// Close the release channel when you want to clean up nicely.
func CreatePidNode(zconn zookeeper.Conn, zkPath string, contents string, done chan struct{}) error {
// On the first try, assume the cluster is up and running, that will
// help hunt down any config issues present at startup
if _, err := zconn.Create(zkPath, []byte(contents), zookeeper.FlagEphemeral, zookeeper.WorldACL(PERM_FILE)); err != nil {
if ZkErrorEqual(err, zookeeper.ErrNodeExists) {
err = zconn.Delete(zkPath, -1)
}
if err != nil {
return fmt.Errorf("zkutil: failed deleting pid node: %v: %v", zkPath, err)
}
_, err = zconn.Create(zkPath, []byte(contents), zookeeper.FlagEphemeral, zookeeper.WorldACL(PERM_FILE))
if err != nil {
return fmt.Errorf("zkutil: failed creating pid node: %v: %v", zkPath, err)
}
}
go func() {
for {
_, _, watch, err := zconn.GetW(zkPath)
if err != nil {
if ZkErrorEqual(err, zookeeper.ErrNoNode) {
_, err = zconn.Create(zkPath, []byte(contents), zookeeper.FlagEphemeral, zookeeper.WorldACL(zookeeper.PermAll))
if err != nil {
log.Warningf("failed recreating pid node: %v: %v", zkPath, err)
} else {
log.Infof("recreated pid node: %v", zkPath)
continue
}
} else {
log.Warningf("failed reading pid node: %v", err)
}
} else {
select {
case event := <-watch:
if ZkEventOk(event) && event.Type == zookeeper.EventNodeDeleted {
// Most likely another process has started up. However,
// there is a chance that an ephemeral node is deleted by
// the session expiring, yet that same session gets a watch
// notification. This seems like buggy behavior, but rather
// than race too hard on the node, just wait a bit and see
// if the situation resolves itself.
log.Warningf("pid deleted: %v", zkPath)
} else {
log.Infof("pid node event: %v", event)
}
// break here and wait for a bit before attempting
case <-done:
log.Infof("pid watcher stopped on done: %v", zkPath)
return
}
}
select {
// No one likes a thundering herd, least of all zookeeper.
case <-time.After(5*time.Second + time.Duration(rand.Int63n(55e9))):
case <-done:
log.Infof("pid watcher stopped on done: %v", zkPath)
return
}
}
}()
return nil
}
func DeleteRecursive(zconn zookeeper.Conn, zkPath string, version int) error {
// version: -1 delete any version of the node at path - only applies to the top node
err := zconn.Delete(zkPath, int32(version))
if err == nil {
return nil
}
if !ZkErrorEqual(err, zookeeper.ErrNotEmpty) {
return err
}
// Remove the ability for other nodes to get created while we are trying to delete.
// Otherwise, you can enter a race condition, or get starved out from deleting.
_, err = zconn.SetACL(zkPath, zookeeper.WorldACL(zookeeper.PermAdmin|zookeeper.PermDelete|zookeeper.PermRead), int32(version))
if err != nil {
return err
}
children, _, err := zconn.Children(zkPath)
if err != nil {
return err
}
for _, child := range children {
err := DeleteRecursive(zconn, path.Join(zkPath, child), -1)
if err != nil && !ZkErrorEqual(err, zookeeper.ErrNoNode) {
return fmt.Errorf("zkutil: recursive delete failed: %v", err)
}
}
err = zconn.Delete(zkPath, int32(version))
if err != nil && !ZkErrorEqual(err, zookeeper.ErrNotEmpty) {
err = fmt.Errorf("zkutil: nodes getting recreated underneath delete (app race condition): %v", zkPath)
}
return err
}
func recursiveDelete(c *zk.Conn, path string) error {
children, _, err := c.Children(path)
if err != nil && err != zk.ErrNoNode {
logError("err finding children of %s", path)
return err
}
for _, child := range children {
err := recursiveDelete(c, path+"/"+child)
if err != nil && err != zk.ErrNoNode {
logError("err deleting %s", child)
return err
}
}
// get version
_, stat, err := c.Get(path)
if err != nil && err != zk.ErrNoNode {
logError("err getting version of %s", path)
return err
}
if err := c.Delete(path, stat.Version); err != nil && err != zk.ErrNoNode {
return err
}
return nil
}
func DeleteNodebyData(path string, conn *zk.Conn, data []byte) error {
children, _, err := conn.Children(path)
if err != nil {
glog.Warning("Could not list children")
return err
}
for _, child := range children {
child_path := path + "/" + child
child_data, _, err := conn.Get(child_path)
if err != nil {
glog.Warning("Could not get data for %s", child_path)
continue
}
if bytes.Compare(data, child_data) == 0 {
for i := 0; i < 5; i++ {
_, stats, _ := conn.Get(child_path)
err = conn.Delete(child_path, stats.Version)
if err == nil || err == zk.ErrNoNode {
return nil
}
}
glog.Error("Could not delete matched node %s", child_path)
}
}
return nil
}
// notify every Comet node to migrate
func notifyMigrate(conn *zk.Conn, migrateLockPath, znode, key string, update bool, nodeWeightMap map[string]int) (err error) {
// try lock
if _, err = conn.Create(migrateLockPath, []byte("1"), zk.FlagEphemeral, zk.WorldACL(zk.PermAll)); err != nil {
log.Error("conn.Create(\"/gopush-migrate-lock\", \"1\", zk.FlagEphemeral) error(%v)", err)
return
}
// call comet migrate rpc
wg := &sync.WaitGroup{}
wg.Add(len(cometNodeInfoMap))
for node, nodeInfo := range cometNodeInfoMap {
go func(n string, info *CometNodeInfo) {
if info.Rpc == nil {
log.Error("notify migrate failed, no rpc found, node:%s", n)
wg.Done()
return
}
r := info.Rpc.Get()
if r == nil {
log.Error("notify migrate failed, no rpc found, node:%s", n)
wg.Done()
return
}
reply := 0
args := &CometMigrateArgs{Nodes: nodeWeightMap}
if err = r.Call(CometServiceMigrate, args, &reply); err != nil {
log.Error("rpc.Call(\"%s\") error(%v)", CometServiceMigrate, err)
wg.Done()
return
}
log.Debug("notify node:%s migrate succeed", n)
wg.Done()
}(node, nodeInfo)
}
wg.Wait()
// update znode info
if update {
var data []byte
data, err = json.Marshal(cometNodeInfoMap[key])
if err != nil {
log.Error("json.Marshal() node:%s error(%v)", key, err)
return
}
if _, err = conn.Set(znode, data, -1); err != nil {
log.Error("conn.Set(\"%s\",\"%s\",\"-1\") error(%v)", znode, string(data), err)
return
}
}
// release lock
if err = conn.Delete(migrateLockPath, -1); err != nil {
log.Error("conn.Delete(\"%s\") error(%v)", migrateLockPath, err)
}
return
}
func zk_unreginster(path string, c *zk.Conn, exit chan os.Signal) {
for sig := range exit {
log.Warnf("zk_unreginster | received ctrl+c(%v)\n", sig)
err := c.Delete(path, -1)
log.Infof("zk_unreginster | path :%+v", path)
if err != nil {
log.Warnf("zk_unreginster | Delete returned: %+v", err)
}
os.Exit(0)
}
}
func bench(conn *zk.Conn, value []byte, ratio float64) {
choice := rand.Float64()
atomic.AddInt32(&reqCnt, 1)
if choice >= ratio {
// write
conn.Delete("/_test_load_obliterator", 0)
conn.Create("/_test_load_obliterator", value, 1, zk.WorldACL(zk.PermAll))
} else {
// read
conn.Get("/_test_load_obliterator")
}
}
func deleteIfExists(path string, conn *zk.Conn) error {
_, stat, err := conn.Get(path)
if err != nil && err != zk.ErrNoNode {
return err
}
if err != zk.ErrNoNode {
if err := conn.Delete(path, stat.Version); err != nil {
return err
}
}
return nil
}
func zkDelete(conn *zk.Conn, zpath string) error {
if zpath != "" {
_, stat, _ := conn.Get(zpath)
err := conn.Delete(zpath, stat.Version)
if err != nil {
return err
}
index := strings.LastIndex(zpath, "/")
if index != -1 {
return zkDelete(conn, zpath[:index])
} else {
return nil
}
} else {
return nil
}
}
func zkDeleteChildren(conn *zk.Conn, path string) {
children, _, _ := conn.Children(path)
// Leaf
if len(children) == 0 {
fmt.Println("Deleting ", path)
err := conn.Delete(path, -1)
if err != nil {
log.Panic(err)
}
return
}
// Branches
for _, name := range children {
zkDeleteChildren(conn, path+"/"+name)
}
return
}
func (zk *ZookeeperClusterRunner) deleteRecursively(client *zkClient.Conn, key string) {
if key != "/zookeeper" {
children, _, err := client.Children(key)
Ω(err).ShouldNot(HaveOccured(), "Failed to fetch children for '%s'", key)
for _, child := range children {
childPath := key + "/" + child
if key == "/" {
childPath = key + child
}
zk.deleteRecursively(client, childPath)
}
if key != "/" {
err = client.Delete(key, -1)
}
Ω(err).ShouldNot(HaveOccured(), "Failed to delete key '%s'", key)
}
return
}
func markTerminated(conn *zk.Conn, hss *zzk.HostServiceState) {
ssPath := zzk.ServiceStatePath(hss.ServiceId, hss.ServiceStateId)
_, stats, err := conn.Get(ssPath)
if err != nil {
glog.V(0).Infof("Unable to get service state %s for delete because: %v", ssPath, err)
return
}
err = conn.Delete(ssPath, stats.Version)
if err != nil {
glog.V(0).Infof("Unable to delete service state %s because: %v", ssPath, err)
return
}
hssPath := zzk.HostServiceStatePath(hss.HostId, hss.ServiceStateId)
_, stats, err = conn.Get(hssPath)
if err != nil {
glog.V(0).Infof("Unable to get host service state %s for delete becaus: %v", hssPath, err)
return
}
err = conn.Delete(hssPath, stats.Version)
if err != nil {
glog.V(0).Infof("Unable to delete host service state %s", hssPath)
}
}
func DeleteNode(conn *zk.Conn, fpath string) error {
log.Debug("delete znode %s", fpath)
err := conn.Delete(fpath, -1)
return err
}
Key: "/restaurant/menu/dinner",
Value: []byte{},
TTL: 0,
Dir: true,
}))
})
It("deletes the expired entries", func() {
_, _, err := client.Get("/restauraunt/menu/breakfast")
Ω(err).Should(HaveOccured())
})
})
Context("when the node is empty", func() {
BeforeEach(func() {
err := client.Delete("/restaurant/menu/dinner/first_course", -1)
Ω(err).ShouldNot(HaveOccured())
})
It("should return an empty list without erroring", func() {
nodes, err := adapter.List("/restaurant/menu/dinner")
Ω(nodes).Should(HaveLen(0))
Ω(err).ShouldNot(HaveOccured())
})
})
})
Context("when the node exists, but is not a directory", func() {
It("should return an error", func() {
nodes, err := adapter.List("/restaurant/menu/breakfast")
Ω(nodes).Should(BeEmpty())
func Delete(conn *zk.Conn, zkConf conf.Zookeeper, appId string) error {
path := concatPath(zkConf.Path, appId)
return conn.Delete(path, -1)
}
func (a *HostAgent) processServiceState(conn *zk.Conn, shutdown <-chan int, done chan<- stateResult, ssId string) {
procFinished := make(chan int, 1)
var attached bool
for {
var hss zzk.HostServiceState
hssStats, zkEvent, err := zzk.LoadHostServiceStateW(conn, a.hostId, ssId, &hss)
if err != nil {
errS := fmt.Sprintf("Unable to load host service state %s: %v", ssId, err)
glog.Error(errS)
done <- stateResult{ssId, errors.New(errS)}
return
}
if len(hss.ServiceStateId) == 0 || len(hss.ServiceId) == 0 {
errS := fmt.Sprintf("Service for %s is invalid", zzk.HostServiceStatePath(a.hostId, ssId))
glog.Error(errS)
done <- stateResult{ssId, errors.New(errS)}
return
}
var ss dao.ServiceState
ssStats, err := zzk.LoadServiceState(conn, hss.ServiceId, hss.ServiceStateId, &ss)
if err != nil {
errS := fmt.Sprintf("Host service state unable to load service state %s", ssId)
glog.Error(errS)
// This goroutine is watching a node for a service state that does not
// exist or could not be loaded. We should *probably* delete this node.
hssPath := zzk.HostServiceStatePath(a.hostId, ssId)
err = conn.Delete(hssPath, hssStats.Version)
if err != nil {
glog.Warningf("Unable to delete host service state %s", hssPath)
}
done <- stateResult{ssId, errors.New(errS)}
return
}
var service dao.Service
_, err = zzk.LoadService(conn, ss.ServiceId, &service)
if err != nil {
errS := fmt.Sprintf("Host service state unable to load service %s", ss.ServiceId)
glog.Errorf(errS)
done <- stateResult{ssId, errors.New(errS)}
return
}
glog.V(1).Infof("Processing %s, desired state: %d", service.Name, hss.DesiredState)
switch {
case hss.DesiredState == dao.SVC_STOP:
// This node is marked for death
glog.V(1).Infof("Service %s was marked for death, quitting", service.Name)
if attached {
err = a.terminateAttached(conn, procFinished, &ss)
} else {
err = a.terminateInstance(conn, &ss)
}
done <- stateResult{ssId, err}
return
case attached:
// Something uninteresting happened. Why are we here?
glog.V(1).Infof("Service %s is attached in a child goroutine", service.Name)
case hss.DesiredState == dao.SVC_RUN &&
ss.Started.Year() <= 1 || ss.Terminated.Year() > 2:
// Should run, and either not started or process died
glog.V(1).Infof("Service %s does not appear to be running; starting", service.Name)
attached, err = a.startService(conn, procFinished, ssStats, &service, &ss)
case ss.Started.Year() > 1 && ss.Terminated.Year() <= 1:
// Service superficially seems to be running. We need to attach
glog.V(1).Infof("Service %s appears to be running; attaching", service.Name)
attached, err = a.attachToService(conn, procFinished, &ss, &hss)
default:
glog.V(0).Infof("Unhandled service %s", service.Name)
}
if !attached || err != nil {
errS := fmt.Sprintf("Service state %s unable to start or attach to process", ssId)
glog.V(1).Info(errS)
a.terminateInstance(conn, &ss)
done <- stateResult{ssId, errors.New(errS)}
return
}
glog.V(3).Infoln("Successfully processed state for %s", service.Name)
select {
case <-shutdown:
glog.V(0).Info("Agent goroutine will stop watching ", ssId)
err = a.terminateAttached(conn, procFinished, &ss)
if err != nil {
glog.Errorf("Error terminating %s: %v", service.Name, err)
}
done <- stateResult{ssId, err}
return
case <-procFinished:
glog.V(1).Infof("Process finished %s", ssId)
attached = false
continue
case evt := <-zkEvent:
if evt.Type == zk.EventNodeDeleted {
glog.V(0).Info("Host service state deleted: ", ssId)
err = a.terminateAttached(conn, procFinished, &ss)
if err != nil {
glog.Errorf("Error terminating %s: %v", service.Name, err)
}
done <- stateResult{ssId, err}
return
}
glog.V(1).Infof("Host service state %s received event %v", ssId, evt)
continue
}
}
}
func (z *Zookeeper) deleteNode(conn *zk.Conn, keyPath string) error {
log.Debugf("Deleting zk://%s", keyPath)
return conn.Delete(keyPath, -1)
}
