The situation is as follows:
I have a substantial number of tables, with each a substantial number of columns. I need to deal with this old and to-be-deprecated data
select column_name
from user_tab_columns
where table_name='Table_name' and num_nulls>=1;
Just by simple query you will get those two columns.
I am not an expert in SQL procedures, hence giving general idea using SQL queries and a PHP/python script.
use SHOW TABLES
or some other query on INFORMATION_SCHEMA
database to get all tables in your database MY_DATABASE
do a query to generate a statement to get all column names in a particular table, this will be used in next query.
SELECT Group_concat(Concat( "MAX(", column_name, ")" )) FROM information_schema.columns WHERE table_schema = 'MY_DATABSE' AND table_name = 'MY_TABLE' ORDER BY table_name,ordinal_position
You will get an output like MAX(column_a),MAX(column_b),MAX(column_c),MAX(column_d)
Use this output to generate final query :
SELECT Max(column_a), Max(column_b), Max(column_c), Max(column_d) FROM MY_DATABASE.MY_TABLE
The output would be :
MAX(column_a) MAX(column_b) MAX(column_c) MAX(column_d)
NULL 1 NULL 1
NULL
are the ones which have all values NULL
SQL Fiddle Demo Link
I have created 4 tables. Three for demo and one nullcolumns
is the compulsory part of solution. Among three tables, only salary
and dept
have columns with all values null (you may have a look at their script).
The compulsory table and the procedure are given at the end
You can copy paste and run (the compulsory part or all) as sql (just you have to change the delimiter to //) in your desired database on your localhost and then --- call get();
and see the results
CREATE TABLE IF NOT EXISTS `dept` (
`did` int(11) NOT NULL,
`dname` varchar(50) DEFAULT NULL,
PRIMARY KEY (`did`)
) ENGINE=InnoDB DEFAULT CHARSET=latin1;
INSERT INTO `dept` (`did`, `dname`) VALUES
(1, NULL),
(2, NULL),
(3, NULL),
(4, NULL),
(5, NULL);
CREATE TABLE IF NOT EXISTS `emp` (
`id` int(11) NOT NULL AUTO_INCREMENT,
`ename` varchar(50) NOT NULL,
`did` int(11) NOT NULL,
PRIMARY KEY (`ename`),
KEY `deptid` (`did`),
KEY `id` (`id`)
) ENGINE=InnoDB DEFAULT CHARSET=latin1 AUTO_INCREMENT=6 ;
INSERT INTO `emp` (`id`, `ename`, `did`) VALUES
(1, 'e1', 4),
(2, 'e2', 4),
(3, 'e3', 2),
(4, 'e4', 4),
(5, 'e5', 3);
CREATE TABLE IF NOT EXISTS `salary` (
`EmpCode` varchar(50) NOT NULL,
`Amount` int(11) DEFAULT NULL,
`Date` int(11) DEFAULT NULL
) ENGINE=InnoDB DEFAULT CHARSET=latin1;
INSERT INTO `salary` (`EmpCode`, `Amount`, `Date`) VALUES
('1', 344, NULL),
('2', NULL, NULL);
------------------------------------------------------------------------
------------------------------------------------------------------------
CREATE TABLE IF NOT EXISTS `nullcolumns` (
`Table_Name` varchar(100) NOT NULL,
`Column_Name` varchar(100) NOT NULL
) ENGINE=InnoDB DEFAULT CHARSET=latin1;
--Only one procedure Now
CREATE PROCEDURE get(dn varchar(100))
BEGIN
declare c1 int; declare b1 int default 0; declare tn varchar(30);
declare c2 int; declare b2 int; declare cn varchar(30);
select count(*) into c1 from information_schema.tables where table_schema=dn;
delete from nullcolumns;
while b1<c1 do
select table_name into tn from information_schema.tables where
table_schema=dn limit b1,1;
select count(*) into c2 from information_schema.columns where
table_schema=dn and table_name=tn;
set b2=0;
while b2<c2 do
select column_name into cn from information_schema.columns where
table_schema=dn and table_name=tn limit b2,1;
set @nor := 0;
set @query := concat("select count(*) into @nor from ", dn,".",tn);
prepare s1 from @query;
execute s1;deallocate prepare s1;
if @nor>0 then set @res := 0;
set @query := concat("select ((select max(",cn,") from ", dn,".",tn,")
is NULL) into @res");
prepare s1 from @query;
execute s1;deallocate prepare s1;
if @res=1 then
insert into nullcolumns values(tn,cn);
end if; end if;
set b2=b2+1;
end while;
set b1=b1+1;
end while;
select * from nullcolumns;
END;
You can easily execute stored procedure easily as sql in your phpmyadin 'as it is' just change the Delimiters (at the bottom of SQL quesry box) to // Then
call get();
And Enjoy :)
You can see Now the table nullcolumns
showing all columns having 100/100 null values along with the table Names
In procedure code if @nor>0
restricts that no empty table should be included in results you can remove that restriction.
You can avoid using a procedure by dynamically creating (from the INFORMATION_SCHEMA.COLUMNS table) a string that contains the SQL you wish to execute, then preparing a statement from that string and executing it.
The SQL we wish to build will look like:
SELECT * FROM (
SELECT 'tableA' AS `table`,
IF(COUNT(`column_a`), NULL, 'column_a') AS `column`
FROM tableA
UNION ALL
SELECT 'tableB' AS `table`,
IF(COUNT(`column_b`), NULL, 'column_b') AS `column`
FROM tableB
UNION ALL
-- etc.
) t WHERE `column` IS NOT NULL
This can be done using the following:
SET group_concat_max_len = 4294967295; -- to overcome default 1KB limitation
SELECT CONCAT(
'SELECT * FROM ('
, GROUP_CONCAT(
'SELECT ', QUOTE(TABLE_NAME), ' AS `table`,'
, 'IF('
, 'COUNT(`', REPLACE(COLUMN_NAME, '`', '``'), '`),'
, 'NULL,'
, QUOTE(COLUMN_NAME)
, ') AS `column` '
, 'FROM `', REPLACE(TABLE_NAME, '`', '``'), '`'
SEPARATOR ' UNION ALL '
)
, ') t WHERE `column` IS NOT NULL'
)
INTO @sql
FROM INFORMATION_SCHEMA.COLUMNS
WHERE TABLE_SCHEMA = DATABASE();
PREPARE stmt FROM @sql;
EXECUTE stmt;
DEALLOCATE PREPARE stmt;
See it on sqlfiddle.
You can take advantage of the behavior of COUNT aggregate function regarding NULLs. By passing the field as argument, the COUNT function returns the number of non-NULL values while COUNT(*) returns the total number of rows. Thus you can calculate the ratio of NULL to "acceptable" values.
I will give an example with the following table structure:
CREATE TABLE `t1` (
`id` int(10) unsigned NOT NULL AUTO_INCREMENT,
`col_1` int(10) unsigned DEFAULT NULL,
`col_2` int(10) unsigned DEFAULT NULL,
PRIMARY KEY (`id`)
) ;
-- let's fill the table with random values
INSERT INTO t1(col_1,col_2) VALUES(1,2);
INSERT INTO t1(col_1,col_2)
SELECT
IF(RAND() > 0.5, NULL ,FLOOR(RAND()*1000),
IF(RAND() > 0.5, NULL ,FLOOR(RAND()*1000) FROM t1;
-- run the last INSERT-SELECT statement a few times
SELECT COUNT(col_1)/COUNT(*) AS col_1_ratio,
COUNT(col_2)/COUNT(*) AS col_2_ratio FROM t1;
You can write a function that automatically constructs a query from the INFORMATION_SCHEMA database by passing the table name as input variable. Here's how to obtain the structure data directly from INFORMATION_SCHEMA tables:
SET @query:=CONCAT("SELECT @column_list:=GROUP_CONCAT(col) FROM (
SELECT CONCAT('COUNT(',c.COLUMN_NAME,')/COUNT(*)') AS col
FROM INFORMATION_SCHEMA.COLUMNS c
WHERE NOT COLUMN_KEY IN('PRI') AND TABLE_SCHEMA=DATABASE()
AND TABLE_NAME='t1' ORDER BY ORDINAL_POSITION ) q");
PREPARE COLUMN_SELECT FROM @query;
EXECUTE COLUMN_SELECT;
SET @null_counters_sql := CONCAT('SELECT ',@column_list, ' FROM t1');
PREPARE NULL_COUNTERS FROM @null_counters_sql;
EXECUTE NULL_COUNTERS;
I think you can do this with GROUP_CONCAT and GROUP BY:
select length(replace(GROUP_CONCAT(my_col), ',', ''))
from my_table
group by my_col
(untested)
EDIT: the docs don't seem to state that GROUP_CONCAT needs a corresponding GROUP BY, so try this:
select
length(replace(GROUP_CONCAT(col_a), ',', '')) as len_a
, length(replace(GROUP_CONCAT(col_b), ',', '')) as len_b
, length(replace(GROUP_CONCAT(col_c), ',', '')) as Len_c
from my_table