Re: Analyzing RUNSQLSTMT jobs

Sadly, still stuck at 7.2.

And not sure this would help. It's not a query running for 8 hours, it is
a procedure with SQL that instead of taking a split second to complete its
task like normal, it is taking several seconds. Called many 1000s of
times, adds up to the job taking 8 hours instead of 2 minutes. But maybe
it could find something like that?


On Tue, Jul 13, 2021 at 11:38 AM Timothy P Clark <timclark@xxxxxxxxxx>
wrote:

I could have not set this up better if I were paying you... And now that
you've pushed the soapbox conveniently into place, I'll take just a minute
to get up on it and say that scenarios like this are *exactly* the reason
that IBM delivered the Query Supervisor in the most recent TRs for IBM i
7.3 and 7.4.

With QS, you can set up a threshold for query elapsed time (e.g. 6 hours)
and then have an exit program that dumps any information about the query
or the job you may be interested in. For example, I would start by dumping
the cached query plan for the query and then comparing it to a fast
running plan. Did the plan change? Was an index advised? You could also
dump the job stack to show whether you may be waiting for a resource. Or
start job watcher for the job or dump Mimix status or...

In fact, you can probably set this up in under 15 minutes...

CALL QSYS2.ADD_QUERY_THRESHOLD(
THRESHOLD_NAME=>'6 hour queries',
THRESHOLD_TYPE=>'ELAPSED TIME',
THRESHOLD_VALUE=>6*60*60,
LONG_COMMENT=>'Detects queries that run for more than 6 hours');

Then go to

https://www.ibm.com/docs/en/i/7.4?topic=programs-exit-program-dump-plan-cache-information-query
and pick your favorite programming language. Copy and paste the sample
program and build instructions.

Next time a query runs for 6 hours, you'll have a plan cache snapshot in
the SUPERVISOR library containing only the query/queries that ran long.
You can use the SQL performance center in ACS to import the snapshot and
then Visual Explain the query.

And if you want to automatically kill the query, there is sample code for
that, too:
https://www.ibm.com/docs/en/i/7.4?topic=programs-exit-program-end-query

Hope that helps!

Tim Clark
DB2 for IBM i / SQL Optimizer

"MIDRANGE-L" <midrange-l-bounces@xxxxxxxxxxxxxxxxxx> wrote on 07/13/2021
08:56:26 AM:

From: "Rick Rauterkus" <rick.a.rauterkus@xxxxxxxxxx>
To: "Midrange Systems Technical Discussion"

<midrange-l@xxxxxxxxxxxxxxxxxx>,

Date: 07/13/2021 08:57 AM
Subject: Re: [EXTERNAL] Re: Analyzing RUNSQLSTMT jobs
Sent by: "MIDRANGE-L" <midrange-l-bounces@xxxxxxxxxxxxxxxxxx>

We have had something very similar happen 3 or 4 times over the past

year.

Something that runs every day and takes about 2 minutes to complete (an
inventory report to a customer), will suddenly one day take 8 hours to
run. The job runs, and makes steady progress, but just really slowly.
Looking at the call stack during a slow run, it shows the same spot

almost

every time, a procedure in a service program that is using embedded SQL

to

extract some data from a file. This procedure is called for every item.
During the last long run, we killed the job, and then immediately
resubmitted it, and it ran in its normal 2 minutes.

The only theory we could muster up is that for some reason when the

first

call to that procedure happened, the access path that the SQL normally

used

was not available for some reason, so it ended up with a very much less
efficient one. And was stuck with it for the duration of the job. As

to

why it wouldn't be available, maybe the system was rebuilding the index?
Maybe Mimix had it locked for whatever it was doing? We have found

nothing

to explain this either. We are considering changing the embedded SQL to
RPG and see if the issue ever comes up again.


On Tue, Jul 13, 2021 at 7:58 AM Alan Shore via MIDRANGE-L <
midrange-l@xxxxxxxxxxxxxxxxxx> wrote:

Hi everyone
Here is my latest quirk/conundrum
I ran the reorgs this past weekend and the SQL query that was causing

me

heartache ran EXTRMELY quick on Sunday night, with no problems
However, the run on Monday night was STILL running in batch Tuesday
morning – 11 hours later
Now heres the next kicker
I took a copy of the query that is being run, and changed it to create

a

file in a different library and ran it interactively
It finished in under 2 minutes
I have NO idea what to look at or even where to start
Does anyone have ANY ideas or hunches
As always – ALL answers gratefully accepted

Alan Shore
Solutions Architect
IT Supply Chain Execution

[cid:image001.png@01D777C5.3A4B1220]

60 Orville Drive
Bohemia, NY 11716
Phone [O] : (631) 200-5019
Phone [C] : (631) 880-8640
E-mail : ASHORE@xxxxxxxxxxxxxxxxxxxx

‘If you're going through hell, keep going.’
Winston Churchill

From: MIDRANGE-L [mailto:midrange-l-bounces@xxxxxxxxxxxxxxxxxx] On

Behalf

Of Alan Shore via MIDRANGE-L
Sent: Tuesday, July 6, 2021 11:57 AM
To: Midrange Systems Technical Discussion

<midrange-l@xxxxxxxxxxxxxxxxxx>

Cc: Alan Shore <ashore@xxxxxxxx>
Subject: RE: [EXTERNAL] Re: Analyzing RUNSQLSTMT jobs

Thanks for your reply Mark

Re-use deleted records is NOT set on the files that I am looking at. I
would DEFINITELY need to perform a RE-ORG on 3 files in particular

THEN

change them to re-use deleted records



Heres another question - hopefully you know the answer.

On at least 2 of the 3 files, there are some join logicals of BOTH of
these 3 files

I was thinking of running 3 separate jobs at the same time doing a

re-org

on these 3 files

However - because of the join logicals - will that cause some conflict
between the 2 re-orgs?




Alan Shore
Solutions Architect
IT Supply Chain Execution

[cid:image001.png@01D7725D.FBFDACB0]

60 Orville Drive
Bohemia, NY 11716
Phone [O] : (631) 200-5019
Phone [C] : (631) 880-8640
E-mail : ASHORE@xxxxxxxxxxxxxxxxxxxx<

mailto:ASHORE@xxxxxxxxxxxxxxxxxxxx>

‘If you're going through hell, keep going.’
Winston Churchill

From: MIDRANGE-L [mailto:midrange-l-bounces@xxxxxxxxxxxxxxxxxx] On

Behalf

Of Mark Waterbury
Sent: Tuesday, July 6, 2021 11:52 AM
To: Alan Shore via MIDRANGE-L <midrange-l@xxxxxxxxxxxxxxxxxx<mailto:
midrange-l@xxxxxxxxxxxxxxxxxx>>
Subject: Re: [EXTERNAL] Re: Analyzing RUNSQLSTMT jobs

Alan,

A "mass purge" of database records could impact your applications
performance if those physical files are not using REUSEDLT(*YES), so

that a

RGZPFM is needed to reorganize the PF member(s) to reclaim the space

used

by the deleted records.

After such a RGZPFM, any and all Logical File access paths over those

PF

members need to be updated; if the Logical Files (or physical files

with

keys) were not set to MAINT(*IMMED), this access path rebuild will

occur on

the next OPEN of that file. This could significantly slow down your
applications, the first time each file is used after such a "re-org."

HTH,

Mark S. Waterbury

On Tuesday, July 6, 2021, 11:30:44 AM EDT, Alan Shore via MIDRANGE-L

<

midrange-l@xxxxxxxxxxxxxxxxxx<mailto:midrange-l@xxxxxxxxxxxxxxxxxx

<mailto:

midrange-l@xxxxxxxxxxxxxxxxxx%3cmailto:midrange-l@xxxxxxxxxxxxxxxxxx>>>

wrote:

Thanks for your reply Vern
I am wading my way through the document that Jack sent, and it's a LOT

of

information to soak in
My problem is that a job that's presently running is still running

from

last night
Its been running for about 5 weeks or so - taking different times to
complete - but NOTHING like it is today
However - in my analysis I have discovered that I believe a purge was
recently run on a couple of the files
There are quite a number of deleted rows in these files
My question is - would the deleted rows impact SQL queries?
Forgive my ignorance as I haven't had a decent nights sleep in quite a

few

days and Im not thinking straight

Alan Shore
Solutions Architect
IT Supply Chain Execution

60 Orville Drive
Bohemia, NY 11716
Phone [O] : (631) 200-5019
Phone [C] : (631) 880-8640
E-mail : ASHORE@xxxxxxxxxxxxxxxxxxxx<

mailto:ASHORE@xxxxxxxxxxxxxxxxxxxx

<

mailto:ASHORE@xxxxxxxxxxxxxxxxxxxx%3cmailto:ASHORE@xxxxxxxxxxxxxxxxxxxx>>

'If you're going through hell, keep going.' --Winston Churchill

--
This is the Midrange Systems Technical Discussion (MIDRANGE-L) mailing list
To post a message email: MIDRANGE-L@xxxxxxxxxxxxxxxxxx
To subscribe, unsubscribe, or change list options,
visit: https://lists.midrange.com/mailman/listinfo/midrange-l
or email: MIDRANGE-L-request@xxxxxxxxxxxxxxxxxx
Before posting, please take a moment to review the archives
at https://archive.midrange.com/midrange-l.

Please contact support@xxxxxxxxxxxxxxxxxxxx for any subscription related
questions.

Help support midrange.com by shopping at amazon.com with our affiliate
link: https://amazon.midrange.com