Re: Resurrecting a failed node (Oracle MM replication)

i have been involved in a similar recovery (with two instances having multi-master replication). what we did was:

• create the destroyed database completely new (no recovery at all)
• exported all data from an instance that survived and imported it in the crashed instand
• reesteblish the replication

of cause our way required exclusive access to ALL instances during this period (our crash occurred on a friday, so we could do it during the weekend)

if the update volume is not too big there would be an alternative to recover as you describe and then find the difference for each table with a helper procedure, which is provided in one of the packages (DBMS_DEFER i think) this will only work for small tables and update volumes, since this function is very slow.

the last alternative i see is recover all instance to the same point in time (and loose all data until 1 hour before earthcrake)

olaf

jim_nash_at_yahoo.com (jim nash) wrote in message news:<6ef12c.0208121418.735e3ee_at_posting.google.com>...
> Consider a scenario where there are replicated Oracle databases at 3
> network nodes, say san francisco, new york and london, operating as
> peers in a Multi-Master replication environment. Say timestamp based
> conflict resolution is used for essentially all conflicts.
>
> Now say there is a disaster at San Francisco (you know those 'quakes),
> and a redo log recovery is required, which brings the SF database back
> to 1 hour prior to the point of the failure. That is, the last hour's
> worth of updates aplied to the SF node database are lost.
>
> My question is how can/does the SF node database get back in sync with
> the database at the other two nodes?
>
> Thanks,
>
> - Jim
Received on Wed Aug 14 2002 - 03:54:13 CDT