Hi Bill,

I've had a good look at the Ehcache/RMI replicated setup as documented on 
Confluence
and think that may be our best bet in the long term. I'll have a chat with UNE 
and follow up
to the list with any questions.

Thanks for all of the info.

regards,
Steve



On 21/01/2012, at 2:14 AM, William G. Thompson, Jr. wrote:

> On Fri, Jan 20, 2012 at 9:21 AM, Marvin Addison
> <[email protected]> wrote:
>>> This is caused by the service ticket being written to one node of the MySQL 
>>> cluster,
>>> but then the other CAS node happens to get to do the service ticket 
>>> validation,
>>> and it happens to use the other DB node, the ticket isn't found, because 
>>> database
>>> replication hasn't been fast enough to bring the ticket over.
>> 
>> I assume you're using traditional master-slave replication where no
>> amount of tweaks will _guarantee_ replication on a timescale that
>> would work in all cases.  I believe you MUST use multi-master MySQL
>> replication in order to make this work properly.  Bill Thompson
>> mentioned recently that he'd worked on such a setup and it worked
>> admirably, so perhaps he can weigh in here with more information.
>> 
> 
> The HA CAS setup we helped build for Lamar University is a multi-node
> multi-site active/active architecture.  The ticket registry is a
> multi-master MySQL with synchronous replication.  The solution
> actually included two CAS clusters...one in each data center with some
> fancy routing to make sure validate request ultimately land in the
> right cluster.  This solution is working well for Lamar.  They
> happened to be in hurricane alley and frequently use the alternate
> data center.  The decision to go with JPA and MySQL was mostly due to
> Lamar's experience and expertise with MySQL.
> 
> We've also helped University of New England (Australia) with a similar
> HA CAS setup, but based on Ehcache (w/ simple RMI replication, not
> Terracotta).  They have been running in production since January 2011
> and report that their 2 node Ehcache setup has been running
> "flawlessly".
> 
> I'd recommend seriously considering your HA requirements and the
> possibility of running a single node deployment, especially if you
> running on virtualized hardware with inherent redundancy and
> scalability.
> 
> If you are convinced you need multi-node deployment, I'd try out Ehcache.
> 
> Happy to go into further details with you.  I should also point out
> that Unicon of course provides consulting and support for CAS
> deployments if you're looking for more direct help.
> 
> Best,
> Bill
> 
> -- 
> You are currently subscribed to [email protected] as: 
> [email protected]
> To unsubscribe, change settings or access archives, see 
> http://www.ja-sig.org/wiki/display/JSG/cas-user
> 


-- 
You are currently subscribed to [email protected] as: 
[email protected]
To unsubscribe, change settings or access archives, see 
http://www.ja-sig.org/wiki/display/JSG/cas-user

Reply via email to