RDF System Management Manual

Managing RDF
HP NonStop RDF System Management Manual—524388-003
5-29
Access to Backup Databases in a Consistent State
If the specified timestamp is not at least five minutes greater than the current time,
RDFCOM aborts the command and displays the following message:
The specified timestamp must be at least five minutes greater
than the current time.
The STOP UPDATE command itself is logged to the EMS event log under the general
RDF message 835. As each updater shuts down because it has reached a
commit/abort record generated after the specified timestamp, it logs the following
shutdown message to the EMS event log:
785 Redo pass ending on reaching timestamp timestamp.
Because some audit data will not have been applied, the updaters must reexamine that
audit data when they restart. Consequently, when restarting the updaters after a STOP
UPDATE to a timestamp, the initial RTD times of the updaters will be greater than the
duration of time they were actually stopped.
The example that follows illustrates the effect of a STOP UPDATE, TIMESTAMP
command. In the example, tn indicates a transid, and the timestamp below reflects the
time of most recent commit or abort record in the audit trail.
RDFCOM; STOP UPDATE, TIMESTAMP 20JUN2004 12:00
commit update update update update update update commit update update commit commit
t0 t1 t2 t3 t1 t2 t3 t1 t2 t3 t2 t3
11:50 11:50 11:50 11:50 11:50 11:50 11:50 11:59 11:59 11:59 12:01 12:02
The updater applies the updates associated with t1 because that transaction
committed before the specified timestamp. It does not, however, apply the updates
associated with t2 and t3 because those transactions did not commit until after the
specified timestamp. The updater does not actually shut down until it encounters the
commit record for t2 (12:01). As soon as it encounters a commit or abort record whose
timestamp is greater than the specified timestamp, the updater will have processed all
audit data for transactions that committed or aborted up to the specified timestamp.
Thus, the database is in a completely consistent state with regard to transaction
boundaries.
The updater keeps track of the first record it could not apply because its transaction
committed or aborted after the specified timestamp. In the above example, the
updater’s restart position in the image trail will be the first update record for t2. Thus,
when restarted, the updater will backtrack to that record. This practice ensures that no
audit data is lost during the STOP UPDATE to a timestamp operation.
If you erroneously set the timestamp too far into the future (for example, 26DEC3000),
the only way to correct this mistake is to enter a STOP RDF command, restart RDF,
and reenter the STOP UPDATE command with the correct timestamp.
See also the description of the STOP UPDATE command in section 8.