RE: Unpredicatable sequence for reading logical file on V5R3

[Evan Harris <spanner@xxxxxxxxxx>]

Hi Booth

The documented behaviour for restore of logical files (from the iSereis 
Information Centre) is that:

Unless you do something to change the restoration, the iSeries system:
   * Restores the logical file with the largest number of keys first
   * Does not build unnecessary access paths
Which can have interesting side effects when files are restored in a 
different order from that in which they are built. This phenomenon is 
described when talking about implicitly shared access paths. See

http://publib.boulder.ibm.com/iseries/v5r2/ic2924/index.htm?info/dbp/rbafomst111.htm

for an example or search on implicitly shared access path.

If this is the cause of the originally reported problem (and it sounds like 
the most likely candidate) then the problem will occur even if the files 
are restored onto the old system (as suggested elsewhere) as the issue 
manifests itself due to the restore operation.

Since this behaviour has been documented for some time the problem is not a 
restore or OS issue it is - as Neil recognised - a database design issue. 
Recreating the access paths in the order they were created should do the 
job, which equates to IBM's 'unless you do something...' clause from the 
documentation.

Regards
Evan Harris

> Is there any requirement that a restore actually make the files identical?
> Or, just that the files meet the specs.
>
> I wonder what would happen if you reorganized the file, as you said the
> desired order is key order, anyway.
>
> Which leads to the obvious remark....  The program has a k on the F-spec,
> right?  They haven;t just been lucky all this time?
>
> ---------------------------------
>
> Booth Martin