hbase-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Jeff Hodges (JIRA)" <j...@apache.org>
Subject [jira] Commented: (HBASE-1388) Implement performance tests using Vpork
Date Mon, 29 Jun 2009 21:44:47 GMT

    [ https://issues.apache.org/jira/browse/HBASE-1388?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12725362#action_12725362
] 

Jeff Hodges commented on HBASE-1388:
------------------------------------

Uh, okay, so some misconceptions need to be cleared up.

1) Neither I nor the code I wrote speak for the vpork point of view. The first time I even
looked at vpork's codebase was Saturday. I didn't even know groovy beforehand.

2) I don't know that anyone who's actually working on vpork thinks that thrift should be The
Only Way to access anything. I honestly have no friggin' clue how that little meme started.
Can we stop that?

3) I wrote the Thrift client because I could not find any documentation that described what
was needed to make `new HBaseConfiguration()` Just Work. There was also no documentation describing
how to lay out the Configuration that could be passed to HbaseConfiguration as an argument.
It wasn't even clear that you could work on a remote table without the whole set of hbase
configs available.

4) People are trying to find all the pain points of using a variety of dbs, including the
random reads and writes. vpork helps with getting Real Numbers for those including, for certain
folks concerned with lots of data and lots of accesses, the 99% case (which I'd like to be
99.9%, but working on that). Hence, why it's nice. So, yeah, it's not testing everything hbase
does and I'm not sure, as-is, it's meant to. Just because it's an HBase pain point doesn't
mean it's a bad test.

To reiterate, the Thrift interface was picked because it worked, worked easily, worked remotely
(meaning, to a whole cluster) and the API for it described exactly how to use it. (Well, modulo
the Mutations stuff. That took another minute of digging through the API.) No other reasons.

I'll take a look at the hbase-site.xml and zoo.cfg files and see if I can't hack those requirements
into something that fits the vpork config model. St.Ack mentioned maybe working on this. That'd
be nice.

> Implement performance tests using Vpork
> ---------------------------------------
>
>                 Key: HBASE-1388
>                 URL: https://issues.apache.org/jira/browse/HBASE-1388
>             Project: Hadoop HBase
>          Issue Type: Wish
>          Components: test
>            Reporter: Johan Oskarsson
>            Priority: Trivial
>
> It would be useful to have a suite of performance tests so that HBase can be compared
to other similar projects. This could then be automated on the same hardware (perhaps even
EC2?).
> There is a project called Vpork, developed by Jon Travis to test Voldemort that would
be a good starting point. It is written in Groovy and can use the Java client libs. This version
has been slightly refactored and also supports Cassandra:
> http://github.com/johanoskarsson/vpork/tree/master

-- 
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.


Mime
View raw message