Build HBase from source with Ambari deploy

[#1]

I'd like to use Ambari but be able to build HBase from source as we need a custom co-processor. Is there a documented way of doing this or do I have to turn into a full-custom install?

[#2]

You can deploy HBase as it ships with HDP and then deploy the jar for Co-processor to all RS with pdsh or Ansible or anything else. Editing hbase-site will be handled via Ambari configs. You don't need to rebuild HBase source.

[#3]

@Artem we do need to rebuild HBase since we need to run custom co-processors from Thrift. We want to do something like is described here: https://issues.apache.org/jira/browse/HBASE-5600 "We could create a thrift method to take the name of the class, method, and an array of params and then call coprocessorExec". If there's a better way to run a custom co-processor and access it from Thrift I'm all ears. But this brings me back to a good way to run a custom build of HBase with Ambari.

[#4]

You're probably going to have a rough time trying to build a Thrift server inside of a coprocessor.

Why not just create a Thrift server external to HBase which can access your custom coprocessor via the standard protobuf RPCs?

[#5 · Accepted solution]

End of the day, Ambari calls yum install (or the equivalent zypper/apt) to install the hbase packages. Post install if you replace the jars or other libraries then it will be transparent to Ambari (e.g. in /usr/hdp/2.6.0.0-457/hbase/lib). Of course this means that if you do Add Host then the same steps have to repeated.

[#6]

That's what I was thinking. Was just wondering if there was a way to define a custom package somehow. Thanks!

[#7]

@Vasco Pinho this is an interesting thread and I'm curious which path you'll take here, please accept the answer that best addresses your question. Thanks.