apache/hadoop · error · UnsupportedOperationException

Killing tasks in LocalJobRunner is not supported

Error message

Killing tasks in LocalJobRunner is not supported

What it means

LocalJobRunner.killTask(TaskAttemptID, boolean) unconditionally throws UnsupportedOperationException: the local runner executes all tasks on threads of one Job thread, so individual task attempts cannot be killed or failed independently. Job-level kill is supported — killJob() sets the Job.killed flag and interrupts the worker thread.

Source

Thrown at hadoop-mapreduce-project/hadoop-mapreduce-client/hadoop-mapreduce-client-common/src/main/java/org/apache/hadoop/mapred/LocalJobRunner.java:814

    return job.status;

  }

  public void killJob(org.apache.hadoop.mapreduce.JobID id) {
    jobs.get(JobID.downgrade(id)).killed = true;
    jobs.get(JobID.downgrade(id)).interrupt();
  }

  public void setJobPriority(org.apache.hadoop.mapreduce.JobID id,
      String jp) throws IOException {
    throw new UnsupportedOperationException("Changing job priority " +
                      "in LocalJobRunner is not supported.");
  }
  
  /** Throws {@link UnsupportedOperationException} */
  public boolean killTask(org.apache.hadoop.mapreduce.TaskAttemptID taskId,
      boolean shouldFail) throws IOException {
    throw new UnsupportedOperationException("Killing tasks in " +
    "LocalJobRunner is not supported");
  }

  public org.apache.hadoop.mapreduce.TaskReport[] getTaskReports(
      org.apache.hadoop.mapreduce.JobID id, TaskType type) {
    return new org.apache.hadoop.mapreduce.TaskReport[0];
  }

  public org.apache.hadoop.mapreduce.JobStatus getJobStatus(
      org.apache.hadoop.mapreduce.JobID id) {
    Job job = jobs.get(JobID.downgrade(id));
    if(job != null)
      return job.status;
    else 
      return null;
  }
  
  public org.apache.hadoop.mapreduce.Counters getJobCounters(

View on GitHub (pinned to 2add963021)

Solutions

  1. Kill the whole job instead (JobClient.killJob) when running locally
  2. Run the test on a real or mini YARN cluster (MiniYARNCluster / mapreduce.framework.name=yarn) when per-task kill semantics are required
  3. Guard the call behind a framework-name check before invoking

Example fix

// before
client.killTask(attemptId, false); // throws in local mode

// after
if ("local".equals(conf.get("mapreduce.framework.name", "local"))) {
  client.killJob(jobId); // only job-level kill exists locally
} else {
  client.killTask(attemptId, false);
}
Defensive patterns

Strategy: validation

Validate before calling

if (!"local".equals(conf.get("mapreduce.framework.name", "local"))) {
  client.killTask(attemptId, false);
} else {
  client.killJob(jobId); // local runner: job-level kill only
}

Try / catch

try {
  client.killTask(attemptId, shouldFail);
} catch (UnsupportedOperationException e) {
  LOG.warn("Per-task kill unsupported locally; killing job instead");
  client.killJob(jobId);
}

Prevention

When it happens

Trigger: Calling JobClient.killTask(attemptId, shouldFail), killTaskAttempt, or 'hadoop job -kill-task' / '-fail-task' against a job whose framework is local.

Common situations: Test harnesses that mimic cluster administration (kill a mapper to test retry logic) but run under LocalJobRunner; porting cluster-operations tooling to a local dev setup; retry-behavior tests that must instead run on MiniYARNCluster.

Related errors

AI-assisted analysis of apache/hadoop@2add963021 (2026-08-22). Data as JSON: /api/errors/ad160d90fee7a868. Report an issue: GitHub.