SQL/MP Installation and Management Guide

Performing Recovery Operations
HP NonStop SQL/MP Installation and Management Guide—523353-004
11-12
Database Recovery After a Disk or Node (System)
Failure
Database Recovery After a Disk or Node (System) Failure
When a disk or node (system) fails, often SQL catalogs tables and database files on
the disk or node are left in a crash-open state. To recover the database, both the
catalogs and the files must be recovered to a consistent state.
Depending upon the situation, choose an appropriate method from this list to achieve
the desired result:
•
Use the file recovery method to recover the database, starting with online dumps
(files containing copies of consistent catalogs and objects saved by the TMF
DUMP FILES command). The file recovery function starts with the saved files and
updates transactions to the last consistent point in the audit trails.
•
Use the file recovery method with a specified TIME option to recover a database to
a given consistent time, as described under File Recovery With the TIME Option
on page 11-14.
•
Use the file recovery method to recover files that cannot be recovered by volume
recovery because the audit trails are missing or damaged in some way. In some
cases, the damage could also prevent file recovery to the most recent point.
Volume Recovery
TMF volume recovery is invoked automatically by the TMF commands START TMF,
and is invoked as needed thereafter when a volume becomes accessible. Volume
recovery uses the audit trails to roll back incomplete transactions and return the
database to the last consistent state.
Volume recovery might fail to recover a volume or a file. Some of the recoverable
cases follow:
•
A volume becomes unavailable during the volume recovery operation. When you
bring up the volume, TMF automatically restarts volume recovery to the last
recovery point in the database.
•
A file is corrupted or inconsistent in such a way that volume recovery cannot apply
the audit trail information. If volume recovery fails to recover a file, FILEINFO
displays setting of the REDONEEDED and UNDONEEDED flags. For tables,
indexes, and Enscribe files, the information appears after the modification
timestamp of the table. For views, the information appears after the open states
LABEL QUESTIONABLE and DEFINITION INVALID if they appear in the display.
Normally, volume recovery recovers such files when the volume is started for
transaction processing. If, however, the volume is already started and the file is still
marked with REDONEEDED or UNDONEEDED, you must recover the file by using
file recovery.