something about the cbo was bothering me

A query similar to this was getting a full table scan. Now we know the NDR in archived, deleted and flag is 1. My initial thought was because of this we would get a full table scan. I was wrong, because we are using owner like ‘S%’ this would trigger the full table scan.

image

As you see this first run did a full table scan.

image

Now look at what happens when we change owner like ‘S%’ to owner = ‘SYSTEM’. We now get an index range scan. Much better.

image

1 – filter(“ARCHIVED”=0 AND “DELETED”=0 AND “FLAG”=’1′)

2 – access(“OWNER”=’SYSTEM’)

Foreign Keys and indexes

 

In this test case we are going to use two tables with a foreign key.  The question is, if we create a foreign key, do we need to add an index on that foreign key column?

OPS$ORACLE@VETDEV > create table bigtable as select * from dba_objects;

Table created.

OPS$ORACLE@VETDEV > create table owners as select * from dba_users;

Table created.

OPS$ORACLE@VETDEV > analyze table bigtable compute statistics;

Table analyzed.

OPS$ORACLE@VETDEV > analyze table owners compute statistics;
Table analyzed.

OPS$ORACLE@VETDEV > alter table bigtable add primary key (object_id);

Table altered.

OPS$ORACLE@VETDEV > alter table owners add primary key (username);

Table altered.

OPS$ORACLE@VETDEV > alter table bigtable add constraint bigtable_fk1
  2* foreign key (owner) references owners(username);
Table altered.

We are going to write a query against both tables joining them on the foreign key. Start by setting autotrace on with explain and statistics.  In the following explain plan we are cutting out statistics to focus on what jumps out at us. Even though we have a foreign key constraint we still got a full table scan on both big table and owners.

TC1

Okay, lets try and limit the number of rows returned by adding a filter predicate.

set autotrace trace

select b.owner, b.object_type, o.account_status
from owners o
,   bigtable b
where o.username = b.owner
and b.owner = ‘SUZANNE’;

Now first you will notice we put the filter predicate on b.owner and not o.username.  We know from the data distribution we have multiple owners and unique usernames. Now what does the explain plan look like.

image
I’m about to get off on a tangent; but bare with me.  There is something interesting going on here.  Look at the index unique scan against the username primary key column and the filter predicate on bigtable.owner.  What happened?  We are going to use the 10053 trace to find out what the optimized did.

alter session set events ‘10053’;

set autotrace trace

select b.owner, b.object_type, o.account_status
from owners o
,   bigtable b
where o.username = b.owner
and b.owner = ‘SUZANNE’;

Lets find the trace file we just generated and dig through it.  The trace file VETDEV_mmon_6474.trc will tell us just what the optimizer did.

[oracle@owirdb1 trace]$ ls -ltr
total 600
-rw-r—– 1 oracle oinstall 594496 Jan 16 07:00 alert_VETDEV.log
-rw-r—– 1 oracle oinstall    384 Jan 16 12:02 VETDEV_mmon_6474.trm
-rw-r—– 1 oracle oinstall   4111 Jan 16 12:02 VETDEV_mmon_6474.trc
[oracle@owirdb1 trace]$

There is a bunch of information in this file and it’s beyond the scope of what I wanted to discuss; what we are looking for is how did the optimizer transform our query.

**************************
Predicate Move-Around (PM)
**************************
PM:     PM bypassed: Outer query contains no views.
PM:     PM bypassed: Outer query contains no views.
query block SEL$1 (#0) unchanged
FPD: Considering simple filter push in query block SEL$1 (#0)
“O”.”USERNAME”=”B”.”OWNER” AND “B”.”OWNER”=’SUZANNE’
try to generate transitive predicate from check constraints for query block SEL$1 (#0)
finally: “O”.”USERNAME”=”B”.”OWNER” AND “B”.”OWNER”=’SUZANNE’ AND “O”.”USERNAME”=’SUZANNE’

Check this out, the optimizer rewrote the query to reduce the number of rows returned in the filter.

OPS$ORACLE@VETDEV > select count(*) from owners where username = ‘SUZANNE’;

  COUNT(*)
———-
         1

OPS$ORACLE@VETDEV > select count(*) from bigtable where owner = ‘SUZANNE’;

  COUNT(*)
———-
       199

So it makes more sense to return one row from owners and using that for the filter as apposed to using owners from bigtable.

Well what would happen if we put an index on the foreign key?

OPS$ORACLE@VETDEV > create index bigtable_idx1 on bigtable(owner);

Index created.

OPS$ORACLE@VETDEV > analyze index bigtable_idx1 compute statistics;

Index analyzed.

Now lets run the test again and without the filter predicate.
TC1
Same issue, but why. Well lets look at the query and ask what is getting returned.  Well, there are no filter predicates, so every row is going to be returned.  This is shown under predicate information and the rows column.  Therefore in this case, the full table scan with a hash join is the most efficient way to get the data. 
What would happen if we add a filter predicate to the query?
TC1
Well that’s nice, we have the index range scan on bigtable_idx1 and an index unique scan on the owners primary key index. 
So, what did we learn, 1) not having an index on the foreign key pretty much guarantees a full table scan on the joined table. 2) having an index does not guarantee the index will be used.  If the database determines that it would be cheaper to do a full table scan then that is what it’s going to do. 3) a filter predicate to the query reduced the total number of rows returned to the point where it would be cheaper to do an index unique scan. And 4) the optimizer will re-write a query to improve performance.

How do You succeed?

This does not have much to do with Oracle consulting, this is more some life lessons on how do succeed at work. I have learned many lessons in life, one of them is the freedom that comes from being independent and not having to depend on others. 

1) If you want to do a lot of things real bad then try multitasking.  The fact is humans are not good executing more then one task at a time.  Humans are good at switching from one task to another.  Think about writing an email and talking to someone at the same time. You can’t do both.  There is a lot of issues with rapid task switching.  I have called this context switching for many years. Context switching is moving from one line of thought to another line of thought.  Every time you do a context switch, you have to change your line of thought.  When multitasking, mistakes will increase and time to complete a task will increase. In my business Oraclewizard, inc I wear every hat, I’m accounts receivable, accounts payable, the receiving department, sales department and quite a few other departments. If I’m working on the company books and I think of something that needs to be done, I write it down in my todo book and keep with the current task until it’s complete. Between tasks I review my todo book, re-prioritize and start the next task.

“People can’t multitask very well, and when people say they can, they’re deluding themselves,” “The brain is very good at deluding itself.” neuroscientist Earl Miller.

I have learned over the years that if I’m designing a database system to step away from the computer, sit down at a desk with templates and a lot of paper then start the design.  When I’m attached to a computer, email and the internet is constantly trying to get my attention. The key is when I’m doing something that requires all my attention, I remove all the distractions.  There is no law that says “you must answer that text, email or phone call right now.”

2) Always improve yourself.

Spend one hour a day doing professional development.  My morning always starts with studying for an hour.  Note, not everything revolves around Oracle in my world.  Yes Oracle is a big port of my professional life; however I am also a pilot, technical diver and public speaker.

Attend at least one professional conference a year.  In my professional area, there are many opportunities to attend conferences. These are great ways to meet people in your arena and network.

If you really want to improve yourself, then volunteer to speak at an event.  A good friend of mine drilled into me, “Proper Preparation Prevents Poor Performance.”  Okay that is a cliché, but it’s true.

3) Your employer is your customer and the customer of your customer / employer is your customer. If you are the customer, how do you want to be treated? If you want to get ahead at work, provide good customer service to your employer.

4) You own your brand.  What is your brand?  Your brand is what do people think of you.  Way back when your brand was limited to maybe a few hundred people.  However; with the internet and mass communications you can influence your brand world wide. 

One of the things I like about “my brand” is when asked to do something, I don’t make excuses, I figure out how to do it.  Case in point, our Connect Direct / Connect Enterprise expert left the contract where I was working.  A program manger asked me to take it over.  Connect Direct / Connect Enterprise if more about secure data transmission then data storage / manipulation.  I could have have pushed back and told the program manager that was not my specialty but instead I accepted the challenge and made the program manger aware that there may be some rocky places along the way; but I will give it my best shot. 

You can also influence your brand by helping people when they have problems. People will remember it.

5) The success you enjoy will be from learning from your mistakes. To quote my fourteen year old niece “OMG you did that.” Let me tell you I have made some doozies of mistakes.  There was the time I truncated sys’s tables and could not figure out why the database did not start back up.  Then there was the time I ripped an employee of a customer a new one because her incompetence that was having an impact on me doing my job. And there were a lot of other mistakes.  But the key is each time I made a mistake, I learned from it.  You will make mistakes and some will go down in history but if you analyze the mistake and learn from it, you will climb another rung on the ladder of success.

6) Don’t get overwhelmed by your good ideas. This still happens to me, but I have gotten better at writing down my “good ideas” then letting them stew for a week or two before I decide wither they are worth starting. Frequently after a week or so, what seemed like a good idea at the time turns out to not be such a good idea.  The ones that are good ideas, I can take to the next level and figure out what it is going to take to get an idea off the ground.

7) Set goals and work on your goals daily.  This almost goes without saying; but I’m surprised by the number of people who don’t set goals.  Goals should be realistic and you should be able to measure the results.