0b823c5 test: stop CI flakes in socket, daemon election, web reconnect, and QUIC tests · vev

M internal/adapters/ipc/mux_test.go +10 -3


@@ -259,7 +259,14 @@ func createStaleMuxSocket(t *testing.T, path string) os.FileInfo {
259	259		// recovered race-safely and the carriage binds and serves.
260	260		func TestListenMuxStaleSocketRecovered(t *testing.T) {
261	261		path := muxTestPath(t)
262		-	stale := createStaleMuxSocket(t, path)
	262	+	createStaleMuxSocket(t, path)
	263	+	// Mark the stale file: a filesystem such as ext4 may hand the freed inode
	264	+	// number straight to the recovered socket, so inode identity alone cannot
	265	+	// prove the file was replaced.
	266	+	marked := time.Unix(1, 0)
	267	+	if err := os.Chtimes(path, marked, marked); err != nil {
	268	+	t.Fatalf("mark stale socket: %v", err)
	269	+	}
263	270
264	271		mustListenMux(t, path)
265	272

@@ -267,8 +274,8 @@ func TestListenMuxStaleSocketRecovered(t *testing.T) {
267	274		if err != nil {
268	275		t.Fatalf("stat recovered socket: %v", err)
269	276		}
270		-	if os.SameFile(stale, recovered) {
271		-	t.Fatal("stale socket inode was not replaced by the recovered listener")
	277	+	if recovered.ModTime().Equal(marked) {
	278	+	t.Fatal("stale socket was not replaced by the recovered listener")
272	279		}
273	280
274	281		// The recovered carriage serves a same-user connection.

 // recovered race-safely and the carriage binds and serves.
 func TestListenMuxStaleSocketRecovered(t *testing.T) {
 	path := muxTestPath(t)
-	stale := createStaleMuxSocket(t, path)
+	createStaleMuxSocket(t, path)
+	// Mark the stale file: a filesystem such as ext4 may hand the freed inode
+	// number straight to the recovered socket, so inode identity alone cannot
+	// prove the file was replaced.
+	marked := time.Unix(1, 0)
+	if err := os.Chtimes(path, marked, marked); err != nil {
+		t.Fatalf("mark stale socket: %v", err)
+	}
 	mustListenMux(t, path)
 	if err != nil {
 		t.Fatalf("stat recovered socket: %v", err)
 	}
-	if os.SameFile(stale, recovered) {
+	if recovered.ModTime().Equal(marked) {
-		t.Fatal("stale socket inode was not replaced by the recovered listener")
+		t.Fatal("stale socket was not replaced by the recovered listener")
 	}
 	// The recovered carriage serves a same-user connection.

M internal/adapters/quicnettest/proxy_test.go +6 -0


@@ -54,6 +54,12 @@ func serveUDP(t testing.T, echo bool) (net.UDPConn, <-chan datagram) {
54	54		t.Fatalf("listen udp server: %v", err)
55	55		}
56	56		t.Cleanup(func() { _ = conn.Close() })
	57	+	// Flows burst hundreds of MTU-sized datagrams. The default receive buffer
	58	+	// (about 200 KiB) holds too few of them when a loaded machine delays the
	59	+	// reader, and the kernel then drops datagrams the test is still counting.
	60	+	if err := conn.SetReadBuffer(4 << 20); err != nil {
	61	+	t.Fatalf("size udp server buffer: %v", err)
	62	+	}
57	63		packets := make(chan datagram, 4096)
58	64		go func() {
59	65		buf := make([]byte, 2048)


@@ -54,6 +54,12 @@ func serveUDP(t testing.T, echo bool) (net.UDPConn, <-chan datagram) {
54	t.Fatalf("listen udp server: %v", err)	54		t.Fatalf("listen udp server: %v", err)
55	}	55		}
56	t.Cleanup(func() { _ = conn.Close() })	56		t.Cleanup(func() { _ = conn.Close() })
		57	+	// Flows burst hundreds of MTU-sized datagrams. The default receive buffer
		58	+	// (about 200 KiB) holds too few of them when a loaded machine delays the
		59	+	// reader, and the kernel then drops datagrams the test is still counting.
		60	+	if err := conn.SetReadBuffer(4 << 20); err != nil {
		61	+	t.Fatalf("size udp server buffer: %v", err)
		62	+	}
57	packets := make(chan datagram, 4096)	63		packets := make(chan datagram, 4096)
58	go func() {	64		go func() {
59	buf := make([]byte, 2048)	65		buf := make([]byte, 2048)

M internal/app/broker_daemon_start_test.go +7 -0


@@ -429,6 +429,13 @@ func TestDaemonStartConcurrentRequestsShareOneElection(t *testing.T) {
429	429		require.NoError(t, err)
430	430		require.True(t, acquired, "the test must hold the spawn lock")
431	431		t.Cleanup(release)
	432	+	// The daemon being started takes lifecycle ownership at once and keeps it
	433	+	// for its whole run, as the real daemon does. Without it, a caller that
	434	+	// dialed just before the daemon was published could win the spawn lock
	435	+	// once it is released and elect itself.
	436	+	daemonOwner, err := daemonLifecycleProbe.TryAcquire(filepath.Dir(carriage))
	437	+	require.NoError(t, err)
	438	+	t.Cleanup(func() { _ = daemonOwner.Release() })
432	439
433	440		// The target is resolved once: require.* is not goroutine-safe, and the
434	441		// resolved target is immutable and shared by every caller.


@@ -429,6 +429,13 @@ func TestDaemonStartConcurrentRequestsShareOneElection(t *testing.T) {
429	require.NoError(t, err)	429		require.NoError(t, err)
430	require.True(t, acquired, "the test must hold the spawn lock")	430		require.True(t, acquired, "the test must hold the spawn lock")
431	t.Cleanup(release)	431		t.Cleanup(release)
		432	+	// The daemon being started takes lifecycle ownership at once and keeps it
		433	+	// for its whole run, as the real daemon does. Without it, a caller that
		434	+	// dialed just before the daemon was published could win the spawn lock
		435	+	// once it is released and elect itself.
		436	+	daemonOwner, err := daemonLifecycleProbe.TryAcquire(filepath.Dir(carriage))
		437	+	require.NoError(t, err)
		438	+	t.Cleanup(func() { _ = daemonOwner.Release() })
432		439
433	// The target is resolved once: require.* is not goroutine-safe, and the	440		// The target is resolved once: require.* is not goroutine-safe, and the
434	// resolved target is immutable and shared by every caller.	441		// resolved target is immutable and shared by every caller.

M internal/app/web_daemon_broker_test.go +3 -1


@@ -635,7 +635,9 @@ func TestWebGatewayTabReturnsToPickerOnBrokerLoss(t *testing.T) {
635	635		t.Fatalf("broker loss ended the browser run: %v", tab.readerErr())
636	636		default:
637	637		}
638		-	require.GreaterOrEqual(t, calls.Load(), int32(2), "a lost connection must be retried, never fatal")
	638	+	// A repaint can land before the first reconnect attempt, so wait for it.
	639	+	require.Eventually(t, func() bool { return calls.Load() >= 2 }, brokerTestWait, 5*time.Millisecond,
	640	+	"a lost connection must be retried, never fatal")
639	641		}
640	642
641	643		// TestWebCompositionUsesTheBrokerConnector pins the production composition

 		t.Fatalf("broker loss ended the browser run: %v", tab.readerErr())
 	default:
 	}
-	require.GreaterOrEqual(t, calls.Load(), int32(2), "a lost connection must be retried, never fatal")
+	// A repaint can land before the first reconnect attempt, so wait for it.
+	require.Eventually(t, func() bool { return calls.Load() >= 2 }, brokerTestWait, 5*time.Millisecond,
+		"a lost connection must be retried, never fatal")
 }
 // TestWebCompositionUsesTheBrokerConnector pins the production composition