Merge pull request #20894 from ahrtr/20251106_learner

Fix the issue that `--force-new-cluster` can't clean up learner after creating v2 snapshot

Author: ahrtr

server/storage/util.go

@@ -121,6 +121,9 @@ func GetEffectiveNodeIDsFromWALEntries(lg *zap.Logger, snap *raftpb.Snapshot, en
 		for _, id := range snap.Metadata.ConfState.Voters {
 			ids[id] = true
 		}
+		for _, id := range snap.Metadata.ConfState.Learners {
+			ids[id] = true
+		}
 	}
 	for _, e := range ents {
 		if e.Type != raftpb.EntryConfChange {

tests/e2e/force_new_cluster_test.go

@@ -18,6 +18,7 @@ package e2e
 
 import (
 	"encoding/json"
+	"strings"
 	"testing"
 	"time"
 
@@ -29,6 +30,7 @@ import (
 	"go.etcd.io/etcd/server/v3/storage/schema"
 	"go.etcd.io/etcd/tests/v3/framework/config"
 	"go.etcd.io/etcd/tests/v3/framework/e2e"
+	"go.etcd.io/etcd/tests/v3/framework/testutils"
 )
 
 // TestForceNewCluster verified that etcd works as expected when --force-new-cluster.
@@ -111,6 +113,71 @@ func TestForceNewCluster_MemberCount(t *testing.T) {
 	require.Len(t, members, 1)
 }
 
+// TestForceNewCluster_AddLearner_MemberCount verifies that `--force-new-cluster`
+// should always be able to clean up all other members, including learners.
+func TestForceNewCluster_AddLearner_MemberCount(t *testing.T) {
+	e2e.BeforeTest(t)
+
+	testCases := []struct {
+		name      string
+		snapcount int
+	}{
+		{
+			name:      "no snapshot after adding learner",
+			snapcount: 0,
+		},
+		{
+			name:      "create a snapshot after adding learner",
+			snapcount: 5,
+		},
+	}
+
+	for _, tc := range testCases {
+		t.Run(tc.name, func(t *testing.T) {
+			cfg := e2e.NewConfig(e2e.WithClusterSize(3))
+			epc, err := e2e.NewEtcdProcessCluster(t.Context(), t, e2e.WithConfig(cfg), e2e.WithSnapshotCount(uint64(tc.snapcount)), e2e.WithKeepDataDir(true))
+			require.NoError(t, err)
+
+			t.Log("Adding a learner member")
+			testutils.ExecuteWithTimeout(t, 1*time.Minute, func() {
+				for {
+					_, aerr := epc.StartNewProc(t.Context(), nil, t, true)
+					if aerr != nil {
+						if strings.Contains(aerr.Error(), "etcdserver: unhealthy cluster") {
+							time.Sleep(1 * time.Second)
+							continue
+						}
+					}
+					break
+				}
+			})
+
+			for i := 0; i < tc.snapcount; i++ {
+				werr := epc.Etcdctl().Put(t.Context(), "foo", "bar", config.PutOptions{})
+				require.NoError(t, werr)
+			}
+			require.NoError(t, epc.Close())
+
+			m := epc.Procs[0]
+			t.Logf("Forcibly create a one-member cluster with member: %s", m.Config().Name)
+			m.Config().Args = append(m.Config().Args, "--force-new-cluster")
+			require.NoError(t, m.Start(t.Context()))
+
+			t.Log("Restarting the member")
+			require.NoError(t, m.Restart(t.Context()))
+			defer func() {
+				t.Log("Closing the member")
+				require.NoError(t, m.Close())
+			}()
+
+			t.Log("Checking member count")
+			resp, merr := m.Etcdctl().MemberList(t.Context(), false)
+			require.NoError(t, merr)
+			require.Len(t, resp.Members, 1)
+		})
+	}
+}
+
 func mustReadMembersFromBoltDB(t *testing.T, dataDir string) []*membership.Member {
 	dbPath := datadir.ToBackendFileName(dataDir)
 	db, err := bbolt.Open(dbPath, 0o400, &bbolt.Options{ReadOnly: true})