Gints Plivna blog

2011-03-12T02:21:00.005+02:00

Indexes cannot be used for inequalities?

For years and years I thought it was true. Now I'd say it's only partially true. OK it is true for b-tree indexes (at least there isn't any reasonable effect of using an index for "not eaquals to"), but for bitmaps it's a different answer though. And I mean generally YES, they can be used :)

OK let's look at examples. I'll create a table and populate with almost all the same values but one. That's to make things more attractive for CBO to get the one and only value.

SQL> CREATE TABLE test (
2    id NUMBER PRIMARY KEY,
3    txt VARCHAR2(10) NOT NULL);

Table created.

SQL> INSERT INTO test
2  SELECT rownum, 'AAA'
3  FROM DUAL CONNECT BY LEVEL < 10000;

9999 rows created.

SQL> INSERT INTO test
2  VALUES (10000, 'BBB');

1 row created.

So now I have 9999 of "AAA" and only one "BBB".
I won't explain with examples what one can get with b-tree indexes, the best is INDEX FULL SCAN and the example by Richard Foote is here.

OK but what if we try bitmap index? Let's create it:

SQL> CREATE BITMAP INDEX txt_idx ON test (txt) COMPUTE STATISTICS;

Index created.

Now let's see what we get for the query we know should get only 1 row:

SQL> SELECT * FROM test
 2  WHERE txt <> 'AAA';

       ID TXT
---------- ----------
    10000 BBB

--------------------------------------------------------------------------
| Id  | Operation         | Name | Rows  | Bytes | Cost (%CPU)| Time     |
--------------------------------------------------------------------------
|   0 | SELECT STATEMENT  |      |     1 |     7 |     7   (0)| 00:00:01 |
|*  1 |  TABLE ACCESS FULL| TEST |     1 |     7 |     7   (0)| 00:00:01 |
--------------------------------------------------------------------------
Predicate Information (identified by operation id):
---------------------------------------------------

  1 - filter("TXT"<>'AAA')

Statistics
----------------------------------------------------------
        23  consistent gets
         1  rows processed

Bang! Full scan. Not very promising. OK what if we try to hint the query?
At first let's give a soft hint - how about using an index?


SQL> SELECT /*+ index(test) */ * FROM test
  2  WHERE txt <> 'AAA';

        ID TXT
---------- ----------
     10000 BBB

--------------------------------------------------------------------------------
| Id  | Operation                   | Name        | Rows  | Bytes | Cost (%CPU)|
--------------------------------------------------------------------------------
|   0 | SELECT STATEMENT            |             |     1 |     7 |   826   (0)|
|*  1 |  TABLE ACCESS BY INDEX ROWID| TEST        |     1 |     7 |   826   (0)|
|   2 |   INDEX FULL SCAN           | SYS_C009499 | 10000 |       |    26   (0)|
--------------------------------------------------------------------------------
Predicate Information (identified by operation id):
---------------------------------------------------
   1 - filter("TXT"<>'AAA')
Statistics
----------------------------------------------------------
         39  consistent gets
          1  rows processed

WoW! That's actually annoying. Bad plan, many consistent gets. Oracle is using primary key index, but not my brand new bitmap index! I'm disappointed.
OK now let's hammer it with the precise hint!

SQL> SELECT /*+ index(test txt_idx) */ * FROM test
  2  WHERE txt <> 'AAA';

        ID TXT
---------- ----------
     10000 BBB

-----------------------------------------------------------------------------
| Id  | Operation                    | Name    | Rows  | Bytes | Cost (%CPU)|
-----------------------------------------------------------------------------
|   0 | SELECT STATEMENT             |         |     1 |     7 |    36   (0)|
|   1 |  TABLE ACCESS BY INDEX ROWID | TEST    |     1 |     7 |    36   (0)|
|   2 |   BITMAP CONVERSION TO ROWIDS|         |       |       |            |
|*  3 |    BITMAP INDEX FULL SCAN    | TXT_IDX |       |       |            |
-----------------------------------------------------------------------------
Predicate Information (identified by operation id):
---------------------------------------------------
   3 - filter("TXT"<>'AAA')

Statistics
----------------------------------------------------------
          2  consistent gets
          1  rows processed

So that's it!! At last we get it! Only 2 consistent gets. And the cost is much less than using that stupid primary key index full scan. So that means Oracle even didn't consider using this plan at least for the query with just index hint. I looked at 10053 trace and found following for the query with just hint "index(test)":
The beginning looks promising:


Table Stats::
  Table: TEST  Alias:  A
    #Rows: 10000  #Blks:  20  AvgRowLen:  7.00
Index Stats::
  Index: SYS_C009499  Col#: 1    (NOT ANALYZED)
    LVLS: 1  #LB: 25  #DK: 100  LB/K: 1.00  DB/K: 1.00  CLUF: 800.00
    User hint to use this index
  Index: TXT_IDX  Col#: 2
    LVLS: 0  #LB: 1  #DK: 2  LB/K: 1.00  DB/K: 1.00  CLUF: 2.00
    User hint to use this index

However in access path analysis Oracle simply doesn't even consider my bitmap index.


SINGLE TABLE ACCESS PATH
  Single Table Cardinality Estimation for TEST[A]
  Table: TEST  Alias: A
    Card: Original: 10000.000000  Rounded: 1  Computed: 1.00  Non Adjusted: 1.00
  Access Path: index (FullScan)
    Index: SYS_C009499
    resc_io: 826.00  resc_cpu: 6932309
    ix_sel: 1.000000  ix_sel_with_filters: 1.000000
    Cost: 826.39  Resp: 826.39  Degree: 1
  Best:: AccessPath: IndexRange
  Index: SYS_C009499
         Cost: 826.39  Degree: 1  Resp: 826.39  Card: 1.00  Bytes: 0

Things however get brighter if we are just counting rows, then we don't need hint or anything, just run the select:

SQL> SELECT count(*) FROM test
  2  WHERE txt <> 'AAA';

  COUNT(*)
----------
         1

------------------------------------------------------------------------------
| Id  | Operation                     | Name    | Rows  | Bytes | Cost (%CPU)|
------------------------------------------------------------------------------
|   0 | SELECT STATEMENT              |         |     1 |     4 |     1   (0)|
|   1 |  SORT AGGREGATE               |         |     1 |     4 |            |
|   2 |   BITMAP CONVERSION COUNT     |         |     1 |     4 |     1   (0)|
|*  3 |    BITMAP INDEX FAST FULL SCAN| TXT_IDX |       |       |            |
------------------------------------------------------------------------------
Statistics
----------------------------------------------------------
          3  consistent gets
          1  rows processed

So it is more or less OK that table full scan is better (i.e. cost less than) bitmap index full scan and then bitmap conversion to rowids and then access table by calculated rowids. However it is quite strange that for simple index hint Oracle doesn't even consider this access path however as we can see from CBO trace understands that bitmap index is one of the hinted ones.
Of course the usual problems with bitmap indexes remains - locking, concurrency issues etc.
And yea I just arrived from Richard's seminar in Tallinn. Many new things, fluent story, cool!! :)

2010-02-17T16:58:00.002+02:00

Unbreakable Oracle documentation

Yes, that's the fact - everyone knowing the real situation will start Homeric laughter. It is down, down, down and down again. To be true sometimes it is also up and running. Unfortunately "sometimes" seems to be more and more rarely. Just now the popular entry point tahiti.oracle.com shows following:

Access Denied
You don't have permission to access "http://www-portal-stage.oracle.com/splash/www/index.html" on this server.
Reference #18.3c55293e.1266419217.19d36a

Interesting - is tooooo much to ask Oracle to provide just static htmls (OK there is advanced thing - search feature, however I could also easily use Google instead, at least it is less unbreakable ;) for its users continuously without interruptions more than not?
Or after the grand results of maintenance income there isn't any single $ left for an admin to monitor the resource? Or probably the last reliable box has gone to super functional and alike working MOS?

Chris Warticki in his blog wrote what to do if MOS is down then what? Unfortunately the blog now is (seems to be censored and) closed and for a little while pages are only accessible from google's cache. Most of the people in oracle-l list say they have downloaded oracle documentation locally in a recent discussion about oracle documentation. However it is bit odd for a company providing leading database of the world having income with nine zeroes being unable to provide simple html files without continued interruptions...

Probably that's the reason I more and more have to do something with SQL Server :)

2009-02-07T19:32:00.010+02:00

Woodwork

Time after time I'm doing something different than databases and Oracle :) One of these things is woodwork. So around the New Year I finished the most recent one - a shelf. Requirements, design and implementation fully by me. Here it is just attached to the wall.

And now full with stuff. Most of pottery in this shelf is produced by my cousin and her husband.

The hobby started already in school years when we had special woodwork lessons and this probably is one of the best results from these times - a turned wood vessel with amber decor on the top.

Nowadays there are fantastic possibilities to make something from wood on his own compared to Soviet and early post-Soviet times. Tools, paints, lacquers, various components and semi-finished products in shops are far and away more. For example the shelf above also was created using ground glued board, so actually the creation process is not complicated.
One of the most interesting tasks was to restore a few Soviet style chairs as old as myself (i.e ~30 years). Here is a photo with one old chair and one already restored.

2009-01-31T22:46:00.002+02:00

Security through obscurity - object names

This post was inspired by question in one of Latvian forums - can we use only numbers in column names. Of course we can do it using quotes but that's starting point into dangerous path of how to write unmaintainable code. But why stop only with tables like C34 and columns 1, 2, 3? There are ways beyond that!
So let's start with the same table names differing only in upper/lower case.

SQL> create table A (big number);
Table created.
SQL> create table "a" (small number);
Table created.

So what are our tables A and a:

SQL> desc A
 Name                    Null?    Type
 ----------------------- -------- -------
 BIG                              NUMBER
SQL> desc a
 Name                    Null?    Type
 ----------------------- -------- -------
 BIG                              NUMBER

A bit strange, isn't it? But that is because all nonquoted identifiers are converted to uppercase so we have to be precise:

SQL> desc "a"
 Name                    Null?    Type
 ----------------------- -------- ------
 SMALL                            NUMBER

But we can make things even more obscure! For example how about table and/or column name as space?

SQL> create table " " (A number, "a" number);
Table created.
SQL> select table_name from user_tables;
TABLE_NAME
------------------------------
a

A
3 rows selected.

Previous example with 2 rows instead of reported three and next with lost columns definitely could make some nervous people start searching for Oracle bugs.

SQL> create table b (" " number, "  " date);
Table created.
SQL> desc b
 Name                    Null?    Type
 ----------------------- -------- ------
                                  NUMBER
                                  DATE

But we can make things better. Why should we use only symbols on keybord? Remember the nice possibility of ALT key and numeric keypad keys. ALT + these keys gives us symbols according to their ASCII values, so just use your imagination. For example ALT+223 gives me this:

SQL> create table abnormal ("▀" number);
Table created.
SQL> desc abnormal
 Name                    Null?    Type
 ----------------------- -------- -------
 ▀                                NUMBER

Nice little black square.
Next level might be making some columns a bit like hidden.

SQL> create table persons (name varchar2(10), "SYSDATE" date);
Table created.
SQL> insert into persons values ('John', sysdate-1);
1 row created.
SQL> select sysdate from dual;
SYSDATE
---------
31-JAN-09
SQL> select name, sysdate from persons;
NAME       SYSDATE
---------- ---------
John       31-JAN-09

So where our column sysdate was gone? We inserted value sysdate-1 and got back the same sysdate? Although our column was in upper case, so our query column should be used? Naah, we should enclose it in quotes.

SQL> select name, "SYSDATE" from persons;
NAME       SYSDATE
---------- ---------
John       30-JAN-09

So next time when you are starting to think about table name like C12, numeric column names, column names in quotes or something like that, don't be shy! Use the complete power of obscurity, so making your db more secure. If you'd try hard enough, you'd reach perfection when obscurity is so great that nobody, even you, won't be able to understand anything ;)
And I wasn't even suggesting to replace DUAL in your schema ;)

2009-01-09T00:52:00.002+02:00

About experience

Just a few days ago Ryan Watson mentioned one of my articles in his blog post. It really wouldn't be anything important to write about but the first comment by Kevin really amused me and the comment was "Who is this guy and why is he qualified to comment on databases."

Blahhh... My answer was "[..] why does it matter and isn't better just to make reasonable pros and/or cons about the subject?" and Kevin answered with "Experts carry more weight than a novice; a senior Oracle admin with 20 years of experience has more experience to draw on. This is why a senior DBA will make $100 k and novice $20 k."

This was something so immediate and direct statement that let me started to think more about this subject. How much experience means to me? How much it means to other people (virtually) around me? Probably we all know one of the brightest stars in this area good old Don with one of his latest performances here, but undoubtely for all of us - if we don't know other people and they say something, their experience somehow matters. So how much does it matter for me?

1. I definitely have positive experience with some people and know their knowledge and writing style. They have credit in my eyes and I read each their article/statement/whatever else with more caution than on average. Even if the beginning of the message is somehow blurry or overall quality is not that perfect this time. They have to "apply efforts" to their articles to make me angry and sceptic.
2. I definitely have negative experience with a few people and know their usually blurry and washy posts without any resultant technical value. They have really minus credits in my eyes and even if this time they have written something technically sound I'm quite suspect about that. They have to write something really, really good to attract my attention other than sigh "usual crap".
3. I definitely have neutral experience with quite a lot of people. Let's say "nothing special" :) I read their articles only when they are directly related to my particular current needs.
4. People whom I've met the first time or haven't made any particular impression. These are the most complex part. What is my attitude to them? Analyzing my usual behaviour the conclusion is - I start reading the article with some interest and few starting paragraphs should attract my attention. They must be interesting or suitable for my needs otherwise I jump to next article. BUT. I've never (OK almost never :) read the author's CV, last rows about the author how cool he is or whatever else describing his experience. The content of the article is what matters for me.

So people I don't know win over people with negative credit (whose articles I usually even don't read), but lose to people with positive credit (whose articles I usually try to read until the and to find out something valuable).

And returning to the initial comments - what does 20 years of experience mean to me? Most probably nothing. At least initially. 20 years ago there was Oracle 6, how does it help for today's work? Most probably in almost no way. I'd say that a few years for an inquiring and smart mind is enough to deserve more than a person with 20 years experience stalled in Oracle 8. Actually he wins without any doubt at all.

The (quality of) content is what matters - this is the conclusion. At least for people I don't know :)

2008-11-23T15:00:00.003+02:00

Advanced Oracle Troubleshooting by Tanel Poder..

.. is over. His seminar in Riga was real fun for two days. It was different from the big guys' seminars I've attended before (for example, Thomas Kyte and Jonathan Lewis). Not in the way that it was or wasn't somehow better or not, but in the way that Tanel spoke more about:

internal memory structures (x$ tables), process stacks, Oracle kernel functions and their mapping to SQL statement execution plan steps;
understanding structure of library cache, process state objects, buffer cache;
dumping cursors, trace buffers, process stack;
systematic troubleshooting a session and finding the root cause of slowness and what it is doing now;
and many more subjects.

Lots of live demos, nothing like an Oracle documentation rephrase, narrative explanation of quite advanced technical things - these were the main characteristics of this seminar.

Some previous understanding of Operating systems basic functions, commands and Oracle architecture however is a must, otherwise you simply will be overwhelmed by new information and won't be able to follow the main idea. On the other hand I'm absolutely not a sysadmin and/or true DBA, but I understood let's say 95% of the material provided ;)

So the conclusion is - I definitely recommend it for the people interested in these topics.

2008-11-20T22:34:00.003+02:00

Data modeling

Time after time it is worth to remember (or learn) basic theory. One of them, which is very important when creating new systems, is data modeling. I'm quite sure most of the DBMS people have seen one or another data model with tables, columns, primary and foreign keys. However the theory of data modeling says that this is only one level out of three. Unfortunately I've seen quite many cases when people have forgotten two another levels. So let's remember what they are.

Conceptual data model

Conceptual data model contains entities, relationships and attributes. One can draw them using different notations for example ER modeling or UML, but the idea remains the same - this model doesn't know anything about the underlying DBMS. This model is the same for Oracle, SQL Server, DB2, MySQL, whatever else. So why is it important?

It speaks in business language and uses terms only from business users, not any programmer's or IT person's hacks.
It is normalized, every piece of information is stored only once.
It displays common structures and tries to aggregate them, it is easier to find them in such model.
It is readable (after a short introduction probably) by every sensible customer.
It doesn't contain any reference to any particular DBMS, so one can easily reuse it for ultimately any DBMS.

So the question - have you used it? If not - why? Are you aware of possible data waste dangers?

Logical data model

This is the model people usually use. It contains tables, columns, primary keys, foreign keys, unique keys. Some characteristics of it are:

It is created on the basis of conceptual data model, but it might not correspond 1:1 to it. For example some entities might be split, some unioned.
It may contain some programmer's hacks for example different flag columns for easier data selection.
It is aware of used DBMS. For example data types for columns are DBMS specific.
It may have denormalized columns (i.e. the same facts stored in more than one column) for performance reasons.
Table and column names might not exactly mimic business terms i.e. they might be abbreviated, changed to conform some naming rules and/or DBMS restrictions, although for easier readability it is worth to retain some dependency.
It is based on conceptual data model but created keeping performance and most common possible SQL statements in mind.

I'm quite sure you have used it, so the only question is - have you created it keeping performance and business critical SQL in mind?

Physical data model

What's that? Is there anything remaining to model at all? Yes it is. Heap table (i.e. table type which is used by default) is not the only available table type in Oracle. Partitioning is not the only option, which can be used in Oracle. So what is modeled here?

Table type - heap, index organized, clustered, sorted hash cluster, single table hash cluster there are many possible options.
Partitioning, compression, encryption - do you need it, can you afford it?
Indexes - will you use only b-tree or something more?
Will you use object tables?
What tablespaces will be created and on what devices/files?
Here comes the real strength (or weakness) of your chosen DBMS - if it offers all these features - why not at least consider them?

If you are DBA you have definitely used some of the features mentioned above, but have you thought about non-default ones? Have you systematically gone through at least the most business critical tables and considered what kind of table type to use, what kind of indexes create?

Not that I'm suggesting to enforce anything just for the sake of completeness ;) but I'm sure I've read something like that somewhere "If you have only heap tables and b-tree indexes, most probably your schema is not optimal".

Conclusion

Sometimes conceptual modeling is called logical and logical physical and then of course question remains how should be called the real "physical modeling", but it is not so important. The important thing is - whatever DBMS you choose for a new application it is worth to go through all the data model steps. Go through the conceptual modeling to better understand your requirements, to better confirm them with your customer, to better find out common patterns. Of course logical modeling cannot be avoided if we'd like to get at least one table, however quite many people forget about referential integrity constraints and rely only on application enforcing it. Sighhhh... And at last don't forget about physical modeling because when your table will contain (hundreds) millions of rows in production it would be much harder to change it's type, move it somewhere else etc.

2008-10-14T18:32:00.018+03:00

Mortgage calculator using SQL Model clause

Recently we had local Latvian Oracle day conference. This year it was quite big event with 5 parallel sessions and more than 400 participants. I also was one of the presenters telling about analytic functions and SQL Model clause. During the process of understanding deeper and better Oracle SQL Model clause I've tried make also some a bit complex examples myself. Nowadays the actual problem is mortgage and of course the possibility to pay monthly payments (fortunately not for me :) ) on the background of world wide financial crisis.
So here is fully functional mortgage calculator using SQL Model clause. There are two variants:

Adjustable rate mortgage with variable payment each month and

Fixed rate mortgage with fixed monthly payment.

I've used this site to get formula for fixed monthly payment required to fully amortize a loan over a term of fixed months at a fixed monthly interest rate.
OK and now the scripts.


set ver off                                                     
set lines 120                                                   
undefine rem_loan year_int_rate term                            
SELECT m+1 month,                                               
            to_char(rem_loan, '99999999.00') rem_loan,          
            to_char(loan_paid_tot, '99999999.00') loan_paid_tot,
            to_char(mon_int, '999999.00') mon_int,              
            to_char(tot_int, '99999999.00') tot_int,            
            to_char(mon_paym, '99999999.00') mon_paym,          
            to_char(mon_paym_tot, '99999999.00') mon_paym_tot,  
            to_char(grand_tot, '99999999.00') grand_tot         
FROM dual
MODEL    
DIMENSION BY (-1 m)                                                                         
MEASURES (&&rem_loan rem_loan,                                                              
  round(&&rem_loan*&&year_int_rate/100/12,2) mon_int,                                       
  ceil(&&rem_loan/&&term*100)/100 mon_paym,                                                 
  (&&rem_loan/&&term*100)/100 loan_paid_tot,                                                
  round(&&rem_loan*&&year_int_rate/12/100,2) tot_int,                                       
  ceil(&&rem_loan/&&term*100)/100 + round(&&rem_loan*&&year_int_rate/100/12,2) mon_paym_tot,
  ceil(&&rem_loan/&&term*100)/100 + round(&&rem_loan*&&year_int_rate/100/12,2) grand_tot    
)
RULES ITERATE (&&term) UNTIL (round(loan_paid_tot[iteration_number], 2) = &&rem_loan) (
  rem_loan[iteration_number] = rem_loan[iteration_number -1] - mon_paym[iteration_number - 1],       
  mon_int[iteration_number] = round(rem_loan[iteration_number]*&&year_int_rate/100/12,2),            
  mon_paym[iteration_number] = least(ceil(&&rem_loan/&&term*100)/100, rem_loan[iteration_number]),   
  loan_paid_tot[iteration_number] = loan_paid_tot[iteration_number - 1] + mon_paym[iteration_number],
  tot_int[iteration_number] = tot_int[iteration_number - 1] + mon_int[iteration_number],             
  mon_paym_tot[iteration_number] = mon_paym[iteration_number] + mon_int[iteration_number],           
  grand_tot[iteration_number] = grand_tot[iteration_number - 1] + mon_paym_tot[iteration_number]);

Result for loan of 1000 money units (lats, euros, dollars, pounds whatever) for 10 % year rate with term of 7 months looks as follows:


Enter value for rem_loan: 1000
Enter value for year_int_rate: 10
Enter value for term: 7

     MONTH REM_LOAN     LOAN_PAID_TO MON_INT    TOT_INT      MON_PAYM     MON_PAYM_TOT GRAND_TOT
---------- ------------ ------------ ---------- ------------ ------------ ------------ ------------
         0      1000.00       142.86       8.33         8.33       142.86       151.19       151.19
         1       857.14       285.72       7.14        15.47       142.86       150.00       301.19
         2       714.28       428.58       5.95        21.42       142.86       148.81       450.00
         3       571.42       571.44       4.76        26.18       142.86       147.62       597.62
         4       428.56       714.30       3.57        29.75       142.86       146.43       744.05
         5       285.70       857.16       2.38        32.13       142.86       145.24       889.29
         6       142.84      1000.00       1.19        33.32       142.84       144.03      1033.32

For fixed payment each month the script is as follows:


SELECT m+1 month,                                               
            to_char(rem_loan, '99999999.00') rem_loan,          
            to_char(loan_paid_tot, '99999999.00') loan_paid_tot,
            to_char(mon_int, '999999.00') mon_int,              
            to_char(tot_int, '99999999.00') tot_int,            
            to_char(mon_paym, '99999999.00') mon_paym,          
            to_char(mon_paym_tot, '99999999.00') mon_paym_tot,  
            to_char(grand_tot, '99999999.00') grand_tot         
FROM dual
MODEL    
DIMENSION BY (-1 m)                                                                 
MEASURES (&&rem_loan rem_loan,                                                      
  round(&&rem_loan*&&year_int_rate/100/12,2) mon_int,                               
  ceil(&&rem_loan*(&&year_int_rate/100/12*POWER((1+&&year_int_rate/100/12),&&term))/
    (POWER((1+&&year_int_rate/100/12),&&term)-1)*100)/100 -                         
    round(&&rem_loan*&&year_int_rate/100/12,2) mon_paym,                            
  ceil(&&rem_loan*(&&year_int_rate/100/12*POWER((1+&&year_int_rate/100/12),&&term))/
    (POWER((1+&&year_int_rate/100/12),&&term)-1)*100)/100 -                         
    round(&&rem_loan*&&year_int_rate/100/12,2) loan_paid_tot,                       
  round(&&rem_loan*&&year_int_rate/12/100,2) tot_int,                               
  ceil(&&rem_loan*(&&year_int_rate/100/12*POWER((1+&&year_int_rate/100/12),&&term))/
    (POWER((1+&&year_int_rate/100/12),&&term)-1)*100)/100 mon_paym_tot,             
  ceil(&&rem_loan*(&&year_int_rate/100/12*POWER((1+&&year_int_rate/100/12),&&term))/
    (POWER((1+&&year_int_rate/100/12),&&term)-1)*100)/100 grand_tot                 
)
RULES ITERATE (&&term) UNTIL (round(loan_paid_tot[iteration_number], 2) = &&rem_loan) (              
  rem_loan[iteration_number] = rem_loan[iteration_number -1] - mon_paym[iteration_number - 1],       
  mon_int[iteration_number] = round(rem_loan[iteration_number]*&&year_int_rate/100/12,2),            
  mon_paym_tot[iteration_number] = least(                                                            
    ceil(&&rem_loan*(&&year_int_rate/100/12*POWER((1+&&year_int_rate/100/12),&&term))/               
      (POWER((1+&&year_int_rate/100/12),&&term)-1)*100)/100,                                         
    rem_loan[iteration_number] + mon_int[iteration_number]),                                         
  mon_paym[iteration_number] = mon_paym_tot[iteration_number] - mon_int[iteration_number],           
  loan_paid_tot[iteration_number] = loan_paid_tot[iteration_number - 1] + mon_paym[iteration_number],
  tot_int[iteration_number] = tot_int[iteration_number - 1] + mon_int[iteration_number],             
  grand_tot[iteration_number] = grand_tot[iteration_number - 1] + mon_paym_tot[iteration_number]);

Result for the same loan of 1000 money units (lats, euros, dollars, pounds whatever) for 10 % year rate with term of 7 months looks as follows:


     MONTH REM_LOAN     LOAN_PAID_TO MON_INT    TOT_INT      MON_PAYM     MON_PAYM_TOT GRAND_TOT
---------- ------------ ------------ ---------- ------------ ------------ ------------ ------------
         0      1000.00       139.33       8.33         8.33       139.33       147.66       147.66
         1       860.67       279.82       7.17        15.50       140.49       147.66       295.32
         2       720.18       421.48       6.00        21.50       141.66       147.66       442.98
         3       578.52       564.32       4.82        26.32       142.84       147.66       590.64
         4       435.68       708.35       3.63        29.95       144.03       147.66       738.30
         5       291.65       853.58       2.43        32.38       145.23       147.66       885.96
         6       146.42      1000.00       1.22        33.60       146.42       147.64      1033.60

During my research I've found invaluable these resources:

Oracle official documentation;

Rob van Wijk blog;

Volder blog.

2008-07-14T23:42:00.002+03:00

One of 364 000 articles

According to google I've written another one of ~364 000 articles about SQL join types. So the question is why anyone should read it?
Here are a few reasons:

It is 25 pages long;

It contains join types metamodel or at least my attempt to create one;

It tries to describe what are different join types and how they relate to each other;

It contains visual pictures of cross, inner, and left, right, full outer joins and this time not with Venn diagramms, which I personally think are suitable for absolutely different operations (set operations);

To describe theory it contains 44 examples, which are tested on Oracle, SQL Server and MySQL;

It contains several links to other resources.

Because it is 25 pages I've decided not to put it here but it is on my website.

I have one wish though - if You are reading this article and have access to another DBMS (than already mentioned in article i.e. Oracle, SQL Server, MySQL) and have a bit free time, please run examples on your DBMS and post results either here or send me to mail gints.plivna@gmail.com. Then I will put them in original article along with contributor's name.

2008-07-05T16:28:00.008+03:00

Latvian Nationwide Song and Dance Celebration in Riga

Today is the beginning of 24th song festival - unique celebration of choral songs, which started 135 years ago with the first song festival. In these times under the rule of Russian Czar, later they continued during the short freedom period as well as even in Soviet times. This year about 35 000 singers and dancers will come to Riga to sing and dance in various concerts. This means more than 2% of all Latvians are coming together to sing and dance and many many more are coming to see them on the concerts or via TV.

Brief history

The first All Latvian Song Festival took place in Riga, year 1873. There were 1003 singers in these times. Under the Russian Czar there were 5 festivals. During the Latvian independence there were 4 festivals and participant count raised to more than 14 000 people. During the Soviet times these festivals partially become the weapon of Soviet propaganda, however these were like two-edged sword, for example, the 100th anniversary of song festival in 1973 was some kind of emotional uprising against Soviet regime. Since 1990 song festivals are held in independent state again and the participant count has raised to impressive more than 2% of all Latvians. The learning process of songs and dances is continued all the time and it is worth to mention that noone pays a single santīms (centime) to participants both for preparation process lasting a few years and also the very festival, although of course there is governmental material support for the organizational purposes of the festival.

Events

During the festival there are many concerts and other kind of events. I'll mention a few of them. The first is participant parade. In the picture below one can look at one of the best choirs in the world "Kamēr", which has won many international awards and also takes part in this festival.

The second one is main dance concert. In recent times more than 10 000 dancers are taking part in it.

The last and biggest event is the final concert with all the singers in one big choir. It is very impressive moment.

Some links

Gaismas pils (Castle of Light) - one of the favourite Song Festival songs of all time, one of the symbols of resistance against oppressors
The official website of Latvian Nationwide Song and Dance Celebration in Riga, July 5-12, 2008
Photo gallery of the current festival
History of Song Festivals

2008-06-27T15:55:00.054+03:00

Trip to Estonia

During my holidays I've been in a 4 days trip to Estonia. For Oracle adepts this is the country where Tanel Poder comes from ;)
The trip was made up by myself and my wife, during it we drove ~1700 km. The general idea was to drive around all Estonia (it is not very big country) and look at some (hopefully) most interesting places. Here are 26 most interesting photos:

A view from Munamagi hill, the highest point in Baltics

Tartu Dom Cathedral, it is being rebuilt now

Interesting fountain in Tartu, second largest Estonian city

Rainbow over Peipsi lake

There is 8 km long street through the old villages along the Peipsi lake. It looks like that all lengthways

Ivangorod castle over the river in Russia. This castle is just in the opposite side of Narva river from the Narva castle

Narva castle. This castle is just in the opposite side of Narva river from the Ivangorod castle

Aija along with bells in Narva castle

Northern coast (Gulf of Finland) mostly consists of limestone rocks up to 56 metres high

Sunset over Gulf of Finland

Valaste waterfall is the highest waterfall of Estonia

Estonia has some kind of oil shale, this hill is man-made and consists of slag of oil shales. It is 116 metres high. The picture was taken from a bit lower hill of the same origin, just older one.

Symbol of Rakvere town

Bikes at Palmse manor, which is in the very centre of Lahemaa national park, a beautiful sightseeing place

Jagala waterfall near from Tallinn

Tallinn has more than 2km long old city wall with many towers and everything in very well preserved

Tallinn old town

There was a bit rainy when we crossed over to Saaremaa island (actually Muhu island, which is connected to Saaremaa by dam). The sail is very short just ~20 minutes.

13th century painting in Liiva church, Muhu island

Maasilinn stronghold in Saaremaa

Angla windmills. There were multitude of windmills in Saremaa (~800) these are just a few of forever gone old might.

It is very easy to get to this Kaali meteoritic lake, it is just 17 km from Kuressaare, centre of Saaremaa

A passage in Kuressaare castle, Saremaa

Toilett of Middle Ages in Kuressaare castle, Saremaa

Organ of Kihelkonna church, Saremaa

Kihelkonna church, Saremaa. It was possible to climb up to the tower and look around through the windows

If you enjoyed these you can look also at some of my photos from Latvia.

2008-06-26T14:20:00.002+03:00

New improved price list

Obviously fluctuations of $ is affecting also Oracle ;) and as a result we have new global price list. Changes have affected also all Oracle database editions except Express, which is for free. So now Standard Edition One license is for 5 800 (was 4 995), Standard Edition is for 17 500 (was 15 000) and Enterprise Edition 47 500 (was 40 000) per processor. Personal Edition named user plus licence now is 460 US dollars (was 400). All new prices are here. These are the bad news.

The good news are that Oracle has created very informative Pricing and Licensing Rules, Tools page (at least I was not aware of that), which has many links to other valuable documents regarding Oracle pricing and licensing, for example, Oracle product prices in other currencies using Oracle exchange rates.

So get accustomed to new prices ;) and don't forget that there are other editions than Enterprise as well as there are such thing like term licensing (licenses for 1 till 5 years starting with 20% and ending with 70% of list price)!

2008-05-22T00:52:00.002+03:00

SQL Merge and Sequence gaps

I hope You my dear reader isn't one of the people cursed by requirement for gapless sequences ;)
Yesterday I just found another nail in the coffin for oracle sequence objects used as gapless sequence generators. Of course there are many causes not to do that, just like simple rollback, users changing their minds, network connection error, flushing cache etc. But I didn't know at least till yesterday, that MERGE statements generate gaps with quite big guarantee. It seems that Merge statement is generating the same number of calls for sequence next value as row count of source table or subquery.
Let's look at example:

CREATE TABLE target (pk number, data varchar2(10));
CREATE TABLE source (pk_s number, data_s varchar2(10));
INSERT INTO target VALUES (1, 'a');
INSERT INTO target VALUES (3, 'c');
INSERT INTO source VALUES (1, 'a');
INSERT INTO source VALUES (2, 'b');
INSERT INTO source VALUES (3, 'c');
INSERT INTO source VALUES (4, 'd');
COMMIT;
CREATE SEQUENCE seq START WITH 5;

So now we have 2 rows in target table, and 4 rows in source table. We'll compare them using data and data_s column. We'd insert only two rows, because another 2 already exists and even won't use WHEN MATCHED clause for update.

MERGE INTO target
USING source
ON (data = data_s)
WHEN NOT MATCHED THEN
INSERT VALUES (seq.nextval, data_s);
SQL> SELECT * FROM target;
PK DATA
---------- ----------
1 a
3 c
7 d
8 b

Although the sequence should have started from 5 new pk values are 7 and 8. Let's add another row into source and look what happens:

INSERT INTO source VALUES (5, 'e');
MERGE INTO target
USING source
ON (data = data_s)
WHEN NOT MATCHED THEN
INSERT VALUES (seq.nextval, data_s);
SQL> SELECT * FROM target;
PK DATA
---------- ----------
1 a
3 c
7 d
8 b
13 e

So now the value is 13 i.e. 8 + count of rows from source table (5). Now let's add another row and delete all previous.

INSERT INTO source VALUES (6, 'f');
DELETE source WHERE pk_s <=5;
MERGE INTO target
USING source
ON (data = data_s)
WHEN NOT MATCHED THEN
INSERT VALUES (seq.nextval, data_s);
SQL> SELECT * FROM target;
PK DATA
---------- ----------
1 a
3 c
7 d
8 b
13 e
14 f

As we can see pk is incremented only by 1.
After my experiments I found also metalink note "Merge Increments SEQUENCE.NEXTVAL for Both Insert and Update" (Note:554656.1), which also assures this as expected situation and not bug. So another reason not to worry about sequence values, just uniqueness of them ;)

2008-05-16T18:16:00.002+03:00

Minus All and Intersect All

Today I rediscovered Oracle Mix ideas and remembered my surprise that there are also Minus all and Intersect all operators in SQL standard just like Union and Union all. I'll briefly remind You what exactly they were. Let's imagine we have two sets:

T1: 1, 2, 2, 2, 3, 4, 4
T2: 2, 3, 4, 4, 4, 5

Then T1 INTERSECT T2 is 2, 3, 4
T1 INTERSECT ALL T2 would be 2, 3, 4, 4

T1 MINUS T2 is 1
T1 MINUS ALL T2 would be 1, 2, 2

So by default Intersect and Minus have DISTINCT operator removing all non-unique values. But in case we add keyword ALL then all values and their cardinalities are counted as well. I think these operators could be quite handy in process of analysing two data sets and finding what exactly the difference is. So if You are thinking the same way then I kindly ask you to support my idea about intersect all and minus all with your voice in Oracle Mix.

Thank You in advance for any decision :) and I hope sooner or later (better sooner of course ;) to find that in New features guide.

2008-05-07T19:49:00.007+03:00

DBMSes are different...

Recently had to do some work with SQL Server. As my experience with it is quite limited I've got a few interesting impressions, which probably are absolutely trivial for SQL Server adepts :)

Do you know that SELECT query can create new tables?

OK in SQL Server it can. So I had to create backup of a table and firstly was quite frustrated not finding anything like CTAS (CREATE TABLE ... AS SELECT). I was on a wrong track however. The right syntax is SELECT INTO as in following example:

CREATE TABLE persons (
prs_id INT IDENTITY NOT NULL PRIMARY KEY,
prs_name VARCHAR(30) NOT NULL) ;

INSERT INTO persons (prs_name) VALUES ('GINTS');
INSERT INTO persons (prs_name) VALUES ('JOE');

SELECT *
INTO persons_bkp
FROM persons;

SELECT * FROM persons_bkp;
-------------------------
1 GINTS
2 JOE

So till now I was under the impression that SELECT is DML statement however obviously that's not true at least for SQL Server.

Do you know that SQL Server can handle recursive queries?

At least since version 2005 it can. Long ago I have heard that SQL Server has nothing similar to START WITH CONNECT BY. This site comparing SQL Server and Oracle seemed to approve my thoughts. However DBMSes are different and one has to look for different ways accomplishing the same goals. And in SQL Server one can use Common table expressions (aka query factoring clause or with clause in Oracle world) to get hierarchical results. And BTW here SQL Server "outfunctions" Oracle because it allows to recursively reference the same common table in its definition, which is not allowed by Oracle. Here are two articles if you are interested in how exactly it is done:

So in conclusion these two features which are possible in both DBMSes, but are implemented in completely different ways supports my previous article that all databases are different :)

2008-01-30T23:39:00.000+02:00

The year has gone..

..since I started blogging. Actually I started it because of real anger against Oracle. The reason was one of the most famous problems in OTN history - inability to change e-mails ;) However about a month or two later it was resolved.
It seems the last year Oracle has made advances towards the community and:
1) listened to it (above mentioned case) as well as for example the case with AWR and ASH licensing. Probably all these steps were already on their way but I'm quite sure blog entries accelerated the result.
2) created public resources (oracle wiki) probably not veeeeery actively used until now.
3) granted free access for active members of public community to Oracle open world.
4) most probably made other positive steps I cannot remember or simply haven't heard about.

Last year was also big improvement in so called Oracle blogosphere - orana.info was founded along with some other aggregators, which can be found on oracle wiki.

For me it was a year of realizing the fact that there are tremendous resources about databases and Oracle in English, but just few in my mother tongue Latvian. So as I don't think English is the only language in the world even for databases :) I started to blog about databases in Latvian. Quite a bit of free time goes there.

So in my little anniversary I wish Oracle remember and support its community (probably sponsor a trip to Open world next year for bloggers khe khe :))) and myself to continue work about databases and Oracle resources especially in Latvian!

2008-01-25T12:12:00.006+02:00

SQL Join Types

New enhanced 25 pages long article with 44 examples and visual pictures describing SQL join types can be found here.

I'm studying a bit SQL joins and trying to make clear at least for me what they are and what the relationships among them are. As I've always thought that picture is worth hundred words I've tried to create ER diagram describing metainfo about joins. So the question for everyone is - is this diagram OK? And aren't there obvious bugs? :)
Probably from the strict viewpoint of Set theory and relational algebra it is not OK, but I'm more concerned of practical SQL usage and understanding. In case you know similar diagramm or whatever else visual explanation of relationships among join types please leave comment with link.
So the diagram is as follows (please click on it to get bigger):

Now for those that aren’t familiar with all these join types and terms in diagram above I'll try to give SQL examples (without data and results) for each one of them. If you'd like to have also examples with data cehck one of the links at the very end of this post.
Assume we have 2 tables - TableA (id, title) and TableB (id, description).
What the result are for each one of them you can find in other places for example:

Cross join:

SELECT * FROM TableA CROSS JOIN TableB

Join with restriction:

It is either natural join or qualified join. Each join with restriction has one cross join which is the degraded case of it.

Natural join:

Natural joins can be Inner or one of Outer joins, but they always are equi joins. Without explicitly specified inner or outer it is inner join. More about natural joins in my previous blog post.

SELECT * FROM TableA NATURAL JOIN TableB

Qualified join:

Qualified joins can be written either specifying join columns (named columns join) or explicitly writing join condition.

Join type:

Could not find anything better to somehow unite inner and various outer joins under one roof.

Inner join:

SELECT * FROM TableA NATURAL INNER JOIN TableB
SELECT * FROM TableA INNER JOIN TableB USING (id)
SELECT * FROM TableA INNER JOIN TableB ON (tablea.id = tableb.id)

Outer join:

Outer join is one of left, right or full outer join.

Left outer join:

SELECT * FROM TableA NATURAL LEFT OUTER JOIN TableB
SELECT * FROM TableA LEFT OUTER JOIN TableB USING (id)
SELECT * FROM TableA LEFT OUTER JOIN TableB ON (tablea.id = tableb.id)

Right outer join and Full outer join:

Take the same examples as from left outer join and just replace left with right or outer.

Qualified join expression:

There are two ways how more explicitly write join condition either specifying join columns or explicitly writing join condition.

Named columns join:

This join is written with "USING" clause. Named columns join always is equi join.

SELECT * FROM TableA INNER JOIN TableB USING (id)

Conditional join:

This join is the real one that I'd like to suggest using. With explicitly specifying what the join condition is. This also is the only join where one can use not only equi join but also another predicate operator than "=".

SELECT * FROM TableA INNER JOIN TableB ON (tablea.id = tableb.id)

Predicate operator type:

Based on predicate operator type joins can be classified as either equi or theta (nonequi) joins.

Equi join:

Equi join is join where operator used in join condition is equivalence ("=").
Both natural joins and Named columns joins are always equi joins. Examples:

SELECT * FROM TableA NATURAL LEFT OUTER JOIN TableB
SELECT * FROM TableA FULL OUTER JOIN TableB USING (id)
SELECT * FROM TableA INNER JOIN TableB ON (tablea.id = tableb.id)

Theta (Nonequi) join:

Equi join is join where operator used in join condition is something other than equivalence ("=").

SELECT * FROM TableA INNER JOIN TableB ON (tablea.id <>

I've not showed here such beasts like UNION JOINS, SEMI JOINS and ANTI JOINS because the first one (UNION join) is neither in SQL standard nor in any database I worked with.
SEMI JOINS and ANTI JOINS are somehow less interesting for me.

Further reading:

SQL und relationale Algebra (in German with for everyone understandable examples)
Relational algebra, joins and join like operation from wikipedia
Why I think natural joins are most unnatural joins ever!

2008-01-13T14:26:00.000+02:00

Some photos

I've been tagged by APC. Usually I don't respond to chain letters but this will be a bit of exception. However I won't tag anyone further. Looking at usual curses happening after NOT forwarding chain letters I hope all my Oracle installations won't silently transform to SQL Server ones ;)
So I won't tell anything about me, I'll just publish some of my best photos taken in various places in Latvia.


Lilacs in Dobele garden	Garden if front of National Opera

Nest of Lesser Spotted Eagle	Airing trail in heath

A lake in fog	Sun set view over my country landed property

Open air museum in Atte	Easter eggs by my wife(colouring of course :)

National Opera House in Riga	Rundale Palace

Riga Dom Cathedral - biggest church in Baltics	House Of Blackheads and St.Peter church

Railroad bridge at night	Art Noveau in Riga

2008-01-10T22:08:00.001+02:00

MINUS ALL and INTERSECT ALL in Oracle Revisited

Yesterday I wrote an article explaining that MINUS ALL and INTERSECT ALL is not possible in Oracle.
However today after a bit thinking I got insight - multiset operations! Yes these should be the right thing!
Since 10g Release 1 Oracle supports MULTISET UNION/EXCEPT/INTERSECT for both ALL/DISTINCT. I have only to make some types, a bit type casting and everything should be ok!

So I've started with the simple tables I've shown yesterday.


CREATE TABLE t1 AS SELECT mod(rownum, 1000) rn
FROM dba_source WHERE rownum <=50000;
CREATE TABLE t2 AS SELECT * FROM t1;

Now I can query both tables and try to compare then. To use multiset operations I need to cast subquery as nested table. To be able to use such type casting I have to create two additional types. The first one is simple Object type containing records I'd like to compare. Second one is object table type, to which I'll cast my subqueries. So they are as follows:


CREATE TYPE t1_type
AS OBJECT ( rn NUMBER);
/
CREATE TYPE t1_tab_type
AS TABLE OF t1_type;
/

So far so good. Let's start building queries step by step. The first one just builds nested table from subquery.


SQL> select cast(multiset(
  2    select rn from t1 where rownum <=3
  3  ) as t1_tab_type)
  4  from dual
  5  /

CAST(MULTISET(SELECTRNFROMT1WHEREROWNUM<=3)AST1_TAB_TYPE)(RN)
---------------------------------------------------------------
T1_TAB_TYPE(T1_TYPE(1), T1_TYPE(2), T1_TYPE(3))

The second one already uses multiset intersect to get desired result.


SQL> select col1 multiset intersect distinct col2
  2  from (
  3    select cast(multiset(
  4      select rn from t1 where rownum <=3
  5    ) as t1_tab_type) col1,
  6    cast(multiset(
  7      select rn from t2 where rownum <=3
  8    ) as t1_tab_type) col2
  9    from dual
 10  )
 11  /

COL1MULTISETINTERSECTDISTINCTCOL2(RN)
---------------------------------------------------
T1_TAB_TYPE(T1_TYPE(1), T1_TYPE(2), T1_TYPE(3))

Now we just need to add one final step - cast nested table rows back to normal rows. Here us helps select from table construct:


SQL> select * from table (
  2    select col1 multiset intersect distinct col2
  3    from (
  4      select cast(multiset(
  5        select rn from t1 where rownum <=3
  6      ) as t1_tab_type) col1,
  7      cast(multiset(
  8        select rn from t2 where rownum <=3
  9      ) as t1_tab_type) col2
 10      from dual
 11    )
 12  )
 13  /

        RN
----------
         1
         2
         3

Voila! Everything works!
However then I was torn with doubt - how it will work when not used as a toy but on real volumes? I created million rows worth table with wide column and got - yeahh I got following:

ORA-22813: operand value exceeds system limits

Not pleasing.

Then I started to play with the tables defined above and got quite awful results. What is the performance of normal MINUS set operator?


SQL> select * from t1
   2 minus
   3 select * from t2 where rn <=998;
        RN
----------
       999
Elapsed: 00:00:00.10

What is the equivalent for multiset except distinct? I was a bit unpleasantly surprised again:


SQL> select * from table (
   2   select col1 multiset except distinct col2
   3   from (
   4     select cast(multiset(
   5       select rn from t1
   6     ) as t1_tab_type) col1,
   7     cast(multiset(
   8       select rn from t2 where rn <= 998
   9     ) as t1_tab_type) col2
   10     from dual
   11   )
   12 )
   13 /
           RN
   ----------
          999
   Elapsed: 00:01:17.98

However for multiset except all I was simply shocked!


SQL> select * from table (
   2   select col1 multiset except all col2
   3   from (
   4     select cast(multiset(
   5       select rn from t1
   6     ) as t1_tab_type) col1,
   7     cast(multiset(
   8       select rn from t2 where rn <= 998
   9     ) as t1_tab_type) col2
   10     from dual
   11   )
   12 )
   13 /
          RN
  ----------
         999
         ...
         999
  50 rows selected.
  Elapsed: 00:06:05.39

6 minutes for 2 * 50K rows!? Hmm is it implemented using bubble sort or what? :O

So normal SQL minus needed 0.10 seconds, multiset minus distinct 1 minute 18 seconds, but multiset except all more than 6 minutes.

Yeahh I got another strong argument NOT to use nested tables and their operations ;)

2008-01-09T18:51:00.000+02:00

MINUS ALL and INTERSECT ALL in Oracle

Recently I was a bit studying SQL standard and a lot to my surprise found that there exists not only UNION ALL set operator, but also INTERSECT ALL and EXCEPT ALL (in Oracle version it would be MINUS ALL).

So you'd say Oracle haven't them and it is also (although partially - somehow minus has been forgotten) documented in Oracle Support for Optional Features of SQL/Foundation:2003.

Hmm, yea but a while ago I was doing several data migrations and then such set operators would really helped me. Now a bit what these non-existant (at least for Oracle) set operators are doing?

Imagine we have tables T1 and T2 with data as follows:

T1: 1, 2, 2, 2, 3, 4, 4

T2: 2, 3, 4, 4, 4, 5

So what is the result for SELECT * FROM T1 MINUS SELECT * FROM T2?

What would be the result for SELECT * FROM T1 MINUS ALL SELECT * FROM T2?

1, 2, 2

What is the result for SELECT * FROM T1 INTERSECT SELECT * FROM T2?

2, 3, 4

What would be the result for SELECT * FROM T1 INTERSECT ALL SELECT * FROM T2?

2, 3, 4, 4

As you can see ALL for EXCEPT/MINUS and INTERSECT retains cardinality and doesn't keep only unique records. Let's hope Oracle someday will stop this gap and give us this functionality :)

2007-12-08T22:29:00.000+02:00

Oracle wiki

I've thought for a while about creating an article about the resources available in Oracle problem solving process. Meanwhile Oracle wiki was born and already exists for about two months. As it is supposed to be shared and editable resource by everyone in Oracle community I thought that it would be a perfect place for my eventual article for two reasons:
1) more people hopefully will read it and probably get some help;
2) more people can add their suggestions how they solve Oracle related problems.
So my first contribution to Oracle wiki is How to solve Oracle Database related problems. Everyone is invited to correct it and/or add some other resources. As well as for other Oracle wiki pages of course.

2007-11-26T23:43:00.000+02:00

How to choose database

Most of us (I mean readers of Oracle related blogs) most probably have already decided which DBMS to use. However the world is changing and sooner or later either you'll have to choose a new DBMS or defend your favourite one.
Quite often we hear such argument like DB X is cheaper than DB Y, therefore we have to take this one. And cheaper usually means only license cost. However license cost is only one part of direct expenses not to speak even about indirect expenses.
So i've written an article trying to explain more than 10 criteria which should be analyzed before final decision. Article contains also a list of more than 30 databases with links to their sites one can choose from.

2007-11-04T19:34:00.000+02:00

The curse of gapless sequences

Gapless sequences are one of the best examples of bad requirements. Initially it was thought as an article for my blog but the result was too big and I've put it on my site.

2007-10-25T15:12:00.000+03:00

Hints "stronger" than db parameters?

Actually I've tested only one parameter so the plural in title is questionable generalization :)

Recently battling with ORA-04030: out of process memory errors I found at least for me quite interesting fact.
USE_HASH hint is more powerful than parameter _hash_join_enabled = false (for 10g) or hash_join_enabled = false (for 9i).
I was quite suprised because without this test case I'd thought completely opposite. Of course neither hash_join_enabled nor _hash_join_enabled should be often altered to different value than default "true", but in case of "false" and your code having hints use_hash, you'll have hash joins regardless of above mentioned db parameter setting. It might be quite nasty surprise in case you have eliminated hash joins because of possible bugs, for example.
I've tested this behaviour for 10.2.0.1.0 SE and 10.2.0.1.0 EE on Windows as well as 10.2.0.3.0 SE on AIX and 9.2.0.7.0 EE on windows, all had the same behaviour.
Here is my test case and 10046 event trace execution plan results on 10.2.0.1.0 SE/Windows (all other tests were similar):


SQL> select * from v$version;
BANNER
----------------------------------------------------------------
Oracle Database 10g Release 10.2.0.1.0 - Production
PL/SQL Release 10.2.0.1.0 - Production
CORE 10.2.0.1.0 Production
TNS for 32-bit Windows: Version 10.2.0.1.0 - Production
NLSRTL Version 10.2.0.1.0 - Production

SQL> create table a as select rownum rn, object_name
  2  from all_objects;

Table created.

SQL> alter session set events '10046 trace name context forever, level 1';

Session altered.

SQL> select count(*) from (
  2    select *
  3    from a a1, a a2
  4    where a1.rn = a2.rn);

  COUNT(*)
----------
     60141

Execution plan:
STAT #24 id=1 cnt=1 pid=0 pos=1 obj=0 op='SORT AGGREGATE (cr=556 pr=73 pw=0 time=212727 us)'
STAT #24 id=2 cnt=60141 pid=1 pos=1 obj=0 op='HASH JOIN (cr=556 pr=73 pw=0 time=406508 us)'
STAT #24 id=3 cnt=60141 pid=2 pos=1 obj=176936 op='TABLE ACCESS FULL A (cr=278 pr=73 pw=0 time=61652 us)'
STAT #24 id=4 cnt=60141 pid=2 pos=2 obj=176936 op='TABLE ACCESS FULL A (cr=278 pr=0 pw=0 time=60192 us)'


SQL> alter session set "_hash_join_enabled" = false;

Session altered.

SQL> select count(*) cnt from (
  2    select *
  3    from a a1, a a2
  4    where a1.rn = a2.rn);

       CNT
----------
     60141

Execution plan:
STAT #12 id=1 cnt=1 pid=0 pos=1 obj=0 op='SORT AGGREGATE (cr=556 pr=0 pw=0 time=374302 us)'
STAT #12 id=2 cnt=60141 pid=1 pos=1 obj=0 op='MERGE JOIN (cr=556 pr=0 pw=0 time=470826 us)'
STAT #12 id=3 cnt=60141 pid=2 pos=1 obj=0 op='SORT JOIN (cr=278 pr=0 pw=0 time=88667 us)'
STAT #12 id=4 cnt=60141 pid=3 pos=1 obj=176936 op='TABLE ACCESS FULL A (cr=278 pr=0 pw=0 time=44 us)'
STAT #12 id=5 cnt=60141 pid=2 pos=2 obj=0 op='SORT JOIN (cr=278 pr=0 pw=0 time=231092 us)'
STAT #12 id=6 cnt=60141 pid=5 pos=1 obj=176936 op='TABLE ACCESS FULL A (cr=278 pr=0 pw=0 time=60174 us)'

SQL> select count(*) cnt from (
  2    select /*+ full(a1) full(a2) use_hash(a1 a2)*/ *
  3    from a a1, a a2
  4    where a1.rn = a2.rn);

       CNT
----------
     60141

Execution plan:
STAT #24 id=1 cnt=1 pid=0 pos=1 obj=0 op='SORT AGGREGATE (cr=556 pr=73 pw=0 time=214116 us)'
STAT #24 id=2 cnt=60141 pid=1 pos=1 obj=0 op='HASH JOIN (cr=556 pr=73 pw=0 time=399462 us)'
STAT #24 id=3 cnt=60141 pid=2 pos=1 obj=176936 op='TABLE ACCESS FULL A (cr=278 pr=73 pw=0 time=60803 us)'
STAT #24 id=4 cnt=60141 pid=2 pos=2 obj=176936 op='TABLE ACCESS FULL A (cr=278 pr=0 pw=0 time=60188 us)'

2007-10-17T23:41:00.000+03:00

All latvian Oracle (and not only) users - UNITE!

I'd like to announce two resources for people that like relational databases and speak Latvian.
The first one is discussion group for Latvian Oracle User Group. The near future plans are described here (in Latvian).
If you speak Latvian and work with Oracle then it is your duty :) to join it! Of course one can join even if he doesn't speak Latvian, however there will be a small problem - everyone will understand you, but it is doubtful that you will understand anything except some keywords like Oracle, SQL, etc :)

The second resource is mine. And it is more general, it is blog about relational databases (in Latvian).
Recently I've searched Internet and was very unpleasantly surprised - I've found very few resources devoted to relational databases. So not to make only complaints I've started to blog about relational databases in my mother tongue. I'm of course not living in illusions that I can do it only myself and therefore anyone that is at least a little bit eager to help me in this process and has some knowledge either in general SQL, Oracle, MS SQL Server, MySQL, PosgreSQL or another relational db is encouraged to join me and become author in this blog.

2007-10-02T11:43:00.000+03:00

Natural joins are evil

Motto:
If you like time bombs use them ;)

During my teaching courses I once again stumble upon ANSI SQL NATURAL JOIN syntax. And once again I suspect that creators of them had in mind something like how to write unmaintainable code. Or probably they liked time bombs ;)
OK for those unfamiliar with Natural joins let's firstly see what they mean:
"A natural join is based on all columns in the two tables that have the same name. It selects rows from the two tables that have equal values in the relevant columns."
So it joins two tables if column names are the same. There isn't any explicit join condition. And here is the biggest problem. The work has been done behind the scenes, and that is one of the biggest advantages of writing unmaintainable code.
So let's assume we have following tables:
CREATE TABLE persons (
person_id NUMBER(10) NOT NULL,
address_id NUMBER(10) NOT NULL,
surname VARCHAR2(40) NOT NULL,
first_name VARCHAR2(40) NOT NULL,
remarks VARCHAR2(1000),
CONSTRAINT person_pk PRIMARY KEY (person_id));

CREATE TABLE addresses (
address_id NUMBER(10) NOT NULL,
country VARCHAR2(40) NOT NULL,
city VARCHAR2(40),
text VARCHAR2(100) NOT NULL,
CONSTRAINT address_pk PRIMARY KEY (address_id));

ALTER TABLE persons ADD CONSTRAINT person_address_fk
FOREIGN KEY (address_id) REFERENCES addresses (address_id);

Lets add a few rows:

INSERT INTO addresses
VALUES (1, 'LATVIA', 'RIGA', 'BRIVIBAS STREET 100');

INSERT INTO persons
VALUES (1, 1, 'PLIVNA', 'GINTS', 'Actually he lives in another place');

So what would be the SQL query to get persons and their countries using NATURAL JOIN syntax?

SQL> SELECT surname, first_name, country
2 FROM persons
3 NATURAL JOIN addresses;
SURNAME
----------------------------------------
FIRST_NAME
----------------------------------------
COUNTRY
----------------------------------------
PLIVNA
GINTS
LATVIA

1 row selected.

SQL>

Nice one - we got one row as expected.
So assume times goes by and as per our new requirements we have to add some description field also in addresses table. To be consistent and following practices that differ from how to write unmaintainable code we use the same column name and data type as in persons.

ALTER TABLE addresses ADD remarks VARCHAR2(1000);

So what our query is showing now?

SQL> SELECT surname, first_name, country
2 FROM persons
3 NATURAL JOIN addresses;

no rows selected

SQL>

WoW! Where the row has gone? We didn't change our data!
Yeahhhh, everything works as expected. We added column remarks with the same name as in persons table and now join is being done using both columns. And of course data in persons.remarks column doesn't match with addresses.remarks. So our time bomb has blown up and we silently introduced a nasty bug in our system. Code is as valid as before from the compilation viewpoint, so we won't get compilation errors. It simply gives wrong result. Invalid SQL would be much better, then we'd spotted bug much easier, but this time we got exactly very nasty and in the beginning absolutely unclear bug.

There are also other things to consider:
1) natural joins between two tables without at least one common column results in cartesian product.
CREATE TABLE numbers AS
SELECT rownum id
FROM all_source;

Lets see how many rows table numbers have and how many rows we got for natural join between them:
SQL> SELECT count(*) FROM numbers;
COUNT(*)
----------
549290

1 row selected.

SQL> SELECT count(*) FROM (
2 SELECT * FROM persons
3 NATURAL JOIN numbers);
COUNT(*)
----------
549290

1 row selected.

SQL>

2) with different data types but the same column names there might be type conversion problems and quite strange error messages (see that it is NATURAL OUTER LEFT or RIGHT JOIN):

ALTER TABLE addresses MODIFY remarks NUMBER (15);

SQL> SELECT surname, first_name, country
2 FROM persons
3 NATURAL LEFT JOIN addresses;
SURNAME
----------------------------------------
FIRST_NAME
----------------------------------------
COUNTRY
----------------------------------------
PLIVNA
GINTS

1 row selected.

SQL> ed
Wrote file afiedt.buf
1 SELECT surname, first_name, country
2 FROM persons
3* NATURAL RIGHT JOIN addresses
SQL> /
SELECT surname, first_name, country
*
ERROR at line 1:
ORA-01722: invalid number

SQL>

So what to do?
1) Do not use this syntax. This is the most efficient and most obvious solution.
2) Use the same schema object naming conventions as me :) Natural joins simply never work using my favourite naming conventions.

Similar threads:
1) Natural Join and Index Monitoring in asktom.
2) Thread should one use ANSI join syntax when writing an Oracle application from Oracle-l list.

2007-09-16T01:34:00.000+03:00

Avoiding trouble

Following post could be seen absolutely obvious mechanism to avoid unnecessary trouble in development process. But it was something new for me when several years ago I've firstly read it from the long term oracle-l list ex member Rachel Carmichael (unfortunately I've lost this e-mail) and now I've seen it is something new for many developers. In one sentence it is:
Inform your boss (customer, colleague or other responsible person) about possible trouble in a written and provable way.
OK now let's explain a bit more. There are three steps in this process:
1) Understand possibility for eventual trouble;
2) Fight for your right decision;
3) If you lost the battle inform responsible person about it explaining why you think there is possibility for trouble.
Let's see what each of these steps mean.

Understand possibility for eventual trouble

Here is the trickiest part. The first tricky thing is to understand that there might be possible trouble. Of course that comes with experience ;) Probably the first time you'll miss the eventual trouble but next time you'll be already smarter. The second tricky thing is to avoid never ending wretch syndrome. And here of course the main factor is your common sense - decide when to warn responsible person and when to not use false warnings.

Fight for your right decision

Almost always there is a possibility to fight for your decision. Of course here is also one tricky thing - you have to explain and motivate your decision. You can enforce your decision with test cases, examples, resources from books, Internet etc. You can find supporters among your colleagues and sometimes even usual opponents. Even if you'll loose at least you'll have clear conscience :)

Inform responsible person about it

This is the easy part. If you have decided that you should warn anybody then most of the time there is no need to seek for a notary. Usually a simple e-mail saved in your sent mails will be enough. You can of course consider possibility to add some more recipients to TO: or CC: fields. If it is a meeting protocol with customer a footnote with warning can be a solution. Of course if the possible consequences are quite serious you have to ensure that your insurance is at least as serious. One thing of course remains - the motivation has to be kept. Also in your written email, meeting protocol or whatever else.

The result

With above mentioned mechanism you've got two points:
1) Once again informed about your decision.
It might be possible that your boss simply missed your viewpoint. Probably he thought you are not serious enough. Probably he misunderstood you. Written text with appropriate explanation and motivation decrease such probability to the minimum.
2) Cover your a$$ (sorry for that expression but I couldn't resist :) in case of failure.
You will be able to point to your written message and say - I already warned you. Without it there are always place for different interpretations. Even the most ordinary thing - your boss probably simply fairly and honestly forgot about your warnings. With such cover even if you need to clean up the mess you won't pay for it. Your boss, your customer or whatever else responsible party being warned will pay for it.

Have I used that?

Yes, I have. I was lucky enough to have knowing and smart bosses till now so I haven't necessity to use such scenario (3rd step) inside my company. Either I could convince my bosses or the problem was out of my responsibility. But in communication with customers I've used this technique quite much along with catching their bad requirements. Sometimes that convinced them at last change their minds, sometimes not, but at least I and my company had very strong argument in case of failure.

2007-09-03T09:38:00.000+03:00

Unrealistic optimism or just pure stupidity?

A while ago I've written a blog entry Where bad performance starts explaining that we as IT professionals should prevent customers from stupid requirements or at least strongly warn them that their desires will make much harder life both for customers and developers.

Some days ago I've spotted a nice estimate of a work task that takes the problem described above to the next level i.e. the core of problem is initiated even before the real project begins. Task was much bigger containing of many subtasks including:
1) three separated subtasks for three different reports. Each report was quite precisely explained;
2) a task formulated as follows: "Up to 20 reports defined more accurately in analysis phase". No more information was supplied, no more information was known to anyone.

So how do you think - what were the estimates for analysis and implementation? According to estimate analysis of each separated report should take 1, 2 and 3 workdays. Implementation estimate of each of them was 2, 3.5 and 3.5 workdays. And now the most interesting part: analysis and implementation of up to 20 unknown reports was estimated 15 days for analysis and 20 days for implementation.

So for known and quite precisely defined reports the average analysis and implementation work was estimated 2 and 3 days respectively for each report. BUT for completely unknown reports without any further information and/or restriction estimate was 0.75 days for analysis and 1 day for implementation.
BAHHH!
So what to think? What was thinking both analyst and programmer making such estimates? What was thinking PM allowing such estimates to be published and giving them to customer?Are they afraid of big numbers? Why are they being so optimistic? It is a real mystery...
I'd be afraid to give any estimate for such purely defined task at all. And if I'd give I'd take at least average or even probably maximum of other similar tasks and also warn my boss about quite big risk to give estimates of unknown tasks. Probably I've seen too much projects fail or at least delay due to overly optimistic estimates...

2007-08-27T18:12:00.000+03:00

Oracle Express Edition and MS SQL Server Express Edition

Today I was listening to a course from the dark side ;) One quite small fact (I was not aware of) about SQL Server 2005 Express Edition compared to Oracle 10g XE caught my attention though.
These versions are similar in their limitations i.e.

1 CPU
1 GB RAM
4 GB user data

But there is one difference not in favour of Oracle - for SQL Server Express Edition one can create up to 16 instances on a single box, but for Oracle XE only one. Of course in general it is not the main decisive factor to choose between SQL Server and Oracle, but I can imagine some scenarios where more than one instance of Oracle XE on a single box would be nice, for example, simultaneous development of different applications on XE.

2007-08-21T21:31:00.000+03:00

A new experience

I've completed my first Oracle certified training course in Mebius Latvia. This time as a trainer though.
For already a few years I feel necessity somebody to tell what I've learned and how to do or not to do certain things in Oracle ;)

The first step was my website where I could put some articles telling some experience stories. Now I've completed the next step – directly telling people how to work with Oracle. The content of my first course as a trainer was quite trivial – 10g PL/SQL fundamentals. Although describing more than 250 slides in two days, even quite easy slides, was a completely new experience for me. I hope I've completed it more or less so that my trainees understood what I was talking about ;) In fact I was really satisfied with them, because all had more or less experience with Oracle and programming as such, so there wasn't necessity to tell for example what the "IF" statement and logical values are. In the very beginning I frankly told them that questions help not only trainees to deeper understand the covered material but questions also help the trainer to feel better and understand that at least some trainees are following the lesson material ;) So they really followed my plea and time after time we had even discussions about one or another topic. So I'm thankful to my trainees, I hope they understood at least main things I told them and I'm looking forward to my next course after a while.

2007-08-09T17:50:00.000+03:00

Flashback for source code

Recent OTN forum question made me think that I'm quite lucky. Everywhere I've worked we had one or another source control software for source code. All scripts for schema creation as well as procedural units were put and maintained there. Sometimes reading about other people's problems that they've lost the only (sic!) version of their procedure dropping the schema or just recreating the procedure gives me the shudders. So for everyone that yet hasn't understood - keeping procedural units only in the db means it is just matter of time when you'll lose them. Either all of them or just one because of erroneous create or replace but sooner or later the time will come ;)

OK but what if you are unlucky and just realized that you have overwritten the last and only version of your source code unit (procedure, package, function, etc)? If you are at least on 9i (10g as well of course) you are not completely lost. Flashback that works on regular user's tables works in the same way also on internal tables holding source code. So the only problem remains how to get it and how to get it fast! Fast because it is just like with lost extremities - surgeon can put it back only if you are acting fast. And fast because the old data can simply be overwritten if Oracle feels the necessity. And now what's the mechanism is?

There are 2 scenarios:
1. query directly from base tables owned by sys or
2. query using provided dba/all/user views i.e. dba/all/user_source.

In any case for ordinary user by default it is not possible. You need sys access to do that or grant privilege from sys to any other user. Even DBA role is not enough.

1st scenario is as follows:

1) get object_id (from dba/all/user_objects)
SELECT object_id
FROM user_objects
WHERE object_name = 'yourname'
AND object_type = 'yourtype';

2) get text from source$ as of timestamp
SELECT source
FROM sys.source$
AS OF timestamp timestamp_before_error
WHERE obj# = object_id_from_1st_step
ORDER BY line;

1st step can be done by everyone. 2nd step needs flashback on source$.

2nd scenario is as follows:

simply select text from user/all/dba_source as of timestamp.
SELECT text
FROM dba_source
AS OF timestamp timestamp_before_error
WHERE name = 'yourname'
AND owner = 'yourowner'
AND type = 'yourtype'
ORDER BY line;

Of course this can be done by sys. I was not able to do that by other user because user_source for example is based on source$, obj$ and fixed table x$joxfs. I was able to grant flashback on source$, obj$ to another user but for fixed tables it is not possible (quite understandable as these are actually memory structures). But it seems on the other hand without flashback one cannot make queries from view that is based both on ordinary tables and fixed tables.
Above experiments were performed both on 9.2.0.7.0 and 10.2.0.1.0 EE Oracle.

2007-06-15T01:19:00.000+03:00

gplivna.eu domain has lived already a year now

My provider has provided a wonderful feature - collecting various statistics about visitors of my site. I really suggest everyone to find such provider or at least use Google analytics to collect various stats about your site.

I started my site a bit more than a year ago on May 2006. In the first month I was the only visitor testing how it looks like and adding some content. In the second month (June 2006) I had already 65 unique ip addresses visited my site. It was a real advancement :) Compared that to today - this is normal weekday rate for my site. Of course on weekends attendance drops.

Statistics provided by my Internet provider gave me interesting possibility to create a world map. All countries having at least one person visiting my site has been coloured more or less black. More black - more visitors. Colouring the map was interesting task - I even had to look in world atlas for some countries, shame on me - but just using skeleton map I was not sure exactly which country Ghana, Dominican Republic and Guatemala are. The most lovely were countries like Indonesia, Canada, Philippines and Russia - having so many islands and each one has to be coloured separately.

So what are the lessons I've learned this year?

1. Don't make fun with Google! It can give you so many hits. And it also can be annoyed and give you nothing. Annoyance for my site was because of two reasons:
1) changing content too rapidly
2) having the same articles in 2 different formats - html and doc.
So the medicine was think more about content and change it only once and place deny on robots.txt for doc files.

2. Google webmasters tool is really useful. One can submit sitemaps, so decreasing the time for Google to catch your new page, view restricted pages by robots.txt and a bunch of other features.

3. All other search engines unfortunately at least for me gives almost nothing. Even yahoo is almost dead for my site.

4. Placing valuable content is as critical as thinking about titles and words in your articles at least for Google.

5. Writing articles that hopefully are usable also for other people aren't so easy therefore I even more admire people having written books with precise technical info like Thomas Kyte and Jonathan Lewis for example.

So one month May 2007 gave almost the same stats as all previous year together (actually half of the year because June was the first month for wide public). Let's hope the trend in the future will be the same :)

2007-05-07T23:04:00.000+03:00

Deferred constraint real life scenario

Motto:
It is worth to know oracle features even if you have never used them.

Some days ago one of our developers came in and started to tell that our testers found another bug. This time bug is raised in Oracle by violated unique key. Of course developer didn't catch it in unit testing because it was a bit tricky but our testers were smart and nasty :) enough to dig it out.
So we had a form that may have many child records for a previously chosen parent record. To feel something concrete let's choose for the parent an order for tax discount - suppose you have submitted it in written form to your local government (not that I'm suggesting you to do that :))). But of course your local government doesn't want to give you tax discount and therefore as always asks you many documents explaining reason for it. But there is one restriction - each document must be of different kind e.g. one proving that you have 5 children, second proving that your rent has risen by 25% etc.
So as I strongly believe that business constraints should be enforced in db as much as possible I've created UK in supporting documents for two columns:
1) FK to order table and
2) document type.
Unfortunately in the form where all documents are entered one can change all attributes and our evil testers of course found the most evil one - swapped type of document A to document B and document B to document A.
So where are the problem you ask? Yea there aren't any problem if one looks on the start point (docs A and B for one order) and finish point (docs B and A for the same order), but somehow situation in the middle is very unpleasant. There is a moment when we have 2 documents with type B simultaneously. And here testers got the error as below:

SQL> create table parent (par_id number not null constraint par_pk primary key);
Table created.
SQL> ed
Wrote file afiedt.buf
1 create table child (cld_id number not null constraint cld_pk primary key,
2 cld_par_id number not null constraint cld_par_fk references parent (par_id),
3 cld_type varchar2(3) not null,
4 cld_data varchar2(100),
5* constraint cld_uk unique (cld_par_id, cld_type))
SQL> /
Table created.
SQL> insert into parent values (1);
1 row created.
SQL> insert into child values (1, 1, 'A', 'Doc type A');
1 row created.
SQL> insert into child values (2, 1, 'B', 'Doc type B');
1 row created.
SQL> commit;
Commit complete.
SQL> update child set cld_type = 'B' where cld_id = 1;
update child set cld_type = 'B' where cld_id = 1
*
ERROR at line 1:ORA-00001: unique constraint (GINTS.CLD_UK) violated

So what to do?
My immediate response was - deferred constraints! I definitely knew that I don't want to loose the constraint forever because I'd like to enforce as much as possible integrity constraints in DB. So how it looks with deferred constraints?

SQL> alter table child drop constraint cld_uk;
Table altered.
SQL> ed
Wrote file afiedt.buf
1 alter table child add constraint cld_uk unique (cld_par_id, cld_type)
2* deferrable initially deferred
SQL> /
Table altered.
SQL> update child set cld_type = 'B' where cld_id = 1;
1 row updated.
SQL> update child set cld_type = 'A' where cld_id = 2;
1 row updated.
SQL> commit;
Commit complete.

Everything works!
How does it look like now?

SQL> select cld_id, cld_par_id, cld_type from child;
CLD_ID CLD_PAR_ID CLD
---------- ---------- ---
1 1 B
2 1 A

Let's try some more experiments. Let's update only one row back and try to commit:

SQL> update child set cld_type = 'A' where cld_id = 1;
1 row updated.
SQL> commit;
commit
*
ERROR at line 1:
ORA-02091: transaction rolled back
ORA-00001: unique constraint (GINTS.CLD_UK) violated

Bahhh!!! We got expected error!
The only thing - one has to remember that deferred constraints rolls back ALL transaction since it started even if the changed rows are completely valid and unrelated to the deferred constraint for example let's try to insert one completely valid row in parent table at first and then try to do nasty things with deferred constraint:

SQL> insert into parent values (2);
1 row created.
SQL> update child set cld_type = 'A' where cld_id = 1;
1 row updated.
SQL> select * from parent;
PAR_ID
----------
1
2
SQL> commit;
commit
*
ERROR at line 1:
ORA-02091: transaction rolled back
ORA-00001: unique constraint (GINTS.CLD_UK) violated
SQL> select * from parent;
PAR_ID
----------
1

Our inserted row into parent table also has gone. With normal (i.e. not deferred) constraint Oracle rolls back only changes made by the latest statement not all statements since transaction started. One has to remember this behaviour when using deferred constraints.

Quite popular use of deferred constraints seems to be update of primary keys (and after that also foreign keys) but as I'm always (OK almost always) using surrogates and primary key update for me is something absolutely awful I've never had such need and I'd never recommend that to anyone.

2007-04-15T22:40:00.000+03:00

Statistics is power! ... And the goat...

Hmmm, you say absolutely stupid title? OK let's see whether you think the same after you read all this stuff.

No I'm not going to write about statistics on Oracle tables, although of course we must have stats on Oracle tables for CBO to operate as well as possible. I'm going to write a bit about a very old problem that now is at least partially resolved. And particularly - inablity to change e-mail in Oracle forums. In my previous blog entry I expressed hope that sometime in the future I'd be able to write another entry about it. Now the time has come. To be precise it came already 3 weeks ago on March 25.

And I was agreeably surprised that the time came so fast. According to this thread the problem started at July 2002. Since then many people many times write about this problem in feedback forum but the best what we got were promises about the bright future. Just like Communism that according to Khrushchev had to be built by 1980 but never started. And surprise, surprise! Just about two months after I published my article summarizing some statistics and counting all the desires, complaints and statements about inability to change the e-mail in 500 most recent threads we got it!

You see - my ego is fantastically high! ;)

OK I'll go down some steps.

The most I'm dare to think of - is that my article was one of the drops that broke up the big rock. But of course important drop that probably showed people the real size of the problem.

You see - my ego is fantastically high anyway even after stepping down! ;)

So I personally learned at least one thing - statistics is power! Even outside CBO!

And where's the goat you ask?
OK here is an old story (most probably already known to you in some variation):

A poor man comes to the rabbi complaining that his family has only one small room, many kids, and almost no money.
The rabbi says, "Take all your money, buy a goat, and keep the goat in your room. Come back in a month."
"But, rabbi, we don't have enough space even for us," the man said
"Just do what I say," the rabbi replied.
A month later the man comes back complaining that the goat smells and breaks everything.
"Sell the goat and come back in a month," the rabbi tells him.
A month later the man comes back to the rabbi with flowers.
"Thank you, rabbi! We're so happy the goat is out, now we have more room and some money!"

By objective considerations rabbi was a bit of cheater and poor man of course was stupid.

So we had a goat in our room for more than four years - much longer than poor man. Of course after removing our own "goat" the feeling is quite releasing and pleasing but we have to understand that it was removing the goat instead of giving a present.

2007-03-26T12:18:00.000+03:00

Where bad performance starts

"My application is slow" - who haven't heard this very common complaint? Both from customers and developers (sic!) that created the application. No this won't be story how to diagnose what is slow and/or what to do to make it run faster. This will be a short story - how to try to not reach this complaint.

So where are the roots of bad performance? Many times the roots are very simple - allowing customer to promote and what's worse - accept bad requirements from customer. Bad and many times completely unnecessary requirements. The primary task of system analyst is of course to analyze, formalize, systematize and record customer requirements. But that should be done using sound criticism of initial requirements. And although the primary goal of specification is to record WHAT is needed and although specification should be implementation independent I'm sure that a really good system analyst has to keep in mind at least the vision HOW specified requirements will be implemented. Otherwise it is possible to result in a SRS describing business illogic instead of business logic (as was quite precisely named by a user in this thread).

Why illogical and unnecessary requirements are bad and should be thrown away instead of just implementing them - probably someone sometime would need them? There are at least following reasons to do that:
1) starting from requirements gathering through development and design until the testing and deployment development staff is wasting their time to implement them resulting in wasted man-hours and $$$,
2) every added functionality makes every application at least a little bit more harder to understand for customer and support which also results in wasted man-hours and $$$,
3) every added functionality needs some system resources either in CPU cycles, I/O calls or increased memory space. Everybody knows that resources are finite and so they simply are stolen from other users and useful functionality,
4) every added functionality usually takes some space on the screen. Of course it is not so critical today with common 17' and 19' monitors, but needless functionality many times can be as annoying as advertisements.

What to do to eliminate unnecessary requirements, if customer asks something crazy or you think that every other similar product already has it? There are at least following possibilities starting from the best:
1) frankly tell the customer that the requirement is bad explaining why it should be eliminated and what bad will happen if it would be created. You can even a bit overstate that :) This to my mind is the best solution, because even if customer and/or somebody else would force to implement you this particular bad feature you'd later have a wonderful (of course it should be written on paper or in e-mail!) argument that you have already predicted that!
2) use some scheme of prioritization, for example a nice one (just coincidence) called like very well known city - MoSCoW. I've never seen a project where all requirements REALLY are with the same priority. So mark the foolish requirements "could have", "won't have", "never have" (OK the latter is my own addition :). Thanks God even "could have" usually guarantees the "never have" because - how many software projects have you seen where the actual implementation is ahead of the schedule and budgeted expenses? :) So usually there is no time left for "could have's" that are ugly and undesirable from development perspective.
3) cash more $$$ for the particular feature. This is the most risky solution, because if customer is spiteful you'd cash him 3 times more than normal and would not deliver a good functionality. I in the place of customer would be 3 times angrier. And most of them are.

How to minimize the impact of the unnecessary requirement? If you have to implement it, of course you have to waste your and your customer resources to implement and accept it. The only possibility to minimize the impact is to make it configurable so that after the day (or week or hour) of intensive annoyance customer is able to switch it off. Of course this is also a solution for a pure marketing driven feature, which is absolutely necessary to sell your product but absolutely needless and harmful in real production.

What are examples of unnecessary functionality? Although very common and accustomed functionality of forums the precise threads count, view count and count of found searches actually is nothing more than satisfying the curiosity. What you'd lose if you'd don't know that this particular Database-General Oracle forum wouldn't display that it has Messages: 209,897 - Threads: 48,569 and searching for term Oracle it found 453 Messages?
Or this Oracle FAQ forum for SQL & PL/SQL Newbies has 44090 messages and 11735 topics and here searching for Oracle you'd get 84 results? Of course I assume that they are not counting these statistical numbers each time the user requests the page, but eventually this requirement can be very disastrous if implemented without caution. I really don't care whether the forum has 209,897 or 185,786 messages, probably the only thing that I care is to see the comparable activity that can be very approximate to judge in which forum there is more probability to get answer. And what I do care the most is to get the forum page as fast as possible. This doesn't mean I'm asking to remove this functionality. This means I'm explaining that for real work this functionality is not needed and if it takes significant additional time it is even harmful to my mind. The same can be said about returning the precise count of search results even if they are some quite big number. Do you really care whether your search resulted in 5000 or 5647 hits? I personally don't unless it is statistical report which is absolutely different subject than everyday search.

Another case study. Another quite common incorrectly used potentially very offensive feature is to put wildcards '%' around the search word. As a result we get potential resource intensive and long lasting full scan of table instead of minimal index scan. Actually I think it is really rare occasions when customer really needs to search for something like %A%. What does %A% really mean? A found document where the only symbol left in person name is "A"? A completely absentminded customer only remembering that his product name contains symbol "A"? Both customer and developer will be desperate in this situation - customer because the application is slow, developer because customer constantly complains that application is slow.

So as a result for the latest case study I thought of writing a basic article how to create indexes in Oracle and put it on my website but after searching the web I abandoned my idea, because so much articles have already been created and I doubt that mine would be better than already existing :) So here is a very small test case how to effectively search using wildcards either in the end (using the most basic index) or in the start (using function based index) but of course not both. So be aware in the following example how both searches for DBA_DB% and %DB_LINKS uses index range scan. This example was included mostly because I haven't seen any website mentioning the possibility to search using wildcard in the beginning of the string (probably haven't searched long enough :)

SQL> create table src (txt varchar2(100) NOT NULL);
Table created.

SQL> insert into src select distinct object_name from dba_objects;
22474 rows created.

SQL> create unique index src_uk1 on src(txt);
Index created.

SQL> create unique index src_uk2 on src(reverse(txt));
Index created.

SQL> exec dbms_stats.gather_table_stats(user, 'src')
PL/SQL procedure successfully completed.

SQL> set autot on explain

SQL> select * from src where txt like 'DBA_DB%';
TXT
--------------------------------------------------------------------------------
DBA_DB_LINKS
Execution Plan
----------------------------------------------------------
0 SELECT STATEMENT Optimizer=FIRST_ROWS (Cost=2 Card=1 Bytes=22)
1 0 INDEX (RANGE SCAN) OF 'SRC_UK1' (UNIQUE) (Cost=2 Card=1 Bytes=22)

SQL> select * from src where reverse(txt) like reverse('%DB_LINKS');
TXT
--------------------------------------------------------------------------------
ORA_KGLR7_DB_LINKS
DBA_DB_LINKS
ALL_DB_LINKS
USER_DB_LINKS
Execution Plan
----------------------------------------------------------
0 SELECT STATEMENT Optimizer=FIRST_ROWS (Cost=3 Card=1 Bytes=22)
1 0 TABLE ACCESS (BY INDEX ROWID) OF 'SRC' (Cost=3 Card=1 Bytes=22)
2 1 INDEX (RANGE SCAN) OF 'SRC_UK2' (UNIQUE) (Cost=2 Card=1)

And here are many links mentioning how to create indexes in Oracle and what one or another access path means:

Oracle® Database Concepts 10g Release 2 (10.2)
Chapter 5 Schema Objects
Overview of Indexes
Overview of Index-Organized Tables
Overview of Application Domain Indexes
http://download-uk.oracle.com/docs/cd/B19306_01/server.102/b14220/schema.htm#i5671

Oracle® Database Application Developer's Guide - Fundamentals 10g Release 2 (10.2)
Chapter 5 Using Indexes in Application Development
http://download-uk.oracle.com/docs/cd/B19306_01/appdev.102/b14251/adfns_indexing.htm

Oracle® Database Performance Tuning Guide 10g Release 2 (10.2)
Chapter 15 Using Indexes and Clusters
http://download-uk.oracle.com/docs/cd/B19306_01/server.102/b14211/data_acc.htm#i17778

Oracle® Database Concepts 10g Release 2 (10.2)
Chapter 18 Partitioned Tables and IndexesT
Overview of Partitioned Indexes
http://download-uk.oracle.com/docs/cd/B19306_01/server.102/b14220/partconc.htm#i461446

When to Use an Index by Cary Millsap
http://www.hotsos.com/e-library/abstract.php?id=5

Indexing options available in Oracle by Brian Hengen
http://download-uk.oracle.com/oowsf2005/137wp.pdf

Why isn't Oracle using my index ?! by Jonathan Lewis
http://www.jlcomp.demon.co.uk/12_using_index.doc

Index Internals by Julian Dyke
http://julian.dyke.users.btopenworld.com/com/Presentations/IndexInternals.ppt#28

Understanding Indexes By Tim Gorman
http://www.evdbt.com/2004_paper_549.doc

Oracle B-Tree Index Internals: Rebuilding The Truth by Richard Foote
http://www.miracleas.dk/images/upload/Docs/Richard%20Foote.pdf

Bitmap index Internals by Julian Dyke
http://julian.dyke.users.btopenworld.com/com/Presentations/BitmapIndexInternals.ppt

Understanding Bitmap indexes by Jonathan Lewis
http://www.jlcomp.demon.co.uk/03_bitmap_1.doc

Bitmap Indexes 3 - Bitmap Join Indexes by Jonathan Lewis
http://www.jlcomp.demon.co.uk/07_bitmap_3.doc - what are Bitmap Join Indexes and when these are useful.

Operations of execution plans by Julian Dyke
http://julian.dyke.users.btopenworld.com/com/Optimisation/Operations/Operations.html

Is this an index I use ? by Jonathan Lewis
http://www.jlcomp.demon.co.uk/index_usage.html - how to find indexes that are not used.

Rebuilding indexes – why, when, how? by Jonathan Lewis
http://www.jlcomp.demon.co.uk/indexes.doc

How high can you go ? by Jonathan Lewis
http://www.jlcomp.demon.co.uk/22_how_high.doc - what is the biggest possible height of a b-tree index?

Expert Oracle Database Architecture 9i and 10g Programming by Thomas Kyte
Chapter 11 Indexes
Table of contents
Book

Cost-Based Oracle Fundamentals by Jonathan Lewis
Chapter 4 Simple B-tree Access
Chapter 8 Bitmap Indexes
Table of contents
Book

2007-01-30T22:55:00.001+02:00

Five Million and Counting - WHAT???

Recently I got one of the regular mails from Oracle (called OTN TechBlast). Title of the first article was "Five Million and Counting" and it was written by Justin Kestelyn, Editor-in-Chief, OTN.
So what is he counting? First paragraph answers that question:
"I can hardly believe it, but OTN membership passed the 5-million benchmark in 2006. But why should you care?"

What is so remarkable in this statement? Yes of course it is a big number. Yes of course it is nice that people around the world are interested in Oracle and its products. But the question remains - What the author is counting?

And the question remains for the on big reason - OTN member cannot change his e-mail for already at least a few years (according to this post since about July 2002). It is surprising in nowadays when

people change their works quite often and the right thing is not to work all the life in one company;
companies merge, crash, change names, domains;
free e-mail servers come and go.

One of the important parts of OTN is forums. Among other forums there exists one called Feedback. The goal of the forum is defined as follows: "Use this forum for feedback about OTN services and problems with OTN accounts.". Actually nothing is said that someone will solve your problem... Anyway people obviously are hoping for the best and continually posting their problems. As the Justin Kestelyn, Editor-in-Chief, OTN hasn't gathered statistics of what the main problem for OTN members is (or at least hasn't indicated that in his article about 5 million members), I've tried to do this job myself.

The source of statistics were 500 most recent threads starting from Mar 6, 2006 6:04 PM till Jan 30, 2007 ~20:00 PM.

So 49 of 500 threads (or 9.8%) were only about this very problem asking how to change email, asking for possibility to do that or asking to delete old account because they've created new with lack of possibility to change email. When looking at such statistics like thread views and replies on threads the situation was even more dramatic:

79632 out of 329834 views (24.1%) were on above mentioned threads;
412 out of 1684 replies (including original posts) were to above mentioned threads.

So 24.5% or about one fourth of all posted messages were about only one problem.

I can at least try to convert that to lost $$.

Assuming each view took 10 seconds (which is very very modest assumption due to fact that forum response times cannot be called very short :) employees of Oracle and its partners and clients have used 79632 views * 10 sec = 796320 person seconds = 13272 person minutes = 221 person hours = 28 person days = more than one person month to just watch threads about this particular topic. Of course that is not including

time to formulate and write responses including OTN members from Oracle;
people not knowing about such forums and trying to find out how to change their e-mails;
people wasting time for creating new accounts;
people wasting time for asking to delete old accounts because they have created new ones;
forums support group wasting time to delete old accounts;
time wasted by other Oracle employees according to this message;
people getting angry and tearing their hairs wondering how such basic functionality cannot be included;
people loosing trust to Oracle seeing how their feedback is (not) being listened to;
6 hours of my time to gather above mentioned statistics and write these paragraphs :)

Yeahhh, as we can see even Oracle ACEs are suffering from this forced duplication of individualities (see here and here) so what can we say?

Anyway I hope that I will waste another hour or so to write another article that this nightmare has been ended!

P.S. Here is the list of the above mentioned 49 of 500 threads speaking about the same problem along with thread views and replies (the first is one of the most famous in forum history :)

http://forums.oracle.com/forums/thread.jspa?threadID=332623&tstart=0 32374 260

http://forums.oracle.com/forums/thread.jspa?threadID=466383&tstart=0 83 2

http://forums.oracle.com/forums/thread.jspa?threadID=465156&tstart=0 129 2

http://forums.oracle.com/forums/thread.jspa?threadID=315641&tstart=0 785 6

http://forums.oracle.com/forums/thread.jspa?threadID=457141&tstart=50 489 1

http://forums.oracle.com/forums/thread.jspa?threadID=455907&tstart=50 273 1

http://forums.oracle.com/forums/thread.jspa?threadID=454522&tstart=50 257 1

http://forums.oracle.com/forums/thread.jspa?threadID=449311&tstart=50 392 1

http://forums.oracle.com/forums/thread.jspa?threadID=448357&tstart=50 406 2

http://forums.oracle.com/forums/thread.jspa?threadID=296695&tstart=100 1342 9

http://forums.oracle.com/forums/thread.jspa?threadID=228506&tstart=100 4593 15

http://forums.oracle.com/forums/thread.jspa?threadID=442434&tstart=100 418 2