server: Depend only on cluster version to detect downgrade

Problem with old code was that during downgrade only members with
downgrade target version were allowed to join. This is unrealistic as
it doesn't handle any members to disconnect/rejoin.

server/etcdserver/version/version_test.go

@@ -62,6 +62,68 @@ func TestUpgradeThreeNodes(t *testing.T) {
 	assert.Equal(t, newCluster(lg, 3, V3_7), c)
 }
 
+func TestDowngradeSingleNode(t *testing.T) {
+	lg := zaptest.NewLogger(t)
+	c := newCluster(lg, 1, V3_6)
+	c.StepMonitors()
+	assert.Equal(t, newCluster(lg, 1, V3_6), c)
+
+	assert.NoError(t, c.Version().DowngradeEnable(context.Background(), &V3_5))
+	c.StepMonitors()
+	assert.Equal(t, V3_5, c.clusterVersion)
+
+	c.ReplaceMemberBinary(0, V3_5)
+	c.StepMonitors()
+
+	assert.Equal(t, newCluster(lg, 1, V3_5), c)
+}
+
+func TestDowngradeThreeNode(t *testing.T) {
+	lg := zaptest.NewLogger(t)
+	c := newCluster(lg, 3, V3_6)
+	c.StepMonitors()
+	assert.Equal(t, newCluster(lg, 3, V3_6), c)
+
+	assert.NoError(t, c.Version().DowngradeEnable(context.Background(), &V3_5))
+	c.StepMonitors()
+	assert.Equal(t, V3_5, c.clusterVersion)
+
+	c.ReplaceMemberBinary(0, V3_5)
+	c.StepMonitors()
+	c.ReplaceMemberBinary(1, V3_5)
+	c.StepMonitors()
+	c.ReplaceMemberBinary(2, V3_5)
+	c.StepMonitors()
+
+	assert.Equal(t, newCluster(lg, 3, V3_5), c)
+}
+
+func TestNewerMemberCanReconnectDuringDowngrade(t *testing.T) {
+	lg := zaptest.NewLogger(t)
+	c := newCluster(lg, 3, V3_6)
+	c.StepMonitors()
+	assert.Equal(t, newCluster(lg, 3, V3_6), c)
+
+	assert.NoError(t, c.Version().DowngradeEnable(context.Background(), &V3_5))
+	c.StepMonitors()
+	assert.Equal(t, V3_5, c.clusterVersion)
+
+	c.ReplaceMemberBinary(0, V3_5)
+	c.StepMonitors()
+
+	c.MemberCrashes(2)
+	c.StepMonitors()
+	c.MemberReconnects(2)
+	c.StepMonitors()
+
+	c.ReplaceMemberBinary(1, V3_5)
+	c.StepMonitors()
+	c.ReplaceMemberBinary(2, V3_5)
+	c.StepMonitors()
+
+	assert.Equal(t, newCluster(lg, 3, V3_5), c)
+}
+
 func newCluster(lg *zap.Logger, memberCount int, ver semver.Version) *clusterMock {
 	cluster := &clusterMock{
 		lg:             lg,
@@ -71,6 +133,7 @@ func newCluster(lg *zap.Logger, memberCount int, ver semver.Version) *clusterMoc
 	majorMinVer := semver.Version{Major: ver.Major, Minor: ver.Minor}
 	for i := 0; i < memberCount; i++ {
 		m := &memberMock{
+			isRunning:      true,
 			cluster:        cluster,
 			serverVersion:  ver,
 			storageVersion: majorMinVer,
@@ -113,22 +176,34 @@ func (c *clusterMock) Version() *Manager {
 func (c *clusterMock) MembersVersions() map[string]*version.Versions {
 	result := map[string]*version.Versions{}
 	for i, m := range c.members {
-		result[fmt.Sprintf("%d", i)] = &version.Versions{
-			Server:  m.serverVersion.String(),
-			Cluster: c.clusterVersion.String(),
+		if m.isRunning {
+			result[fmt.Sprintf("%d", i)] = &version.Versions{
+				Server:  m.serverVersion.String(),
+				Cluster: c.clusterVersion.String(),
+			}
 		}
 	}
 	return result
 }
 
 func (c *clusterMock) ReplaceMemberBinary(mid int, newServerVersion semver.Version) {
-	MustDetectDowngrade(c.lg, &c.members[mid].serverVersion, &c.clusterVersion, c.downgradeInfo)
+	MustDetectDowngrade(c.lg, &c.members[mid].serverVersion, &c.clusterVersion)
 	c.members[mid].serverVersion = newServerVersion
 }
 
+func (c *clusterMock) MemberCrashes(mid int) {
+	c.members[mid].isRunning = false
+}
+
+func (c *clusterMock) MemberReconnects(mid int) {
+	MustDetectDowngrade(c.lg, &c.members[mid].serverVersion, &c.clusterVersion)
+	c.members[mid].isRunning = true
+}
+
 type memberMock struct {
 	cluster *clusterMock
 
+	isRunning      bool
 	isLeader       bool
 	serverVersion  semver.Version
 	storageVersion semver.Version
@@ -174,8 +249,9 @@ func (m *memberMock) GetStorageVersion() *semver.Version {
 	return &m.storageVersion
 }
 
-func (m *memberMock) UpdateStorageVersion(v semver.Version) {
+func (m *memberMock) UpdateStorageVersion(v semver.Version) error {
 	m.storageVersion = v
+	return nil
 }
 
 func (m *memberMock) TriggerSnapshot() {