MacLochlainns Weblog

Michael McLaughlin's Technical Blog

Site Admin

Using a Sparse Index

with 2 comments

My vacation from my blog is officially over. The question that I’m answering today is: How can you pass a set of non-sequential ID values to a function and return a result set? You can solve the problem by passing an ADT (Attribute Data Type) or UDT (User Defined Type) variable into a subquery of a cursor. The subquery leverages the TABLE function to translate the ADT or UDT into SQL result set, which is equivalent to a comma-delimited list of values.

You can also solve this problem with Native Dynamic SQL (NDS). However, the person who posed the question didn’t want to use NDS to build out a variable length list of comma-delimited numbers.

You need to create three object types for this example. They are:

  • a list of numbers
  • a record structure, declared as an object type without methods
  • a list of the record structure

These are the SQL commands to create the required data types:

CREATE OR REPLACE
  TYPE list_ids IS TABLE OF NUMBER;
/
CREATE OR REPLACE
  TYPE item_struct IS OBJECT
  ( item_id       NUMBER
  , item_title    VARCHAR2(80)
  , release_date  DATE );
/
CREATE OR REPLACE
  TYPE item_struct_list IS TABLE OF item_struct;
/

Next, you create a nonsynchronous function. It takes a sparsely populated list of values that map to the surrogate key of the column, which is typically the table’s primary key column. It returns a collection of the item_struct object type. This type of function is an object-table function.

The code follows:

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
CREATE OR REPLACE
  FUNCTION nonsynchronous 
  ( pv_list_ids  LIST_IDS ) RETURN item_struct_list IS
    /* Declare a record data structure list. */
    lv_struct_list  ITEM_STRUCT_LIST := item_struct_list();
 
    /* Declare a sparsely indexed list of film items. */
    CURSOR get_items
    ( cv_list_ids  LIST_IDS ) IS
      SELECT   item_id AS item_id
      ,        item_title
      ||       CASE
                 WHEN item_subtitle IS NOT NULL THEN
                   ': '|| item_subtitle
               END AS item_title
      ,        release_date AS release_date
      FROM     item
      WHERE    item_id IN (SELECT *
                           FROM   TABLE(cv_list_ids))
      ORDER BY item_id;
BEGIN
  /* Lood through the sparsely populated list of numbers. */
  FOR i IN get_items(pv_list_ids) LOOP
    lv_struct_list.EXTEND;
    lv_struct_list(lv_struct_list.COUNT) := item_struct( item_id      => i.item_id
                                                       , item_title   => i.item_title
                                                       , release_date => i.release_date );
  END LOOP;
 
  /* Return the record structure list. */
  RETURN lv_struct_list;
END;
/

The foregoing nonsynchronous function uses a nested query that transforms to a result set on lines 18 and 19. In the execution block of the program, it uses a call to the item_struct structure to capture and assign row values to an element of the lv_struct_list variable.

You can now test the nonsynchronous function with the following query:

COL item_id      FORMAT 9999  HEADING "Item|ID #"
COL item_title   FORMAT A40   HEADING "Item Title"
COL release_date FORMAT A11   HEADING "Release|Date"
SELECT   *
FROM     TABLE(nonsynchronous(list_ids(1002, 1013, 1007)));

The query returns the record set as an ordered list in the result set, like:

Item					       Release
 ID # Item Title			       Date
----- ---------------------------------------- -----------
 1002 Star Wars I: Phantom Menace	       04-MAY-99
 1007 RoboCop				       24-JUL-03
 1013 The DaVinci Code			       19-MAY-06

I hope this answers the question about how to get results sets with sparsely populated ID values.

Written by maclochlainn

May 11th, 2016 at 1:37 am

TurboTax Bug

with 4 comments

It was quite annoying to find that TurboTax couldn’t send me a text message to confirm my order of a second state. However, I made the mistake of clicking the “Confirm my account a different way (takes longer)” radio button to get to their web page.

After I got two-step verification enabled on their web site, now it’s impossible to order the second state software. It appears that once you click that button, the software writes it to a file and never prompts you for text, email, or other verification. That seems like a bug to me, what do you think?

TurboTaxBug

Hope this helps others …

Written by maclochlainn

December 23rd, 2015 at 1:31 pm

Posted in Uncategorized

IT Salary Thought

with 2 comments

During the holidays, I check salaries for my students and the IT industry overall. I’m never surprised by the reality, after all salaries pay for return on skills and effort. Here’s my annual look, which some may find unkind but reality is seldom kind.

Before looking at IT salaries, it seems like a good opportunity to first look at the overall job market for Millennials in the United States. AOL provides a great graphic of the median income for Millennials (those born between 1981 and 1997), which is $18,000 to $43,000 a year:

millennial-median-income-state-map

That’s a stark contrast to Forbes’ statistics on the top college baccalaureate degrees. In fact, the top five with the highest salary are between $58 to $67 thousand a year. They are:

  1. Computer Science ………… $66,800
  2. Engineering ………………… $65,000
  3. Mathematics & Statistics … $60,300
  4. Economics ………………….. $58,600
  5. Finance ……………………… $58,000

Computer science, applied computer science, and information technology are probably lumped into the first category. Information systems, exposure without real skills, is a management degree and probably opens positions equivalent to the business degree at $50 thousand a year. More or less, that’s a nine thousand dollar difference between having real skills and being able to talk the game and supervise technical resources. (The 10 hottest IT skills for 2015 are listed in Computerworld.)

There’s no surprise that Ruby, Objective C (iPhone, iPad, Mac OS X), Python, Java, C++ are at the top of the pyramid. Starting salaries in the Salt Lake area are higher for programmers college than they are for other computer science skill sets. In fact, my informal contacts peg them as starting at $70+ thousand. That’s higher than Forbes average for computer science. Here’s a visual on experienced programmers by language:

2015LanguageSalaries

It seems fair to say that a computer science, applied computer science, and information technology degree with an emphasis in real programming skills is the best bet to pay off student loans. However, some will wait for politicians to do that for them, but really that’s quite unlikely, isn’t it?

Reality is always blunt. Reality also seems to frequently differs from what politicians say. After all, politicians pander to audiences, which generally means they say a great deal of nonsense. Nonsense like economics doesn’t matter, everyone should earn the same regardless of their education, skills, or work ethic. Aldous Huxley said it more elegantly when he said, “That all men are equal is a proposition to which, at ordinary times, no sane human being has ever given his assent.”

Written by maclochlainn

December 21st, 2015 at 6:24 pm

Fedora LAMP Steps

without comments

I posted earlier in the year how to configure a Fedora instance to test PHP code on a local VM. However, I’ve got a few questions on how to find those posts. Here’s a consolidation with links on those steps:

  1. Go to this blog post and install the httpd and php libraries with the yum installer.
  2. In the same blog post as step 1 (you can put the sample PHP code into the /var/www/html directory for testing), connect to the yum shell and remove the php-mysql library and then install the mysqlnd library.
  3. Go to this blog post and install the php-gd libraries, which enable you to render PNG images stored as binary streams in MySQL.

As always, I hope that helps.

Written by maclochlainn

December 9th, 2015 at 9:44 am

REGEXP_LIKE Behavior

with one comment

Often, the biggest problem with regular expressions is that those who use them sometimes don’t use them correctly. A great example occurs in the Oracle Database with the REGEXP_LIKE function. For example, some developer use the following to validate whether a string is a number but it only validates whether the first character is a number.

1
2
3
4
5
6
7
8
9
10
11
12
13
14
DECLARE
  lv_input  VARCHAR2(100);
BEGIN
  /* Assign input value. */
  lv_input := '&input';
 
  /* Check for numeric string. */
  IF REGEXP_LIKE(lv_input,'[[:digit:]]') THEN
    dbms_output.put_line('It''s a number.');
  ELSE
    dbms_output.put_line('It''s a string.');
  END IF;
END;
/

When they test numbers it appears to works, it even appears to work when the test string start with number, but it fails with any string that starts with a character. That’s because the REGEXP_LIKE function on line 8 only checks the first character, but the following checks all the characters in the string.

8
  IF REGEXP_LIKE(lv_inputs(i),'[[:digit:]]{'||LENGTH(lv_inputs(i))||'}') THEN

You can also fix it with the following non-Posix solution:

8
  IF REGEXP_LIKE(lv_input,'[[0-9]]') THEN

You can add a collection to the program and use it to test single-digit, double-digit, and string with a leading integer. Save the program as test.sql and you can test three conditions with one call.

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
DECLARE
  /* Declare the local collection type. */
  TYPE inputs IS TABLE OF VARCHAR2(100);
 
  /* Declare a local variable of the collection type. */
  lv_inputs  INPUTS;
BEGIN
  /* Assign the inputs to the collection variable. */
  lv_inputs := inputs('&1','&2','&3');
 
  /* Read through the collection and print whether it's an number or string. */
  FOR i IN 1..lv_inputs.COUNT LOOP
    IF REGEXP_LIKE(lv_inputs(i),'[[:digit:]]{'||LENGTH(lv_inputs(i))||',}') THEN
      dbms_output.put_line('It''s a number.');
    ELSE
      dbms_output.put_line('It''s a string.');
    END IF;
  END LOOP;
END;
/

You can run the test.sql program like this:

SQL> @test.sql 1 12 1a

It prints:

It's a number.
It's a number.
It's a string.

As always, I hope this helps those looking for a solution.

Written by maclochlainn

September 30th, 2015 at 7:23 pm

Create MySQL Index

without comments

Indexes are separate data structures that provide alternate pathways to finding data. They can and do generally speed up the processing of queries and other DML commands, like the INSERT, UPDATE, REPLACE INTO, and DELETE statements. Indexes are also called fast access paths.

In the scope of the InnoDB Database Engine, the MySQL database maintains the integrity of indexes after you create them. The upside of indexes is that they can improve SQL statement performance. The downside is that they impose overhead on every INSERT, UPDATE, REPLACE INTO, and DELETE statement, because the database maintains them by inserting, updating, or deleting items for each related change in the tables that the indexes support.

Indexes have two key properties—usability and visibility. Indexes are both usable and visible by default. That means they are visible to the MySQL cost-based optimizer and usable when statements run against the tables they support.

You have the ability to make any index invisible, in which case queries and DML statements won’t use the index because they won’t see it. However, the cost-based optimizer still sees the index and maintains it with any DML statement change. That means making an index invisible isn’t quite like making the index unusable or like dropping it temporarily. An invisible index becomes overhead and thus is typically a short-term solution to run a resource-intensive statement that behaves better without the index while avoiding the cost of rebuilding it after the statement runs.

It is also possible to make an index unusable, in which case it stops collecting information and becomes obsolete and the database drops its index segment. You rebuild the index when you change it back to a usable index.

Indexes work on the principal of a key. A key is typically a set of columns or expressions on which you can build an index, but it’s possible that a key can be a single column. An index based on a set of columns is a composite, or concatenated, index.

Indexes can be unique or non-unique. You create a unique index anytime you constrain a column by assigning a primary key or unique constraint, but they’re indirect indexes. You create a direct unique index on a single column with the following syntax against two non-unique columns:

1
2
CREATE INDEX common_lookup_u1
ON common_lookup (common_lookup_table) USING BTREE;

You could convert this to a non-unique index on two columns by using this syntax:

1
2
CREATE INDEX common_lookup_u1
ON common_lookup (common_lookup_table, common_lookup_column) USING BTREE;

Making the index unique is straightforward;, you only need to add a UNIQUE key wordk to the CREATE INDEX statement, like

1
2
3
4
CREATE UNIQUE INDEX common_lookup_u1
ON common_lookup ( common_lookup_table
                 , common_lookup_column
                 , common_lookup_type) USING BTREE;

Most indexes use a B-tree (balanced tree). A B-tree is composed of three types of blocks—a root branch block for searching next-level blocks, branch blocks for searching other branch blocks, or and leaf blocks that store pointers to row values. B-trees are balanced because all leaf-blocks are at the same level, which means the length of search is the same to any element in the tree. All branch blocks store the minimum key prefix required to make branching decisions through the B-tree.

Written by maclochlainn

September 29th, 2015 at 6:41 pm

SQL*Plus Tricks

with 3 comments

Have you ever wondered how to leverage substitution variables in anonymous block programs? There are several tricks that you can use beyond passing numeric and string values to local variable. The generic default appears to take a number unless you cast it as a string but that’s not really the whole story. The first two are standard examples of how to use numeric and string substitution values.

The following accept a numeric substitution value:

1
2
3
4
5
6
7
8
9
10
DECLARE
  lv_input  NUMBER;
BEGIN
  /* Assign substitution value to local variable. */
  lv_input := &input;
 
  /* Print the local variable. */
  dbms_output.put_line('['||lv_input||']');
END;
/

The following accept a string substitution value, casts the input as a string, assigns the string value to a 4,000 character length local variable, checks whether the 4,000 character length is greater than 10, and assigns the first 10 characters to the lv_parse_input variable:

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
DECLARE
  lv_unparsed_input  VARCHAR2(4000);
  lv_parsed_input    VARCHAR2(10);
BEGIN
  /* Assign substitution value to local variable. */
  lv_unparsed_input := '&input';
 
  /* Check size of input value. */
  IF LENGTH(lv_unparsed_input) > 10 THEN
    lv_parsed_input := SUBSTR(lv_unparsed_input,1,10);
  END IF;
 
  /* Print the local variable. */
  dbms_output.put_line('Print {lv_parsed_input}: ['||lv_parsed_input||']');
END;
/

Next, let’s examine two tricks. The first passes a case insensitive variable name and the second passes a case sensitive variable name as a parameter to an anonymous block program.

This declares an anonymous block program that uses a substitution value as a variable name:

1
2
3
4
5
6
7
DECLARE
  mine  VARCHAR2(10) := 'Default';
BEGIN
  /* Print the local variable's value. */
  dbms_output.put_line('Print {mine} variable value: ['||&input||']');
END;
/

When you run the anonymous block, you’re prompted for an input variable. You provide a case insensitive variable name as the input value:

Enter value for input: MINE
old   5:   dbms_output.put_line('['||&input||']');
new   5:   dbms_output.put_line('['||MINE||']');
Print {mine} variable value: [Default]

The downside of this approach, yields an ORA-06550 and PLS-00201 exception. Neither of these can be caught because Oracle raises the errors during parsing when the variable name isn’t a 100% case insensitive match. The same type of problem occurs in the next example when the input variable isn’t a 100% case sensitive match.

You can rewrite the program to handle case insensitive variables like this:

1
2
3
4
5
6
7
DECLARE
  "Mine"  VARCHAR2(10) := 'Default';
BEGIN
  /* Print the local variable's value. */
  dbms_output.put_line('Print {mine} variable value: ['||"&input"||']');
END;
/

When you run the anonymous block, you’re prompted for an input variable. You provide a case sensitive variable name as the input value:

Enter value for input: Mine
old   5:   dbms_output.put_line('['||&input||']');
new   5:   dbms_output.put_line('['||"Mine"||']');
Print {Mine} variable value: [Default]

Hope this helps those looking for a solution.

Written by maclochlainn

September 24th, 2015 at 1:19 am

Using CALIBRATE_IO

without comments

Using Oracle’s Resource Manager requires you to understand the IO dynamics. The first step requires you to run the CALIBRATE_IO procedure from the DBMS_RESOURCE_MANAGER package.

Oracle provides some great examples about how to use the CALIBRATE_IO procedure of the DBMS_RESOURCE_MANAGER package in the Oracle Database Database PL/SQL Packages and Types Reference. The CALIBRATE_IO procedure returns the best answer when you provide a valid number of files, which you can capture by querying the V$ASM_DISK view.

The following code queries the view and assigns the value to a session level variable:

CLEAR BREAKS
CLEAR COLUMNS
CLEAR COMPUTES
 
VARIABLE files NUMBER
 
BEGIN
  SELECT COUNT(DISTINCT name) disks
  INTO :files
  FROM v$asm_disk;
END;
/

When you have the number of files, you can calibrate the IO with the following anonymous block. The query should always work but just in case the NVL function on line 9 assigns the default number of files.

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
DECLARE
  lv_num_physical_disks BINARY_INTEGER; — v$asm_disk
  lv_max_latency BINARY_INTEGER := 10;
  lv_max_iops BINARY_INTEGER;
  lv_max_mbps BINARY_INTEGER;
  lv_actual_latency BINARY_INTEGER;
BEGIN
  /* Assign actual files to anonymous block variable. */
  lv_num_physical_disks := NVL(:files,2);
 
  /* Run the calibrate_io procedure. */
  DBMS_RESOURCE_MANAGER.CALIBRATE_IO(
      num_physical_disks => lv_num_physical_disks
    , max_latency => lv_max_latency
    , max_iops => lv_max_iops
    , max_mbps => lv_max_mbps
    , actual_latency => lv_actual_latency);
END;
/

You can query the results like this:

SELECT max_iops
,      max_mbps
,      max_pmbps
,      latency
,      num_physical_disks
FROM   dba_rsrc_io_calibrate;

It should show results like these:

MAX_IOPS MAX_MBPS MAX_PMBPS LATENCY NUM_PHYSICAL_DISKS
-------- -------- --------- ------- ------------------
    8894      443       294       9                 18

Hope this helps those using the CALIBRATE_IO procedure of the DBMS_RESOURCE_MANAGER package.

Written by maclochlainn

August 31st, 2015 at 8:59 pm

Free Oracle Tuning Book

with 2 comments

Quick Start Guide to Oracle Query TuningWho can resist a free Rich Nimeiec book on SQL Tuning? O.K., those who know everything can resist. If you’re like me, this is an opportunity to learn from Rich. Click on the book image or this link to get a free copy, or if you want to pay $10 for a copy click here to buy Quick Start Guide to Oracle Query Tuning: Tips for DBAs and Developers from Amazon.com.

The book is four chapters long, is a 129 pages in length, and is in a PDF format. The outline is:

  1. Query Tuning: Developer and Beginning DBA
  2. Query Tuning: Basics for DBAs and Developers
  3. Advanced Performance Tuning
  4. Tips for Tuning When You Have Everything Tuned

Enjoy reading it. His more comprehensive book is Oracle Database 11g Release 2 Performance Tuning Tips & Techniques (Oracle Press) and it’s $30, but it’s written for an advanced audience (more or less OCA or higher).

Written by maclochlainn

August 31st, 2015 at 11:24 am

Use an object in a query?

without comments

Using an Oracle object type’s instance in a query is a powerful capability. Unfortunately, Oracle’s SQL syntax doesn’t make it immediately obvious how to do it. Most get far enough to put it in a runtime view (a subquery in the FROM clause), but then they get errors like this:

SELECT	 instance.get_type()
         *
ERROR AT line 4:
ORA-00904: "INSTANCE"."GET_TYPE": invalid identifier

The problem is how Oracle treats runtime views, which appears to me as a casting error. Somewhat like the ORDER BY clause irregularity that I noted in July, the trick is complete versus incomplete syntax. The following query fails and generates the foregoing error:

1
2
3
4
SELECT instance.get_type() AS object_type
,      instance.to_string() AS object_content
FROM  (SELECT dependent()AS instance
       FROM   dual);

If you add a table alias, or name, to the runtime view on line 4, it works fine:

1
2
3
4
SELECT cte.instance.get_type() AS object_type
,      cte.instance.to_string() AS object_content
FROM  (SELECT dependent() AS instance
       FROM   dual) cte;

That is the trick. You use an alias for the query, which assigns the alias like a table reference. The reference lets you access instance methods in the scope of a query. Different columns in the query’s SELECT-list may return different results from different methods from the same instance of the object type.

You can also raise an exception if you forget the open and close parentheses for a method call to a UDT, which differs from how Oracle treats no argument functions and procedures. That type of error would look like this:

SELECT cte.instance.get_type AS object_type
       *
ERROR AT line 1:
ORA-00904: : invalid identifier

It is an invalid identifier because there’s no public variable get_type, and a method is only found by using the parenthesis and a list of parameters where they’re required.

The object source code is visible by clicking on the expandable label below.

As always, I hope this helps those solving problems.

Written by maclochlainn

August 22nd, 2015 at 5:23 pm