[jira] [Commented] (ATLAS-40) Support Accumulo as persistent store for graph repository

["Venkatesh Seetharam (JIRA)" <ji...@apache.org>]

    [ https://issues.apache.org/jira/browse/ATLAS-40?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14609382#comment-14609382 ] 

Venkatesh Seetharam commented on ATLAS-40:
------------------------------------------

Hi [~rweeks], thanks for taking this up. 

bq. Would this improvement be welcome?
of course, this is a great addition

bq. If so, can somebody suggest the best way to proceed?
IMO, option 1 would be a larger effort as compared to option 2.
There are several issues that needs to be addressed if you decide to pursue 1:

* search DSL translates to gremlin for titan and not sure what would that be for option 1
* search indexing depends on titan's capability today, not sure how that would map in option 1

On a positive note, option 2 does not take advantage of the native capabilities of the backing store. In Titan, its the same data model across backing stores which is bad IMO.

The least path to resistance would be to pursue option 2 IMHO.

bq. I was looking at option #1 originally but it looks like some parts of the code (eg. GraphBackedMetadataRepository have explicit dependencies on TitanGraph
You could have a separate implementation of the MetadataRepository itself, no? I originally used TinkerPop but then until tinker pop 3 was released, we were forced to use TitanGraph many times and thought lets depend on it instead.

Makes sense?


> Support Accumulo as persistent store for graph repository
> ---------------------------------------------------------
>
>                 Key: ATLAS-40
>                 URL: https://issues.apache.org/jira/browse/ATLAS-40
>             Project: Atlas
>          Issue Type: Improvement
>    Affects Versions: 0.5-incubating
>            Reporter: Russ Weeks
>            Priority: Minor
>             Fix For: 0.6-incubating
>
>
> I'd like to work on supporting [Accumulo|http://accumulo.apache.org] as well as HBase as a persistent store for the code in org.apache.atlas.repository.
> I can see two ways that this could work:
> # Define a new GraphProvider and provide an implementation of the Tinkerpop blueprints Graph which is backed by Accumulo - there are several good options here like [Calrissian|https://github.com/calrissian/accumulo-recipes/tree/master/store/graph-store] or the [JHU APL implementation|https://github.com/JHUAPL/AccumuloGraph].
> # Implement a new [Titan storage backend|https://github.com/thinkaurelius/titan/wiki/Storage-Backend-Overview] that writes to Accumulo. Any changes to Atlas should be limited to Titan configuration (probably some Accumulo-specific parameters).
> Would this improvement be welcome? If so, can somebody suggest the best way to proceed? I was looking at option #1 originally but it looks like some parts of the code (eg. [GraphBackedMetadataRepository|https://github.com/apache/incubator-atlas/blob/6b33bcf647264131cf5b37838485675d2ed9c017/repository/src/main/java/org/apache/atlas/repository/graph/GraphBackedMetadataRepository.java#L87] have explicit dependencies on TitanGraph, not the Tinkerpop Graph interface.



--
This message was sent by Atlassian JIRA
(v6.3.4#6332)