Re: High Availability Recommendation

I was a poster in the previously mentioned thread.

We currently have 25 repositories on a GFS filesystem being served on
two nodes via hardware load balancer for high availability using
Subversion 1.4.2, Apache 2.2.3 and mod_dav_svn.

We've not yet seen any evidence of corruption in any way, shape or
form. We've been using this solution for about 3 months.

-- 
-- Tom Mornini, CTO
-- Engine Yard, Ruby on Rails Hosting
-- Reliability, Ease of Use, Scalability
-- (866) 518-YARD (9273)
On Jan 31, 2007, at 1:42 PM, Justin Johnson wrote:
>> >>> > > I am using FSFS.  I was planning on having the filesystem  
>> mounted on
>> >>> > > two systems at the same time, but only so we can quickly  
>> fail over if
>> >>> > > one goes down.  Won't the repository be corrupted if two  
>> servers are
>> >>> > > actively writing to the same repository at the same time?   
>> If not, I
>> >>> > > am confused as to how.
>> >>> > >
>> >>> >
>> >>> > It's been explained to me that the with FSFS format, the  
>> file locking with
>> >> the Subversion binaries the transaction files in the FSFS  
>> directories keep
>> >> things in check.  But this means putting a set of Subversion  
>> binaries on
>> > each
>> >> server and not the NFS mount.  However, the binaries don't  
>> require much
>> >> configuration.  As you said, hooks scripts and other stuff  
>> involving the
>> >> repositories themselves is where all the configuration occurs.
>> >>>
>> >>> Wow, that would be great!  Can someone else confirm this for  
>> me and
>> >>> explain how it works or point me to something I can read about  
>> it?
>> >>> Why would this work only if the binaries aren't on the NFS  
>> mount?  I
>> >>> can see the need for different configuration files in Apache  
>> due to
>> >>> different host names for the servers.  Is there some other  
>> reason the
>> >>> binaries would need to be off of the NFS mount?
>> >>
>> >> In addition to needing confirmation about whether multiple  
>> servers can
>> >> write to a repository on the same disk NFS mounted on all of the
>> >> servers, I am also concerned about NFS in general.  The notes at
>> >> http://svn.collab.net/repos/svn/trunk/notes/fsfs in the "Locking"
>> >> section indicate that there might be some issues with NFS.  Can  
>> anyone
>> >> confirm or deny this?  Are any issues with with FSFS  
>> repositories and
>> >> NFS/NAS, any special NFS mount options, performance issues, etc.
>> >>
>> >> Thanks,
>> >> Justin
>> >
>> > Yeah, I had mentioned that as well one time back in a posting  
>> here, I was
>> > quickly told that wasn't the case any more and that the book  
>> needs to be
>> > updated there because of FSFS.
>> >
>> > Johnathan
>>
>> Here's an earlier thread that dealt with clustering of  
>> Subversion.  In particular check the responses from Ryan Schmidt  
>> and Garrett Rooney.  Hope this helps,
>>
>> http://subversion.tigris.org/servlets/BrowseList? 
>> list=users&by=thread&from=518292
>>
>
> Thanks for your research Johnathan.
>
> Is there anyone on this list who can confirm that multiple servers
> with the same disk mounted (NFS/NAS or SAN) can be load balanced where
> they potentially simultaneously write to the same repository without
> corrupting it?  This would be wonderful, but I'm not comfortable doing
> so without someone on the Subversion project saying it would work.
> Some clarification on Garrett Rooney's comments on that thread about
> dav related files or Tom Mornini's use of AoE/GFS would be helpful as
> well.  Maybe I'll have to reply to that post to get more information.
>
> Thanks,
> Justin
>
> ---------------------------------------------------------------------
> To unsubscribe, e-mail: users-unsubscribe@subversion.tigris.org
> For additional commands, e-mail: users-help@subversion.tigris.org
>
---------------------------------------------------------------------
To unsubscribe, e-mail: users-unsubscribe@subversion.tigris.org
For additional commands, e-mail: users-help@subversion.tigris.org