8.4. Known Issues for YARN

  • BUG-158341: YARN and/or mapred client should add tokens for default filesystem.

    Problem: As noticed in BUG-15360 if jobs are using webhdfs then they could run into an issue where the job only has webhdfs tokens and yarn jobs would fail as it tries to talk to hdfs over default fs since tokens for default fs as not part of the job.

  • BUG-15376: {YARN-1892] CS fast scheduling patch ends up causing excessive logging.

    Problem: Seeing about 1 GB of logs per hour.

  • BUG-15360: In HDFS HA mode, Distcp/SLive with webhdfs on secure cluster fails with Client cannot authenticate via:[TOKEN, KERBEROS] error.

  • BUG-13231: YARN RM won't failover if the RPC port is unreachable.

    Problem: YARN does not have a service similar to HDFS where the zkfc process monitors the health of the NameNode. Thus, if the RPC port gets blocked the RM service will not failover.

  • BUG-12327: [Yarn-90] NM cannot detect when bad disks become healthy again.

    Problem: If you start NM with good log-dir, then rename the directory away, the NM will become unhealthy. If you then rename the directory away, NM will be unhealthy. If you rename the directory back, then wait for some period of time (120 sec), NM won't return to healthy state.

  • BUG-7531: Hadoop metrics link does not contain correct content.

    Problem: In the Resource Manager UI Tools section, clicking on Logs and Metrics opens pages that do not contain correct information.