hbase-user mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Stack <st...@duboce.net>
Subject Re: servlet container (embedded jetty)
Date Fri, 23 Jul 2010 15:43:43 GMT
On Fri, Jul 23, 2010 at 8:16 AM, Mark Laffoon
<mlaffoon@semanticresearch.com> wrote:
> I have a couple noobish questions about the servlet container.
>
> 1. As I understand it from reading docs and poking around in the code
> hadoop and hbase both use an embedded jetty instance to host their admin
> pages.
>

Yes

> 2. If I have hadoop master and hbase master running on the same machine,
> does that mean I have two jetty instances instantiated?
>

Yes.

> 2a. What hadoop and hbase processes use an embedded jetty? Obviously
> namenode, hbase master, and jobtracker. What about regionserver,
> tasktracker?
>

The regionserver yes, the tasktracker does too at least to transfer
files during MR (last I checked).  The datanode doesn't have an
instance.

> 3. Any idea what the approximate footprint of an embedded jetty instance
> is?
>

Guessing, I'd say its "small"


> 4. If I have a servlet container (jetty, tomcat, whatever) running on the
> machine for other reasons, how tough is it to get hadoop and hbase to use
> it instead of the embedded jetty?
>

We've not done the WAR packaging.  The pieces are all there -- a
web.xml, etc., see hbase-webapps when you download a binary -- so
shouldn't be hard bundling them up.  Hard  part, as for any WAR, is
configuring it.  HBase/Hadoop config is in the conf directory.  You'd
need to get this conf on the webapp CLASSPATH so it was pointing at
proper cluster.

Historically, IIUC hadoop/hbase doesn't go all the way and make a WAR
because wants to ship w/ precompiled jsps... so on deploy you can get
away w/ JRE rather than JDK (Not so sure how this holds up any more).

>
>
> This boils down to: I'm going to have various processes using a servlet
> container on my servers. Should I worry about trying to share the same
> container across all the hadoop/hbase processes as well?
>

Unless a compelling reason, I'd say avoid the unification.
Hadoop/HBase are non-standard in their use of java technologies.
You'll be surprised at every turn.  It'd be time sink for little
benefit.

Others might have different opinions.

St.Ack

>
>
> Thanks,
>
> Mark
>
>

Mime
View raw message