Comments on: Simulation

By: khailey

khailey — Wed, 21 Jan 2009 20:12:07 +0000

Stats are only collected for SQL that has been found running but yes it’s collected for all the sql that has bee found running. This should probably be limited to the most executed or at least those execute recently.
I these enhancements are on the todo list but I won’t be writing any more code as I’m now working on a production version of similar code at Embarcadero, though I can post any improvemtns people want to submit.

Interesting error on sash_sqlstat. Oracle doesn’t complain creating a table with two fields with same name.
Changed and uploaded a version with just one field.

By: Bart Cozijnsen

Bart Cozijnsen — Tue, 06 Jan 2009 21:13:59 +0000

Great Job,

I get the feeling that there is a restriction missing
in procedure get_sqlstats of V2 targ_1_pkg.sql.
It is now inserting all rows from gv$sql.
Is this correct?

In repo_1_tables.sql table sash_sqlstat contains a field:
elasped.
Also the correct field elapsed is there.

best regards Bart

By: khailey

khailey — Wed, 15 Oct 2008 21:56:17 +0000

Linux version of ASHMON available !!

http://www.perfvision.com/ftp/ashmon_linux.zip

still needs a small change to soft link tile0.7.8 to /usr/share/tcl :
cd /usr/share/tcl8.4
ln -s /opt/ashmon/lib/tile0.7.8 .
lrwxrwxrwx 1 root root 24 2008-10-14 12:27 tile0.7.8 -> /usr/share/tcl/tile0.7.8

Thanks to Simon Coter from Oracle Italy who got it working during my 2 day class this past week in Milan.

By: khailey

khailey — Thu, 31 Jul 2008 23:50:34 +0000

Thanks for the input. Turns out there was sql_id and sql_child_number were reversed in the insert into the repository. Should be fixed now with the latest version.
.
Yes the enhancement to collect new users would be useful. At this point I have no plans to do enhancements though I will look at fixes. If anyone wants to enhance the package, the code is welcome, and I’ll check it in on the site.
Thanks
Kyle
http://perfvision.com

By: Carel-Jan

Carel-Jan — Thu, 31 Jul 2008 21:49:28 +0000

Thanks for the good work!
I just upgraded from ash simulator V1 to V2 at a set of five 10.2.0.2 SE-One databases on Linux, using one central repository.

I no longer get any SQLs in the repository. I found that most of the SQL-ids inserted have a value 0. Do you know why this can happen?

Further, I have a slight enhancement. I experienced that a user that is added to that database after the installation of the simulator will not be added to the repository. To deal with this, I have added the get_users call to the get_all procedure in the package. It works fine for me, but is there a reason why this hasn’t been done, or why I shouldn’t do this?

By: khailey

khailey — Thu, 08 May 2008 17:16:27 +0000

1) Andrew -yep - sash_curhost was wrong. Should have been sash_target. Changed.
2) added back in the use of gv$ as a workaround to ORA-02070
3) on 9i I as getting ora-22804 when selecting OBJECT_OWNER from v$sqlplan so I just changed it select ‘unknown’. I could revisit this to see if it works on 10g or maybe I could select values into a variable and then do the insert instead of doing the insert into the repository as a select from the target.

By: andrew henry

andrew henry — Thu, 08 May 2008 14:14:24 +0000

If I run your script for the target DB (targ_1_pkg.sql) on either a 9.2.0.8 or 10.2.0.4 database (win32) I get compilation errors on the package body. I run the script as sys, the database link works fine and the sashnow view gets created without issues.

I just ran it through Toad to see where it was failing:

SASH.sash_curhost@SASHREPO;

This object; sah_curhost, is not created in your script, but is referenced in the package body. Do you have the definition for this table?

By: khailey

khailey — Wed, 07 May 2008 20:05:22 +0000

andrew henry -
1)sleep is a variable - I just changed it to v_sleep to be less confusing
2)sys.v$database access - script needs to be run as SYS
3)gv$ - just changed these to v$
———————————–
Checinski -
Thanks for the good info.
1) There seems to be a bug on 9.2 querying v$sql_plan, so the work around is just to not collect the plan info on 9.2. I could add a version check for this, but that’s way down my list of things to do. Anyone else want to add a version check in the collection package?
2) For v1, sash_targets should be filled in with the target dbid automatically when installing sash_pkg on each target.
For v2, you see 0 rows in v$active_session_history its probably because the dbid filter isn’t set up.
For example
select count(*) from sash_all;
227
select count(*) from v$active_session_history;
0
@repo_5_curdb
HOST
——————————
KYLEHPD
KYLEHPLAP
Enter value for host: KYLEHPD
select count(*) from v$active_session_history;
227
3) yes, the sql plans, text and stats don’t get purged, only SASH does. Still to be done is a purge routine for SQL plans, text and stats. An easy purging method is to just delete from
sash_sqltxt
sash_sqlplans
sash_sqlstats
where sql_id in (select sql_id from sash_sqlids where last_found < sysdate - 7 )
or soemthing to that effect.
Deletes are expensive, instead of automating this deleting, I was still thinking of some way to partition the data, but didn’t come up with a method yet, so deleting is the only way right now.

By: Checinski

Checinski — Wed, 07 May 2008 14:29:37 +0000

On Standard Edition 9i, ashmon (via s-ash) is useful to complement eom for tuning oracle databases.

Here are some remarks from my experiences with the scripts:

-Use of s-ash for tuning database 9.2.0.7 works fine if you modify the package sash_pkg not to query v$sql_plan.
Otherwise you will get an error “ORA-07445″
This bug seems to be fixed in 9.2.0.8 (Metalink 4434689.8).

-Before execute script hostview.sql ( filtering) you have to verify that the field “host ” (table SASH_TARGETS) is correctly filled.for each dbid.

-default tablespace of user sash increases in size:
You have periodically to purge some tables ( eg: SASH_SQLSTATS_ALL)

By: andrew henry

andrew henry — Wed, 07 May 2008 11:49:24 +0000

I installed the V.2 objects but I have problems with the targ_1_pkg.sql script. The package body seems to have a number of compilation errors.

At the end, you use a dbms_lock.sleep inbuilt but instead of specifying seconds, you enter the string sleep:

dbms_lock.sleep(sleep);

Is this correct? Should it not be (example):

dbms_lock.sleep(3);

Also, you reference SYS.v$database, amongst others, but I could not get this to work manually when prefixing the schema owner, and as v$ objects are all views, I simply omitted SYS. If you want to see v$ views as a normal user, I just grant SELECT_CATALOG_ROLE to be able to see all v$ views.

Also, there were a number of v$ views that were prefixed with a ‘g’:

select … from … gv$instance;

Is this correct or a typo? could not find any sash segments with this prefix.

Still would not compile after correcting these.

I did look at the poor mans partitioning though and it looks like it will work fine, but I have not tested.