How to exclude ORA- errors from alerts in Grid Control

There is a Note: 330996.1 on Metalink how to filter
certain alert.log errors from appearing in alerts.

There is example there how to ignore certain errors
from coming to alerts of database.

Unfortunately proposed expression doesn't work - might be some
characters hidden in html code of the page or might be simply missed.
Since it is based on regular expressions there must be clause
to ignore certain patterns using symbol ^.

After some testing I came with the expression like this
.*ORA-0[^(3333|2222|1111)].*
that allows to ignore ORA-01111, ORA-02222, ORA-03333 messages.

Also I removed "Generic Alert Log Error Status" metric to elude messages
about ignored errors.

The pattern was tested on EM GC Rel.4 on 10.1.0.4 on Linux.


Have a good day!

ORA-1873 accessing the SQL_RESPONSE_COLLECT metric

So you upgraded your databases to 10.2.0.4 and also recreated hc...dat file not
to hang Grid Control agent, enabled health check, started agent -
everything to fix consequences of bug https://metalink.oracle.com/metalink/plsql/showdoc?db=Bug&id=5872000 (see post http://dbadailytelegraph.blogspot.com/2008/08/killing-health-check.html)
and executed tail -f emagent.trc to see if some errors coming - 1, 2, 3 minutes,
everything looks to be alright.

Don't hurry up! :(

The errors are coming! And they are from anonymous PL/SQL block that runs
to measure "SQL Response Time (%)" metric and it scheduled to gather
information every 5 minutes by default:

ORA-06512: at line 7
2008-09-15 22:12:17,815 Thread-4064263072 ERROR engine: [oracle_database,test,sql_response_collect] : nmeegd_GetMetricData failed : ORA-01873: the leading precision of the interval is too small
ORA-06512: at "DBSNMP.MGMT_RESPONSE", line 1013
ORA-06512: at "DBSNMP.MGMT_RESPONSE", line 923
ORA-06512: at line 7

2008-09-15 22:12:17,816 Thread-4064263072 WARN collector: Error exit. Error message: ORA-01873: the leading precision of the interval is too small
ORA-06512: at "DBSNMP.MGMT_RESPONSE", line 1013
ORA-06512: at "DBSNMP.MGMT_RESPONSE", line 923
ORA-06512: at line 7


The search on Metalink leads to https://metalink.oracle.com/metalink/plsql/ml2_documents.showDocument?p_database_id=NOT&p_id=6737060.8
and affected (confirmed) versions of databases are 10.2.0.3 and 10.2.0.4
and it will be fixed in 11.1.0.7 and 11.2

Disabling the metric will help to avoid errors but it can be important metric at some stage.


Have a good day!

using EM Grid Control templates to change metrics

If you use EM Grid Control and have a lot of databases added there you
definitely have templates to manage metrics depending on what type of
databases you monitor. And of course you apply changes not only through
GUI but using emcli as well.

Are they the same? They should be but at least recently I found that
using emcli to apply metrics works I would say more correctly and allows
to avoid some errors on a stage of targets selection.

The code to apply template is simple:

./emcli apply_template -name="test template" -targets="test1:oracle_database" -copy_flags="1"

where copy_flags is 0 or 1 depending
on you willing to rewrite metrics or not.

Well, looks like it is not difficult to create or change metrics on number of
targets but what about to delete some of them based on criteria or at least
those named the same.

I stuck with it when the metric to monitor tablespace's free size percentage
was added wrongly:
* The metric "Tablespace Space Used (%)" named %UNDO%
with thresholds 110 and 110 for warning and critical;
* The template with the metric was applied on dozens of databases.

Even metrics were added and had 110 % they did not work properly -
something was wrong either with name (even there is a tip to use wildcards
for the same pattern) or with per cent which is more than 100.
And the task is just to avoid paging on undo tablespaces.

After many tests of different configurations for metric names and thresholds
I found out that only 100 per cent works for thresholds and name should
not be with % symbols.

The template was changed to have a new metric name
UNDOTBS01 (100 and 100 %%) and I was ready to apply it using emcli but
what about old supposed-to-be-working ones?
How to delete them all at once or to generate a script for emcli?

Unfortunately emcli has no such command to change existing metrics
on target and I was forced to delete wrong metrics one by one manually
through Grid Control interface.
It was not quick and I spent about an hour routinely clicking the same buttons.
And I still have no idea how to do it massively.
To trace EM actions? To search Metalink?
To use tricks and tips in existing EM version?
I will try and

you have a good day!


All said before is for EM Grid Control 10.2.0.3
10.2.0.4 came with tips that
"Wildcard characters for object names are not supported."
"The "warning and critical threshold combination"
should be either (0,0), (NULL, NULL)
or (0 < warning threshold < 100, 0 < critical threshold < 100,
warning threshold <>

Attempt to apply metric for non-existing tablespace
will be executed successfully but it will not be there after update.

"Refreshed" database files naming

I have been doing many "refreshes" last days.
One of scenarios is to refresh a development
or test database from backup of production database.
The databases are on ASM and names consist dev
or test or prod in them.

Nothing difficult about it - get backup restored to disk,
start instance in nomount mode, restore controlfiles,
catalog backupsets, restore and recover database and open it.
And rename a database after the procedure using nid utility.

But the thing is that if you restore a database from
backup to OMF destination as ASM or file system,
oracle will use database name (if db_unique_name
is not set to something different) to name folders
where datafiles are going to be.
That means files will appear in folders named as source database
and will have "prod" or "prd" (or whatever the pattern is)
in folder names.

It can be nothing bad with it, but if you are going to restore
the same database under different name on the same server
you will get problem. Not to mention violation of all standards.
And also even after successful open of restored database
and its renaming you will have all files staying at the same place.

So if you would like to avoid it put db_name as source one
to you initialization file and add db_unique_name set to the name
of what database will be renamed to later. This will help
to restore datafiles to right location and you will need
just to rename the database afterwards.

Have a good day!

wrong results for 10.2.0.4 with cursor_sharing='SIMILAR'

If you running Oracle 10.2.0.4 using cursor_sharing='SIMILAR'
you probably already aware about a bug that causes database
to produce wrong results for some type of sql statements.

If you have not run into it yet, try to execute the following
statement with cursor_sharing set to SIMILAR (either on system
or session level)

select t1.d1, t2.n1
from
(select 1 as d1 from dual d1, dual d2) t1,
(select 1 as d1, 2 as n1 from dual) t2
where t2.d1(+) = t1.d1
group by t1.d1, t2.n1;


If you got 1 and null in output - this is it, you hit the one.

Further change of cursor_sharing to 'EXACT' will return
1 and 2 and if after that you want to get bad results again
(changing to 'SIMILAR' back) nothing happens - from now on
you will see good results (for this particular cursor) until
you bounce a database.

The bug 7277101 is also applicable for 11.1.0.6 and
unfortunately has not fixed yet.


Have a good day!

Killing health check

Do you think health check can kill?
Well... at least it can make oracle agent "not running" and left hung
agent processes.

It is not a common situation for all versions of agents and databases.
The problem affects oracle agent 10.2.0.4 and either 10.2.0.3 or
11.1.0.6 databases on Linux x86-64 platform.

The symptoms like this: you just upgraded you agents up to
10.2.0.4 version, saw fine ("green") status in Grid Control
and later (or next day) you see agents down and not running.
What happened?

Most likely, you hit the problem described in Note:566607.1
and agent opened too many hc_database_name.dat files
(which are in $ORACLE_HOME/dbs) and crashed.

That is because it was trying to open file using
a slightly different structure to access health check files
(1544 bytes instead of 1552 as it was before) but failed
and left file used. It continues until "too many files opened" error.

As a remedy patch for Bug 5872000 can be applied to database
or health check of instance can be disabled in Grid Control.

Probably the best one will be to apply patch but
it requires to bounce database and that makes me thinking
that I should prepare upgrades even of agents
carefully since it can even lead to patch a database!


Have a good day!

Silent upgrade of Oracle Agent to 10.2.0.4

Recently I got a task to upgrade RAC-aware
oracle agents from 10.2.0.3 to 10.2.0.4.

I presumed that installation available on OTN for mass
deployments will do the trick but I failed on that assumption.

Finally I recognized that upgrade should be done with patch
3731593 which is available on Metalink.

Due to security reasons there is only access to command line
on the server and that led me to use silent installation of the
patch. It was not the first time when I used it so I started to
put necessary values to the copy of patchset.rsp file.

Looking through the contents of the file I discovered
with pleasure that there is no entry for ORACLE_HOME_NAME
variable there. It is not a problem to find out it (/etc/oraInst.loc shows
location of oracle inventory and there is ContentsXML/inventory.xml
file there with information about oracle home paths and names)
but I can't see the reason to use it during installation.

Anyway, I made a file for silent installation and executed installer
(stopped agents on each host before it) - got an error

3731593/Disk1$ ./runInstaller -silent responseFile /patch/3731593/Disk1/response/patchagent10204.rsp
...
OUI-10202:No response file is specified for this session.
A silent session requires inputs from a response file or
from the command line. Specify the response file and re-run the installer.

Ok sounds familiar. How about to add "=" after responseFile parameter
Ran it, got an error again

./runInstaller -silent responseFile=/patch/3731593/Disk1/response/patchagent10204.rsp
...
SEVERE:Values for the following variables could not
be obtained from the command line or response file(s):
 ORACLE_HOME(OracleHome),
 ORACLE_HOME_NAME(OracleHomeName)
Silent install cannot continue.

Great! ORACLE_HOME_NAME had appeared!
And even ORACLE_HOME was not recognized :(

Finally I added those 2 variables to command line
and successfully executed upgrade of oracle agent
(remote installation also was completed without
problems)

./runInstaller -silent responseFile=/patch/3731593/Disk1/response/patchagent10204.rsp
ORACLE_HOME="/u01/app/oracle/product/10.2.0/agent10g"
ORACLE_HOME_NAME="agent10g"

And moreover the installation did not ask to run any root.sh scripts
which saved SA's time :)

After upgrade everything went smoothly - new versions of the agents
had been discovered in Grid Control, files were uploaded and
monitoring had resumed. Lucky me!

And the great news - I got Troubleshooting Oracle Performance book!

What a good day!
Have a good one too!