func (e *UnregisterSupervisorExecutor) Execute(t *Task) error {
if e.arg.Host == "" {
return errors.New("Please specify a host to unregister")
}
// TODO(edanaher): With appropriate confirmation (--force?), tear down containers on the host and delete
// metadata. Until then, cowardly refuse to tear down supervisors with containers.
listResponse, err := supervisor.List(e.arg.Host)
if err != nil {
return err
}
containerCount := len(listResponse.Containers)
if containerCount > 0 {
plural := ""
if containerCount > 1 {
plural = "s"
}
return errors.New(fmt.Sprintf("Supervisor still has %d running container%s", +containerCount, plural))
}
supervisor.Teardown(e.arg.Host, []string{}, true)
err = datamodel.Supervisor(e.arg.Host).Delete()
if err != nil {
return err
}
e.reply.Status = StatusOk
return nil
}
func (e *RegisterSupervisorExecutor) Execute(t *Task) error {
if e.arg.Host == "" {
return errors.New("Please specify a host to register")
}
// check health of to be registered supervisor
health, err := supervisor.HealthCheck(e.arg.Host)
if err != nil {
e.reply.Status = StatusError
return err
}
if health.Status != StatusOk && health.Status != StatusFull {
e.reply.Status = health.Status
return errors.New("Status is " + health.Status)
}
if health.Region != Region {
e.reply.Status = "Region Mismatch"
return errors.New("Supervisor Region (" + health.Region + ") does not match Manager Region (" + Region + ")")
}
err = datamodel.Supervisor(e.arg.Host).Touch()
if err != nil {
e.reply.Status = StatusError
return err
}
// try to push all ip groups to this new supervisor
if err := netsec.UpdateSupervisor(e.arg.Host); err != nil {
return err
}
if err == nil {
e.reply.Status = StatusOk
} else {
e.reply.Status = StatusError
}
return err
}
func cleanup(removeContainerFromHost bool, deployedContainers []*Container, t *Task) {
// kill all references to deployed containers as well as the container itself
for _, container := range deployedContainers {
supervisor.Teardown(container.Host, []string{container.ID}, false)
if instance, err := datamodel.GetInstance(container.ID); err == nil {
instance.Delete()
} else {
t.Log(fmt.Sprintf("Failed to clean up instance %s: %s", container.ID, err.Error()))
}
DeleteAppShaFromEnv(container.App, container.Sha, container.Env)
if removeContainerFromHost {
datamodel.Supervisor(container.Host).RemoveContainer(container.ID)
}
}
}
func (e *TeardownExecutor) Execute(t *Task) error {
hostMap, err := getContainerIDsToTeardown(t, e.arg)
if err != nil {
return err
}
if e.arg.All {
tl := datamodel.NewTeardownLock(t.ID)
if err := tl.Lock(); err != nil {
return err
}
defer tl.Unlock()
} else if e.arg.Env != "" {
tl := datamodel.NewTeardownLock(t.ID, e.arg.App, e.arg.Sha, e.arg.Env)
if err := tl.Lock(); err != nil {
return err
}
defer tl.Unlock()
} else if e.arg.Sha != "" {
tl := datamodel.NewTeardownLock(t.ID, e.arg.App, e.arg.Sha)
if err := tl.Lock(); err != nil {
return err
}
defer tl.Unlock()
} else if e.arg.App != "" {
tl := datamodel.NewTeardownLock(t.ID, e.arg.App)
if err := tl.Lock(); err != nil {
return err
}
defer tl.Unlock()
}
tornContainers := []string{}
for host, containerIDs := range hostMap {
if e.arg.All {
t.LogStatus("Tearing Down * from %s", host)
} else {
t.LogStatus("Tearing Down %v from %s", containerIDs, host)
}
ihReply, err := supervisor.Teardown(host, containerIDs, e.arg.All)
if err != nil {
return errors.New(fmt.Sprintf("Error Tearing Down %v from %s : %s", containerIDs, host,
err.Error()))
}
tornContainers = append(tornContainers, ihReply.ContainerIDs...)
for _, tornContainerID := range ihReply.ContainerIDs {
t.LogStatus("%s has been removed from host %s; removing zookeeper record about the container",
tornContainerID, host)
err := datamodel.DeleteFromPool([]string{tornContainerID})
if err != nil {
t.Log("Error removing %s from pool: %v", tornContainerID, err)
}
datamodel.Supervisor(host).RemoveContainer(tornContainerID)
instance, err := datamodel.GetInstance(tornContainerID)
if err != nil {
continue
}
last, _ := instance.Delete()
if last {
t.LogStatus("%s is the last one of its kind [app: %s SHA: %s Env: %s]",
tornContainerID, instance.App, instance.Sha, instance.Env)
DeleteAppShaFromEnv(instance.App, instance.Sha, instance.Env)
}
t.LogStatus("Successfully teardown %s", tornContainerID)
}
}
e.reply.ContainerIDs = tornContainers
return nil
}
func deployToHostsInZones(deps map[string]DepsType, manifest *Manifest, sha, env string,
hosts map[string][]string, zones []string, t *Task) ([]*Container, error) {
deployedContainers := []*Container{}
// fetch the app
zkApp, err := datamodel.GetApp(manifest.Name)
if err != nil {
return nil, err
}
// first check if zones have enough hosts
for _, zone := range zones {
// fail if zone has no hosts
if hosts[zone] == nil || len(hosts[zone]) == 0 {
return nil, errors.New(fmt.Sprintf("No hosts available for app %s in zone %s", manifest.Name, zone))
}
}
// now that we know that enough hosts are available
t.LogStatus("Deploying to zones: %v", zones)
respCh := make(chan *DeployZoneResult, len(zones))
for _, zone := range zones {
go deployToZone(respCh, deps, manifest, sha, env, hosts[zone], zone)
}
numResults := 0
status := "Deployed to zones: "
for result := range respCh {
deployedContainers = append(deployedContainers, result.Containers...)
if result.Error != nil {
err = result.Error
t.Log(err.Error())
status += result.Zone + ":FAIL "
} else {
status += result.Zone + ":SUCCESS "
}
t.LogStatus(status)
numResults++
if numResults >= len(zones) { // we're done
close(respCh)
}
}
if err != nil {
cleanup(false, deployedContainers, t)
return nil, err
}
// set ports on zk supervisor - can't do this in parallel. we may deploy to the same host at the same time
// and since we don't lock zookeeper (maybe we should), this would result in a race condition.
t.LogStatus("Updating Zookeeper")
for _, cont := range deployedContainers {
datamodel.Supervisor(cont.Host).SetContainerAndPort(cont.ID, cont.PrimaryPort)
}
// we're good now, so lets move on
t.LogStatus("Updating Router")
deployedIDs := make([]string, len(deployedContainers))
count := 0
for _, cont := range deployedContainers {
deployedIDs[count] = cont.ID
count++
}
err = datamodel.AddToPool(deployedIDs)
if err != nil { // if we can't add the pool, clean up and fail
cleanup(true, deployedContainers, t)
return nil, errors.New("Update Pool Error: " + err.Error())
}
if zkApp.Internal {
// reserve router port if needed and add app+env
_, _, err = datamodel.ReserveRouterPortAndUpdateTrie(zkApp.Internal, manifest.Name, sha, env)
if err != nil {
datamodel.DeleteFromPool(deployedIDs)
cleanup(true, deployedContainers, t)
return nil, errors.New("Reserve Router Port Error: " + err.Error())
}
} else {
// only update trie
_, err = datamodel.UpdateAppEnvTrie(zkApp.Internal, manifest.Name, sha, env)
if err != nil {
datamodel.DeleteFromPool(deployedIDs)
cleanup(true, deployedContainers, t)
return nil, errors.New("Reserve Router Port Error: " + err.Error())
}
}
return deployedContainers, nil
}