From 8e16b880c8d161e6174ec7da712ca77b2c64a19a Mon Sep 17 00:00:00 2001
From: binwiederhier <phil@heckel.io>
Date: Sat, 31 Oct 2026 01:55:42 +0000
Subject: [PATCH] Close the cluster when the server stops

Server.stop never closed the cluster, so a stopped node stayed registered and its
heartbeat kept running against the closed database, logging from leaked goroutines.
In CI that raced with captureLog's unsynchronized buffer in a later test. captureLog
now uses syncBuffer as well.
---
 server/server.go              |  3 +++
 server/server_admin_test.go   |  4 ++--
 server/server_cluster_test.go | 21 +++++++++++++++++++++
 3 files changed, 26 insertions(+), 2 deletions(-)

diff --git a/server/server.go b/server/server.go
index ccd8af6c..9111fbec 100644
--- a/server/server.go
+++ b/server/server.go
@@ -617,6 +617,9 @@ func (s *Server) stop() {
 	if s.tail != nil {
 		s.tail.Close() // Before the databases close under it
 	}
+	if s.cluster != nil {
+		s.cluster.Close() // Before the databases close, so it can deregister the node and its heartbeat stops
+	}
 	s.closeDatabases()
 	if s.ban != nil {
 		s.ban.Close()
diff --git a/server/server_admin_test.go b/server/server_admin_test.go
index 9378f7bd..40362c0d 100644
--- a/server/server_admin_test.go
+++ b/server/server_admin_test.go
@@ -1577,8 +1577,8 @@ func (b *syncBuffer) String() string {
 }
 
 func captureLog(t *testing.T, f func()) string {
-	var buf bytes.Buffer
-	log.SetOutput(&buf)
+	buf := &syncBuffer{} // Background goroutines of other tests' servers may log meanwhile, too
+	log.SetOutput(buf)
 	log.SetLevel(log.InfoLevel)
 	defer func() {
 		log.SetLevel(log.ErrorLevel)                  // As in TestMain
diff --git a/server/server_cluster_test.go b/server/server_cluster_test.go
index 47fe9804..c3e4b08e 100644
--- a/server/server_cluster_test.go
+++ b/server/server_cluster_test.go
@@ -237,6 +237,27 @@ func TestServer_Cluster_Role(t *testing.T) {
 	require.Equal(t, cluster.RolePassive, members[0].Role)
 }
 
+func TestServer_Cluster_StopClosesCluster(t *testing.T) {
+	// Stopping the server must stop the cluster: it deregisters the node, so a node starting
+	// after it no longer sees it, and its heartbeat does not outlive the server's database
+	schemaDSN := dbtest.CreateTestPostgresSchema(t)
+	newNode := func(nodeID string) *Server {
+		conf := newTestConfig(t, schemaDSN)
+		conf.ClusterNodeID = nodeID
+		conf.ClusterListen = "127.0.0.1:1" // Enables clustering; not bound since Run() is not called
+		conf.ClusterSecret = "s3cret"
+		conf.ClusterAdvertiseURL = "http://127.0.0.1:1"
+		return newTestServer(t, conf)
+	}
+	sA := newNode("node-a")
+	require.Len(t, sA.cluster.Members(), 1)
+	sA.Stop()
+	sB := newNode("node-b")
+	members := sB.cluster.Members()
+	require.Len(t, members, 1)
+	require.Equal(t, cluster.NodeID("node-b"), members[0].NodeID)
+}
+
 func TestServer_Cluster_ActiveWithPassivePeer(t *testing.T) {
 	// The setup for trying clustering beside production: the production server is the only
 	// active node, a second one runs passive. The active one leads right away, both deliver