IDP Cluster going forward (JDK 7)

[Paul Hethmon]

On 5/28/13 8:43 PM, "Cantor, Scott" <cantor.2 at osu.edu> wrote:

>On 5/28/13 8:38 PM, "Paul Hethmon" <paul.hethmon at clareitysecurity.com>
>wrote:
>
>>At the end of the day, I punted on it. Took the skeleton code from the
>>Infinispan version, gutted Infinispan, and plugged in a Hibernate based
>>DB
>>storage engine. I do intend to publish that one under the Apache license,
>>just haven't made the time to get it done. I have found this one pretty
>>robust in usage. It intentionally eats any DB errors since the result of
>>that to the user is having to login again. So having high availability as
>>well as session persistence were primary goals. I can take down every
>>node
>>in a cluster and then bring them back up again preserving all of the
>>session information.
>
>Is the database itself replicated? If not, how would you patch the
>database host without taking it all down?

At this point, I'm leaving that up to the DB administer to make sure the
DB is resilient.

In practice for us, it's not a major concern as we schedule maintenance
windows for DB related tasks with the expectation there is some down time
involved. Now that is typically at 3 AM local time to the users, so not
too many people to complain (though there are some).

Even with a single DB server, the storage service itself will default to
local in-memory storage if it can't talk to the DB server. For me, high
availability was the most important criteria. If I lose a session because
of DB connectivity, then I cause that user to have to login again. But
they can login again and access services instead of getting stopped
because the Infinispan cluster froze (which I saw way too frequently).

Paul