apache/hadoop · error · IOException

Unknown TaskAttempt {}

Error message

Unknown TaskAttempt {}

What it means

verifyAndGetAttempt throws this IOException when the parent task resolves but task.getAttempt(attemptID) returns null — the attempt number does not exist for that task (attempts are numbered 0..n-1 per task, bounded by mapreduce.map.maxattempts / mapreduce.reduce.maxattempts).

Source

Thrown at hadoop-mapreduce-project/hadoop-mapreduce-client/hadoop-mapreduce-client-app/src/main/java/org/apache/hadoop/mapreduce/v2/app/client/MRClientService.java:233

      return job;
    }
 
    private Task verifyAndGetTask(TaskId taskID, 
        JobACL accessType) throws IOException {
      Task task =
          verifyAndGetJob(taskID.getJobId(), accessType, true).getTask(taskID);
      if (task == null) {
        throw new IOException("Unknown Task " + taskID);
      }
      return task;
    }

    private TaskAttempt verifyAndGetAttempt(TaskAttemptId attemptID, 
        JobACL accessType) throws IOException {
      TaskAttempt attempt = verifyAndGetTask(attemptID.getTaskId(), 
          accessType).getAttempt(attemptID);
      if (attempt == null) {
        throw new IOException("Unknown TaskAttempt " + attemptID);
      }
      return attempt;
    }

    @Override
    public GetCountersResponse getCounters(GetCountersRequest request) 
      throws IOException {
      JobId jobId = request.getJobId();
      Job job = verifyAndGetJob(jobId, JobACL.VIEW_JOB, true);
      GetCountersResponse response =
        recordFactory.newRecordInstance(GetCountersResponse.class);
      response.setCounters(TypeConverter.toYarn(job.getAllCounters()));
      return response;
    }
    
    @Override
    public GetJobReportResponse getJobReport(GetJobReportRequest request) 
      throws IOException {

View on GitHub (pinned to 2add963021)

Solutions

1. Enumerate real attempts via task.getAttempts().keySet() rather than constructing TaskAttemptId with a guessed number
2. Re-query the task report to learn which attempts exist before fetching attempt-level counters/diagnostics
3. Treat 'Unknown TaskAttempt' as a transient miss during task churn and retry once after a report refresh

// before attempt-level queries, learn which attempts exist
TaskReport report = ...; // from job.getTaskReports(type)
for (TaskAttemptID attemptId : report.getRunningTaskAttemptIds().isEmpty()
        ? completedAttemptIdsFromTaskReport(report)
        : report.getRunningTaskAttemptIds()) {
  queryAttempt(attemptId);
}

When it happens

Trigger: Client requests attemptId with attempt number >= attempts actually launched so far (race: asking about attempt 1 while only attempt 0 exists); attempt id string concatenated with a stale attempt counter; querying after the attempt was finalized.

Common situations: Progress dashboards computing attempt ids arithmetically instead of reading TaskReport.getCurrentAttemptAttemptIds / attempt lists; races between speculative-execution attempt launch and client queries.

Related errors

• Unknown Job {}
• Unknown Task {}
• Only 'KILLED' is allowed as a target state.
• too many failures downloading events
• Could not connect to History server.

AI-assisted analysis of apache/hadoop@2add963021 (2026-08-22). Data as JSON: /api/errors/97a16a6cd4935c48. Report an issue: GitHub.