How to fix CPU usage problem in 12c due to DBMS_FEATURE_AWR

I love my job because I always have suprises. This week’s surprise has been another problem related to SQL Plan Directives in 12c. Because it is a common problem that potentially affects ALL the customers, I am glad to share the solution on my blog 😀

Symptom of the problem: High CPU usage on the server

My customer’s DBA team has spotted a consistent high CPU utilisation on its servers:

spd_awr_high_cpu_sar

Everyday, at the same time, and for 20-40 minutes, the servers hosting the Oracle databases run literally out of CPU.

spd_awr_high_cpu_em

 

Troubleshooting

Ok, it would be too easy to give the solution now. If you cannot wait, jump at the end of this post. But what I like more is to explain how I came to it.

First, I gave a look at the processes consuming CPU. Most of the servers have many consolidated databases on them. Surprisingly, this is what I have found:

spd_awr_high_cpu_m001It seems that the source of the problem is not a single database, but all of them. Isn’t it? And I see another pattern here: the CPU usage comes always from the [m001] process, so it is not related to a user process.

My customer has Diagnostic Pack so it is easy to go deeper, but you can get the same result with other free tools like s-ash, statspack and snapper. However, this is what I have found in the Instance Top Activity:

spd_awr_high_cpu_instOk, everything comes from a single query with sql_id auyf8px9ywc6j. This is the full sql_text:

It looks like something made by a DBA, but it comes from the MMON.

Looking around, it seems closely related to two PL/SQL calls that I could find in the SQL Monitor and that systematically fail every day:

spd_cpu_sql_monitorDBMS_FEATURE_AWR function calls internally the SQL auyf8px9ywc6j.

The MOS does not know anything about that query, but the internet does:

spd_awr_franckOh no, not Franck again! He always discovers new stuff and blogs about it before I do 🙂

In his blog post, he points out that the query fails because of error ORA-12751 (resource plan limiting CPU usage) and that  it is a problem of Adaptive Dynamic Sampling. Is it true?

What I like to do when I have a problematic sql_id, is to run sqld360 from Mauro Pagano, but the resulting zip file does not contain anything useful, because actually there are no executions and no plans.

During the execution of the statement (or better, during the period with high CPU usage), there is an entry in v$sql, but no plans associated:

And this is very likely because the statement is still parsing, and all the time is due to the Dynamic Sampling. But because the plan is not there yet, I cannot check it in the DBMS_XPLAN.DISPLAY_CURSOR.

I decided then to trace it with those two statements:

At the next execution I see indeed the Adaptive Dynamic Sampling in the trace file, the errror due to the exhausted CPU in the resource plan, and the directives that caused the Adaptive Dynamic Sampling:

 

 

So, there are some SQL Plan Directives that force the CBO to run ADS for this query.

This query touches three tables, so instead of relying on the DIRECTIVE_IDs, it’s better to get the directives by object name:

Solution

At this point, the solution is the same already pointed out in one of my previous blog posts: disable the directives individually!

This very same PL/SQL block must be run on ALL the 12c databases affected by this Adaptive Dynamic Sampling problem on the sql_id auyf8px9ywc6j.

If you have just migrated the database to 12c, it would make even more sense to programmatically “inject” the disabled SQL Plan Directives into every freshly created or upgraded 12c database (until Oracle releases a patch for this non-bug).

It comes without saying that the next execution has been very quick, consuming almost no CPU and without using ADS.

HTH

Ludovico

 

Oracle Database on ACFS: a perfect marriage?

Update: I will give this presentation at UKOUG Tech15, Wed 9 December at 14:30.

This presentation has had a very poor score in selections for conferences (no OOW, no DOAG) but people liked it very much at Paris Oracle Meetup.  The Database on ACFS is mainstream now, thanks to the new ODA releases. Having some knowledge about why and how you should run (not) Databases on ACFS is definitely worth a read.

Slides

Demo 1 recording

Demo 2 recording

Demo script (DB ACFS clone from Standby Database)

 

Comments are, as always, very appreciated 🙂

Ludo

Generating graphs massively from Windows Performance Counters logs

Windows Performance Monitor is an invaluable tool when you don’t have external enterprise monitoring tools and you need to face performance problems, whether you have a web/application server, a mail server or a database server.

But what I don’t personally like of it is what you get in terms of graphing. If you schedule and collect a big amount of performance metrics you will likely get lost in adding/removing such metrics from the graphical interface.

What I’ve done long time ago (and I’ve done again recently after my old laptop has been stolen 🙁 ) is to prepare a PHP script that parse the resulting CSV file and generate automatically one graph for each metric that could be found.

Unfortunately, most of Windows Sysadmin between you will disagree that I’ve done this using a Linux Box. But I guess you can use my script if you install php inside cygwin. The other tool you need, is rrdtool, again I use it massively to resolve my graphing needs.

How to collect your data

Basically you need to create any Data Collector within the Performance Monitor that generates a log file. You can specify directly a CSV file (Log format: Comma separated) or generate a BLG file and convert it later (Log format: Binary). System dumps are not used, so if you use the standard Performace template, you can delete it from your collection.

Remember that the more counters you take, the more the graph generation will take. The script does not run in parallel, so it will use only one core. Generally:

Where (Speed factor) is depending on both the CPU speed and the disk speed because of the huge number of syncs required to update several thousands of files. I’ve tried to reduce the number of rrdupdates by queuing several update values in a single command line and I’ve noticed an important increase of performances, but I know it’s not enough.

Converting a BLG (binary) log into a CSV log

Just use the relog tool:

 Generating the graphs

Transfer the CSV on the box where you have the php and rrdtool configured, then run:

 

generated_graphs

Now it’s done! 

The script generate a folder with the name of the server (LUDO in my example) and a subfolder for each class of counters (as you see in Performance Monitor).

Inside each folder you will have a PNG (and an rrd) for each metric.

 

generated_graph_cpu

 

Important: The RRD are generated with a single round-robin archive with a size equal to the number of samples. If you want to have the rrd to store your historical data you’ll need to modify the script. Also, the size of the graph will be the same as the number of samples (for best reading), but limited to 1000 to avoid huge images.

Future Improvements

Would be nice to have a prepared set of graphs for standard graphs with multiple metrics (e.g. CPU user, system and idle together) and additional lines like regressions…

Download the script: process_l_php.txt and rename it with a .php extension.

Hope you’ll find it useful!

Cheers

Ludo

Oracle capacity planning with RRDTOOL

RRDize everything, chapter 2

Oracle Database Server has the most powerful system catalog that allows to query almost any aspect inside an oracle instance.
You can query many v$ fixed views at regular intervals and populate many RRD files through rrdtool: space usage, wait events. system statistics and so on…

Since release 10.1 Oracle has introduced Automatic Workload Repository, a finer version of old good Statspack.
No matter if you are using AWR or statspack, you can rely on their views to collect data for your RRDs.

If you are administering a new instance and you haven’t collected its statistics so far, you can query (as example) the DBA_HIST_BG_EVENT_SUMMARY view to gather all AWR data about wait events. Historical views could be useful also to collect historical data once a week rather than query the fixed views every few minutes doing the hard work twice (you and AWR).

The whole process of gathering performance data and update rrd files can be resumed into the following steps:

– connect to the database
– query the AWR’s views
– build and execute an rrdtool update command
– check if rrd file exists or create it
– update the rrd file

The less rrdtool update commands you will execute, the better the whole process will perform.
Do it in a language you are comfortable with and that supports easily connection descriptors.

Since I’m very comfortable with php, I did it this way.

This is a very basilar script that works greatly for me with good performances:

Depending on how many different wait events you have, you’ll have a certain number of rrd files:

As you can see, they are not so big…

Once you have your data in rrd files, it’s quite simple to script even complex plots with several datasources. Everything depends on the results you want.
This script stack all my wait events for a certain instance: it takes the directory containing all the rrds as first argument and the number of hours we want to be plotted as second argument:

The resulting command is very long:

This is the resulting graph:
Graph plotted with rrdtool displaying Oracle instance Wait Events

OHHHHHHHHHHHH COOOOL!!!
😉

Any comment is appreciated! thanks

How to collect Oracle Application Server performance data with DMS and RRDtool

RRDize everything, chapter 1

If you are managing some Application Server deployments you should have wondered how to check and collect performance data.
As stated in documentation, you can gather performance metrics with the dmstool utility.
AFAIK, this can be done from 9.0.2 release upwards, but i’m concerned DMS will not work on Weblogic.

Mainly, you should have an external server that acts as collector (it could be a server in the Oracle AS farm as well): copy the dms.jar library from an Oracle AS installation to your collector and use it as you would use dmstool:

There are three basilar methods to get data:

Get all metrics at once:

Get only the interesting metrics:

Get metrics included into specific DMS tables:

What youraddress:// is, it depends on the component you are trying to connect:

If you are trying to connect to the OHS (Apache), be careful to allow remote access from the collector by editing the dms.conf file.

Now that you can query dms data, you should store it somewhere.
Personally, I did a first attempt with dmstool -dump format=xml. I wrote a parser in PHP with SimpleXML extension and I did a lot of inserts into a MySQL database. After a few months the whole data collected from tens of servers was too much to be mantained…
To avoid the maintenance of a DWH-grade database I investigated and found RRDTool. Now I’m asking how could I live without it!

I then wrote a parser in awk that parse the output of the dms.jar call and invoke an rrdtool update command.
I always use dms.jar -table command. The output has always the same format:

So I written an awk file that works for me.
use it this way:

And this is the code for update_metric_rrd:

Once you have all your rrd files populated, it’s easy to script automatic reporting. You would probably want a graph with the request count served by your Apache cluster, along with its linear regression:

This is the result:
OHS request completed
OHHHHHHHHHHHH!!!! COOL!!!!

That’s all for DMS capacity planning. Stay tuned, more about rrdtool is coming!

JBoss Portal and MySQL scalability: What The…???

I found several queries running on a MySQL 5.0 database like this one:

This query is related to JBoss Portal and does a full scan on table JBP_OBJECT_NODE.

It has bad performances (>0.8 sec) with just a few records:

mysql> select count(*) from JBP_OBJECT_NODE;
+———-+
| count(*) |
+———-+
|    33461 |
+———-+

If I rewrite the query using an inner join (à la Oracle, please forgive me) instead of a subquery I get an index scan:

With 30k records the execution time falls down from 0.8 secs to 0.01 secs…
That’s NOT all! I found this open bug:

https://jira.jboss.org/jira/browse/JBPORTAL-2040

With many users registered in, the JBoss Portal Admin console tooks over a minute to show a single page…

I don’t like portals…