apache/hadoop · critical · YarnRuntimeException

"Not creating intermediate history logDir: [" + doneDirPath

Error message

"Not creating intermediate history logDir: [" + doneDirPath + "] based on conf: mapreduce.am.create-intermediate-jh-base-dir. Either set to true or pre-create this directory with appropriate permissions"

What it means

At MR ApplicationMaster init, JobHistoryEventHandler checks the job-history intermediate done directory (mapreduce.jobhistory.intermediate-done-dir) on its filesystem. If the directory does not exist and mapreduce.am.create-intermediate-jh-base-dir is false (the default), the AM refuses to run and throws YarnRuntimeException('Not creating intermediate history logDir: ... Either set to true or pre-create this directory...'), failing the job before any task starts. The design expects the JobHistoryServer or deployment to have created that base directory.

Source

Thrown at hadoop-mapreduce-project/hadoop-mapreduce-client/hadoop-mapreduce-client-app/src/main/java/org/apache/hadoop/mapreduce/jobhistory/JobHistoryEventHandler.java:226

            + MRJobConfig.MR_AM_CREATE_JH_INTERMEDIATE_BASE_DIR);
          mkdir(
              doneDirFS,
              doneDirPath,
              new FsPermission(
            JobHistoryUtils.HISTORY_INTERMEDIATE_DONE_DIR_PERMISSIONS
                .toShort()));
          // TODO Temporary toShort till new FsPermission(FsPermissions)
          // respects
        // sticky
      } else {
          String message = "Not creating intermediate history logDir: ["
                + doneDirPath
                + "] based on conf: "
                + MRJobConfig.MR_AM_CREATE_JH_INTERMEDIATE_BASE_DIR
                + ". Either set to true or pre-create this directory with" +
                " appropriate permissions";
        LOG.error(message);
        throw new YarnRuntimeException(message);
      }
      }
    } catch (IOException e) {
      LOG.error("Failed checking for the existence of history intermediate " +
      		"done directory: [" + doneDirPath + "]");
      throw new YarnRuntimeException(e);
    }

    //Check/create user directory under intermediate done dir.
    try {
      doneDirPrefixPath =
          FileContext.getFileContext(conf).makeQualified(new Path(userDoneDirStr));
      mkdir(doneDirFS, doneDirPrefixPath, JobHistoryUtils.
          getConfiguredHistoryIntermediateUserDoneDirPermissions(conf));
    } catch (IOException e) {
      LOG.error("Error creating user intermediate history done directory: [ "
          + doneDirPrefixPath + "]", e);
      throw new YarnRuntimeException(e);

View on GitHub (pinned to 2add963021)

Solutions

Pre-create the directory with history-appropriate permissions as part of deployment: hdfs dfs -mkdir -p <intermediate-done-dir> ; hdfs dfs -chmod 1771/1777 <intermediate-done-dir> (match your cluster's policy)
Or set mapreduce.am.create-intermediate-jh-base-dir=true so the AM creates the base dir itself (fine for single-user/test clusters; shared clusters should pre-create it)
Verify the exact configured value: mapreduce.jobhistory.intermediate-done-dir and the staging dir (yarn.app.mapreduce.am.staging-dir) resolve where you expect
Confirm the JobHistoryServer is deployed and created its directory tree; check the AM logs for the doneDirPath in the message

Example fix

# before: dir absent, AM refuses to start
# after: pre-create with history-server-friendly perms, or let the AM create it
hdfs dfs -mkdir -p /mr-history/done_intermediate
hdfs dfs -chmod 1777 /mr-history/done_intermediate
# or in mapred-site.xml
<property><name>mapreduce.am.create-intermediate-jh-base-dir</name><value>true</value></property>

Defensive patterns

Strategy: validation

Validate before calling

// pre-flight in the job client before submit
Path doneDir = new Path(conf.get("mapreduce.jobhistory.intermediate-done-dir",
    "/tmp/hadoop-yarn/staging/history/done_intermediate"));
FileSystem fs = doneDir.getFileSystem(conf);
if (!fs.exists(doneDir)
    && !conf.getBoolean("mapreduce.am.create-intermediate-jh-base-dir", false)) {
  throw new IOException("Pre-create " + doneDir
      + " or set mapreduce.am.create-intermediate-jh-base-dir=true");
}

Try / catch

Catch YarnRuntimeException at AM init in integration tests and assert the message contains 'Not creating intermediate history logDir' -- then fail fast with the two remediation options instead of a generic AM launch failure.

Prevention

Bake 'hdfs dfs -mkdir -p <intermediate-done-dir>' plus permissions into cluster deployment
For test clusters set mapreduce.am.create-intermediate-jh-base-dir=true
Whenever you change mapreduce.jobhistory.intermediate-done-dir, migrate the HDFS layout in the same change

When it happens

Trigger: Fresh cluster where the intermediate done dir (default under the staging root, e.g. <staging>/history/done_intermediate on HDFS) was never created; custom mapreduce.jobhistory.intermediate-done-dir pointing to a missing path; permissions let the AM see the dir is absent but policy forbids it to mkdir (flag false); history server not yet deployed when the first job is submitted.

Common situations: New installations or test clusters without a running JobHistoryServer; secure clusters where the pre-created dir has wrong ownership; config drift where staging/intermediate-done-dir moved but the HDFS layout was not migrated.

Related errors

AI-assisted analysis of apache/hadoop@2add963021 (2026-08-22). Data as JSON: /api/errors/f17436615afef83d. Report an issue: GitHub.