02b..3767b986 100644
--- a/pkg/systemlogmonitor/logwatchers/kmsg/log_watcher_linux_test.go
+++ b/pkg/systemlogmonitor/logwatchers/kmsg/log_watcher_linux_test.go
@@ -468,6 +468,45 @@ func TestWatcherRateLimitsRestarts(t *testing.T) {
 	}
 }
 
+// TestStopDoesNotDeadlockWhenLogChannelFull verifies that Stop() returns even
+// when logCh is full and nobody is draining it.
+func TestStopDoesNotDeadlockWhenLogChannelFull(t *testing.T) {
+	now := time.Now()
+
+	// More messages than logCh capacity so watchLoop ends up blocked sending.
+	kmsgs := make([]kmsgparser.Message, 150)
+	for i := range kmsgs {
+		kmsgs[i] = kmsgparser.Message{Message: fmt.Sprintf("msg-%d", i), Timestamp: now}
+	}
+
+	w := &kernelLogWatcher{
+		cfg:        types.WatcherConfig{},
+		startTime:  now.Add(-time.Minute),
+		tomb:       tomb.NewTomb(),
+		logCh:      make(chan *logtypes.Log, 100),
+		kmsgParser: &mockKmsgParser{kmsgs: kmsgs},
+	}
+
+	// Watch but never read logCh, mimicking the log monitor after it has
+	// decided to stop.
+	_, err := w.Watch()
+	assert.NoError(t, err)
+
+	// Let watchLoop fill the channel and block on the send.
+	time.Sleep(300 * time.Millisecond)
+
+	stopped := make(chan struct{})
+	go func() {
+		w.Stop()
+		close(stopped)
+	}()
+	select {
+	case <-stopped:
+	case <-time.After(2 * time.Second):
+		t.Fatal("Stop() deadlocked while logCh was full")
+	}
+}
+
 // TestWatcherProcessesMessageContent verifies watchLoop's per-message
 // handling: empty messages are dropped, and surrounding whitespace is
 // trimmed before forwarding.