func configReplace(cmd *cobra.Command, args []string) {
if crOpts.file == "" {
die("no config file provided (--file/-f option)")
}
config := []byte{}
var err error
if crOpts.file == "-" {
config, err = ioutil.ReadAll(os.Stdin)
if err != nil {
die("cannot read config file from stdin: %v", err)
}
} else {
config, err = ioutil.ReadFile(crOpts.file)
if err != nil {
die("cannot read provided config file: %v", err)
}
}
storePath := filepath.Join(common.StoreBasePath, cfg.clusterName)
kvstore, err := store.NewStore(store.Backend(cfg.storeBackend), cfg.storeEndpoints)
if err != nil {
die("cannot create store: %v", err)
}
e := store.NewStoreManager(kvstore, storePath)
if err = replaceConfig(e, config); err != nil {
die("error: %v", err)
}
}
func spec(cmd *cobra.Command, args []string) {
storePath := filepath.Join(common.StoreBasePath, cfg.clusterName)
kvstore, err := store.NewStore(store.Backend(cfg.storeBackend), cfg.storeEndpoints)
if err != nil {
die("cannot create store: %v", err)
}
e := store.NewStoreManager(kvstore, storePath)
cd, _, err := getClusterData(e)
if err != nil {
die("%v", err)
}
if cd.Cluster == nil {
die("no cluster spec available")
}
if cd.Cluster.Spec == nil {
die("no cluster spec available")
}
specj, err := json.MarshalIndent(cd.Cluster.Spec, "", "\t")
if err != nil {
die("failed to marshall spec: %v", err)
}
stdout("%s", specj)
}
func configGet(cmd *cobra.Command, args []string) {
storePath := filepath.Join(common.StoreBasePath, cfg.clusterName)
kvstore, err := store.NewStore(store.Backend(cfg.storeBackend), cfg.storeEndpoints)
if err != nil {
die("cannot create store: %v", err)
}
e := store.NewStoreManager(kvstore, storePath)
cfg, err := getConfig(e)
if err != nil {
die("error: %v", err)
}
if cfg == nil {
stdout("config is not defined")
os.Exit(0)
}
cfgj, err := json.MarshalIndent(cfg, "", "\t")
if err != nil {
die("failed to marshall config: %v", err)
}
stdout(string(cfgj))
}
func NewClusterChecker(uid string, cfg config) (*ClusterChecker, error) {
storePath := filepath.Join(common.StoreBasePath, cfg.clusterName)
kvstore, err := store.NewStore(store.Config{
Backend: store.Backend(cfg.storeBackend),
Endpoints: cfg.storeEndpoints,
CertFile: cfg.storeCertFile,
KeyFile: cfg.storeKeyFile,
CAFile: cfg.storeCAFile,
SkipTLSVerify: cfg.storeSkipTlsVerify,
})
if err != nil {
return nil, fmt.Errorf("cannot create store: %v", err)
}
e := store.NewStoreManager(kvstore, storePath)
return &ClusterChecker{
uid: uid,
listenAddress: cfg.listenAddress,
port: cfg.port,
stopListening: cfg.stopListening,
e: e,
endPollonProxyCh: make(chan error),
}, nil
}
// Tests standby elected as new master but fails to become master. Then old
// master comes back and is re-elected as master.
func testFailoverFailed(t *testing.T, syncRepl bool) {
dir, err := ioutil.TempDir("", "stolon")
if err != nil {
t.Fatalf("unexpected err: %v", err)
}
defer os.RemoveAll(dir)
clusterName := uuid.NewV4().String()
tks, tss, tstore := setupServers(t, clusterName, dir, 2, 1, syncRepl, false)
defer shutdown(tks, tss, tstore)
storePath := filepath.Join(common.StoreBasePath, clusterName)
sm := store.NewStoreManager(tstore.store, storePath)
master, standbys := waitMasterStandbysReady(t, sm, tks)
standby := standbys[0]
if syncRepl {
if err := WaitClusterDataSynchronousStandbys([]string{standby.uid}, sm, 30*time.Second); err != nil {
t.Fatalf("expected synchronous standby on keeper %q in cluster data", standby.uid)
}
}
if err := populate(t, master); err != nil {
t.Fatalf("unexpected err: %v", err)
}
if err := write(t, master, 1, 1); err != nil {
t.Fatalf("unexpected err: %v", err)
}
// Stop the keeper process on master, should also stop the database
t.Logf("Stopping current master keeper: %s", master.uid)
master.Stop()
// Wait for cluster data containing standby as master
if err := WaitClusterDataMaster(standby.uid, sm, 30*time.Second); err != nil {
t.Fatalf("expected master %q in cluster view", standby.uid)
}
// Stopping standby before reading the new cluster data and promoting
// TODO(sgotti) this is flacky and the standby can read the data and
// publish new state before it's stopped
t.Logf("Stopping current standby keeper: %s", standby.uid)
standby.Stop()
t.Logf("Starting previous master keeper: %s", master.uid)
master.Start()
// Wait for cluster data containing previous master as master
err = WaitClusterDataMaster(master.uid, sm, 30*time.Second)
if !syncRepl && err != nil {
t.Fatalf("expected master %q in cluster view", master.uid)
}
if syncRepl {
if err == nil {
t.Fatalf("expected timeout since with synchronous replication the old master shouldn't be elected as master")
}
}
}
func TestInitialClusterConfig(t *testing.T) {
dir, err := ioutil.TempDir("", "")
if err != nil {
t.Fatalf("unexpected err: %v", err)
}
defer os.RemoveAll(dir)
tstore, err := NewTestStore(dir)
if err != nil {
t.Fatalf("unexpected err: %v", err)
}
if err := tstore.Start(); err != nil {
t.Fatalf("unexpected err: %v", err)
}
if err := tstore.WaitUp(10 * time.Second); err != nil {
t.Fatalf("error waiting on store up: %v", err)
}
defer tstore.Stop()
clusterName := uuid.NewV4().String()
storeEndpoints := fmt.Sprintf("%s:%s", tstore.listenAddress, tstore.port)
storePath := filepath.Join(common.StoreBasePath, clusterName)
kvstore, err := store.NewStore(tstore.storeBackend, storeEndpoints)
if err != nil {
t.Fatalf("cannot create store: %v", err)
}
e := store.NewStoreManager(kvstore, storePath)
tmpFile, err := ioutil.TempFile(dir, "initial-cluster-config.json")
if err != nil {
t.Fatalf("unexpected err: %v", err)
}
defer tmpFile.Close()
tmpFile.WriteString(`{ "synchronous_replication": true }`)
ts, err := NewTestSentinel(dir, clusterName, tstore.storeBackend, storeEndpoints, fmt.Sprintf("--initial-cluster-config=%s", tmpFile.Name()))
if err != nil {
t.Fatalf("unexpected err: %v", err)
}
if err := ts.Start(); err != nil {
t.Fatalf("unexpected err: %v", err)
}
defer ts.Stop()
if err := WaitClusterInitialized(e, 30*time.Second); err != nil {
t.Fatal("expected cluster initialized")
}
cv, _, err := e.GetClusterView()
if err != nil {
t.Fatalf("unexpected err: %v", err)
}
if !*cv.Config.SynchronousReplication {
t.Fatal("expected cluster config with InitWithMultipleKeepers enabled")
}
}
func TestInitialClusterSpec(t *testing.T) {
t.Parallel()
dir, err := ioutil.TempDir("", "")
if err != nil {
t.Fatalf("unexpected err: %v", err)
}
defer os.RemoveAll(dir)
tstore := setupStore(t, dir)
defer tstore.Stop()
clusterName := uuid.NewV4().String()
storeEndpoints := fmt.Sprintf("%s:%s", tstore.listenAddress, tstore.port)
storePath := filepath.Join(common.StoreBasePath, clusterName)
kvstore, err := store.NewStore(tstore.storeBackend, storeEndpoints)
if err != nil {
t.Fatalf("cannot create store: %v", err)
}
e := store.NewStoreManager(kvstore, storePath)
initialClusterSpec := &cluster.ClusterSpec{
InitMode: cluster.ClusterInitModeNew,
SleepInterval: cluster.Duration{Duration: 2 * time.Second},
FailInterval: cluster.Duration{Duration: 5 * time.Second},
ConvergenceTimeout: cluster.Duration{Duration: 30 * time.Second},
SynchronousReplication: true,
}
initialClusterSpecFile, err := writeClusterSpec(dir, initialClusterSpec)
if err != nil {
t.Fatalf("unexpected err: %v", err)
}
ts, err := NewTestSentinel(t, dir, clusterName, tstore.storeBackend, storeEndpoints, fmt.Sprintf("--initial-cluster-spec=%s", initialClusterSpecFile))
if err != nil {
t.Fatalf("unexpected err: %v", err)
}
if err := ts.Start(); err != nil {
t.Fatalf("unexpected err: %v", err)
}
defer ts.Stop()
if err := WaitClusterPhase(e, cluster.ClusterPhaseInitializing, 60*time.Second); err != nil {
t.Fatal("expected cluster in initializing phase")
}
cd, _, err := e.GetClusterData()
if err != nil {
t.Fatalf("unexpected err: %v", err)
}
if !cd.Cluster.Spec.SynchronousReplication {
t.Fatal("expected cluster spec with SynchronousReplication enabled")
}
}
// tests that a master restart with changed address for both keeper and
// postgres (without triggering failover since it restart before being marked
// ad failed) make the slave continue to sync using the new address
func TestMasterChangedAddress(t *testing.T) {
t.Parallel()
dir, err := ioutil.TempDir("", "stolon")
if err != nil {
t.Fatalf("unexpected err: %v", err)
}
defer os.RemoveAll(dir)
clusterName := uuid.NewV4().String()
tks, tss, tstore := setupServers(t, clusterName, dir, 2, 1, false, false)
defer shutdown(tks, tss, tstore)
storePath := filepath.Join(common.StoreBasePath, clusterName)
sm := store.NewStoreManager(tstore.store, storePath)
master, standbys := waitMasterStandbysReady(t, sm, tks)
if err := populate(t, master); err != nil {
t.Fatalf("unexpected err: %v", err)
}
if err := write(t, master, 1, 1); err != nil {
t.Fatalf("unexpected err: %v", err)
}
// Wait standby synced with master
if err := waitLines(t, master, 1, 60*time.Second); err != nil {
t.Fatalf("unexpected err: %v", err)
}
// Restart the keeper process on master with new keeper and postgres
// addresses (in this case only the port is changed)
t.Logf("Restarting current master keeper %q with different addresses", master.uid)
master.Stop()
storeEndpoints := fmt.Sprintf("%s:%s", tstore.listenAddress, tstore.port)
master, err = NewTestKeeperWithID(t, dir, master.uid, clusterName, pgSUUsername, pgSUPassword, pgReplUsername, pgReplPassword, tstore.storeBackend, storeEndpoints)
tks[master.uid] = master
if err := master.Start(); err != nil {
t.Fatalf("unexpected err: %v", err)
}
if err := master.WaitRole(common.RoleMaster, 30*time.Second); err != nil {
t.Fatalf("unexpected err: %v", err)
}
if err := write(t, master, 2, 2); err != nil {
t.Fatalf("unexpected err: %v", err)
}
// Wait standby synced to master with changed address
if err := waitLines(t, standbys[0], 2, 60*time.Second); err != nil {
t.Fatalf("unexpected err: %v", err)
}
}
// Tests standby elected as new master but fails to become master. Then old
// master comes back and is re-elected as master.
func testFailoverFailed(t *testing.T, syncRepl bool) {
dir, err := ioutil.TempDir("", "stolon")
if err != nil {
t.Fatalf("unexpected err: %v", err)
}
defer os.RemoveAll(dir)
clusterName := uuid.NewV4().String()
tks, tss, tstore := setupServers(t, clusterName, dir, 2, 1, syncRepl, false)
defer shutdown(tks, tss, tstore)
storeEndpoints := fmt.Sprintf("%s:%s", tstore.listenAddress, tstore.port)
storePath := filepath.Join(common.StoreBasePath, clusterName)
kvstore, err := store.NewStore(tstore.storeBackend, storeEndpoints)
if err != nil {
t.Fatalf("cannot create store: %v", err)
}
e := store.NewStoreManager(kvstore, storePath)
master, standbys, err := getRoles(t, tks)
if err != nil {
t.Fatalf("unexpected err: %v", err)
}
standby := standbys[0]
if err := populate(t, master); err != nil {
t.Fatalf("unexpected err: %v", err)
}
if err := write(t, master, 1, 1); err != nil {
t.Fatalf("unexpected err: %v", err)
}
// Stop the keeper process on master, should also stop the database
t.Logf("Stopping current master keeper: %s", master.id)
master.Stop()
// Wait for cluster data containing standy as master
if err := WaitClusterDataMaster(standby.id, e, 30*time.Second); err != nil {
t.Fatalf("expected master %q in cluster view", standby.id)
}
// Stopping standby before reading the new cluster data and promoting
t.Logf("Stopping current stanby keeper: %s", master.id)
standby.Stop()
t.Logf("Starting previous master keeper: %s", master.id)
master.Start()
// Wait for cluster data containing previous master as master
if err := WaitClusterDataMaster(master.id, e, 30*time.Second); err != nil {
t.Fatalf("expected master %q in cluster view", master.id)
}
}
func NewSentinel(uid string, cfg *config, stop chan bool, end chan bool) (*Sentinel, error) {
var initialClusterSpec *cluster.ClusterSpec
if cfg.initialClusterSpecFile != "" {
configData, err := ioutil.ReadFile(cfg.initialClusterSpecFile)
if err != nil {
return nil, fmt.Errorf("cannot read provided initial cluster config file: %v", err)
}
if err := json.Unmarshal(configData, &initialClusterSpec); err != nil {
return nil, fmt.Errorf("cannot parse provided initial cluster config: %v", err)
}
log.Debug("initialClusterSpec dump", zap.String("initialClusterSpec", spew.Sdump(initialClusterSpec)))
if err := initialClusterSpec.Validate(); err != nil {
return nil, fmt.Errorf("invalid initial cluster: %v", err)
}
}
storePath := filepath.Join(common.StoreBasePath, cfg.clusterName)
kvstore, err := store.NewStore(store.Config{
Backend: store.Backend(cfg.storeBackend),
Endpoints: cfg.storeEndpoints,
CertFile: cfg.storeCertFile,
KeyFile: cfg.storeKeyFile,
CAFile: cfg.storeCAFile,
SkipTLSVerify: cfg.storeSkipTlsVerify,
})
if err != nil {
return nil, fmt.Errorf("cannot create store: %v", err)
}
e := store.NewStoreManager(kvstore, storePath)
candidate := leadership.NewCandidate(kvstore, filepath.Join(storePath, common.SentinelLeaderKey), uid, store.MinTTL)
return &Sentinel{
uid: uid,
cfg: cfg,
e: e,
candidate: candidate,
leader: false,
initialClusterSpec: initialClusterSpec,
stop: stop,
end: end,
UIDFn: common.UID,
// This is just to choose a pseudo random keeper so
// use math.rand (no need for crypto.rand) without an
// initial seed.
RandFn: rand.Intn,
sleepInterval: cluster.DefaultSleepInterval,
requestTimeout: cluster.DefaultRequestTimeout,
}, nil
}
func NewStore() (*store.StoreManager, error) {
storePath := filepath.Join(common.StoreBasePath, cfg.clusterName)
kvstore, err := store.NewStore(store.Config{
Backend: store.Backend(cfg.storeBackend),
Endpoints: cfg.storeEndpoints,
CertFile: cfg.storeCertFile,
KeyFile: cfg.storeKeyFile,
CAFile: cfg.storeCAFile,
SkipTLSVerify: cfg.storeSkipTlsVerify,
})
if err != nil {
return nil, fmt.Errorf("cannot create store: %v", err)
}
return store.NewStoreManager(kvstore, storePath), nil
}
func testFailover(t *testing.T, syncRepl bool) {
dir, err := ioutil.TempDir("", "stolon")
if err != nil {
t.Fatalf("unexpected err: %v", err)
}
defer os.RemoveAll(dir)
clusterName := uuid.NewV4().String()
tks, tss, tstore := setupServers(t, clusterName, dir, 2, 1, syncRepl, false)
defer shutdown(tks, tss, tstore)
storePath := filepath.Join(common.StoreBasePath, clusterName)
sm := store.NewStoreManager(tstore.store, storePath)
master, standbys := waitMasterStandbysReady(t, sm, tks)
standby := standbys[0]
if syncRepl {
if err := WaitClusterDataSynchronousStandbys([]string{standby.uid}, sm, 30*time.Second); err != nil {
t.Fatalf("expected synchronous standby on keeper %q in cluster data", standby.uid)
}
}
if err := populate(t, master); err != nil {
t.Fatalf("unexpected err: %v", err)
}
if err := write(t, master, 1, 1); err != nil {
t.Fatalf("unexpected err: %v", err)
}
// Stop the keeper process on master, should also stop the database
t.Logf("Stopping current master keeper: %s", master.uid)
master.Stop()
if err := standby.WaitRole(common.RoleMaster, 30*time.Second); err != nil {
t.Fatalf("unexpected err: %v", err)
}
c, err := getLines(t, standby)
if err != nil {
t.Fatalf("unexpected err: %v", err)
}
if c != 1 {
t.Fatalf("wrong number of lines, want: %d, got: %d", 1, c)
}
}
func NewClusterChecker(id string, cfg config) (*ClusterChecker, error) {
storePath := filepath.Join(common.StoreBasePath, cfg.clusterName)
kvstore, err := store.NewStore(store.Backend(cfg.storeBackend), cfg.storeEndpoints)
if err != nil {
return nil, fmt.Errorf("cannot create store: %v", err)
}
e := store.NewStoreManager(kvstore, storePath)
return &ClusterChecker{
id: id,
listenAddress: cfg.listenAddress,
port: cfg.port,
stopListening: cfg.stopListening,
e: e,
endPollonProxyCh: make(chan error),
}, nil
}
func configPatch(cmd *cobra.Command, args []string) {
if len(args) > 1 {
die("too many arguments")
}
if cpOpts.file == "" && len(args) < 1 {
die("no patch provided as argument and no patch file provided (--file/-f option)")
}
if cpOpts.file != "" && len(args) == 1 {
die("only one of patch provided as argument or patch file must provided (--file/-f option)")
}
config := []byte{}
if len(args) == 1 {
config = []byte(args[0])
} else {
var err error
if cpOpts.file == "-" {
config, err = ioutil.ReadAll(os.Stdin)
if err != nil {
die("cannot read config file from stdin: %v", err)
}
} else {
config, err = ioutil.ReadFile(cpOpts.file)
if err != nil {
die("cannot read provided config file: %v", err)
}
}
}
storePath := filepath.Join(common.StoreBasePath, cfg.clusterName)
kvstore, err := store.NewStore(store.Backend(cfg.storeBackend), cfg.storeEndpoints)
if err != nil {
die("cannot create store: %v", err)
}
e := store.NewStoreManager(kvstore, storePath)
if err = patchConfig(e, config); err != nil {
die("failed to patch config: %v", err)
}
}
func NewPostgresKeeper(id string, cfg config, stop chan bool, end chan error) (*PostgresKeeper, error) {
storePath := filepath.Join(common.StoreBasePath, cfg.clusterName)
kvstore, err := store.NewStore(store.Backend(cfg.storeBackend), cfg.storeEndpoints)
if err != nil {
return nil, fmt.Errorf("cannot create store: %v", err)
}
e := store.NewStoreManager(kvstore, storePath)
p := &PostgresKeeper{id: id,
dataDir: cfg.dataDir,
e: e,
listenAddress: cfg.listenAddress,
port: cfg.port,
pgListenAddress: cfg.pgListenAddress,
pgPort: cfg.pgPort,
stop: stop,
end: end,
}
return p, nil
}
func TestInitUsers(t *testing.T) {
t.Parallel()
dir, err := ioutil.TempDir("", "")
if err != nil {
t.Fatalf("unexpected err: %v", err)
}
defer os.RemoveAll(dir)
tstore := setupStore(t, dir)
defer tstore.Stop()
storeEndpoints := fmt.Sprintf("%s:%s", tstore.listenAddress, tstore.port)
// Test pg-repl-username == pg-su-username but password different
clusterName := uuid.NewV4().String()
tk, err := NewTestKeeper(t, dir, clusterName, "user01", "password01", "user01", "password02", tstore.storeBackend, storeEndpoints)
if err != nil {
t.Fatalf("unexpected err: %v", err)
}
if err := tk.StartExpect(); err != nil {
t.Fatalf("unexpected err: %v", err)
}
defer tk.Stop()
if err := tk.cmd.Expect("provided superuser name and replication user name are the same but provided passwords are different"); err != nil {
t.Fatalf("expecting keeper reporting provided superuser name and replication user name are the same but provided passwords are different")
}
// Test pg-repl-username == pg-su-username
clusterName = uuid.NewV4().String()
storePath := filepath.Join(common.StoreBasePath, clusterName)
sm := store.NewStoreManager(tstore.store, storePath)
initialClusterSpec := &cluster.ClusterSpec{
InitMode: cluster.ClusterInitModeP(cluster.ClusterInitModeNew),
SleepInterval: &cluster.Duration{Duration: 2 * time.Second},
FailInterval: &cluster.Duration{Duration: 5 * time.Second},
ConvergenceTimeout: &cluster.Duration{Duration: 30 * time.Second},
}
initialClusterSpecFile, err := writeClusterSpec(dir, initialClusterSpec)
if err != nil {
t.Fatalf("unexpected err: %v", err)
}
ts, err := NewTestSentinel(t, dir, clusterName, tstore.storeBackend, storeEndpoints, fmt.Sprintf("--initial-cluster-spec=%s", initialClusterSpecFile))
if err != nil {
t.Fatalf("unexpected err: %v", err)
}
if err := ts.Start(); err != nil {
t.Fatalf("unexpected err: %v", err)
}
defer ts.Stop()
if err := WaitClusterPhase(sm, cluster.ClusterPhaseInitializing, 30*time.Second); err != nil {
t.Fatal("expected cluster in initializing phase")
}
tk2, err := NewTestKeeper(t, dir, clusterName, "user01", "password", "user01", "password", tstore.storeBackend, storeEndpoints)
if err != nil {
t.Fatalf("unexpected err: %v", err)
}
if err := tk2.StartExpect(); err != nil {
t.Fatalf("unexpected err: %v", err)
}
defer tk2.Stop()
if err := tk2.cmd.ExpectTimeout("replication role added to superuser", 60*time.Second); err != nil {
t.Fatalf("expecting keeper reporting replication role added to superuser")
}
// Test pg-repl-username != pg-su-username and pg-su-password defined
clusterName = uuid.NewV4().String()
storePath = filepath.Join(common.StoreBasePath, clusterName)
sm = store.NewStoreManager(tstore.store, storePath)
ts2, err := NewTestSentinel(t, dir, clusterName, tstore.storeBackend, storeEndpoints, fmt.Sprintf("--initial-cluster-spec=%s", initialClusterSpecFile))
if err != nil {
t.Fatalf("unexpected err: %v", err)
}
if err := ts2.Start(); err != nil {
t.Fatalf("unexpected err: %v", err)
}
defer ts2.Stop()
if err := WaitClusterPhase(sm, cluster.ClusterPhaseInitializing, 60*time.Second); err != nil {
t.Fatal("expected cluster in initializing phase")
}
tk3, err := NewTestKeeper(t, dir, clusterName, "user01", "password", "user02", "password", tstore.storeBackend, storeEndpoints)
if err != nil {
t.Fatalf("unexpected err: %v", err)
}
if err := tk3.StartExpect(); err != nil {
t.Fatalf("unexpected err: %v", err)
}
defer tk3.Stop()
if err := tk3.cmd.ExpectTimeout("superuser password set", 60*time.Second); err != nil {
t.Fatalf("expecting keeper reporting superuser password set")
}
if err := tk3.cmd.ExpectTimeout("replication role created role=user02", 60*time.Second); err != nil {
t.Fatalf("expecting keeper reporting replication role user02 created")
}
}
func TestProxyListening(t *testing.T) {
dir, err := ioutil.TempDir("", "")
if err != nil {
t.Fatalf("unexpected err: %v", err)
}
defer os.RemoveAll(dir)
clusterName := uuid.NewV4().String()
tstore, err := NewTestStore(dir)
if err != nil {
t.Fatalf("unexpected err: %v", err)
}
storeEndpoints := fmt.Sprintf("%s:%s", tstore.listenAddress, tstore.port)
tp, err := NewTestProxy(dir, clusterName, tstore.storeBackend, storeEndpoints)
if err != nil {
t.Fatalf("unexpected err: %v", err)
}
if err := tp.Start(); err != nil {
t.Fatalf("unexpected err: %v", err)
}
defer tp.Stop()
log.Printf("test proxy start with store down. Should not listen")
// tp should not listen because it cannot talk with store
if err := tp.WaitNotListening(10 * time.Second); err != nil {
t.Fatalf("expecting tp not listening due to failed store communication, but it's listening.")
}
tp.Stop()
if err := tstore.Start(); err != nil {
t.Fatalf("unexpected err: %v", err)
}
if err := tstore.WaitUp(10 * time.Second); err != nil {
t.Fatalf("error waiting on store up: %v", err)
}
defer func() {
if tstore.cmd != nil {
tstore.Stop()
}
}()
storePath := filepath.Join(common.StoreBasePath, clusterName)
kvstore, err := store.NewStore(tstore.storeBackend, storeEndpoints)
if err != nil {
t.Fatalf("cannot create store: %v", err)
}
e := store.NewStoreManager(kvstore, storePath)
pair, err := e.SetClusterData(cluster.KeepersState{},
&cluster.ClusterView{
Version: 1,
Config: &cluster.NilConfig{
SleepInterval: &cluster.Duration{5 * time.Second},
KeeperFailInterval: &cluster.Duration{10 * time.Second},
},
ProxyConf: &cluster.ProxyConf{
// fake pg address, not relevant
Host: "localhost",
Port: "5432",
},
}, nil)
if err != nil {
t.Fatalf("unexpected err: %v", err)
}
// test proxy start with the store up
log.Printf("test proxy start with the store up. Should listen")
if err := tp.Start(); err != nil {
t.Fatalf("unexpected err: %v", err)
}
// tp should listen
if err := tp.WaitListening(10 * time.Second); err != nil {
t.Fatalf("expecting tp listening, but it's not listening.")
}
log.Printf("test proxy error communicating with store. Should stop listening")
// Stop store
tstore.Stop()
if err := tstore.WaitDown(10 * time.Second); err != nil {
t.Fatalf("error waiting on store down: %v", err)
}
// tp should not listen because it cannot talk with the store
if err := tp.WaitNotListening(10 * time.Second); err != nil {
t.Fatalf("expecting tp not listening due to failed store communication, but it's listening.")
}
log.Printf("test proxy communication with store restored. Should start listening")
// Start store
if err := tstore.Start(); err != nil {
t.Fatalf("unexpected err: %v", err)
}
if err := tstore.WaitUp(10 * time.Second); err != nil {
t.Fatalf("error waiting on store up: %v", err)
}
// tp should listen
if err := tp.WaitListening(10 * time.Second); err != nil {
t.Fatalf("expecting tp listening, but it's not listening.")
}
log.Printf("test proxyConf removed. Should continue listening")
// remove proxyConf
pair, err = e.SetClusterData(cluster.KeepersState{},
&cluster.ClusterView{
Version: 1,
Config: &cluster.NilConfig{
SleepInterval: &cluster.Duration{5 * time.Second},
KeeperFailInterval: &cluster.Duration{10 * time.Second},
},
ProxyConf: nil,
}, pair)
if err != nil {
t.Fatalf("unexpected err: %v", err)
}
// tp should listen
if err := tp.WaitListening(10 * time.Second); err != nil {
t.Fatalf("expecting tp listening, but it's not listening.")
}
log.Printf("test proxyConf restored. Should continue listening")
// Set proxyConf again
pair, err = e.SetClusterData(cluster.KeepersState{},
&cluster.ClusterView{
Version: 1,
Config: &cluster.NilConfig{
SleepInterval: &cluster.Duration{5 * time.Second},
KeeperFailInterval: &cluster.Duration{10 * time.Second},
},
ProxyConf: &cluster.ProxyConf{
// fake pg address, not relevant
Host: "localhost",
Port: "5432",
},
}, pair)
if err != nil {
t.Fatalf("unexpected err: %v", err)
}
// tp should listen
if err := tp.WaitListening(10 * time.Second); err != nil {
t.Fatalf("expecting tp listening, but it's not listening.")
}
log.Printf("test clusterView removed. Should continue listening")
// remove whole clusterview
_, err = e.SetClusterData(cluster.KeepersState{}, nil, pair)
if err != nil {
t.Fatalf("unexpected err: %v", err)
}
// tp should listen
if err := tp.WaitListening(10 * time.Second); err != nil {
t.Fatalf("expecting tp listening, but it's not listening.")
}
}
func TestLoweredMaxStandbysPerSender(t *testing.T) {
t.Parallel()
dir, err := ioutil.TempDir("", "stolon")
if err != nil {
t.Fatalf("unexpected err: %v", err)
}
defer os.RemoveAll(dir)
clusterName := uuid.NewV4().String()
initialClusterSpec := &cluster.ClusterSpec{
InitMode: cluster.ClusterInitModeP(cluster.ClusterInitModeNew),
SleepInterval: &cluster.Duration{Duration: 2 * time.Second},
FailInterval: &cluster.Duration{Duration: 5 * time.Second},
ConvergenceTimeout: &cluster.Duration{Duration: 30 * time.Second},
MaxStandbysPerSender: cluster.Uint16P(2),
PGParameters: make(cluster.PGParameters),
}
// Create 3 keepers
tks, tss, tstore := setupServersCustom(t, clusterName, dir, 3, 1, initialClusterSpec)
defer shutdown(tks, tss, tstore)
storeEndpoints := fmt.Sprintf("%s:%s", tstore.listenAddress, tstore.port)
storePath := filepath.Join(common.StoreBasePath, clusterName)
sm := store.NewStoreManager(tstore.store, storePath)
// Wait for clusterView containing a master
masterUID, err := WaitClusterDataWithMaster(sm, 30*time.Second)
if err != nil {
t.Fatal("expected a master in cluster view")
}
master := tks[masterUID]
waitKeeperReady(t, sm, master)
if err := populate(t, master); err != nil {
t.Fatalf("unexpected err: %v", err)
}
if err := write(t, master, 1, 1); err != nil {
t.Fatalf("unexpected err: %v", err)
}
c, err := getLines(t, master)
if err != nil {
t.Fatalf("unexpected err: %v", err)
}
if c != 1 {
t.Fatalf("wrong number of lines, want: %d, got: %d", 1, c)
}
if err := WaitNumDBs(sm, 3, 30*time.Second); err != nil {
t.Fatalf("expected 3 DBs in cluster data: %v", err)
}
// Set MaxStandbysPerSender to 1
err = StolonCtl(clusterName, tstore.storeBackend, storeEndpoints, "update", "--patch", `{ "maxStandbysPerSender" : 1 }`)
if err != nil {
t.Fatalf("unexpected err: %v", err)
}
// Wait for only 1 standby
if err := WaitNumDBs(sm, 2, 30*time.Second); err != nil {
t.Fatalf("expected 2 DBs in cluster data: %v", err)
}
}
func TestFailedStandby(t *testing.T) {
t.Parallel()
dir, err := ioutil.TempDir("", "stolon")
if err != nil {
t.Fatalf("unexpected err: %v", err)
}
defer os.RemoveAll(dir)
clusterName := uuid.NewV4().String()
initialClusterSpec := &cluster.ClusterSpec{
InitMode: cluster.ClusterInitModeP(cluster.ClusterInitModeNew),
SleepInterval: &cluster.Duration{Duration: 2 * time.Second},
FailInterval: &cluster.Duration{Duration: 5 * time.Second},
ConvergenceTimeout: &cluster.Duration{Duration: 30 * time.Second},
MaxStandbysPerSender: cluster.Uint16P(1),
PGParameters: make(cluster.PGParameters),
}
// Create 3 keepers
tks, tss, tstore := setupServersCustom(t, clusterName, dir, 3, 1, initialClusterSpec)
defer shutdown(tks, tss, tstore)
storePath := filepath.Join(common.StoreBasePath, clusterName)
sm := store.NewStoreManager(tstore.store, storePath)
// Wait for clusterView containing a master
masterUID, err := WaitClusterDataWithMaster(sm, 30*time.Second)
if err != nil {
t.Fatal("expected a master in cluster view")
}
master := tks[masterUID]
waitKeeperReady(t, sm, master)
if err := populate(t, master); err != nil {
t.Fatalf("unexpected err: %v", err)
}
if err := write(t, master, 1, 1); err != nil {
t.Fatalf("unexpected err: %v", err)
}
c, err := getLines(t, master)
if err != nil {
t.Fatalf("unexpected err: %v", err)
}
if c != 1 {
t.Fatalf("wrong number of lines, want: %d, got: %d", 1, c)
}
if err := WaitNumDBs(sm, 2, 30*time.Second); err != nil {
t.Fatalf("expected 2 DBs in cluster data: %v", err)
}
cd, _, err := sm.GetClusterData()
if err != nil {
t.Fatalf("unexpected err: %v", err)
}
// Get current standby
var standby *TestKeeper
for _, db := range cd.DBs {
if db.UID == cd.Cluster.Status.Master {
continue
}
standby = tks[db.Spec.KeeperUID]
}
if err := waitLines(t, standby, 1, 30*time.Second); err != nil {
t.Fatalf("unexpected err: %v", err)
}
// Stop current standby. The other keeper should be choosed as new standby
t.Logf("Stopping current standby keeper: %s", standby.uid)
standby.Stop()
// Wait for other keeper to have a standby db assigned
var newStandby *TestKeeper
for _, tk := range tks {
if tk.uid != master.uid && tk.uid != standby.uid {
newStandby = tk
}
}
if err := WaitStandbyKeeper(sm, newStandby.uid, 20*time.Second); err != nil {
t.Fatalf("expected keeper %s to have a standby db assigned: %v", newStandby.uid, err)
}
// Wait for new standby declared as good and remove of old standby
if err := WaitNumDBs(sm, 2, 30*time.Second); err != nil {
t.Fatalf("expected 2 DBs in cluster data: %v", err)
}
}
func NewPostgresKeeper(cfg *config, stop chan bool, end chan error) (*PostgresKeeper, error) {
storePath := filepath.Join(common.StoreBasePath, cfg.clusterName)
kvstore, err := store.NewStore(store.Backend(cfg.storeBackend), cfg.storeEndpoints)
if err != nil {
return nil, fmt.Errorf("cannot create store: %v", err)
}
e := store.NewStoreManager(kvstore, storePath)
p := &PostgresKeeper{
cfg: cfg,
dataDir: cfg.dataDir,
storeBackend: cfg.storeBackend,
storeEndpoints: cfg.storeEndpoints,
debug: cfg.debug,
pgListenAddress: cfg.pgListenAddress,
pgPort: cfg.pgPort,
pgBinPath: cfg.pgBinPath,
pgReplUsername: cfg.pgReplUsername,
pgReplPassword: cfg.pgReplPassword,
pgSUUsername: cfg.pgSUUsername,
pgSUPassword: cfg.pgSUPassword,
pgInitialSUUsername: cfg.pgInitialSUUsername,
sleepInterval: cluster.DefaultSleepInterval,
requestTimeout: cluster.DefaultRequestTimeout,
keeperLocalState: &KeeperLocalState{},
dbLocalState: &DBLocalState{},
e: e,
stop: stop,
end: end,
}
err = p.loadKeeperLocalState()
if err != nil && !os.IsNotExist(err) {
return nil, fmt.Errorf("failed to load keeper local state file: %v", err)
}
if p.keeperLocalState.UID != "" && p.cfg.id != "" && p.keeperLocalState.UID != p.cfg.id {
fmt.Printf("saved id %q differs from configuration id: %q\n", p.keeperLocalState.UID, cfg.id)
os.Exit(1)
}
if p.keeperLocalState.UID == "" {
p.keeperLocalState.UID = cfg.id
if cfg.id == "" {
p.keeperLocalState.UID = common.UID()
log.Info("uid generated", zap.String("id", p.keeperLocalState.UID))
}
if err = p.saveKeeperLocalState(); err != nil {
fmt.Printf("error: %v\n", err)
os.Exit(1)
}
}
log.Info("keeper uid", zap.String("uid", p.keeperLocalState.UID))
err = p.loadDBLocalState()
if err != nil && !os.IsNotExist(err) {
return nil, fmt.Errorf("failed to load db local state file: %v", err)
}
return p, nil
}
func testTimelineFork(t *testing.T, syncRepl, usePgrewind bool) {
dir, err := ioutil.TempDir("", "stolon")
if err != nil {
t.Fatalf("unexpected err: %v", err)
}
defer os.RemoveAll(dir)
clusterName := uuid.NewV4().String()
tks, tss, tstore := setupServers(t, clusterName, dir, 2, 1, syncRepl, usePgrewind)
defer shutdown(tks, tss, tstore)
storeEndpoints := fmt.Sprintf("%s:%s", tstore.listenAddress, tstore.port)
storePath := filepath.Join(common.StoreBasePath, clusterName)
sm := store.NewStoreManager(tstore.store, storePath)
master, standbys := waitMasterStandbysReady(t, sm, tks)
if syncRepl {
if err := WaitClusterDataSynchronousStandbys([]string{standbys[0].uid}, sm, 30*time.Second); err != nil {
t.Fatalf("expected synchronous standby on keeper %q in cluster data", standbys[0].uid)
}
}
if err := populate(t, master); err != nil {
t.Fatalf("unexpected err: %v", err)
}
if err := write(t, master, 1, 1); err != nil {
t.Fatalf("unexpected err: %v", err)
}
// Wait replicated data to standby
if err := waitLines(t, standbys[0], 1, 10*time.Second); err != nil {
t.Fatalf("unexpected err: %v", err)
}
// Add another standby
tk, err := NewTestKeeper(t, dir, clusterName, pgSUUsername, pgSUPassword, pgReplUsername, pgReplPassword, tstore.storeBackend, storeEndpoints)
if err != nil {
t.Fatalf("unexpected err: %v", err)
}
tks[tk.uid] = tk
if err := tk.Start(); err != nil {
t.Fatalf("unexpected err: %v", err)
}
standbys = append(standbys, tk)
// Wait replicated data to standby
if err := waitLines(t, standbys[1], 1, 30*time.Second); err != nil {
t.Fatalf("unexpected err: %v", err)
}
// Stop one standby
t.Logf("Stopping standby[0]: %s", standbys[0].uid)
standbys[0].Stop()
if err := standbys[0].WaitDBDown(60 * time.Second); err != nil {
t.Fatalf("unexpected err: %v", err)
}
// Write to master (and replicated to remaining standby)
if err := write(t, master, 2, 2); err != nil {
t.Fatalf("unexpected err: %v", err)
}
// Wait replicated data to standby[1]
if err := waitLines(t, standbys[1], 2, 10*time.Second); err != nil {
t.Fatalf("unexpected err: %v", err)
}
// Stop the master and remaining standby[1]
t.Logf("Stopping master: %s", master.uid)
master.Stop()
t.Logf("Stopping standby[1]: %s", standbys[1].uid)
standbys[1].Stop()
if err := master.WaitDBDown(60 * time.Second); err != nil {
t.Fatalf("unexpected err: %v", err)
}
if err := standbys[1].WaitDBDown(60 * time.Second); err != nil {
t.Fatalf("unexpected err: %v", err)
}
// Start standby[0].
// If synchronous replication is disabled it will be elected as master but it'll be behind (having only one line).
// If synchronous replication is enabled it won't be elected as master
t.Logf("Starting standby[0]: %s", standbys[0].uid)
standbys[0].Start()
waitKeeperReady(t, sm, standbys[0])
err = standbys[0].WaitRole(common.RoleMaster, 60*time.Second)
if !syncRepl && err != nil {
t.Fatalf("unexpected err: %v", err)
}
if syncRepl {
if err == nil {
t.Fatalf("expected timeout since with synchronous replication the standby shouldn't be elected as master")
}
// end here
return
}
c, err := getLines(t, standbys[0])
if err != nil {
t.Fatalf("unexpected err: %v", err)
}
if c != 1 {
t.Fatalf("wrong number of lines, want: %d, got: %d", 1, c)
}
// Start the other standby, it should be ahead of current on previous timeline and should full resync himself
t.Logf("Starting standby[1]: %s", standbys[1].uid)
standbys[1].Start()
// Standby[1] will start, then it'll detect it's in another timelinehistory,
// will stop, full resync and start. We have to avoid detecting it up
// at the first start. Do this waiting for the number of expected lines.
if err := waitLines(t, standbys[1], 1, 60*time.Second); err != nil {
t.Fatalf("unexpected err: %v", err)
}
if err := standbys[1].WaitRole(common.RoleStandby, 60*time.Second); err != nil {
t.Fatalf("unexpected err: %v", err)
}
}
func TestInitWithMultipleKeepers(t *testing.T) {
t.Parallel()
dir, err := ioutil.TempDir("", "stolon")
if err != nil {
t.Fatalf("unexpected err: %v", err)
}
defer os.RemoveAll(dir)
tstore := setupStore(t, dir)
storeEndpoints := fmt.Sprintf("%s:%s", tstore.listenAddress, tstore.port)
clusterName := uuid.NewV4().String()
storePath := filepath.Join(common.StoreBasePath, clusterName)
sm := store.NewStoreManager(tstore.store, storePath)
initialClusterSpec := &cluster.ClusterSpec{
InitMode: cluster.ClusterInitModeP(cluster.ClusterInitModeNew),
FailInterval: &cluster.Duration{Duration: 10 * time.Second},
ConvergenceTimeout: &cluster.Duration{Duration: 30 * time.Second},
}
initialClusterSpecFile, err := writeClusterSpec(dir, initialClusterSpec)
if err != nil {
t.Fatalf("unexpected err: %v", err)
}
tks := testKeepers{}
tss := testSentinels{}
// Start 3 keepers
for i := uint8(0); i < 3; i++ {
tk, err := NewTestKeeper(t, dir, clusterName, pgSUUsername, pgSUPassword, pgReplUsername, pgReplPassword, tstore.storeBackend, storeEndpoints)
if err != nil {
t.Fatalf("unexpected err: %v", err)
}
if err := tk.Start(); err != nil {
t.Fatalf("unexpected err: %v", err)
}
tks[tk.uid] = tk
}
// Start 2 sentinels
for i := uint8(0); i < 2; i++ {
ts, err := NewTestSentinel(t, dir, clusterName, tstore.storeBackend, storeEndpoints, fmt.Sprintf("--initial-cluster-spec=%s", initialClusterSpecFile))
if err != nil {
t.Fatalf("unexpected err: %v", err)
}
if err := ts.Start(); err != nil {
t.Fatalf("unexpected err: %v", err)
}
tss[ts.uid] = ts
}
defer shutdown(tks, tss, tstore)
// Wait for clusterView containing a master
masterUID, err := WaitClusterDataWithMaster(sm, 30*time.Second)
if err != nil {
t.Fatal("expected a master in cluster view")
}
waitKeeperReady(t, sm, tks[masterUID])
}
func testPartition1(t *testing.T, syncRepl, usePgrewind bool) {
dir, err := ioutil.TempDir("", "stolon")
if err != nil {
t.Fatalf("unexpected err: %v", err)
}
defer os.RemoveAll(dir)
clusterName := uuid.NewV4().String()
tks, tss, tstore := setupServers(t, clusterName, dir, 2, 1, syncRepl, usePgrewind)
defer shutdown(tks, tss, tstore)
storeEndpoints := fmt.Sprintf("%s:%s", tstore.listenAddress, tstore.port)
storePath := filepath.Join(common.StoreBasePath, clusterName)
sm := store.NewStoreManager(tstore.store, storePath)
master, standbys := waitMasterStandbysReady(t, sm, tks)
if syncRepl {
if err := WaitClusterDataSynchronousStandbys([]string{standbys[0].uid}, sm, 30*time.Second); err != nil {
t.Fatalf("expected synchronous standby on keeper %q in cluster data", standbys[0].uid)
}
}
if err := populate(t, master); err != nil {
t.Fatalf("unexpected err: %v", err)
}
if err := write(t, master, 1, 1); err != nil {
t.Fatalf("unexpected err: %v", err)
}
// Freeze the keeper and postgres processes on the master
t.Logf("SIGSTOPping current master keeper: %s", master.uid)
if err := master.Signal(syscall.SIGSTOP); err != nil {
t.Fatalf("unexpected err: %v", err)
}
t.Logf("SIGSTOPping current master postgres: %s", master.uid)
if err := master.SignalPG(syscall.SIGSTOP); err != nil {
t.Fatalf("unexpected err: %v", err)
}
if err := standbys[0].WaitRole(common.RoleMaster, 60*time.Second); err != nil {
t.Fatalf("unexpected err: %v", err)
}
c, err := getLines(t, standbys[0])
if err != nil {
t.Fatalf("unexpected err: %v", err)
}
if c != 1 {
t.Fatalf("wrong number of lines, want: %d, got: %d", 1, c)
}
// Add another standby so we'll have 2 standbys. With only 1 standby,
// when using synchronous replication, the test will block forever when
// writing to the new master since there's not active synchronous
// standby.
tk, err := NewTestKeeper(t, dir, clusterName, pgSUUsername, pgSUPassword, pgReplUsername, pgReplPassword, tstore.storeBackend, storeEndpoints)
if err != nil {
t.Fatalf("unexpected err: %v", err)
}
tks[tk.uid] = tk
if err := tk.Start(); err != nil {
t.Fatalf("unexpected err: %v", err)
}
standbys = append(standbys, tk)
// Wait replicated data to standby
if err := waitLines(t, standbys[1], 1, 30*time.Second); err != nil {
t.Fatalf("unexpected err: %v", err)
}
if err := write(t, standbys[0], 2, 2); err != nil {
t.Fatalf("unexpected err: %v", err)
}
// Make the master come back
t.Logf("Resuming old master keeper: %s", master.uid)
if err := master.Signal(syscall.SIGCONT); err != nil {
t.Fatalf("unexpected err: %v", err)
}
t.Logf("Resuming old master postgres: %s", master.uid)
if err := master.SignalPG(syscall.SIGCONT); err != nil {
t.Fatalf("unexpected err: %v", err)
}
// Wait replicated data to old master
if err := waitLines(t, master, 2, 60*time.Second); err != nil {
t.Fatalf("unexpected err: %v", err)
}
if err := master.WaitRole(common.RoleStandby, 60*time.Second); err != nil {
t.Fatalf("unexpected err: %v", err)
}
}
func update(cmd *cobra.Command, args []string) {
if len(args) > 1 {
die("too many arguments")
}
if updateOpts.file == "" && len(args) < 1 {
die("no cluster spec provided as argument and no file provided (--file/-f option)")
}
if updateOpts.file != "" && len(args) == 1 {
die("only one of cluster spec provided as argument or file must provided (--file/-f option)")
}
data := []byte{}
if len(args) == 1 {
data = []byte(args[0])
} else {
var err error
if updateOpts.file == "-" {
data, err = ioutil.ReadAll(os.Stdin)
if err != nil {
die("cannot read from stdin: %v", err)
}
} else {
data, err = ioutil.ReadFile(updateOpts.file)
if err != nil {
die("cannot read file: %v", err)
}
}
}
storePath := filepath.Join(common.StoreBasePath, cfg.clusterName)
kvstore, err := store.NewStore(store.Backend(cfg.storeBackend), cfg.storeEndpoints)
if err != nil {
die("cannot create store: %v", err)
}
e := store.NewStoreManager(kvstore, storePath)
retry := 0
for retry < maxRetries {
cd, pair, err := getClusterData(e)
if err != nil {
die("%v", err)
}
if cd.Cluster == nil {
die("no cluster spec available")
}
if cd.Cluster.Spec == nil {
die("no cluster spec available")
}
var newcs *cluster.ClusterSpec
if updateOpts.patch {
newcs, err = patchClusterSpec(cd.Cluster.Spec, data)
if err != nil {
die("failed to patch cluster spec: %v", err)
}
} else {
if err := json.Unmarshal(data, &newcs); err != nil {
die("failed to unmarshal cluster spec: %v", err)
}
}
newcs.SetDefaults()
if err := cd.Cluster.UpdateSpec(newcs); err != nil {
die("Cannot update cluster spec: %v", err)
}
// retry if cd has been modified between reading and writing
_, err = e.AtomicPutClusterData(cd, pair)
if err != nil {
if err == libkvstore.ErrKeyModified {
retry++
continue
}
die("cannot update cluster data: %v", err)
}
break
}
if retry == maxRetries {
die("failed to update cluster data after %d retries", maxRetries)
}
}
func setupServers(t *testing.T, clusterName, dir string, numKeepers, numSentinels uint8, syncRepl bool, usePgrewind bool) (testKeepers, testSentinels, *TestStore) {
tstore := setupStore(t, dir)
storeEndpoints := fmt.Sprintf("%s:%s", tstore.listenAddress, tstore.port)
storePath := filepath.Join(common.StoreBasePath, clusterName)
kvstore, err := store.NewStore(tstore.storeBackend, storeEndpoints)
if err != nil {
t.Fatalf("cannot create store: %v", err)
}
e := store.NewStoreManager(kvstore, storePath)
initialClusterSpec := &cluster.ClusterSpec{
InitMode: cluster.ClusterInitModeNew,
SleepInterval: cluster.Duration{Duration: 2 * time.Second},
FailInterval: cluster.Duration{Duration: 5 * time.Second},
ConvergenceTimeout: cluster.Duration{Duration: 30 * time.Second},
SynchronousReplication: syncRepl,
UsePgrewind: usePgrewind,
PGParameters: make(cluster.PGParameters),
}
initialClusterSpecFile, err := writeClusterSpec(dir, initialClusterSpec)
if err != nil {
t.Fatalf("unexpected err: %v", err)
}
tks := map[string]*TestKeeper{}
tss := map[string]*TestSentinel{}
tk, err := NewTestKeeper(t, dir, clusterName, pgSUUsername, pgSUPassword, pgReplUsername, pgReplPassword, tstore.storeBackend, storeEndpoints)
if err != nil {
t.Fatalf("unexpected err: %v", err)
}
tks[tk.id] = tk
t.Logf("tk: %v", tk)
// Start sentinels
for i := uint8(0); i < numSentinels; i++ {
ts, err := NewTestSentinel(t, dir, clusterName, tstore.storeBackend, storeEndpoints, fmt.Sprintf("--initial-cluster-spec=%s", initialClusterSpecFile))
if err != nil {
t.Fatalf("unexpected err: %v", err)
}
if err := ts.Start(); err != nil {
t.Fatalf("unexpected err: %v", err)
}
tss[ts.id] = ts
}
// Start first keeper
if err := tk.Start(); err != nil {
t.Fatalf("unexpected err: %v", err)
}
if err := tk.WaitDBUp(60 * time.Second); err != nil {
t.Fatalf("unexpected err: %v", err)
}
if err := tk.WaitRole(common.RoleMaster, 30*time.Second); err != nil {
t.Fatalf("unexpected err: %v", err)
}
// Wait for clusterView containing tk as master
if err := WaitClusterDataMaster(tk.id, e, 30*time.Second); err != nil {
t.Fatalf("expected master %q in cluster view", tk.id)
}
// Start other keepers
for i := uint8(1); i < numKeepers; i++ {
tk, err := NewTestKeeper(t, dir, clusterName, pgSUUsername, pgSUPassword, pgReplUsername, pgReplPassword, tstore.storeBackend, storeEndpoints)
if err != nil {
t.Fatalf("unexpected err: %v", err)
}
if err := tk.Start(); err != nil {
t.Fatalf("unexpected err: %v", err)
}
if err := tk.WaitDBUp(60 * time.Second); err != nil {
t.Fatalf("unexpected err: %v", err)
}
// Wait for clusterView containing tk as standby
if err := tk.WaitRole(common.RoleStandby, 30*time.Second); err != nil {
t.Fatalf("unexpected err: %v", err)
}
tks[tk.id] = tk
}
return tks, tss, tstore
}
func testInitExisting(t *testing.T, merge bool) {
clusterName := uuid.NewV4().String()
dir, err := ioutil.TempDir("", "")
if err != nil {
t.Fatalf("unexpected err: %v", err)
}
defer os.RemoveAll(dir)
tstore := setupStore(t, dir)
defer tstore.Stop()
storeEndpoints := fmt.Sprintf("%s:%s", tstore.listenAddress, tstore.port)
storePath := filepath.Join(common.StoreBasePath, clusterName)
sm := store.NewStoreManager(tstore.store, storePath)
initialClusterSpec := &cluster.ClusterSpec{
InitMode: cluster.ClusterInitModeP(cluster.ClusterInitModeNew),
SleepInterval: &cluster.Duration{Duration: 2 * time.Second},
FailInterval: &cluster.Duration{Duration: 5 * time.Second},
ConvergenceTimeout: &cluster.Duration{Duration: 30 * time.Second},
PGParameters: cluster.PGParameters{
"archive_mode": "on",
},
}
initialClusterSpecFile, err := writeClusterSpec(dir, initialClusterSpec)
if err != nil {
t.Fatalf("unexpected err: %v", err)
}
ts, err := NewTestSentinel(t, dir, clusterName, tstore.storeBackend, storeEndpoints, fmt.Sprintf("--initial-cluster-spec=%s", initialClusterSpecFile))
if err != nil {
t.Fatalf("unexpected err: %v", err)
}
if err := ts.Start(); err != nil {
t.Fatalf("unexpected err: %v", err)
}
tk, err := NewTestKeeper(t, dir, clusterName, pgSUUsername, pgSUPassword, pgReplUsername, pgReplPassword, tstore.storeBackend, storeEndpoints)
if err != nil {
t.Fatalf("unexpected err: %v", err)
}
if err := tk.Start(); err != nil {
t.Fatalf("unexpected err: %v", err)
}
if err := WaitClusterPhase(sm, cluster.ClusterPhaseNormal, 60*time.Second); err != nil {
t.Fatalf("unexpected err: %v", err)
}
if err := tk.WaitDBUp(60 * time.Second); err != nil {
t.Fatalf("unexpected err: %v", err)
}
if err := populate(t, tk); err != nil {
t.Fatalf("unexpected err: %v", err)
}
if err := write(t, tk, 1, 1); err != nil {
t.Fatalf("unexpected err: %v", err)
}
// Now initialize a new cluster with the existing keeper
initialClusterSpec = &cluster.ClusterSpec{
InitMode: cluster.ClusterInitModeP(cluster.ClusterInitModeExisting),
SleepInterval: &cluster.Duration{Duration: 2 * time.Second},
FailInterval: &cluster.Duration{Duration: 5 * time.Second},
ConvergenceTimeout: &cluster.Duration{Duration: 30 * time.Second},
MergePgParameters: &merge,
ExistingConfig: &cluster.ExistingConfig{
KeeperUID: tk.uid,
},
}
initialClusterSpecFile, err = writeClusterSpec(dir, initialClusterSpec)
if err != nil {
t.Fatalf("unexpected err: %v", err)
}
t.Logf("reinitializing cluster")
// Initialize cluster with new spec
err = StolonCtl(clusterName, tstore.storeBackend, storeEndpoints, "init", "-y", "-f", initialClusterSpecFile)
if err != nil {
t.Fatalf("unexpected err: %v", err)
}
if err := WaitClusterPhase(sm, cluster.ClusterPhaseInitializing, 60*time.Second); err != nil {
t.Fatalf("unexpected err: %v", err)
}
if err := WaitClusterPhase(sm, cluster.ClusterPhaseNormal, 60*time.Second); err != nil {
t.Fatalf("unexpected err: %v", err)
}
if err := tk.WaitDBUp(60 * time.Second); err != nil {
t.Fatalf("unexpected err: %v", err)
}
c, err := getLines(t, tk)
if err != nil {
t.Fatalf("unexpected err: %v", err)
}
if c != 1 {
t.Fatalf("wrong number of lines, want: %d, got: %d", 1, c)
}
pgParameters, err := tk.GetPGParameters()
if err != nil {
t.Fatalf("unexpected err: %v", err)
}
v, ok := pgParameters["archive_mode"]
if merge && v != "on" {
t.Fatalf("expected archive_mode == on got %q", v)
}
if !merge && ok {
t.Fatalf("expected archive_mode empty")
}
cd, _, err := sm.GetClusterData()
// max_connection should be set by initdb
v, ok = cd.Cluster.Spec.PGParameters["archive_mode"]
if merge && v != "on" {
t.Fatalf("expected archive_mode == on got %q", v)
}
if !merge && ok {
t.Fatalf("expected archive_mode empty")
}
tk.Stop()
}
func TestServerParameters(t *testing.T) {
t.Parallel()
dir, err := ioutil.TempDir("", "")
if err != nil {
t.Fatalf("unexpected err: %v", err)
}
defer os.RemoveAll(dir)
tstore, err := NewTestStore(t, dir)
if err != nil {
t.Fatalf("unexpected err: %v", err)
}
if err := tstore.Start(); err != nil {
t.Fatalf("unexpected err: %v", err)
}
if err := tstore.WaitUp(10 * time.Second); err != nil {
t.Fatalf("error waiting on store up: %v", err)
}
storeEndpoints := fmt.Sprintf("%s:%s", tstore.listenAddress, tstore.port)
defer tstore.Stop()
clusterName := uuid.NewV4().String()
storePath := filepath.Join(common.StoreBasePath, clusterName)
sm := store.NewStoreManager(tstore.store, storePath)
initialClusterSpec := &cluster.ClusterSpec{
InitMode: cluster.ClusterInitModeP(cluster.ClusterInitModeNew),
SleepInterval: &cluster.Duration{Duration: 2 * time.Second},
FailInterval: &cluster.Duration{Duration: 5 * time.Second},
ConvergenceTimeout: &cluster.Duration{Duration: 30 * time.Second},
}
initialClusterSpecFile, err := writeClusterSpec(dir, initialClusterSpec)
if err != nil {
t.Fatalf("unexpected err: %v", err)
}
ts, err := NewTestSentinel(t, dir, clusterName, tstore.storeBackend, storeEndpoints, fmt.Sprintf("--initial-cluster-spec=%s", initialClusterSpecFile))
if err != nil {
t.Fatalf("unexpected err: %v", err)
}
if err := ts.Start(); err != nil {
t.Fatalf("unexpected err: %v", err)
}
tk, err := NewTestKeeper(t, dir, clusterName, pgSUUsername, pgSUPassword, pgReplUsername, pgReplPassword, tstore.storeBackend, storeEndpoints)
if err != nil {
t.Fatalf("unexpected err: %v", err)
}
if err := tk.Start(); err != nil {
t.Fatalf("unexpected err: %v", err)
}
if err := WaitClusterPhase(sm, cluster.ClusterPhaseNormal, 60*time.Second); err != nil {
t.Fatalf("unexpected err: %v", err)
}
if err := tk.WaitDBUp(60 * time.Second); err != nil {
t.Fatalf("unexpected err: %v", err)
}
err = StolonCtl(clusterName, tstore.storeBackend, storeEndpoints, "update", "--patch", `{ "pgParameters" : { "unexistent_parameter": "value" } }`)
if err != nil {
t.Fatalf("unexpected err: %v", err)
}
if err := tk.cmd.ExpectTimeout("postgres parameters changed, reloading postgres instance", 30*time.Second); err != nil {
t.Fatalf("unexpected err: %v", err)
}
// On the next keeper check they shouldn't be changed
if err := tk.cmd.ExpectTimeout("postgres parameters not changed", 30*time.Second); err != nil {
t.Fatalf("unexpected err: %v", err)
}
tk.Stop()
// Start tk again, postgres should fail to start due to bad parameter
if err := tk.Start(); err != nil {
t.Fatalf("unexpected err: %v", err)
}
defer tk.Stop()
if err := tk.cmd.ExpectTimeout("failed to start postgres", 30*time.Second); err != nil {
t.Fatalf("unexpected err: %v", err)
}
// Fix wrong parameters
err = StolonCtl(clusterName, tstore.storeBackend, storeEndpoints, "update", "--patch", `{ "pgParameters" : null }`)
if err != nil {
t.Fatalf("unexpected err: %v", err)
}
if err := tk.WaitDBUp(30 * time.Second); err != nil {
t.Fatalf("unexpected err: %v", err)
}
}
func testInitNew(t *testing.T, merge bool) {
clusterName := uuid.NewV4().String()
dir, err := ioutil.TempDir("", "")
if err != nil {
t.Fatalf("unexpected err: %v", err)
}
defer os.RemoveAll(dir)
tstore := setupStore(t, dir)
defer tstore.Stop()
storeEndpoints := fmt.Sprintf("%s:%s", tstore.listenAddress, tstore.port)
storePath := filepath.Join(common.StoreBasePath, clusterName)
sm := store.NewStoreManager(tstore.store, storePath)
initialClusterSpec := &cluster.ClusterSpec{
InitMode: cluster.ClusterInitModeP(cluster.ClusterInitModeNew),
FailInterval: &cluster.Duration{Duration: 10 * time.Second},
ConvergenceTimeout: &cluster.Duration{Duration: 30 * time.Second},
MergePgParameters: &merge,
}
initialClusterSpecFile, err := writeClusterSpec(dir, initialClusterSpec)
if err != nil {
t.Fatalf("unexpected err: %v", err)
}
ts, err := NewTestSentinel(t, dir, clusterName, tstore.storeBackend, storeEndpoints, fmt.Sprintf("--initial-cluster-spec=%s", initialClusterSpecFile))
if err != nil {
t.Fatalf("unexpected err: %v", err)
}
if err := ts.Start(); err != nil {
t.Fatalf("unexpected err: %v", err)
}
tk, err := NewTestKeeper(t, dir, clusterName, pgSUUsername, pgSUPassword, pgReplUsername, pgReplPassword, tstore.storeBackend, storeEndpoints)
if err != nil {
t.Fatalf("unexpected err: %v", err)
}
if err := tk.Start(); err != nil {
t.Fatalf("unexpected err: %v", err)
}
if err := WaitClusterPhase(sm, cluster.ClusterPhaseNormal, 60*time.Second); err != nil {
t.Fatalf("unexpected err: %v", err)
}
if err := tk.WaitDBUp(60 * time.Second); err != nil {
t.Fatalf("unexpected err: %v", err)
}
cd, _, err := sm.GetClusterData()
// max_connection should be set by initdb
_, ok := cd.Cluster.Spec.PGParameters["max_connections"]
if merge && !ok {
t.Fatalf("expected max_connection set in cluster data pgParameters")
}
if !merge && ok {
t.Fatalf("expected no max_connection set in cluster data pgParameters")
}
tk.Stop()
}
func initCluster(cmd *cobra.Command, args []string) {
if len(args) > 1 {
die("too many arguments")
}
data := []byte{}
switch len(args) {
case 1:
data = []byte(args[0])
case 0:
if initOpts.file != "" {
var err error
if initOpts.file == "-" {
data, err = ioutil.ReadAll(os.Stdin)
if err != nil {
die("cannot read from stdin: %v", err)
}
} else {
data, err = ioutil.ReadFile(initOpts.file)
if err != nil {
die("cannot read file: %v", err)
}
}
}
}
storePath := filepath.Join(common.StoreBasePath, cfg.clusterName)
kvstore, err := store.NewStore(store.Backend(cfg.storeBackend), cfg.storeEndpoints)
if err != nil {
die("cannot create store: %v", err)
}
e := store.NewStoreManager(kvstore, storePath)
cd, _, err := e.GetClusterData()
if err != nil {
die("cannot get cluster data: %v", err)
}
if cd != nil {
stdout("WARNING: The current cluster data will be removed")
}
stdout("WARNING: The databases managed by the keepers will be overwrited depending on the provided cluster spec.")
accepted := true
if !initOpts.forceYes {
accepted, err = askConfirmation("Are you sure you want to continue? [yes/no] ")
if err != nil {
die("%v", err)
}
}
if !accepted {
stdout("exiting")
os.Exit(0)
}
cd, _, err = e.GetClusterData()
if err != nil {
die("cannot get cluster data: %v", err)
}
var cs *cluster.ClusterSpec
if len(data) == 0 {
// Define a new cluster spec with initMode "new"
cs = &cluster.ClusterSpec{}
cs.InitMode = cluster.ClusterInitModeNew
} else {
if err := json.Unmarshal(data, &cs); err != nil {
die("failed to unmarshal cluster spec: %v", err)
}
}
cs.SetDefaults()
if err := cs.Validate(); err != nil {
die("invalid cluster spec: %v", err)
}
c := cluster.NewCluster(common.UID(), cs)
cd = cluster.NewClusterData(c)
// We ignore if cd has been modified between reading and writing
if err := e.PutClusterData(cd); err != nil {
die("cannot update cluster data: %v", err)
}
}
func TestProxyListening(t *testing.T) {
t.Parallel()
dir, err := ioutil.TempDir("", "")
if err != nil {
t.Fatalf("unexpected err: %v", err)
}
defer os.RemoveAll(dir)
clusterName := uuid.NewV4().String()
tstore, err := NewTestStore(t, dir)
if err != nil {
t.Fatalf("unexpected err: %v", err)
}
storeEndpoints := fmt.Sprintf("%s:%s", tstore.listenAddress, tstore.port)
tp, err := NewTestProxy(t, dir, clusterName, tstore.storeBackend, storeEndpoints)
if err != nil {
t.Fatalf("unexpected err: %v", err)
}
if err := tp.Start(); err != nil {
t.Fatalf("unexpected err: %v", err)
}
defer tp.Stop()
t.Logf("test proxy start with store down. Should not listen")
// tp should not listen because it cannot talk with store
if err := tp.WaitNotListening(10 * time.Second); err != nil {
t.Fatalf("expecting tp not listening due to failed store communication, but it's listening.")
}
tp.Stop()
if err := tstore.Start(); err != nil {
t.Fatalf("unexpected err: %v", err)
}
if err := tstore.WaitUp(10 * time.Second); err != nil {
t.Fatalf("error waiting on store up: %v", err)
}
defer func() {
if tstore.cmd != nil {
tstore.Stop()
}
}()
storePath := filepath.Join(common.StoreBasePath, clusterName)
sm := store.NewStoreManager(tstore.store, storePath)
cd := &cluster.ClusterData{
FormatVersion: cluster.CurrentCDFormatVersion,
Cluster: &cluster.Cluster{
UID: "01",
Generation: 1,
Spec: &cluster.ClusterSpec{
InitMode: cluster.ClusterInitModeP(cluster.ClusterInitModeNew),
FailInterval: &cluster.Duration{Duration: 10 * time.Second},
},
Status: cluster.ClusterStatus{
CurrentGeneration: 1,
Phase: cluster.ClusterPhaseNormal,
Master: "01",
},
},
Keepers: cluster.Keepers{
"01": &cluster.Keeper{
UID: "01",
Spec: &cluster.KeeperSpec{},
Status: cluster.KeeperStatus{
Healthy: true,
},
},
},
DBs: cluster.DBs{
"01": &cluster.DB{
UID: "01",
Generation: 1,
ChangeTime: time.Time{},
Spec: &cluster.DBSpec{
KeeperUID: "01",
Role: common.RoleMaster,
Followers: []string{"02"},
},
Status: cluster.DBStatus{
Healthy: false,
CurrentGeneration: 1,
},
},
},
Proxy: &cluster.Proxy{
Spec: cluster.ProxySpec{
MasterDBUID: "01",
},
},
}
pair, err := sm.AtomicPutClusterData(cd, nil)
if err != nil {
t.Fatalf("unexpected err: %v", err)
}
// test proxy start with the store up
t.Logf("test proxy start with the store up. Should listen")
if err := tp.Start(); err != nil {
t.Fatalf("unexpected err: %v", err)
}
// tp should listen
if err := tp.WaitListening(10 * time.Second); err != nil {
t.Fatalf("expecting tp listening, but it's not listening.")
}
t.Logf("test proxy error communicating with store. Should stop listening")
// Stop store
tstore.Stop()
if err := tstore.WaitDown(10 * time.Second); err != nil {
t.Fatalf("error waiting on store down: %v", err)
}
// tp should not listen because it cannot talk with the store
if err := tp.WaitNotListening(10 * time.Second); err != nil {
t.Fatalf("expecting tp not listening due to failed store communication, but it's listening.")
}
t.Logf("test proxy communication with store restored. Should start listening")
// Start store
if err := tstore.Start(); err != nil {
t.Fatalf("unexpected err: %v", err)
}
if err := tstore.WaitUp(10 * time.Second); err != nil {
t.Fatalf("error waiting on store up: %v", err)
}
// tp should listen
if err := tp.WaitListening(10 * time.Second); err != nil {
t.Fatalf("expecting tp listening, but it's not listening.")
}
t.Logf("test proxyConf removed. Should continue listening")
// remove proxyConf
cd.Proxy.Spec.MasterDBUID = ""
pair, err = sm.AtomicPutClusterData(cd, pair)
if err != nil {
t.Fatalf("unexpected err: %v", err)
}
// tp should listen
if err := tp.WaitListening(10 * time.Second); err != nil {
t.Fatalf("expecting tp listening, but it's not listening.")
}
t.Logf("test proxyConf restored. Should continue listening")
// Set proxyConf again
cd.Proxy.Spec.MasterDBUID = "01"
pair, err = sm.AtomicPutClusterData(cd, pair)
if err != nil {
t.Fatalf("unexpected err: %v", err)
}
// tp should listen
if err := tp.WaitListening(10 * time.Second); err != nil {
t.Fatalf("expecting tp listening, but it's not listening.")
}
t.Logf("test clusterView removed. Should continue listening")
// remove whole clusterview
_, err = sm.AtomicPutClusterData(nil, pair)
if err != nil {
t.Fatalf("unexpected err: %v", err)
}
// tp should listen
if err := tp.WaitListening(10 * time.Second); err != nil {
t.Fatalf("expecting tp listening, but it's not listening.")
}
}