hbase-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Nitay <nit...@gmail.com>
Subject Re: Should we up the zk default time out from default 10 seconds to twenty or so?
Date Wed, 17 Jun 2009 18:06:06 GMT
Good point, the session timeout and the JNI stuff is disjoint, and we
don't want to see the user behavior you are describing. I agree, let's
bump it up to 30, and in my own testing I can play around with it.

On Wed, Jun 17, 2009 at 10:47 AM, Andrew Purtell<apurtell@apache.org> wrote:
> Actually 30 seconds does not sound unreasonable when considering how many people we see
up on hbase-users@ with problems related to expecting too much from underpowered test clusters.
I think it is more likely that they will overcommit resources and get a session timeout leading
to other problems then complain about recovery taking too long. We can also put up a note
on the troubleshooting page about
> lowering the zk session timeout to achieve faster failover.
>   - Andy
> ________________________________
> From: stack <stack@duboce.net>
> To: hbase-dev@hadoop.apache.org
> Sent: Wednesday, June 17, 2009 9:50:16 AM
> Subject: Should we up the zk default time out from default 10 seconds to  twenty or
> During my uploads on small, anemic cluster -- what folks usually run hbase
> on -- I am seeing that the ten second timeout happens easily enough.  On a
> small cluster, loss of zk lease is pretty catastrophic in terms of region
> shuffle especially if lots of data.  In my hair-trigger upload, I have it
> set to fail on a hiccup, loss of zk lease will usually mean my upload fails.
> On other hand, maybe we should leave it as is so we can make sure we are
> good at riding over zk lease expiration?
> What you all think?
> St.Ack

View raw message