[AMBARI-24681] Cannot deploy cluster without HDFS_CLIENT

• Type: Bug
• Status: Resolved
• Priority: Critical
• Resolution: Fixed
• Affects Version/s: 2.7.0
• Fix Version/s: 2.7.3
• Component/s: ambari-server
• Labels: pull-request-available
• Assignee: Attila Doroszlai
• Reporter: Attila Doroszlai
• Created: 25/Sep/18 04:47
• Updated: 22/Oct/18 21:00
• Resolved: 25/Sep/18 17:45

Description

The attached blueprint cannot be deployed, because hadoop-env.sh is not saved by Ambari, because it requires HDFS_CLIENT component on at least one host:

https://github.com/apache/ambari/blob/8d145e0c04917866fd76690688826cf44065370e/ambari-server/src/main/resources/stack-hooks/before-ANY/scripts/hook.py#L32-L33

Datanode start fails with:

ExecutionFailed: Execution of 'ambari-sudo.sh su hdfs -l -s /bin/bash -c 'ulimit -c unlimited ;  /usr/hdp/3.0.0.0-1634/hadoop/bin/hdfs --config /usr/hdp/3.0.0.0-1634/hadoop/conf --daemon start datanode'' returned 1. ERROR: JAVA_HOME is not set and could not be found.

This appears in output from after-INSTALL hook:

Parameter hadoop_conf_dir is missing or directory does not exist. This is expected if this host does not have any Hadoop components.

Additionally, host groups without HDFS_CLIENT cannot be scaled up.

Issue Links

• links to: GitHub Pull Request #2370