ASF subversion and git services commented on ATLAS-3762:

Commit 23aea76fff6c8f530816319649218583f4dfd091 in atlas's branch 
refs/heads/master from Ashutosh Mestry
[ https://gitbox.apache.org/repos/asf?p=atlas.git;h=23aea76 ]

ATLAS-3762: Improve edge creation using genuine iterator. Part 2

> Entity Creation: Improve Edges Fetch Between Vertices
> -----------------------------------------------------
>                 Key: ATLAS-3762
>                 URL: https://issues.apache.org/jira/browse/ATLAS-3762
>             Project: Atlas
>          Issue Type: Improvement
>            Reporter: Ashutosh Mestry
>            Assignee: Ashutosh Mestry
>            Priority: Major
>         Attachments: ATLAS-3762-Edge-fetch-improvement-gremlin.patch, 
> ATLAS-3762-Improve-Edge-creator-using-Genuine-iterat.patch
> *Background*
> One of the earlier commits replaced vertices and edges fetch with 
> _StreamSupport.stream_. This uses _Collect(toList),_ which causes all 
> contents to be fetched. 
> Using this causes large amount of data to be fetched.
> *Solution*
> Switch to iterators that will use lazy loading.
> *Edge Fetch Refactoring*
> Change the _getEdge_ to iterate on smaller dataset. 
> Here are the scenarios:
> - _fromVertex_ is _hive_table_, _toVertex_ is _hive_column_. This means that 
> outgoing edges from _fromVertex_ will be many more than incoming edges to 
> _toVertex_.
> - _fromVertex_ is _hive_process_execution_, _toVertex_ is _hive_table_. This 
> means that outgoing edges from _fromVertex_ will be fewer than incoming edges 
> _hive_table_.
> Approach:
>  * Search it is a linear search, it will be more efficient to iterate over 
> fewer items than more items.
>  * Fetch count edge items for _fromVertex_ and _toVertex_. If either of the 
> count is 0, return NULL, since it will not result in anything being found.
>  * If either of the counts is not 0, take the one with fewer elements and 
> perform a search.
> [~sidharthkmishra] Thanks for this simple but effective fix.

This message was sent by Atlassian Jira

Reply via email to