Re: Performance hit with trigger program

When I created a message broker that called exit programs to do various things, I used a named AG that was specifically for those exit programs. That way I could leave my files open, and if I needed to access one of those open files from another process it was an easy task to suspend the broker (which also closed down that named AG), do what I needed to do, then resume the broker. I suspect that my exit programs are something like your triggers. You would just need a way to shut them down for a moment while you do what you need to do with the files. Possibly using Alan Campin's Trigger Mediator found here: http://www.think400.dk/downloads.htm. Of course if you do shut down triggers, then they won't fire, and you will miss those transactions. You will have to take that into consideration.



Mark Murphy
STAR BASE Consulting, Inc.
mmurphy@xxxxxxxxxxxxxxx


-----Stuart Rowe <rowestu@xxxxxxxxx> wrote: -----
To: "RPG programming on the IBM i (AS/400 and iSeries)" <rpg400-l@xxxxxxxxxxxx>
From: Stuart Rowe <rowestu@xxxxxxxxx>
Date: 04/07/2016 02:04PM
Subject: Re: Performance hit with trigger program


THOU SHALT NEVER EVER USE THE *DFTACTGRP WITH AN ILE EXECUTABLE (oh I'm
gonna take it in the backside for that one...)

Considering that a trigger program *may* be executed in a multi-threaded
job, you also do not want to use *NEW because when that activation ends the
whole job will end.

IMO *CALLER is the best bet for a trigger handler. CRTBNDRPG DFTACTGRP(*NO)
ACTGRP(*CALLER) will do it.

If you do away with the file and use a queue instead, activation group is
of less concern but still don't ever let it end for a trigger program.

Stu


On Thu, Apr 7, 2016 at 12:57 PM, Dan <dan27649@xxxxxxxxx> wrote:

Jon, can you elaborate on this? I haven't explicitly defined AGs in
several years. The trigger program uses *DFTACTGRP.

Links or pointers to RTFMs are fine.

Thanks!
- Dan

On Thu, Apr 7, 2016 at 12:59 PM, Jon Paris <jon.paris@xxxxxxxxxxxxxx>
wrote:

You don&#8217;t say what AG you are using but that would also potentially have
an impact. It is the constant close/open of the file that is killing you
when setting LR on.

No need to set on LR each time but without it you need to decide how you
are going to close the file.


Jon Paris

www.partner400.com
www.SystemiDeveloper.com

On Apr 7, 2016, at 11:25 AM, Dan <dan27649@xxxxxxxxx> wrote:

I'm testing the trigger app I developed with help from some of you here
these past few days.

I ran an iteration of performance tests. A simple program that reads

(by

key) the OWNER table, updates a field, and updates the OWNER record,

inside

a FOR X = 1 TO 100000 loop. I ran this program 20 times; odd-numbered

runs

without a trigger, even-numbered runs with a trigger. The difference

was

significant. Without a trigger, the 10 runs averaged 3.4 seconds.

With

a

trigger, the 10 runs averaged 198.9 seconds. The 100,000 records is
representative of the batch job volumes they have here. The OWNER

table

is

updated in interactive programs as well.

Reviewing several examples of trigger programs online, I noticed that

all

of them turned on the LR indicator. For testing, I no-op'd the EVAL
*INLR=*ON so the program would compile, but never have LR turned on.
Duplicating the previous test cycle, without a trigger, the 10 runs
averaged 3.6 seconds. With a trigger, the 10 runs averaged 11.8

seconds.

Yeah, about 16 times faster than when turning LR on.

As you can tell by the program source I've included below, this is a

simple

app that basically acts like a journal, so there's no need for
initialization each time it's called. I was concerned that each of the
100,000 calls would put 100,000 instances of the trigger program in the
job's call stack, but that was not the case. The output file in the
trigger program remained open after my test run completed. I'm not

sure

whether this a concern, because there are about 200 interactive users

who

will be doing something throughout their day that updates the OWNER

table.

CAN ANYONE ADVISE WHETHER THERE ARE ANY PROBLEMS WITH THIS APPROACH?

Oh, and for those who will inevitably ask, I was told that we cannot

turn

on journaling. This is my second week here, and I don't have enough

clout

to delve into the why.

fTRG001H o e disk
d pi ExtPgm( 'TRG001R' )
d eTrgBuffer LikeDS( TrgBuffer )
d eTrgBufferLen Like( TrgBufferLen )
/copy TRGBUFCOPY
d TriggerPtr s *
d TriggerRecord e ds ExtName( OWNER )
d Based( TriggerPtr )
d Prefix( T_ )
d OutputRecord e ds ExtName( OWNER )

Select;
When eTrgBuffer.tbEvent = '1'; // Insert
TriggerPtr = %addr( eTrgBuffer ) + eTrgBuffer.tbNewOffset ;
TrgType = 'I';
When eTrgBuffer.tbEvent = '2'; // Delete
TriggerPtr = %addr( eTrgBuffer ) + eTrgBuffer.tbOldOffset ;
TrgType = 'D';
When eTrgBuffer.tbEvent = '3'; // Update
TriggerPtr = %addr( eTrgBuffer ) + eTrgBuffer.tbNewOffset ;
TrgType = 'C';
Endsl;

TrgTmStamp = %timestamp;

OutputRecord = TriggerRecord;
Write TRG001Hr;

If *inLR <> *inLR; // Test for performance by not turning LR

on

*inLR = *on;
Endif;
Return;

copy member TRGBUFCOPY:
d TrgBuffer ds
d tbFileName 10
d tbLibraryName 10
d tbMemberName 10
d tbEvent 1
d tbTime 1
d tbCommitLock 1
d tbFill01 3
d tbCCSID 10i 0
d tbRRN 10i 0
d tbFill02 10i 0
d tbOldOffset 10i 0
d tbOldLength 10i 0
d tbOldNullOff 10i 0
d tbOldNullLen 10i 0
d tbNewOffset 10i 0
d tbNewLength 10i 0
d tbNewNullOff 10i 0
d tbNewNullLen 10i 0
d tbBufChar 1 32767
d tbBufArry 1 Overlay( tbBufChar )
d Dim( %size( tbBufChar ))

d TrgBufferLen s 10i 0
--
This is the RPG programming on the IBM i (AS/400 and iSeries)

(RPG400-L)

mailing list

To post a message email: RPG400-L@xxxxxxxxxxxx
To subscribe, unsubscribe, or change list options,
visit: http://lists.midrange.com/mailman/listinfo/rpg400-l
or email: RPG400-L-request@xxxxxxxxxxxx
Before posting, please take a moment to review the archives
at http://archive.midrange.com/rpg400-l.

Please contact support@xxxxxxxxxxxx for any subscription related

questions.

--
This is the RPG programming on the IBM i (AS/400 and iSeries) (RPG400-L)
mailing list
To post a message email: RPG400-L@xxxxxxxxxxxx
To subscribe, unsubscribe, or change list options,
visit: http://lists.midrange.com/mailman/listinfo/rpg400-l
or email: RPG400-L-request@xxxxxxxxxxxx
Before posting, please take a moment to review the archives
at http://archive.midrange.com/rpg400-l.

Please contact support@xxxxxxxxxxxx for any subscription related
questions.

--
This is the RPG programming on the IBM i (AS/400 and iSeries) (RPG400-L)
mailing list
To post a message email: RPG400-L@xxxxxxxxxxxx
To subscribe, unsubscribe, or change list options,
visit: http://lists.midrange.com/mailman/listinfo/rpg400-l
or email: RPG400-L-request@xxxxxxxxxxxx
Before posting, please take a moment to review the archives
at http://archive.midrange.com/rpg400-l.

Please contact support@xxxxxxxxxxxx for any subscription related
questions.