Takeover And File Recovery; The Effects Of Undoing Network Transactions - HP NonStop RDF J-series RVUs Management Manual

Takeover and File Recovery

When a takeover operation completes in an RDF network environment, the purger logs two
events: one reports a safe MAT position (indicating that all committed data up to that location
was successfully applied to the backup database), and the second (888 or 858) reports whether
or not a File Recovery position is available for use on the primary system. The RDF event 888
reports that a File Recovery is available and it includes the exact sno and RBA to be used for a
File Recovery operation on the primary system. If, however, "kept-commits" have been
encountered during phase 2 processing, a File Recovery position is not available; this is reported
in RDF event 858. This last situation will never occur in an RDF/ZLT environment because a File
Recovery position is always available with RDF/ZLT.
If an RDF event 888 is reported, then the specified File Recovery position is based on both phase
1 and phase 3 processing. Each system logs its own File Recovery position. While that position
can differ from one backup system to the next, the logged position for any single system is correct.
If you supply the returned File Recovery position to the TMF file recovery process on the primary
system, the process recovers the files on the primary database up to that point. If you use File
Recovery to a MAT position on all primary systems in the RDF network, in each case using the
returned File Recovery positions, then your primary distributed database will be consistent
across the RDF network.
You would use the File Recovery position with File Recovery in several situations: Assume you
have had an outage of your primary system, you have executed the RDF takeover operation on
your backup system, and you have resumed business transactions on your backup system.
Assume further that the former primary system has been repaired, it is back online, and you
want to switch your business transactions from the active backup database back to the former
primary database. To do so, you merely execute a planned RDF switchover from the backup to
the newly restored primary.
The problem with doing a planned switchover from backup to primary after an RDF takeover
operation is that some transactions might have committed on the primary system immediately
prior to the unplanned outage, and the outage brought down the extractor before it could send
that data to the backup system. In such a case, when you bring the primary system back up the
two databases are no longer synchronized because the primary database contains committed
transactions that are not in the backup database. Such transactions cannot be recovered.
In the past you would have had to synchronize your entire primary and backup databases. That
could be a lengthy task. Now you can simply use TMF file recovery to a MAT position. If you
execute this operation on your primary system using the MAT position specified in the RDF
event 888 message (see the description of message 888 in
primary database into the exact same state that the backup database was in upon completion of
the RDF takeover. Thus, after file recovery has completed, you can execute a normal planned
switchover from backup to primary.
NOTE: Due to the order transactions that commit on individual systems, file recovery might
not always be possible. If an 888 message is generated, however, it can be trusted.

The Effects of Undoing Network Transactions

Except with RDF/ZLT, phase 3 undo processing within an RDF network environment usually
results in other transactions being undone on every system in the network because the RDF
product is designed to make the safest, and most conservative, assumptions regarding all possible
interrelationships between transactions. This is best illustrated by example.
Consider an RDF network consisting of two RDF subsystem configurations (primary system \A
protected by backup system \X, and primary system \B protected by backup system \Y). Assume
that network transactions originate on both \A and \B, and that they update data on both \A
and \B. Assume further that each system also executes local, non-network, transactions.
300 Network Transactions
Appendix C (page 365)), it brings the