Jon,
I maybe didn't explain correctly my problematic issues... What we need
is the simplest arch to hold a database system with no single point of
failure, but this database (composed by two backends in the simplest
form) should look as only one to our apps. So what I think is that we
MUST use two controllers, and I was looking for the correct
implementation. As I see, two controllers could use a singleDB each (I
could not start this config... I will review this case if it's the
correct for us...). Is that ok?
No, SingleDB can only be used with one controller without replication.
If you need replication, you have to use RAIDb-1 with a <Distribution>
element in the virtual database configuration file so that the
controllers are synchronized. Each controller can have a single database
backend but the virtual database will have a total of 2 replicated
backends that will appear as a single database to the client application.
Then the question should be, if these two database backends, each
controlled by a different controller, are indeed replicated (we can
recover one from the other in case of failure), the database system is
fault tolerant with no single point of failure, and functionally
perfom as a single database to the application... Is this Ok?
Yes, it is. You can look at the distributed RAIDb-1 demo that ships with
Sequoia. You can change the config to have a single backend per
controller and that should be similar to what you need to build.
I will use another small database in each controller for the
RecoveryLog...
Thanks for your interest in Sequoia,
Emmanuel
Emmanuel Cecchet escribió:
Hi Jon,
I'm new to the list and to sequoia, and so I'm trying to understand
and realize the capabilities of this product.
Welcome and thanks for your interest.
We are trying to find a database architecture as the backend for our
Java apps, with the only requisite of four nine availability
(99,99). The load will always be low. My view is from the
infrastructure point (I know that I should pass this problem to our
dev part, I've done it indeed...), and that's why I want to build a
base platform for deploying different client solutions in our
virtualized datacenter: My needs should be just two backend
databases ("mirrored" ideally) controlled by two controllers
replicating everything.
If you need only 2 database backends, they can also be under the same
controller. The controller becomes a single point of failure but this
can be fine if it is co-located with the application and the
application is not clustered. Now if your application is also
replicated, you can have 2 controllers with 1 db each.
I was looking at Sequoia+Postgre as the platform, and have been
reading all the documentation it seems to fit well. I have some doubts:
1) Which should be my simplest configuration? RADIb-1 in each
controller would give me 4 backends... That is supposing that two
controllers cannot control the same two backends, is that allright?
RAIDb-1 is for mirroring so you are on the right path. You are right
that backends cannot be shared between controllers. But one
controller can have multiple backends. So I am not sure if you just
need to replicate 2 databases on different machines which will lead
to 4 backends (2 for each controller as you mentioned) or just 1
database in which case 1 backend per controller is enough.
2) I tried using simpleDB in each controller, but I got errors
saying that no distribution can be done with simpleDB... Is that
allright?
SimpleDB is only meant for 1 database backend only (cluster-wide). If
you replicate a database, you have 2 backends in the cluster and you
need RAIDb-1.
3) In the configs I'm proposing, where should the RecoveryLogs be
stored? In the same backends (maybe only in one of the two?) that
compose the DBclusters?
I usually recommend to have the recovery log local to each
controller. Given the small workload that you will have, a Java
database like Derby or Hsqldb can do it on the controller.
Hosting the recovery log on a database backend is error prone for
administration and when you need to service the database node you
will have to stop your controller that will not be able to work
without its recovery log.
I hope this helps,
Emmanuel
--
Emmanuel Cecchet
FTO @ Frog Thinker
Open Source Development & Consulting
--
Web: http://www.frogthinker.org
email: [EMAIL PROTECTED]
Skype: emmanuel_cecchet
_______________________________________________
Sequoia mailing list
[email protected]
https://forge.continuent.org/mailman/listinfo/sequoia
