Home | About | Sematext search-lucene.com search-hadoop.com
 Search Hadoop and all its subprojects:

Switch to Threaded View
Accumulo, mail # dev - Re: [DISCUSS] Hadoop 2 and Accumulo 1.6.0


Copy link to this message
-
Re: [DISCUSS] Hadoop 2 and Accumulo 1.6.0
Steve Loughran 2013-10-24, 18:05
On 24 October 2013 18:01, John Vines <[EMAIL PROTECTED]> wrote:

> Accumulo isn't going to run in YARN,
Well it can if you want it too - what we are doing with hoya is giving the
option to do without changing the core application.

Some applications (e.g. Samza) are being written from the ground up for
YARN , so they have an Application Master that manages requesting
containers, running code in it, while the code knows it is in an isolated
container (currently: transient dir and cgroup-limited RAM, more limits
coming). But you don't need a complete rewrite to gain from running under
YAR.
Accumulo is very suited to running in YARN; if you look at the list of
things we need from an app to get it to play, it's pretty much complete:
https://github.com/hortonworks/hoya/blob/develop/src/site/markdown/app_needs.md

A key benefit is how you can react to failures. Until now: tserver down:
log it, balance workload. {master, monitor, gc} down: page someone.

In YARN,  all failures become something for the automation to handle: log
it, ask for new instance, balance work back when it is up. You can
downgrade the one-per-application node to that of logged statistics rather
than events that need immediate human intervention,

Detecting a process failure is trivial: it exits and the container does
too. Harder is more byzantine failure modes -what would be ideal would just
be able to go from the ZK data to identifying which service instance needs
to be let go.

What is useful here is getting all the locations of things like web URLs
out of ZK too -as you can't predetermine the host they will be on, and if
you fix the port you may have binding problems.

> it's just a matter of replacing MR
> compatibility with YARN compatibility.
>
>
MRv1 code should work with MR-on-YARN : if not, file a bug report

What you do get long term is the option of talking to other services
running in the cluster -and for them to talk to you

--
CONFIDENTIALITY NOTICE
NOTICE: This message is intended for the use of the individual or entity to
which it is addressed and may contain information that is confidential,
privileged and exempt from disclosure under applicable law. If the reader
of this message is not the intended recipient, you are hereby notified that
any printing, copying, dissemination, distribution, disclosure or
forwarding of this communication is strictly prohibited. If you have
received this communication in error, please contact the sender immediately
and delete it from your system. Thank You.