giraph-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Nitay Joffe" <ni...@apache.org>
Subject Re: Review Request: GIRAPH-453: Pure Hive I/O (nitay)
Date Fri, 22 Feb 2013 03:24:33 GMT


> On Feb. 21, 2013, 6:45 p.m., Maja Kabiljo wrote:
> > This is a lot of great work, Nitay, thanks! I really like that user doesn't have
to extend the whole Input/Output format anymore, that was a lot of code duplication every
time.
> > 
> > Is it possible to provide some examples/tests for this?

Opened https://issues.apache.org/jira/browse/GIRAPH-534 so that we create examples / tests.


> On Feb. 21, 2013, 6:45 p.m., Maja Kabiljo wrote:
> > giraph-hive/src/main/java/org/apache/giraph/hive/common/HiveProfiles.java, lines
29-31
> > <https://reviews.apache.org/r/8611/diff/6/?file=260732#file260732line29>
> >
> >     What is this for? (on some other places too)

It is to allow multiple tables at same time. Basically to do it you need to have some namespacing
for Configuration keys, so these profiles are my way of doing it. I have a cleaner solution
in mind that I will put in another diff which should clean up some of these.


> On Feb. 21, 2013, 6:45 p.m., Maja Kabiljo wrote:
> > giraph-hive/src/main/java/org/apache/giraph/hive/common/HiveProfiles.java, lines
35-36
> > <https://reviews.apache.org/r/8611/diff/6/?file=260732#file260732line35>
> >
> >     Out of curiosity - why do we do this? (why isn't it private)

Sometimes I want to allow inheritance but in this case no need, private it is.


> On Feb. 21, 2013, 6:45 p.m., Maja Kabiljo wrote:
> > giraph-hive/src/main/java/org/apache/giraph/hive/input/edge/HiveEdgeReader.java,
line 154
> > <https://reviews.apache.org/r/8611/diff/6/?file=260735#file260735line154>
> >
> >     Could we have an option to reuse edge objects here?

Good call


- Nitay


-----------------------------------------------------------
This is an automatically generated e-mail. To reply, visit:
https://reviews.apache.org/r/8611/#review16867
-----------------------------------------------------------


On Feb. 21, 2013, 6:17 p.m., Nitay Joffe wrote:
> 
> -----------------------------------------------------------
> This is an automatically generated e-mail. To reply, visit:
> https://reviews.apache.org/r/8611/
> -----------------------------------------------------------
> 
> (Updated Feb. 21, 2013, 6:17 p.m.)
> 
> 
> Review request for giraph.
> 
> 
> Description
> -------
> 
> One particular thing I added was the concept of "profiles", allowing for easily reading
/ writing from multiple tables. This should remove a lot of the cruft around the GiraphHCat*
classes.
> 
> Note in the diff I separated the code so that there would be a Giraph-unrelated Hive-only
portion (under package org.apache.hadoop.hive). Things under this package (and its children)
do not touch any Giraph code, and so can be contributed as an IOFormat back to Hive itself.
> 
> Also note the new (I think improved) interface: Users do not need to actually implement
an XInputFormat anymore. They just create a class the implements the HiveToVertex (HiveToEdge,
VertexToHive) interface, plug that in, and use HiveVertexInputFormat. Should make user code
much cleaner.
> 
> 
> This addresses bug GIRAPH-453.
>     https://issues.apache.org/jira/browse/GIRAPH-453
> 
> 
> Diffs
> -----
> 
>   giraph-accumulo/pom.xml cb9fbc02e6fc8adcb0ec41e0c6aeff75b1ef3f06 
>   giraph-core/src/main/java/org/apache/giraph/comm/netty/NettyClient.java 89ef87fea7a370354156fb7be02ef4249e0a6111

>   giraph-core/src/main/java/org/apache/giraph/conf/GiraphConfiguration.java ddeaeb769b548eb1002ccf8c18ffe048eb096f8d

>   giraph-hbase/pom.xml 7bbbd98c0b3db6878aee4be21eecd821448da7ef 
>   giraph-hcatalog/pom.xml 019f02083012704a997ffe715cefe3adeb153dd9 
>   giraph-hcatalog/src/main/java/org/apache/giraph/io/hcatalog/HCatGiraphRunner.java PRE-CREATION

>   giraph-hcatalog/src/main/java/org/apache/giraph/io/hcatalog/HiveGiraphRunner.java 313bab04c50ed6be7143254de80e36a4ba291516

>   giraph-hcatalog/src/main/java/org/apache/giraph/io/hcatalog/HiveUtils.java c1f76f1a46d1fc9af489a916256884520c138cb4

>   giraph-hive/pom.xml PRE-CREATION 
>   giraph-hive/src/main/assembly/compile.xml PRE-CREATION 
>   giraph-hive/src/main/java/org/apache/giraph/hive/HiveGiraphRunner.java PRE-CREATION

>   giraph-hive/src/main/java/org/apache/giraph/hive/common/HiveProfiles.java PRE-CREATION

>   giraph-hive/src/main/java/org/apache/giraph/hive/common/package-info.java PRE-CREATION

>   giraph-hive/src/main/java/org/apache/giraph/hive/input/edge/HiveEdgeInputFormat.java
PRE-CREATION 
>   giraph-hive/src/main/java/org/apache/giraph/hive/input/edge/HiveEdgeReader.java PRE-CREATION

>   giraph-hive/src/main/java/org/apache/giraph/hive/input/edge/HiveToEdge.java PRE-CREATION

>   giraph-hive/src/main/java/org/apache/giraph/hive/input/edge/package-info.java PRE-CREATION

>   giraph-hive/src/main/java/org/apache/giraph/hive/input/package-info.java PRE-CREATION

>   giraph-hive/src/main/java/org/apache/giraph/hive/input/vertex/HiveToVertex.java PRE-CREATION

>   giraph-hive/src/main/java/org/apache/giraph/hive/input/vertex/HiveVertexInputFormat.java
PRE-CREATION 
>   giraph-hive/src/main/java/org/apache/giraph/hive/input/vertex/HiveVertexReader.java
PRE-CREATION 
>   giraph-hive/src/main/java/org/apache/giraph/hive/input/vertex/package-info.java PRE-CREATION

>   giraph-hive/src/main/java/org/apache/giraph/hive/output/HiveVertexOutputFormat.java
PRE-CREATION 
>   giraph-hive/src/main/java/org/apache/giraph/hive/output/HiveVertexWriter.java PRE-CREATION

>   giraph-hive/src/main/java/org/apache/giraph/hive/output/VertexToHive.java PRE-CREATION

>   giraph-hive/src/main/java/org/apache/giraph/hive/output/package-info.java PRE-CREATION

>   giraph-hive/src/main/java/org/apache/giraph/hive/package-info.java PRE-CREATION 
>   pom.xml c075762cddd7a698c92aaad4017cd74915160e41 
> 
> Diff: https://reviews.apache.org/r/8611/diff/
> 
> 
> Testing
> -------
> 
> Ran on some production jobs and verified results were exactly the same.
> 
> Here's a comparison of performance on real work loads ("base" is hcatalog, "mine" is
hive):
> https://gist.github.com/nitay/880d8fb20d2ac86015d4/raw/6b297fcb287bf8d3dc8175bad217aa86544b4f18/high+school
> 
> Basically we see slight improvement which is expected because I haven't done a lot in
terms of performance yet.
> There are few performance improvement ideas coming, this is just the first working version.
> 
> 
> Thanks,
> 
> Nitay Joffe
> 
>


Mime
  • Unnamed multipart/alternative (inline, None, 0 bytes)
View raw message