Re: Sysmaster - sysprofile - size of values
A monitoring tool reported negative values from sysmaster because counters (e.g. bufreads) exceeded 2 billion, since sysmaster exposes them as 4-byte integers while the engine tracks larger values; the poster's workaround was periodic 'onstat -z'. Jonathan Leffler first said the counters are 8-byte in IDS 10, but another poster showed that on 10.00.TC3ET sysrstcb/syssesprof columns such as upf_bufreads are still INTEGER. Leffler conceded the correction and explained why (sysmaster schema isn't changed after initial release, plus INTEGER8's odd 10-byte internal representation). No fix is recorded beyond resetting counters.
Auto-generated by DrWatson from the posts below — may be imperfect; read the full thread.
mweallans@panacea.co.uk wrote:
> I have recently encountered some interesting problems with my remote
> monitoring software. My software is heavily reliant on the sysmaster
> database and I have observed that many figures show up as negative in
> sysmaster because they are in fact >2billion. The reason is quite
> obvious - sysmaster uses a 4 byte integer whereas the internal values
> are held as 8 byte integers.
>
> Can anyone who has Version 10 runing reassure me that this problem is
> fixed in IDS version 10?
>
> I normally workaround this problem with a periodic onstat -z but on one
> site I have a bufread rate of > 2billion in an 8 hour period. And that
> messes up caching rate calculations etc.
That's quite a read rate...
Yes, the counters are 8-byte quantities in version 10.
--
Jonathan Leffler #include <disclaimer.h>
Email: jleffler@earthlink.net, jleffler@us.ibm.com
Guardian of DBD::Informix v2005.02 -- http://dbi.perl.org/
Well internally they are but on IDS 10.00.TC3ET. sysrstcb (counters for syssesprof) still has upf_bufreads integer, not an int8!
david@smooth1.co.uk wrote: > Well internally they are but on IDS 10.00.TC3ET. > > sysrstcb (counters for syssesprof) still has > > upf_bufreads integer, > > not an int8! Yes - I misspoke. I was going to send out a correction, but I wanted to get my ducks lined up, but you got them lined up quicker. Thanks. I think I know why they weren't made into INTEGER8; in fact, I think I can think of two reasons why that was the case, but only one of them sort of counts. I do not like asymmetry, and this is a piece of wanton asymmetry. Given the constraint that we do not change the schema of sysmaster after the initial (UC1/FC1/TC1) release of the server, that is one legitimate reason for not changing the sysmaster in a 9.40 fix pack; I'm not so convinced about 10.00.UC1. The fact that IDS uses an oddball 10-byte structure rather than 8-byte integers to represent INTEGER8 is the other, less acceptable reason. However, even that is barely an excuse - the counters could be manipulated correctly if they were stored as the 10-byte structures with only a little more work than if they were actually 8-byte integers. Having said that, even a 'little more work' in the main loops is not entirely desirable. -- Jonathan Leffler #include <disclaimer.h> Email: jleffler@earthlink.net, jleffler@us.ibm.com Guardian of DBD::Informix v2005.02 -- http://dbi.perl.org/
Related threads
- IDS not writing to online.log
- Help!!! syntax error
- installclientsdk bug?
- RamDisk tempdbs boot script for Linux