openjpa-users mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "C N Davies" <...@cndavies.com>
Subject RE: configure RemoteCommitProvider for Amazon EC2
Date Mon, 17 May 2010 12:38:33 GMT
Not quite answering your question but my issue was similar with proprietary
OMS.  Our DB is pretty large but not in clustered environment, bad design
decision made many years ago. So our DB will "go away" and "come back"
several times every day without warning, effectively leaving our application
without a DB for short periods. We do have other DBs that can support the
application but without a cluster our application really needs to reconnect
to a new DB when its current DB disappears. I implemented jgroups because it
means I can add new nodes on failure of one node, the new node will connect
to the new DB and the abandoned node can use an RPC connection to the new
node to tell it to take over. 

It's rough as guts but it's surprisingly reliable.

Chris

PS no comments on the design, I inherited it I didn't design it, I just got
it to 96% uptime with no budget and late nights.

-----Original Message-----
From: Oliver Krohne [mailto:mail@oliverkrohne.de] 
Sent: Monday, 17 May 2010 7:21 PM
To: users@openjpa.apache.org
Subject: configure RemoteCommitProvider for Amazon EC2

Hi,

we are going to deploy our app to Amazon EC2.
One issue is that the ip-adresses of an ec instance can change and new ec
instances
can be started which makes the RemoteCommitProvider using TCP unuseable as
all
TCP adresses where messages are sent to must be configured at the point the 
instance is starting. New instances can't be added without reconfiguring the
running ec instances.

Using JMS as transport would work but i don't want a another machine running
and hence a new point of failure.
But I would be thankful for any experience using JMS (openjms or activemq?).

So am thinking of using jgroups were new instances can join a channel, but I
did not found someone 
who has implemented it, maybe in this mailinglist?

The other solution would be using ehcache with jgroups replication and 
just replicate "removes and updates". Does anyone have experience with this
setup?


Thanks,
Oliver=


Mime
View raw message