NonStop SQL/MP Reference Manual

NonStop SQL/MP Reference Manual—142115
A-41
Considerations—ALTER TABLE
The duration of a WITH SHARED ACCESS operation increases with the
number and length of transactions on the node that contains the source partition,
particularly with the number and length of transactions that involve the source
partition and the amount of activity on the audit trail used for the source
partition.
°
TMF audit trail requirements
An operation that uses WITH SHARED ACCESS cannot complete successfully
unless the TMF audit trail generated during the operation is available for reading
later in the operation. If a required audit trail has been overwritten, a WITH
SHARED ACCESS operation cancels changes made to the database and
terminates.
When performed on a source object that has a valid TMF online dump, an
operation that uses WITH SHARED ACCESS generates audit information for
the target object.
Lengthy operations that use WITH SHARED ACCESS might require an
operator to mount tapes of TMF audit dumps. (Requests to mount TMF audit
dump tapes for WITH SHARED ACCESS operations are not distinguishable
from other requests to mount TMF audit dump tapes. Such requests are
generally sent to an operator's console. SQL does not return information about
such requests to the terminal or process that started the operation.)
°
Completing the ALTER TABLE request
When a split operation with the WITH SHARED ACCESS option finishes
successfully, check SQL FILEINFO for the source partition to see if the F flag is
present. For a merge operation with the WITH SHARED ACCESS OPTION,
check the target partition; for a one-way move operation with the WITH
SHARED ACCESS option, check the source and target partitions. If the F flag
is present, the file contains data blocks allocated to obsolete (moved) records.
Use the FUP RELOAD command to reclaim the disk space. For more
information, see the File Utility Program (FUP) Reference Manual.
If ALTER TABLE fails, the original table normally remains intact and
accessible. However, if ALTER TABLE fails because of a CPU failure or
system crash, a newly added, moved, or split partition of the table might
continue to exist—along with the original table—even though it is inaccessible.
After the system becomes available, use CLEANUP to drop the new partition
(or ask the local super ID to do so), then reissue the ALTER TABLE statement.
ALTER TABLE returns an error if there is a problem with the table.
When you add a partition to a table, the PARTNS catalog table and associated
IXPART01 index might become full. To correct the situation, distribute object
and partition definitions across multiple catalogs. For more information about
partition limits and the PARTNS table, see the Limits
on page L-5 entry.
If the SLACK space in the source file is less than the value chosen for the target
file, a MOVE operation can fail with a file full error. To prevent this situation,
check the actual slack amount in the source file (using the FILEINFO
STATISTICS command) and specify EXTENTS and MAXEXTENTS values
for the target file that are sufficient to hold the data.