One of the features discussed around the new Replication features in 11.7 is that it allows you to do maintenance on the primary. Now this is useful if the maintenance is patching up the OS, say, but I know in older versions there's very little you can do to the actual database in terms of maintenance without breaking replication. So, has this list of activities changed with 11.7? Or would you still have to fail back to the primary after the maintenance window and then reseed replication? Because this introduces a window with a single point of failure again.
The list of database utilities that can be run on replication enabled database has not be changed. However, we do put some improvement in 11.7. For example, builddb used to break replication, but now it won't. We understand there are still quite a few other utilities that should be improved, for example, idxbuild, truncate bi etc. And they are on our to-do list.
The failover transition is also more reliable and powerful than previous releases. With the concept of "replication set" introduced in 11.7, If the maintenance is for operations like patching the OS, then replication will not be interrupted, and a single point of failure can also be avoided. The "replication set" involves one source and two targets, and they will be transitioned together. So even if the source is lost, there will still be one (new) source and one target (of the new source) left after transition. This can buy us more time to reseed the second target and reduce the window down to single point of failure.
Dapeng
Thanks for the detailed response Dapeng.
After I transitioned from serverA Source to serverB Target, the former target serverB becomes the new Source.
Can serverA become new Target without backup/restore ?
And subsequently can I transition from new Source serverB to new Target serverA?
Do I need replication set ( 3 databases ) to achieve the above or it could be done with just two?
I think you're talking about the failover/failback scenario:
The source and target can be configured so that after failover transition, the former target becomes the new source; and the former source becomes the new target of the new source automatically. There is no need to create another target to restore the original one source, one target configuration.
To failback to the original server A for the source, another failover transition will be needed.
The failover transition is basically a role switch without breaking replication. This can be done with just two databases.
Replication set will allow the second target transition together with the source and the first target. So after failover, you'll still have one source and two targets without re-bases.
Dapeng
Yes, I am talking about failback scenario.
With just two databases.
In the previous releases of OE we need to re-base the original database to failback.
Do I understand correctly that in 11.7 failback could be done without a re-base, correct?
I have 1.3 TB db and re-base is 24hr+ process. So it is a big deal for us.
I heard there were problems to failover/failback in previous releases, although it "should" work. Could you post more details about the errors you have encountered? For example, steps, transition properties etc.
Thanks,
Dapeng
I am not using OE Replication, not yet. With after-image replication we could not failback without a re-base.
One of the main selling points of OE Replication 11.7 (for us) is that we should be able to switch roles source and target without a re-base. That is what we are looking forward. As to the previous versions I am not really concerned about as I installed 11.7 yesterday.
After I execute "dsrutil -C transition failover" command to switch roles does the former source immediately becomes a target without any other actions? (Yes/No)
Was that the same before version 11.7?