Skip to content

Latest commit

 

History

History
430 lines (349 loc) · 14.7 KB

HadoopMapReduceMaster.md

File metadata and controls

430 lines (349 loc) · 14.7 KB
Raw

PerfectHadoop: MapReduce Application

This project provides a Swift wrapper of MapReduce Application Master REST API:

  • MapReduceApplication(): access a specific map / reduce application running on current server.

RELEASE NOTE

⚠️This is an experimental module and is subject to change in future. ⚠️

Connect to Hadoop Map Reduce Application

To connect to a current active Hadoop Map / Reduce Application by Perfect, initialize a MapReduceApplication() object with an APPLICATION ID, and other essential parameters:

// this connection could possibly do some basic operations
let app = MapReduceApplication(applicationId: "application_12345678_xxxxxx", host: "mapReducer.somedomain.com")

you can also

// add user name if need
let app = MapReduceApplication(applicationId: "application_12345678_xxxxxx", host: "mapReducer.somedomain.com", user: "your user name")

Authentication

If using Kerberos to authenticate, please try codes below:

// set auth to kerberos
let app = MapReduceApplication(applicationId: "application_12345678_xxxxxx", host: "mapReducer.somedomain.com", user: "your user name", auth: .krb5)

Parameters of MapReduceApplication Object

Item Data Type Description
applicationId String the id string of the application to control. Required
service String the service protocol of web request - http / https
host String the hostname or ip address of the Hadoop Map Reduce Application Master host
port Int the port of map reduce application master host, default is 8088
auth Authorization .off or .krb5. Default value is .off
proxyUser String proxy user, if applicable
apibase String use this parameter ONLY the target server has a different api routine other than /ws/v1/mapreduce
timeout Int timeout in seconds, zero means never timeout during transfer

Get General Information

Call checkInfo() to get the general information of a Hadoop MapReduce Application Master in form of a MapReduceApplication.Info structure:

guard let inf = try app.checkInfo() else {
	// something goes wrong here
}
print(inf.startedOn)
print(inf.hadoopVersion)
print(inf.hadoopBuildVersion)
print(inf.hadoopVersionBuiltOn)

Members of MapReduceApplication.Info

Item Data Type Description
appId Int The application id
startedOn Int The time the application started (in ms since epoch)
name string The name of the application
user string The user name of the user who started the application
elapsedTime long The time since the application was started (in ms)

MapReduce Jobs

Call checkJobs() to return an array of Job structure. The jobs resource provides a list of the MapReduce jobs that have finished. It does not currently return a full list of parameters. The simplest form is checkJobs(), which will return all jobs available:

let jobs = try app.checkJobs()
jobs.forEach { j in
  print(j.id)
  print(j.name)
  print(j.queue)
  print(j.state)
}

If succeeded, checkJobs() would probably return a job object as described below:

Data Structure of Job

Item Data Type Description
id String The job id
name String The job name
queue String The queue the job was submitted to
user String The user name
state String the job state - valid values are: NEW, INITED, RUNNING, SUCCEEDED, FAILED, KILL_WAIT, KILLED, ERROR
startTime Int The time the job started (in ms since epoch)
finishTime Int The time the job finished (in ms since epoch)
elapsedTime Int The elapsed time since job started (in ms)
mapsTotal Int The total number of maps
mapsCompleted Int The number of completed maps
reducesTotal Int The total number of reduces
reducesCompleted Int The number of completed reduces
diagnostics String A diagnostic message
uberized Bool Indicates if the job was an uber job - ran completely in the application master
mapsPending Int The number of maps still to be run
mapsRunning Int The number of running maps
reducesPending Int The number of reduces still to be run
reducesRunning Int The number of running reduces
failedReduceAttempts Int The number of failed reduce attempts
killedReduceAttempts Int The number of killed reduce attempts
successfulReduceAttempts Int The number of successful reduce attempts
failedMapAttempts Int The number of failed map attempts
killedMapAttempts Int The number of killed map attempts
successfulMapAttempts Int The number of successful map attempts
acls [ACL] A collection of acls objects

Elements of the acls object

Item Data Type Description
value String The acl value
name String The acl name

Check a Specific Job

It is also possible to check a specific job by a job id:

let job = try app.checkJob(jobId: "job_1484231633049_0005")

See [Data Structure of Job](# Data Structure of Job) as above.

JobAttempt of Job

When you make a request for the list of job attempts, the information will be returned as an array of job attempt objects.

guard let attempts = try app.checkJobAttempts(jobId: "job_1484231633049_0005") else {
	// something goes wrong
}
attempts.forEach { attempt in
	print(attempt.id)
	print(attempt.containerId)
	print(attempt.nodeHttpAddress)
	print(attempt.nodeId)
	print(attempt.startTime)
}

JobAttempt Structure:

Item Data Type Description
id String The job attempt id
nodeId String The node id of the node the attempt ran on
nodeHttpAddress String The node http address of the node the attempt ran on
logsLink String The http link to the job attempt logs
containerId String The id of the container for the job attempt
startTime Long The start time of the attempt (in ms since epoch)

Counters of Job

With the job counters API, you can object a collection of resources that represent al the counters for that job.

guard let js = try app.checkJobCounters(jobId: "job_1484231633049_0005") else {
	// something wrong
}//end guard
js.counterGroup.forEach{ group in
  print(group.counterGroupName)
  group.counters.forEach { counter in
    print(counter.name)
    print(counter.mapCounterValue)
    print(counter.reduceCounterValue)
    print(counter.totalCounterValue)
  }//next counter
}//next group

JobCounter Object

Item Data Type Description
id String The job id
counterGroup [CounterGroup] An array of counter group objects

CounterGroup Object

Item Data Type Description
counterGroupName string The name of the counter group
counter [Counter] An array of counter objects

Counter Object

Item Data Type Description
name String The name of the counter
reduceCounterValue Int The counter value of reduce tasks
mapCounterValue Int The counter value of map tasks
totalCounterValue Int The counter value of all tasks

Check Configuration of Job

Use checkJobConfig() to check configuration of a specific job:

gurard let config = try app.checkJobConfig(jobId: "job_1484231633049_0005") else {
	/// something wrong
}
// print the configuration path
print(config.path)
// check properties of configuration file
for p in config.property {
  print(p.name)
  print(p.value)
  print(p.source)
}

JobConfig Object

A job configuration resource contains information about the job configuration for this job.

Item Data Type Description
path String The path to the job configuration file
property [Property] an array of property object

Property Object

Item Data Type Description
name String The name of the configuration property
value String The value of the configuration property
source String The location this configuration object came from. If there is more then one of these it shows the history with the latest source at the end of the list.

Tasks of a Job

Use checkJobTasks() to collect task information from a specific job:

// get all tasks information from a specific job
let tasks = try app.checkJobTasks(jobId: "job_1484231633049_0005")

// print properties of each task
for t in tasks {
  print(t.progress)
  print(t.elapsedTime)
  print(t.state)
  print(t.startTime)
  print(t.id)
  print(t.type)
  print(t.successfulAttempt)
  print(t.finishTime)
}//next t

Parameters of checkJobTasks()

Calling checkJobTasks() requires a jobId all the time, however, you can also specify an extra parameter to indicate what type of tasks shall be listed in the query results:

  • jobId: a string represents the id of the job
  • taskType: optional, could be .MAP or .REDUCE

JobTask Object

The checkJobTasks() will return an array of JobTask data structure, as described below:

Item Data Type Description
id string The task id
state String The state of the task - valid values are: NEW, SCHEDULED, RUNNING, SUCCEEDED, FAILED, KILL_WAIT, KILLED
type String The task type - MAP or REDUCE
successfulAttempt String The id of the last successful attempt
progress Double The progress of the task as a percent
startTime Int The time in which the task started (in ms since epoch) or -1 if it was never started
finishTime Int The time in which the task finished (in ms since epoch)
elapsedTime Int The elapsed time since the application started (in ms)

Query a Specific JobTask

If both jobId and jobTaskId are available, you can use checkJobTask() to perform a query on a specific job task:

guard let task = try app.checkJobTask(jobId: "job_1484231633049_0005", taskId: "task_1326381300833_2_2_m_0") else {
	// something wrong
}
print(task.progress)

The return value of checkJobTask will be a [JobTask](# JobTask Object) object, as described above.

Task Counters of Job

Method checkJobTaskCounters() can return counters of a specific task, as the example below:

guard let js = try app.checkJobTaskCounters(jobId: "job_1484231633049_0005", taskId: "task_1326381300833_2_2_m_0") else {
	// something wrong
}
// print the id of the job task counters
print(js.id)
// check out each counter group
js.taskCounterGroup.forEach{ group in
  print(group.counterGroupName)
  // print counters in each group.
  group.counters.forEach { counter in
    print(counter.name)
    print(counter.value)
  }
}

JobTaskCounter Object

Item Data Type Description
id String The job id
taskCounterGroup [CounterGroup] An array of counter group objects, See [CounterGroup](# CounterGroup Object)

Task Attempts

Function checkJobTaskAttempts() can return attempts from a specific job task.

let jobTaskAttempts = try app.checkJobTaskAttempts(jobId: "job_1484231633049_0005", taskId: "task_1326381300833_2_2_m_0")
for attempt in jobTaskAttempts {
  print(attempt.id)
  print(attempt.diagnostics)
  print(attempt.assignedContainerId)
  print(attempt.rack)
  print(attempt.state)
  print(attempt.progress)
}//next

Once done, the checkJobTaskAttempts will return an array of TaskAttempt object, as described below:

TaskAttempt Object

Elements of the TaskAttempt object:

Item Data Type Description
id String The task id
rack String The rack
state TaskAttempt.State The state of the task attempt - valid values are: NEW, UNASSIGNED, ASSIGNED, RUNNING, COMMIT_PENDING, SUCCESS_CONTAINER_CLEANUP, SUCCEEDED, FAIL_CONTAINER_CLEANUP, FAIL_TASK_CLEANUP, FAILED, KILL_CONTAINER_CLEANUP, KILL_TASK_CLEANUP, KILLED
type TaskType The type of task - MAP or REDUCE
assignedContainerId String The container id this attempt is assigned to
nodeHttpAddress String The http address of the node this task attempt ran on
diagnostics String A diagnostics message
progress Double The progress of the task attempt as a percent
startTime Int The time in which the task attempt started (in ms since epoch)
finishTime Int The time in which the task attempt finished (in ms since epoch)
elapsedTime Int The elapsed time since the task attempt started (in ms)

Besides, For reduce task attempts you also have the following fields:

Item Data Type Description
shuffleFinishTime Int The time at which shuffle finished (in ms since epoch)
mergeFinishTime Int The time at which merge finished (in ms since epoch)
elapsedShuffleTime Int The time it took for the shuffle phase to complete (time in ms between reduce task start and shuffle finish)
elapsedMergeTime Int The time it took for the merge phase to complete (time in ms between the shuffle finish and merge finish)
elapsedReduceTime Int The time it took for the reduce phase to complete (time in ms between merge finish to end of reduce task)

Query a Specific TaskAttempt

If a taskAttempt id is available, you can also check the specific one by checkJobTaskAttempt(), as demo below:

guard let jobTaskAttempts = try app.checkJobTaskAttempt(jobId: "job_1484231633049_0005", taskId: "task_1326381300833_2_2_m_0", "attempt_1326381300833_2_2_m_0_0") else {
	// something wrong
	print("there is no such a task attempt")
	return
}//end guard
print(attempt.id)
print(attempt.diagnostics)
print(attempt.assignedContainerId)
print(attempt.rack)
print(attempt.state)
print(attempt.progress)

The checkJobTaskAttempt() will return such an [TaskAttempt object](# TaskAttempt Object) as described above.

Attempt State Control

Currently Hadoop MapReduce Application Master has an experimental attempt state control.

Check Attempt

To check a running attempt state, use checkJobTaskAttemptState():

guard let state = try checkJobTaskAttemptState(jobId: "job_1484231633049_0005", taskId: "task_1326381300833_2_2_m_0", "attempt_1326381300833_2_2_m_0_0") else {
	// something wrong
	print("there is no such a task attempt")
	return
}//end guard
print(state)

Kill Attempt

To terminate an attempt, call killTaskAttempt():

try killTaskAttempt(jobId: "job_1484231633049_0005", taskId: "task_1326381300833_2_2_m_0", "attempt_1326381300833_2_2_m_0_0")

Attempt Counters of Task

Method checkJobTaskAttemptCounters() can return counters of a specific task attempt, as the example below:

guard let counters = try app.checkJobTaskAttemptCounters(jobId: "job_1484231633049_0005", taskId: "task_1326381300833_2_2_m_0", "attempt_1326381300833_2_2_m_0_0") else {
	// something wrong
	print("Task Attempt Counters Acquiring failed")
	return 
}
// print the counters' id
print(counters.id)
// enumerate all groups
for group in counters.taskAttemptCounterGroup {
  // print group name
  print(group.counterGroupName)
  // enumerate all counters in this group
  for counter in group.counters {
    print(counter.name)
    print(counter.value)
  }//next counter
}//next group
}

JobTaskAttemptCounters Object

Item Data Type Description
id String The job id
taskAttemptcounterGroup [CounterGroup] An array of counter group objects, See [CounterGroup](# CounterGroup Object)