Guillaume Smet a écrit :
Hi Gérard,
On 1/24/07, Gérard BUNEL <[EMAIL PROTECTED]> wrote:
With the following configuration: 2
controllers having each 1 Virtualdatabase supported by 2 backend
(RAIDB1).
- When All backend of a controller are failed: My understanding is
that Failover mechanism is used and requests should be redirected to
the second controller.
Is my understanding correct ? And if yes is the driver responsible of
this redirection ?
Yes, it's correct but the JDBC driver is only responsible for the load
balancing/failover between the two controllers so it's used for
controller failures not backend failures. Backend failures are
directly managed by the controllers.
I made a test a few weeks ago with only the controller with no backend
in the JDBC URL used by the driver and the queries were directly
rerouted to the controller with active backends from the controller
with only failed backends.
- Let's imagine that we stop the controller
holding failed backends. What happens if we restart this controller and
if its configuration automatically loads the virtualdatabase
configuration. Are the backends considered as disabled ?
Yes, they should be considered disabled as they are not in a consistent
state.
Sequoia tries to enable them if you set autoEnableBackends to true but
won't do it if the backends are not consistent. Sequoia needs to have
a checkpoint for the backend to know which queries of the recovery log
it needs to replay. It won't have one if the backend failed.
For example, if you disable a backend manually, Sequoia will register
a checkpoint in the recovery log so when you reenable it, it will be
able to replay the recovery log from the checkpoint.
- If None of the backend were failed, but
instead we have a controller failure. What happens when restarting the
controller having its configuration automatically loading the
virtualdatabase configuration ?
Controller failed -> backends are not safe anymore so you should
have
the same behaviour than above.
So In the computer shutdown procedure, we should include
virtualdatabase and controller shutdown, Yes ?
My understanding is that if we do not do this, at startup, backends
whould be considered disabled.
I've another question about backup/restore of database.
We have a system that builds ghost images of computer disk, so that in
case of complete disk failure we can restore quickly a complete
configuration.
So when this happen on one controller, I can use the second controller
to synchronize databases.
But now, in the unprobable case where we are really unlucky, where the
tow computers holding the controllers crash simultaneously.
can we restore databases with a daily dump ?
I understand that a Sequoia dump not only backup database but also
records checkpoint in the RecoveryLog. so If we start with an old
configuration (ghosts are built monthly)
Is the dump produced the day before the crach useable as a simple
database backup, even if we do need to re-initialize the cluster ? or
do we also need to do dayly database backup (using MySQL backup for
instance as my backends are MySQL bases) ?
Gerard
--
Guillaume
_______________________________________________
Sequoia mailing list
[email protected]
https://forge.continuent.org/mailman/listinfo/sequoia
--
ALTRAN OUEST
Gérard
BUNEL
Chef
de Projet
____________________________________________________________________
Technopôle
Brest Iroise
Site du Vernis – CS 23866
29238
Brest Cedex 3
Tél : + 33 2 98 05 43 21
Fax : + 33 2 98 05 20 34
www.altran.com
|
_______________________________________________
Sequoia mailing list
[email protected]
https://forge.continuent.org/mailman/listinfo/sequoia
