fix: embedded etcd should only connect to self

clustertest/add_remove_host_test.go

@@ -3,6 +3,7 @@
 package clustertest
 
 import (
+	"fmt"
 	"testing"
 	"time"
 
@@ -198,3 +199,25 @@ func TestForcedHostRemovalWithDatabase(t *testing.T) {
 
 	tLog(t, "test completed successfully")
 }
+
+func TestRollingAddRemove(t *testing.T) {
+	// Regression test for PLAT-581
+	t.Parallel()
+
+	// Initialize a four-host cluster, then remove host-1, and add another host.
+	// The order of adds is important because this bug originates in the
+	// endpoint list for host-2+, so we add each host individually.
+	cluster := NewCluster(t, ClusterConfig{
+		Hosts: []HostConfig{
+			{ID: "host-1"},
+		},
+	})
+	cluster.Init(t)
+	for i := 2; i <= 4; i++ {
+		cluster.Add(t, HostConfig{ID: fmt.Sprintf("host-%d", i)})
+		cluster.Init(t)
+	}
+	cluster.Remove(t, "host-1")
+	cluster.Add(t, HostConfig{ID: "host-5"})
+	cluster.Init(t)
+}

clustertest/cluster_test.go

@@ -101,6 +101,9 @@ func (c *Cluster) Add(t testing.TB, hostCfg HostConfig) {
 func (c *Cluster) Remove(t testing.TB, hostID string) {
 	t.Helper()
 
+	delete(c.hosts, hostID)
+	c.client = hostsClient(t, c.hosts)
+
 	resp, err := c.client.RemoveHost(t.Context(), &controlplane.RemoveHostPayload{
 		HostID: controlplane.Identifier(hostID),
 	})
@@ -114,9 +117,6 @@ func (c *Cluster) Remove(t testing.TB, hostID string) {
 		TaskID: resp.Task.TaskID,
 	})
 	require.NoError(t, err)
-
-	delete(c.hosts, hostID)
-	c.client = hostsClient(t, c.hosts)
 }
 
 // RefreshClient recreates the client with updated host configurations.

server/internal/etcd/embedded.go

@@ -409,7 +409,8 @@ func (e *EmbeddedEtcd) GetClient() (*clientv3.Client, error) {
 	}
 
 	cfg := e.cfg.Config()
-	clientCfg, err := clientConfig(cfg, e.logger, e.etcd.Server.Cluster().ClientURLs()...)
+	// We only want to connect to our own Etcd endpoint.
+	clientCfg, err := clientConfig(cfg, e.logger, e.ClientEndpoints()...)
 	if err != nil {
 		return nil, fmt.Errorf("failed to get client config: %w", err)
 	}