RDF/IMP and IMPX System Management Manual (RDF 1.4+)

Introducing RDF
HP NonStop RDF/IMP and IMPX System Management Manual—524388-001
1-5
Unplanned Outages
In the example illustrated in Table 1-1, a disaster has brought down the primary system
immediately after the commit record for transaction 100 was written to the MAT, but
before the RDF extractor process was able to send the commit record to the backup
system. For transaction 101, a single update was logged in the MAT and sent to the
backup system, but the primary system was brought down before the transaction was
completed.
When the command for a takeover is issued, the updater processes treat all
transactions whose outcomes are not known as aborted transactions. In this scenario,
only the changes related to transactions known with certainty to have been committed
on the primary system are left in the backup database. Therefore, in the example
illustrated in Table 1-1, the audit information associated with transactions 100 and 101
is backed out of the backup database.
Typically, the extractor process sends audit information to the backup system within a
second after it has been written to the MAT on the primary system, so a minimum
number of transactions are lost when a disaster brings down the primary system.
Tips for Executing Fast Business Takeover Operations
The following discussion provides tips on how you can execute an RDF takeover and
resume business activities on your backup system in the shortest time. These tips
may not work for everyone, but you should consider them to see if they might work for
you.
Take online dumps of your backup database as frequently as you take them on your
primary system. In this way, when you need to takeover on your backup system, you
will already have dumps available. Remember, if the RDF UPDATEROPEN attribute
is set to PROTECTED, you must stop the updaters and set it to SHARED before taking
online dumps of your backup database.
If your database is SQL and if your applications form the typical requestor-server
environment where the requestors send requests into your primary system from other
locations, then you could start your servers on your primary and backup systems. You
must take care that you only route requestor work to your primary system. This leaves
your servers running essentially in standby mode on your backup system.
When you lose your primary system because of some unplanned outage, you execute
the RDF Takeover operation. This brings your backup database into a consistent
state, typically within a small number of seconds.
TRANS101—Update 1 TRANS101—Update 1
TRANS100—Commit record
(Primary system fails)
Table 1-1. Audit Information At the Time of a Primary System Failure
Primary database updates
(Sequence in master audit trail file)
Updates sent to the backup
(Sequence in image trail file)