func TestClient(t *testing.T) {
t.Skipf("Test cluster is not set up properly")
zookeeper.EnsureZkFatjar()
basePath := ""
tc, err := zklib.StartTestCluster(1, nil, nil)
if err != nil {
t.Fatalf("could not start test zk cluster: %s", err)
}
defer os.RemoveAll(tc.Path)
defer tc.Stop()
time.Sleep(time.Second)
servers := []string{fmt.Sprintf("127.0.0.1:%d", tc.Servers[0].Port)}
dsnBytes, err := json.Marshal(zookeeper.DSN{Servers: servers, Timeout: time.Second * 15})
if err != nil {
t.Fatal("unexpected error creating zk DSN: %s", err)
}
dsn := string(dsnBytes)
zClient, err := client.New("zookeeper", dsn, basePath, nil)
zzk.InitializeLocalClient(zClient)
conn, err := zzk.GetLocalConnection("/")
if err != nil {
t.Fatal("unexpected error getting connection")
}
h := host.New()
h.ID = "nodeID"
h.IPAddr = "192.168.1.5"
h.PoolID = "default1"
defer func(old func(string, os.FileMode) error) {
mkdirAll = old
}(mkdirAll)
dir, err := ioutil.TempDir("", "serviced_var_")
if err != nil {
t.Fatalf("could not create tempdir: %s", err)
}
defer os.RemoveAll(dir)
c, err := NewClient(h, dir)
if err != nil {
t.Fatalf("unexpected error creating client: %s", err)
}
defer c.Close()
time.Sleep(time.Second * 5)
// therefore, we need to check that the client was added under the pool from root
nodePath := fmt.Sprintf("/storage/clients/%s", h.IPAddr)
glog.Infof("about to check for %s", nodePath)
if exists, err := conn.Exists(nodePath); err != nil {
t.Fatalf("did not expect error checking for existence of %s: %s", nodePath, err)
} else {
if !exists {
t.Fatalf("could not find %s", nodePath)
}
}
}
//SetUpSuite is run before the tests to ensure elastic, zookeeper etc. are running.
func (dt *DaoTest) SetUpSuite(c *C) {
docker.SetUseRegistry(true)
dt.Port = 9202
isvcs.Init()
isvcs.Mgr.SetVolumesDir("/tmp/serviced-test")
esServicedClusterName, _ := utils.NewUUID36()
if err := isvcs.Mgr.SetConfigurationOption("elasticsearch-serviced", "cluster", esServicedClusterName); err != nil {
c.Fatalf("Could not set elasticsearch-serviced clustername: %s", err)
}
esLogstashClusterName, _ := utils.NewUUID36()
if err := isvcs.Mgr.SetConfigurationOption("elasticsearch-logstash", "cluster", esLogstashClusterName); err != nil {
c.Fatalf("Could not set elasticsearch-logstash clustername: %s", err)
}
isvcs.Mgr.Wipe()
if err := isvcs.Mgr.Start(); err != nil {
c.Fatalf("Could not start es container: %s", err)
}
dt.MappingsFile = "controlplane.json"
dt.FacadeTest.SetUpSuite(c)
dsn := coordzk.NewDSN([]string{"127.0.0.1:2181"}, time.Second*15).String()
glog.Infof("zookeeper dsn: %s", dsn)
zClient, err := coordclient.New("zookeeper", dsn, "", nil)
if err != nil {
glog.Fatalf("Could not start es container: %s", err)
}
zzk.InitializeLocalClient(zClient)
dt.zkConn, err = zzk.GetLocalConnection("/")
if err != nil {
c.Fatalf("could not get zk connection %v", err)
}
dt.Dao, err = NewControlSvc("localhost", int(dt.Port), dt.Facade, "/tmp", "rsync", 4979, time.Minute*5, "localhost:5000", MockStorageDriver{})
if err != nil {
glog.Fatalf("Could not start es container: %s", err)
} else {
for i := 0; i < 10; i += 1 {
id := strconv.Itoa(i)
dt.Dao.RemoveService(id, &unused)
}
for i := 100; i < 110; i += 1 {
id := strconv.Itoa(i)
dt.Dao.RemoveService(id, &unused)
}
}
}
func TestServer(t *testing.T) {
t.Skip() // the zookeeper part doesnt work in this test, but does work in real life
zookeeper.EnsureZkFatjar()
basePath := ""
tc, err := zklib.StartTestCluster(1, nil, nil)
if err != nil {
t.Fatalf("could not start test zk cluster: %s", err)
}
defer os.RemoveAll(tc.Path)
defer tc.Stop()
time.Sleep(time.Second)
servers := []string{fmt.Sprintf("127.0.0.1:%d", tc.Servers[0].Port)}
dsnBytes, err := json.Marshal(zookeeper.DSN{Servers: servers, Timeout: time.Second * 15})
if err != nil {
t.Fatal("unexpected error creating zk DSN: %s", err)
}
dsn := string(dsnBytes)
zClient, err := client.New("zookeeper", dsn, basePath, nil)
if err != nil {
t.Fatal("unexpected error getting zk client")
}
zzk.InitializeLocalClient(zClient)
defer func(orig func(nfs.Driver, string, string) error) {
nfsMount = orig
}(nfsMount)
var local, remote string
nfsMount = func(driver nfs.Driver, a, b string) error {
glog.Infof("client is mounting %s to %s", a, b)
remote = a
local = b
return nil
}
// creating a UUID in order to make a unique poolID
// the poolID is somehow being saved on the filesystem (zookeeper config somewhere?)
// making the poolID unique on every run will ensure it is stateless
uuid, err := utils.NewUUID()
if err != nil {
t.Fatal("New UUID could not be created")
}
hostServer := host.New()
hostServer.ID = "nodeID"
hostServer.IPAddr = "192.168.1.50"
hostServer.PoolID = uuid
hostClient1 := host.New()
hostClient1.ID = "nodeID_client1"
hostClient1.IPAddr = "192.168.1.100"
hostClient1.PoolID = uuid
mockNfsDriver := &mockNfsDriverT{
exportPath: "/exports",
exportName: "serviced_var",
}
// TODO: this gets stuck at server.go:90 call to conn.CreateDir hangs
s, err := NewServer(mockNfsDriver, hostServer, path.Join(mockNfsDriver.exportPath, mockNfsDriver.exportName))
if err != nil {
t.Fatalf("unexpected error creating Server: %s", err)
}
conn, err := zzk.GetLocalConnection("/")
if err != nil {
t.Fatalf("unexpected error getting connection: %s", err)
}
shutdown := make(chan interface{})
defer close(shutdown)
go s.Run(shutdown, conn)
// give it some time
time.Sleep(time.Second * 5)
if !mockNfsDriver.syncCalled {
t.Fatalf("sync() should have been called by now")
}
if len(mockNfsDriver.clients) != 0 {
t.Fatalf("Expected number of clients: 0 --- Found: %v (%v)", len(mockNfsDriver.clients), mockNfsDriver.clients)
}
mockNfsDriver.syncCalled = false
tmpVar, err := ioutil.TempDir("", "serviced_var")
if err != nil {
t.Fatalf("could not create tempdir: %s", err)
}
defer os.RemoveAll(tmpVar)
c1, err := NewClient(hostClient1, tmpVar)
if err != nil {
t.Fatalf("could not create client: %s", err)
}
// give it some time
time.Sleep(time.Second * 2)
if !mockNfsDriver.syncCalled {
t.Fatalf("sync() should have been called by now")
}
if len(mockNfsDriver.clients) != 1 {
t.Fatalf("expecting 1 client, got %d", len(mockNfsDriver.clients))
}
if mockNfsDriver.clients[0] != hostClient1.IPAddr {
t.Fatalf("expecting '%s', got '%s'", hostServer.IPAddr, mockNfsDriver.clients[0])
}
shareName := fmt.Sprintf("%s:%s", hostServer.IPAddr, mockNfsDriver.ExportPath())
if remote != shareName {
t.Fatalf("remote should be %s, not %s", remote, shareName)
}
glog.Info("about to call c1.Close()")
c1.Close()
}
func (d *daemon) startAgent() error {
muxListener, err := createMuxListener()
if err != nil {
return err
}
mux, err := proxy.NewTCPMux(muxListener)
if err != nil {
return err
}
agentIP := options.OutboundIP
if agentIP == "" {
var err error
agentIP, err = utils.GetIPAddress()
if err != nil {
glog.Fatalf("Failed to acquire ip address: %s", err)
}
}
rpcPort := "0"
parts := strings.Split(options.Listen, ":")
if len(parts) > 1 {
rpcPort = parts[1]
}
thisHost, err := host.Build(agentIP, rpcPort, "unknown")
if err != nil {
panic(err)
}
myHostID, err := utils.HostID()
if err != nil {
return fmt.Errorf("HostID failed: %v", err)
} else if err := validation.ValidHostID(myHostID); err != nil {
glog.Errorf("invalid hostid: %s", myHostID)
}
go func() {
var poolID string
for {
poolID = func() string {
glog.Infof("Trying to discover my pool...")
var myHost *host.Host
masterClient, err := master.NewClient(d.servicedEndpoint)
if err != nil {
glog.Errorf("master.NewClient failed (endpoint %+v) : %v", d.servicedEndpoint, err)
return ""
}
defer masterClient.Close()
myHost, err = masterClient.GetHost(myHostID)
if err != nil {
glog.Warningf("masterClient.GetHost %v failed: %v (has this host been added?)", myHostID, err)
return ""
}
poolID = myHost.PoolID
glog.Infof(" My PoolID: %v", poolID)
//send updated host info
updatedHost, err := host.UpdateHostInfo(*myHost)
if err != nil {
glog.Infof("Could not send updated host information: %v", err)
return poolID
}
err = masterClient.UpdateHost(updatedHost)
if err != nil {
glog.Warningf("Could not update host information: %v", err)
return poolID
}
glog.V(2).Infof("Sent updated host info %#v", updatedHost)
return poolID
}()
if poolID != "" {
break
}
select {
case <-d.shutdown:
return
case <-time.After(5 * time.Second):
continue
}
}
thisHost.PoolID = poolID
basePoolPath := "/pools/" + poolID
dsn := coordzk.NewDSN(options.Zookeepers, time.Second*15).String()
glog.Infof("zookeeper dsn: %s", dsn)
zClient, err := coordclient.New("zookeeper", dsn, basePoolPath, nil)
if err != nil {
glog.Errorf("failed create a new coordclient: %v", err)
}
zzk.InitializeLocalClient(zClient)
poolBasedConn, err := zzk.GetLocalConnection(zzk.GeneratePoolPath(poolID))
if err != nil {
glog.Errorf("Error in getting a connection based on pool %v: %v", poolID, err)
}
if options.NFSClient != "0" {
nfsClient, err := storage.NewClient(thisHost, path.Join(options.VarPath, "volumes"))
if err != nil {
glog.Fatalf("could not create an NFS client: %s", err)
}
go func() {
<-d.shutdown
glog.Infof("shutting down storage client")
nfsClient.Close()
}()
//loop and log waiting for Storage Leader
nfsDone := make(chan struct{})
go func() {
defer close(nfsDone)
nfsClient.Wait()
}()
//wait indefinitely(?) for storage to work before starting
glog.Info("Waiting for Storage Leader")
nfsUp := false
for !nfsUp {
select {
case <-nfsDone:
nfsUp = true
glog.Info("Found Storage Leader")
break
case <-time.After(time.Second * 30):
glog.Info("Waiting for Storage Leader, will not be available for running services. ")
continue
}
}
} else {
glog.Info("NFS Client disabled")
}
agentOptions := node.AgentOptions{
PoolID: thisHost.PoolID,
Master: options.Endpoint,
UIPort: options.UIPort,
RPCPort: options.RPCPort,
DockerDNS: options.DockerDNS,
VarPath: options.VarPath,
Mount: options.Mount,
FSType: options.FSType,
Zookeepers: options.Zookeepers,
Mux: mux,
UseTLS: options.TLS,
DockerRegistry: dockerRegistry,
MaxContainerAge: time.Duration(int(time.Second) * options.MaxContainerAge),
VirtualAddressSubnet: options.VirtualAddressSubnet,
}
// creates a zClient that is not pool based!
hostAgent, err := node.NewHostAgent(agentOptions)
d.hostAgent = hostAgent
d.waitGroup.Add(1)
go func() {
hostAgent.Start(d.shutdown)
glog.Info("Host Agent has shutdown")
d.waitGroup.Done()
}()
// register the API
glog.V(0).Infoln("registering ControlPlaneAgent service")
if err = d.rpcServer.RegisterName("ControlPlaneAgent", hostAgent); err != nil {
glog.Fatalf("could not register ControlPlaneAgent RPC server: %v", err)
}
if options.ReportStats {
statsdest := fmt.Sprintf("http://%s/api/metrics/store", options.HostStats)
statsduration := time.Duration(options.StatsPeriod) * time.Second
glog.V(1).Infoln("Staring container statistics reporter")
statsReporter, err := stats.NewStatsReporter(statsdest, statsduration, poolBasedConn)
if err != nil {
glog.Errorf("Error kicking off stats reporter %v", err)
} else {
go func() {
defer statsReporter.Close()
<-d.shutdown
}()
}
}
}()
glog.Infof("agent start staticips: %v [%d]", d.staticIPs, len(d.staticIPs))
if err = d.rpcServer.RegisterName("Agent", agent.NewServer(d.staticIPs)); err != nil {
glog.Fatalf("could not register Agent RPC server: %v", err)
}
if err != nil {
glog.Fatalf("Could not start ControlPlane agent: %v", err)
}
// TODO: Integrate this server into the rpc server, or something.
// Currently its only use is for command execution.
go func() {
sio := shell.NewProcessExecutorServer(options.Endpoint, dockerRegistry)
http.ListenAndServe(":50000", sio)
}()
return nil
}
func (d *daemon) startMaster() error {
agentIP := options.OutboundIP
if agentIP == "" {
var err error
agentIP, err = utils.GetIPAddress()
if err != nil {
glog.Fatalf("Failed to acquire ip address: %s", err)
}
}
// This is storage related
rpcPort := "0"
parts := strings.Split(options.Listen, ":")
if len(parts) > 1 {
rpcPort = parts[1]
}
thisHost, err := host.Build(agentIP, rpcPort, d.masterPoolID)
if err != nil {
glog.Errorf("could not build host for agent IP %s: %v", agentIP, err)
return err
}
if err := os.MkdirAll(options.VarPath, 0755); err != nil {
glog.Errorf("could not create varpath %s: %s", options.VarPath, err)
return err
}
volumesPath := path.Join(options.VarPath, "volumes")
if d.networkDriver, err = nfs.NewServer(volumesPath, "serviced_var_volumes", "0.0.0.0/0"); err != nil {
return err
} else {
d.storageHandler, err = storage.NewServer(d.networkDriver, thisHost, volumesPath)
if err != nil {
return err
}
}
if d.dsDriver, err = d.initDriver(); err != nil {
return err
}
if d.dsContext, err = d.initContext(); err != nil {
return err
}
localClient, err := d.initZK(options.Zookeepers)
if err != nil {
glog.Errorf("failed to create a local coordclient: %v", err)
return err
}
zzk.InitializeLocalClient(localClient)
if len(options.RemoteZookeepers) > 0 {
remoteClient, err := d.initZK(options.RemoteZookeepers)
if err != nil {
glog.Warningf("failed to create a remote coordclient; running in disconnected mode: %v", err)
} else {
zzk.InitializeRemoteClient(remoteClient)
}
}
d.facade = d.initFacade()
if d.cpDao, err = d.initDAO(); err != nil {
return err
}
health.SetDao(d.cpDao)
go health.Cleanup(d.shutdown)
if err = d.facade.CreateDefaultPool(d.dsContext, d.masterPoolID); err != nil {
return err
}
if err = d.registerMasterRPC(); err != nil {
return err
}
d.initWeb()
d.addTemplates()
d.startScheduler()
return nil
}
// getEndpoints builds exportedEndpoints and importedEndpoints
func (c *Controller) getEndpoints(service *service.Service) error {
var err error
c.zkInfo, err = getAgentZkInfo(c.options.ServicedEndpoint)
if err != nil {
glog.Errorf("Invalid zk info: %v", err)
return err //ErrInvalidZkInfo
}
glog.Infof(" c.zkInfo: %+v", c.zkInfo)
// endpoints are created at the root level (not pool aware)
rootBasePath := ""
zClient, err := coordclient.New("zookeeper", c.zkInfo.ZkDSN, rootBasePath, nil)
if err != nil {
glog.Errorf("failed create a new coordclient: %v", err)
return err
}
zzk.InitializeLocalClient(zClient)
// get zookeeper connection
conn, err := zzk.GetLocalConnection(zzk.GeneratePoolPath(service.PoolID))
if err != nil {
return fmt.Errorf("getEndpoints zzk.GetLocalConnection failed: %v", err)
}
if os.Getenv("SERVICED_IS_SERVICE_SHELL") == "true" {
// this is not a running service, i.e. serviced shell/run
if hostname, err := os.Hostname(); err != nil {
glog.Errorf("could not get hostname: %s", err)
return fmt.Errorf("getEndpoints failed could not get hostname: %v", err)
} else {
c.dockerID = hostname
}
// TODO: deal with exports in the future when there is a use case for it
sstate, err := servicestate.BuildFromService(service, c.hostID)
if err != nil {
return fmt.Errorf("Unable to create temporary service state")
}
// initialize importedEndpoints
c.importedEndpoints, err = buildImportedEndpoints(conn, c.tenantID, sstate)
if err != nil {
glog.Errorf("Invalid ImportedEndpoints")
return ErrInvalidImportedEndpoints
}
} else {
// get service state
glog.Infof("getting service state: %s %v", c.options.Service.ID, c.options.Service.InstanceID)
sstate, err := getServiceState(conn, c.options.Service.ID, c.options.Service.InstanceID)
if err != nil {
return fmt.Errorf("getEndpoints getServiceState failed: %v", err)
}
c.dockerID = sstate.DockerID
// keep a copy of the service EndPoint exports
c.exportedEndpoints, err = buildExportedEndpoints(conn, c.tenantID, sstate)
if err != nil {
glog.Errorf("Invalid ExportedEndpoints")
return ErrInvalidExportedEndpoints
}
// initialize importedEndpoints
c.importedEndpoints, err = buildImportedEndpoints(conn, c.tenantID, sstate)
if err != nil {
glog.Errorf("Invalid ImportedEndpoints")
return ErrInvalidImportedEndpoints
}
}
return nil
}