RDF System Management Manual

Network Transactions
HP NonStop RDF System Management Manual—524388-003
13-12
Takeover and the RETAINCOUNT Value
If we return to the issue of T
13
in the example further above, note that the commit
sequence differs on \A and \B. When the purger on \A determines that T
12
, T
14
, and
T
15
must be undone, it also determines that the results of T
13
can be kept intact
because T
13
had to have completed on \B before T
12
. Why? The commit records on
\A guarantee that both T
12
and T
13
did indeed commit. Therefore, although the
commit record for T
12
is missing from \B, the commit record for T
13
is present. This
guarantees that T
13
committed prior to T
12
, and that the results of T
13
can be kept
intact on both nodes.
When a purger determines that it can keep the results of a transaction even though
that transaction follows one that must be undone because data for it is missing
elsewhere, the purger logs an RDF 823 event message identifying the particular
transaction.
Takeover and the RETAINCOUNT Value
In order for all systems in an RDF network to execute phase 3 takeover processing
correctly, you must ensure that all image data potentially needed for undo is available
on each system. The way to achieve that is to set the purger’s RETAINCOUNT to an
acceptable value on each system. For a complete discussion about this attribute and
how it works, refer to Section 10, Triple Contingency. (The same RETAINCOUNT
guidelines that apply to a triple contingency environment also apply to an RDF network
environment.)
Note that if you have not set the RETAINCOUNT properly and image files have been
purged that are subsequently needed for phase 3 takeover processing, the takeover
operations on the systems where the image data is missing might fail. In such a case,
your entire distributed database across all RDF backup systems could be
compromised with inconsistent data.
Note. If you have local transactions that do not touch data involved in network transaction
activity, and you do not want these local transactions undone just because data might be
missing for the network transactions during a takeover operation, you are advised to configure
separate RDF subsystems: one to protect just the data involved in network transaction activity,
and the second to protect the non-network data. Of course, both sets of data must have no
dependencies on the other.