Upgraded to HDP 2.5.0 and now using Atlas 0.7.0.2.5 but I don't have or want Hbase, how can I use Atlas without Hbase?

[#1]

I just upgraded my cluster from HDP 2.3.4 to HDP 2.5.0. When doing this, I had to uninstall Atlats (0.5.0.2.3) before the upgrade and then install Atlas (0.7.0.2.5) after the upgrade. Now Atlas is wanting to use Hbase and a lot of the configs require Hbase specific items, but my cluster doesn't have Hbase installed and I do not want to install Hbase. Is there something I can do to get around this? I tried setting the following config items:

atlas.graph.storage.backend = berkeleyje
atlas.graph.index.search.backend = elasticsearch

The Atlas Metadata Server was finally able to come up (as per Ambari), but I cannot access the Atlas Metadata Server Web UI. Any help would be appreciated!

[#2 · Accepted solution]

I dont think so as Atlas has dependencies on other Apache projects and you need to make sure HDFS, Hbase, Kafka and Ambari Infra are all started for Atlas to function.

[#3]

I have installed HDP 2.5 and Atlas 0.7.0.2.5 and I don't need HBase to use Atlas. Currently I'm using Solr (Ambari Infra) as the index engine and Berkeleyje as the storage engine. The real dependency between Atlas and HBase is the AuditRepository, by default use HBase. Is not easy change it, but investigating the source code I found a special property atlas.EntityAuditRepository.impl that you have to set with the value org.apache.atlas.repository.audit.InMemoryEntityAuditRepository (is case sensitive, so copy&paste exactly the name of the property and the value).

@Chad Woodhedad, add the above property as the screenshot shows, restart Atlas services and you will have it, Atlas without HBase 🙂

And now some details about how I found that property:

In this link from GitHub you can see why Atlas by default needs HBase.

And in this link from GitHub you can find the available values that you can use for configure the audit.

Don't worry about the rest of the properties related with HBase. Atlas will use the value of atlas.graph.storage.hbase.table to create the table in the storage backend that you choose (Berkeleyje or Cassandra). With these properties my Atlas services work very well.

I hope this information is helpful and work for you to avoid install HBase in your clusters.