Mailing-List: contact hbase-user-help@hadoop.apache.org; run by ezmlm
Precedence: bulk
Reply-To: hbase-user@hadoop.apache.org
Received-SPF: pass (nike.apache.org: domain of gstathis@gmail.com designates
 74.125.82.176 as permitted sender)
DomainKey-Signature: a=rsa-sha1; c=nofws;
        d=gmail.com; s=gamma;
        h=mime-version:in-reply-to:references:from:date:message-id:subject:to
         :content-type;
        b=pYfMfDDFS+RWeoh91PTF34nCj97SSml6Htfqi5QUeIxEd3eYK0YpLF1f9T4u0+/yYE
         SMaTMXVaybt6E7w1et0F6VKJcXwgKN89X3P9Nso3JbBoxgTD82NunR9L4wc2cGxiqVT4
         wrgCwgC+EWFnI+70W295aWvdSm2T6SoCmUHCw=
MIME-Version: 1.0
In-Reply-To: <7c962aed1003301104n531e0d05u4aee9dd2e2b32e1f@mail.gmail.com>
References: <96c64fa81003291251q1cdb2408gce325edbbc13fa12@mail.gmail.com>
	<7c962aed1003301104n531e0d05u4aee9dd2e2b32e1f@mail.gmail.com>
From: George Stathis <gstathis@gmail.com>
Date: Tue, 30 Mar 2010 20:57:52 -0400
Message-ID: <96c64fa81003301757k13d681d4v9f701e828bdc2982@mail.gmail.com>
Subject: Re: Contrib tableindexed package vs. custom indexes
To: hbase-user@hadoop.apache.org
Content-Type: multipart/alternative; boundary=0016365eec828baece04830e3dcf

--0016365eec828baece04830e3dcf
Content-Type: text/plain; charset=ISO-8859-1

Stack, thanks for getting back to me. I think we are better off going with
the contrib approach and investing our dev time there. It will benefit
everyone in the long term, including us.

-GS

On Tue, Mar 30, 2010 at 2:04 PM, Stack <stack@duboce.net> wrote:

> Sorry George for lack of response.  I think it probably a bit of 3.)
> and then 4.) which is that you know cleanly the options so there is
> nothing really to add.
>
> My sense is that when fellas say roll your own indexes, between the
> lines I think what they are saying is that they do not want to do the
> two updates transactionally -- that they do not want to pay the
> ITHBase tax -- and are ok w/ losing a index add or two.
>
> The preso at ApacheCon was a callout to THBase and ITHBase recognizing
> that its been a contrib for a good while now and that perhaps its
> graduated beyond our designation of it as 'experimental'.
>
> St.Ack
>
>
> On Mon, Mar 29, 2010 at 12:51 PM, George Stathis <gstathis@gmail.com>
> wrote:
> > Hi folks,
> >
> > I've seen some people around the list that recommend rolling one's own
> > indexes. Others say to just go with
> > the org.apache.hadoop.hbase.client.tableindexed package. A quick scan of
> the
> > wiki does not reveal any best practices. Presentations from the devs such
> as
> > the Oakland ApacheCon slides point to the contrib package.
> >
> > Some of the comments in the list seem to note that IndexedTable is not
> very
> > performant; then again, I would assume that a custom index would have to
> > wrap any table+index operations in a transaction anyway. So unless folks
> > forego transactions when rolling their own indexes, I don't see how a
> custom
> > implementation could be that much faster.
> >
> > What do the majority of people here do for indexing? Is there a generally
> > accepted good middle-of-the-road approach offering an acceptable
> compromise
> > between performance and maintainability? I must admit that rolling our
> own
> > indexes does not seem like a viable long term approach to me (from a
> > maintenance POV).
> >
> > I'm interested in people's opinion.
> >
> > -GS
> >
>

--0016365eec828baece04830e3dcf--