Hi Rafael,
If I understood the thread correctly I think that your configuration is
wrong. It looks like you have each controller on each site talking to
both database instead of having each controller to talk to its local
database and controllers talking to each other using the group
communication (Distribution element in the virtual database
configuration file).
The scenario you describe with an unreliable link in the middle will not
work with multi-master replication (all databases consistent at all
times). It looks like what you need is some asynchronous replication
like MySQL replication that does master/slave.
Right now there is no load balancer that allows you to reject a query if
a node failed but the others succeeded. Having an all success/all failed
kind of policy would also require all nodes to be enabled or disabled
simultaneously for that policy to make sense. Right now a node is
automatically disabled when a failure is detected. Rejecting a query
would mean that the system would reject all queries as soon as a node
has failed (at least for writes, reads might still be allowed to go
through to other nodes). This looks like a strong assumption and I don't
know which setups could really benefit from such configuration.
What do you think?
Emmanuel
Thanks Martin
Do you know if is possible to configure the RequestManager to not
execute write statements on backends if one of them fails?
if possible my problem will solved
2008/7/24 Martin Lyness <[EMAIL PROTECTED]
<mailto:[EMAIL PROTECTED]>>:
I think you are describing the Split-Brain scenario where the two
controllers are still operational but they don't see each other,
is this correct? I'm not the sequoia expert but I figure I would
shoot you a quick email to help while you wait on the experts J.
As far as I know this is the proper functionality (i.e. not using
an updating queue), and you should monitor this link so that you
can observe when failure occurs between the controllers and
perform recovery steps accordingly.
It would be interesting to automatically handle this type of
failure, but how would you know that it is a simple network
communication failure versus something more serious and if there
is possible data corruption on the other database.
But like I said I'm not the expert so I'm looking forward to
hearing their responses.
Martin Dale Lyness
MagnetStreet
[EMAIL PROTECTED] <mailto:[EMAIL PROTECTED]>
Office: (763) 450-7503
Cell: (715) 630-0296
------------------------------------------------------------------------
*From:* [EMAIL PROTECTED]
<mailto:[EMAIL PROTECTED]>
[mailto:[EMAIL PROTECTED]
<mailto:[EMAIL PROTECTED]>] *On Behalf Of
*Rafael Rossignol Felipe
*Sent:* Thursday, July 24, 2008 6:45 AM
*To:* [email protected]
<mailto:[email protected]>
*Subject:* [Sequoia] Structural ideas
Hi everybody,
I´m newbie on Sequoia, but i've read lots of documentation, and
made some tests.
I have an Application in two subsidiaries of the same company. In
both subsidiaries i have a jBoss with One application that access
a local database.
something like this:
http://br.geocities.com/rafaelrfelipe/today.png
The problem is: the data must be distributed!
If an insert is executed at subsidiary 1, a select executed in
subsidiary 2 must retrieve all data (includes this specific insert)
Well, i´ve chose use Sequoia to help me make it work.
I made a test configuring one controller with two backends (each
one pointing to a one subsidiary), so i executed a update
command. The update was successful and the data was updated on two
oracles databases.
But there´s a problem, the preferential link is unstable, so i
unplug the net cable from one database server to see the failover
algorithm in action (I though there was a queue of statements that
wait the link become available to start to update after server is
back). So i discovered that if a backend is down, sequoia does not
updates the data after it´s back.
I don´t know if I've configured something wrong or if this feature
do not exist at all (a queue of update statements to be executed
after backend returns of a crash).
Is this feature possible?
I used Raidb 1 algorithm
I think that my final scenery it´s some like this:
http://br.geocities.com/rafaelrfelipe/idea.png
Is this scenery possible?
Thanks for any response
Rafael
--
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
