func main() {
var err error
// Create YarnConfiguration
conf, _ := conf.NewYarnConfiguration()
// Create YarnClient
yarnClient, _ := yarn_client.CreateYarnClient(conf)
// Create new application to get ApplicationSubmissionContext
_, asc, _ := yarnClient.CreateNewApplication()
// Some useful information
queue := "default"
appName := "simple-go-yarn-app"
appType := "GO_HADOOP_UNMANAGED"
unmanaged := true
clc := hadoop_yarn.ContainerLaunchContextProto{}
// Setup ApplicationSubmissionContext for the application
asc.AmContainerSpec = &clc
asc.ApplicationName = &appName
asc.Queue = &queue
asc.ApplicationType = &appType
asc.UnmanagedAm = &unmanaged
// Submit!
err = yarnClient.SubmitApplication(asc)
if err != nil {
log.Fatal("yarnClient.SubmitApplication ", err)
}
log.Println("Successfully submitted unmanaged application: ", asc.ApplicationId)
time.Sleep(1 * time.Second)
appReport, err := yarnClient.GetApplicationReport(asc.ApplicationId)
if err != nil {
log.Fatal("yarnClient.GetApplicationReport ", err)
}
appState := appReport.GetYarnApplicationState()
for appState != hadoop_yarn.YarnApplicationStateProto_ACCEPTED {
log.Println("Application in state ", appState)
time.Sleep(1 * time.Second)
appReport, err = yarnClient.GetApplicationReport(asc.ApplicationId)
appState = appReport.GetYarnApplicationState()
if appState == hadoop_yarn.YarnApplicationStateProto_FAILED || appState == hadoop_yarn.YarnApplicationStateProto_KILLED {
log.Fatal("Application in state ", appState)
}
}
log.Println("Application in state ", appState)
// Create AMRMClient
var attemptId int32
attemptId = 1
applicationAttemptId := hadoop_yarn.ApplicationAttemptIdProto{ApplicationId: asc.ApplicationId, AttemptId: &attemptId}
rmClient, _ := yarn_client.CreateAMRMClient(conf, &applicationAttemptId)
log.Println("Created RM client: ", rmClient)
// Wait for ApplicationAttempt to be in Launched state
appAttemptReport, err := yarnClient.GetApplicationAttemptReport(&applicationAttemptId)
appAttemptState := appAttemptReport.GetYarnApplicationAttemptState()
for appAttemptState != hadoop_yarn.YarnApplicationAttemptStateProto_APP_ATTEMPT_LAUNCHED {
log.Println("ApplicationAttempt in state ", appAttemptState)
time.Sleep(1 * time.Second)
appAttemptReport, err = yarnClient.GetApplicationAttemptReport(&applicationAttemptId)
appAttemptState = appAttemptReport.GetYarnApplicationAttemptState()
}
log.Println("ApplicationAttempt in state ", appAttemptState)
// Register with ResourceManager
log.Println("About to register application master.")
err = rmClient.RegisterApplicationMaster("", -1, "")
if err != nil {
log.Fatal("rmClient.RegisterApplicationMaster ", err)
}
log.Println("Successfully registered application master.")
// Add resource requests
const numContainers = int32(1)
memory := int32(128)
resource := hadoop_yarn.ResourceProto{Memory: &memory}
rmClient.AddRequest(1, "*", &resource, numContainers)
// Now call ResourceManager.allocate
allocateResponse, err := rmClient.Allocate()
if err == nil {
log.Println("allocateResponse: ", *allocateResponse)
}
log.Println("#containers allocated: ", len(allocateResponse.AllocatedContainers))
numAllocatedContainers := int32(0)
allocatedContainers := make([]*hadoop_yarn.ContainerProto, numContainers, numContainers)
for numAllocatedContainers < numContainers {
// Sleep for a while
log.Println("Sleeping...")
time.Sleep(3 * time.Second)
log.Println("Sleeping... done!")
// Try to get containers now...
allocateResponse, err = rmClient.Allocate()
if err == nil {
log.Println("allocateResponse: ", *allocateResponse)
}
for _, container := range allocateResponse.AllocatedContainers {
allocatedContainers[numAllocatedContainers] = container
numAllocatedContainers++
}
log.Println("#containers allocated: ", len(allocateResponse.AllocatedContainers))
log.Println("Total #containers allocated so far: ", numAllocatedContainers)
}
log.Println("Final #containers allocated: ", numAllocatedContainers)
// Now launch containers
containerLaunchContext := hadoop_yarn.ContainerLaunchContextProto{Command: []string{"/bin/date"}}
log.Println("containerLaunchContext: ", containerLaunchContext)
for _, container := range allocatedContainers {
log.Println("Launching container: ", *container, " ", container.NodeId.Host, ":", container.NodeId.Port)
nmClient, err := yarn_client.CreateAMNMClient(*container.NodeId.Host, int(*container.NodeId.Port))
if err != nil {
log.Fatal("hadoop_yarn.DialContainerManagementProtocolService: ", err)
}
log.Println("Successfully created nmClient: ", nmClient)
err = nmClient.StartContainer(container, &containerLaunchContext)
if err != nil {
log.Fatal("nmClient.StartContainer: ", err)
}
}
// Wait for Containers to complete
numCompletedContainers := int32(0)
for numCompletedContainers < numContainers {
// Sleep for a while
log.Println("Sleeping...")
time.Sleep(3 * time.Second)
log.Println("Sleeping... done!")
allocateResponse, err = rmClient.Allocate()
if err == nil {
log.Println("allocateResponse: ", *allocateResponse)
}
for _, containerStatus := range allocateResponse.CompletedContainerStatuses {
log.Println("Completed container: ", *containerStatus, " exit-code: ", *containerStatus.ExitStatus)
numCompletedContainers++
}
}
log.Println("Containers complete: ", numCompletedContainers)
// Unregister with ResourceManager
log.Println("About to unregister application master.")
finalStatus := hadoop_yarn.FinalApplicationStatusProto_APP_SUCCEEDED
err = rmClient.FinishApplicationMaster(&finalStatus, "done", "")
if err != nil {
log.Fatal("rmClient.RegisterApplicationMaster ", err)
}
log.Println("Successfully unregistered application master.")
}
//allocates one "fat" container.
func (yarnScheduler *YARNScheduler) Schedule(pod api.Pod, minionLister MinionLister) (string, error) {
rmClient := yarnScheduler.rmClient
// Add resource requests
// Even though we only allocate one container for now, this could change in the future.
const numContainers = int32(1)
memory := int32(128)
resource := hadoop_yarn.ResourceProto{Memory: &memory}
numAllocatedContainers := int32(0)
const maxAttempts = int(5)
allocationAttempts := 0
//allocatedContainers := make([]*hadoop_yarn.ContainerProto, numContainers, numContainers)
rmClient.AddRequest(1, "*", &resource, numContainers)
for numAllocatedContainers < numContainers && allocationAttempts < maxAttempts {
var allocatedContainers []*hadoop_yarn.ContainerProto
select {
case allocatedContainers = <-yarnScheduler.handler.allocatedContainers:
break
case <-time.After(3 * time.Second):
// Sleep for a while before trying again
allocationAttempts++
log.Println("Sleeping...")
time.Sleep(3 * time.Second)
log.Println("Sleeping... done!")
continue
}
for _, container := range allocatedContainers {
allocatedContainers[numAllocatedContainers] = container
numAllocatedContainers++
log.Println("#containers allocated so far: ", numAllocatedContainers)
log.Printf("pod=%s YARN container=%v", pod.Name, container.GetId())
yarnScheduler.podsToContainersMap[pod.Name] = container.GetId()
host := *container.NodeId.Host
port := *container.NodeId.Port
log.Println("allocated container on: ", host)
//launch a "sleep" container so that the "loop" is completed and YARN doesn't deallocate the container
containerLaunchContext := hadoop_yarn.ContainerLaunchContextProto{Command: []string{"while :; do sleep 1; done"}}
nmClient, err := yarn_client.CreateAMNMClient(host, int(port))
if err != nil {
log.Fatal("Failed to create AMNMClient! ", err)
}
log.Printf("Successfully created nmClient: %v", nmClient)
log.Printf("Attempting to start container on %s", host)
err = nmClient.StartContainer(container, &containerLaunchContext)
if err != nil {
log.Fatal("failed to start container! ", err)
}
//We have the hostname available. return from here.
//This many change in case we allocate more than one container in the future.
return findMinionForHost(host, minionLister)
}
log.Println("#containers allocated: ", len(allocatedContainers))
log.Println("Total #containers allocated so far: ", numAllocatedContainers)
}
log.Println("Final #containers allocated: ", numAllocatedContainers)
return "<invalid_host>", errors.New("unable to schedule pod! YARN didn't allocate a container")
}
func main() {
var err error
// Get applicationAttemptId from environment
applicationAttemptId, err := parseAppAttemptId()
if err != nil {
log.Fatal("parseAppAttemptId: ", err)
}
// Create YarnConfiguration
conf, _ := conf.NewYarnConfiguration()
// Create AMRMClient
rmClient, _ := yarn_client.CreateAMRMClient(conf, applicationAttemptId)
log.Println("Created RM client: ", rmClient)
// Register with ResourceManager
log.Println("About to register application master.")
err = rmClient.RegisterApplicationMaster("", -1, "")
if err != nil {
log.Fatal("rmClient.RegisterApplicationMaster ", err)
}
log.Println("Successfully registered application master.")
// Add resource requests
const numContainers = int32(1)
memory := int32(128)
resource := hadoop_yarn.ResourceProto{Memory: &memory}
rmClient.AddRequest(1, "*", &resource, numContainers)
// Now call ResourceManager.allocate
allocateResponse, err := rmClient.Allocate()
if err == nil {
log.Println("allocateResponse: ", *allocateResponse)
}
log.Println("#containers allocated: ", len(allocateResponse.AllocatedContainers))
numAllocatedContainers := int32(0)
allocatedContainers := make([]*hadoop_yarn.ContainerProto, numContainers, numContainers)
for numAllocatedContainers < numContainers {
// Sleep for a while
log.Println("Sleeping...")
time.Sleep(3 * time.Second)
log.Println("Sleeping... done!")
// Try to get containers now...
allocateResponse, err = rmClient.Allocate()
if err == nil {
log.Println("allocateResponse: ", *allocateResponse)
}
for _, container := range allocateResponse.AllocatedContainers {
allocatedContainers[numAllocatedContainers] = container
numAllocatedContainers++
}
log.Println("#containers allocated: ", len(allocateResponse.AllocatedContainers))
log.Println("Total #containers allocated so far: ", numAllocatedContainers)
}
log.Println("Final #containers allocated: ", numAllocatedContainers)
// Now launch containers
containerLaunchContext := hadoop_yarn.ContainerLaunchContextProto{Command: []string{"/bin/date"}}
log.Println("containerLaunchContext: ", containerLaunchContext)
for _, container := range allocatedContainers {
log.Println("Launching container: ", *container, " ", container.NodeId.Host, ":", container.NodeId.Port)
nmClient, err := yarn_client.CreateAMNMClient(*container.NodeId.Host, int(*container.NodeId.Port))
if err != nil {
log.Fatal("hadoop_yarn.DialContainerManagementProtocolService: ", err)
}
log.Println("Successfully created nmClient: ", nmClient)
err = nmClient.StartContainer(container, &containerLaunchContext)
if err != nil {
log.Fatal("nmClient.StartContainer: ", err)
}
}
// Wait for Containers to complete
numCompletedContainers := int32(0)
for numCompletedContainers < numContainers {
// Sleep for a while
log.Println("Sleeping...")
time.Sleep(3 * time.Second)
log.Println("Sleeping... done!")
allocateResponse, err = rmClient.Allocate()
if err == nil {
log.Println("allocateResponse: ", *allocateResponse)
}
for _, containerStatus := range allocateResponse.CompletedContainerStatuses {
log.Println("Completed container: ", *containerStatus, " exit-code: ", *containerStatus.ExitStatus)
numCompletedContainers++
}
}
log.Println("Containers complete: ", numCompletedContainers)
// Unregister with ResourceManager
log.Println("About to unregister application master.")
finalStatus := hadoop_yarn.FinalApplicationStatusProto_APP_SUCCEEDED
err = rmClient.FinishApplicationMaster(&finalStatus, "done", "")
if err != nil {
log.Fatal("rmClient.RegisterApplicationMaster ", err)
}
log.Println("Successfully unregistered application master.")
}
