Hi,
I picked this up from the archives, having just joined this list (I
work with Nico, and have been on holiday last week ),
Questions at the bottom if you want to skip the explanation, and
please do point us to the archive if you have spent time answering
this before.
In Sakai, we want to run Shindig using the OpenSocial API
implementation (and any other cool bits :) ), and wire the storage
through to the database provided by Sakai (MySQL or Oracle).
Sakai runs inside a Spring container, and has 1 jdbc datasource that
is available via Spring, and we would prefer not to open more
connections as some people in production may already be pushing the
connection limits of their backends (200-1000 front end threads, each
with 1 DB connection on each app server).... so ideally if we can
inject a jdbc datasource somewhere it would be great, but we can be
flexible if Shindig manages its own connections and doesnt do 1 per
request thread. (ie like Jackrabbit)
Nico will be able to expand on any other problems he has found since
I have been away.
----
Cluster,
Sakai normally runs in a sticky session app server cluster on 1 DB
backend, so if there is any caching in the JVM it needs to have some
form of cache invalidation of expiry (the front end Load Balancer
manages request routing to the correct app server), . Similarly, if
there is in session storage, it ideally needs to be small. I can give
you and example of the issues..... Apache Jackrabbit.....
In 1.2 it did heavy in JVM caching over multiple request threads, and
didnt have an out of JVM invalidation mechanism.
In 1.3.1 there was a ClusterNode journal that enabled a
transactionally aware cache invalidation .... and this worked for us
in Sakai.
Longer term, I want to get to mobile http sessions over the whole
cluster, but we need to minimize the session footprint first (we have
some big JSF components that use session storage :( )
So the questions:
Can you forsee:
Any problems with a managed datasource pointing to Oracle or MySQL ?
Any problem with a clustered sticky session app server cluster
connecting to a single DB ?
Anything else that might cause problems (eg version conflicts with
common libraries Hibernate, Spring etc)
Thanks,
Ian
On Thu, Mar 27, 2008 at 7:59 AM, <[EMAIL PROTECTED]>
wrote:
> Hello,
>
> Let me give this a little more context. We are experimenting with
> getting Shinding into our Virtual Learning Environment (Sakai).
What specific areas are you having trouble with?
>
>
> Another question we have is whether anyone has ever tried/succeeded
> running
> Shindig in a clustered environment (we run 6 clustered servers for
now)
You'll have to be more specific here as well. Most large scale
deployments
now are most likely (from my own experience) running multiple
instances of
shindig with some load balancer to route traffic. What sort of
integration
are you trying to do?
--
~Kevin
Ian
