Set empty strings ('') to NULL in the whole database

问题

In my database are many text columns where values are empty strings (''). The empty strings need to be set to NULL. I do not know the exact schemas, tables and columns in this database or rather I want to write a general solution which can be reused.

How would I write a query / function to find all text columns in all tables in all schemas and update all columns with empty strings ('') to NULL?

回答1:

The most efficient way to achieve this:

Run a single UPDATE per table.
Only update nullable columns (not defined NOT NULL) with any actual empty string.
Only update rows with any actual empty string.
Leave other values unchanged.

This related answer has a plpgsql function that builds and runs the UPDATE command using system catalog pg_attribute automatically and safely for any given table:

Replace empty strings with null values

Using the function f_empty2null() from this answer, you can loop through selected tables like this:

DO
$do$
DECLARE
   _tbl regclass;
BEGIN
   FOR _tbl IN
      SELECT c.oid::regclass
      FROM   pg_class c
      JOIN   pg_namespace n ON n.oid = c.relnamespace
      WHERE  c.relkind = 'r'            -- only regular tables
      AND    n.nspname NOT LIKE 'pg_%'  -- exclude system schemas
   LOOP
      RAISE NOTICE $$PERFORM f_empty2null('%');$$, _tbl;
      -- PERFORM f_empty2null(_tbl);  -- uncomment to prime the bomb
   END LOOP;
END
$do$;

Careful! This updates all empty strings in all columns of all user tables in the DB. Be sure that's what you want or it might nuke your database.

You need UPDATE privileges on all selected tables, of course.

As a child safety device I commented the payload.

You may have noted that I use the system catalogs directly, not the information schema (which would work, too). About this:

How to check if a table exists in a given schema
Query to return output column names and data types of a query, table or view

For repeated use

Here is an integrated solution for repeated use. Without safety devices:

CREATE OR REPLACE FUNCTION f_all_empty2null(OUT _tables int, OUT _rows int) AS
$func$
DECLARE
   _typ CONSTANT regtype[] := '{text, bpchar, varchar, \"char\"}';
   _sql text;
   _row_ct int;
BEGIN
   _tables := 0;  _rows := 0;
   FOR _sql IN
      SELECT format('UPDATE %s SET %s WHERE %s'
                  , t.tbl
                  , string_agg(format($$%1$s = NULLIF(%1$s, '')$$, t.col), ', ')
                  , string_agg(t.col || $$ = ''$$, ' OR '))
      FROM  (
         SELECT c.oid::regclass AS tbl, quote_ident(attname) AS col
         FROM   pg_namespace n
         JOIN   pg_class     c ON c.relnamespace = n.oid
         JOIN   pg_attribute a ON a.attrelid = c.oid
         WHERE  n.nspname NOT LIKE 'pg_%'      -- exclude system schemas
         AND    c.relkind = 'r'                -- only regular tables
         AND    a.attnum >= 1                  -- exclude tableoid & friends
         AND    NOT a.attisdropped             -- exclude dropped columns
         AND    NOT a.attnotnull               -- exclude columns defined NOT NULL!
         AND    a.atttypid = ANY(_typ)         -- only character types
         ORDER  BY a.attnum
         ) t
      GROUP  BY t.tbl
   LOOP
      EXECUTE _sql;
      GET DIAGNOSTICS _row_ct = ROW_COUNT;     -- report nr. of affected rows
      _tables := _tables + 1;
      _rows := _rows + _row_ct;
   END LOOP;
END
$func$  LANGUAGE plpgsql;

Call:

SELECT * FROM pg_temp.f_all_empty2null();

Returns:

 _tables | _rows
---------+---------
 23      | 123456

Note how I escaped both table and columns names properly!

c.oid::regclass AS tbl, quote_ident(attname)  AS col

Consider:

Table name as a PostgreSQL function parameter

Careful! Same warning as above.
Also consider the basic explanation in the answer I linked above:

Replace empty strings with null values

回答2:

The simplest way to go about it is manually, table by table. For each table, do something like this:

START TRANSACTION;
UPDATE tablename SET 
    stringfield1 = NULLIF(stringfield1, ''),
    stringfield2 = NULLIF(stringfield2, '');
<do some selects to make sure everything looks right>
COMMIT;

That will rewrite every row in the table, but it will only make one pass over the table. It might or might not be impractical for you.

You might instead want to do it field by field using a WHERE clause, to reduce the number of updates, like this.

START TRANSACTION;
UPDATE tablename SET stringfield1 = NULL WHERE stringfield1 = '';
<do some selects to make sure everything looks right>
COMMIT;

That will only rewrite the rows that need to be rewritten, but will require multiple passes over each table.

回答3:

i think the code below is a generalised one. and you can use it any time anywhere:

    Declare @Query varchar(1000)
    declare @AllDatabaseTables table(id int,Table_Name varchar(50))
    Insert into @AllDatabaseTables select Table_Name from information_schema.tables

    declare @Table_Name varchar(50)

    declare @i int=1
    While @i<=(Select Count(*) from @AllDatabaseTables)
    BEGIN
    Select @Table_Name=Table_Name from @AllDatabaseTables Where id=@i

    Declare @ColumnTable table(id int,ColumnName varchar(100))
    Insert into @ColumnTable Select COLUMN_NAME from information_schema.columns Where Table_Name=@Table_Name and DATA_TYPE='varchar' --if the datatype is varchar type
    Declare @ColumnName varchar(50)
    Declare @k int=1
    While @k<=(Select count(*) from @ColumnTable)
        BEGIN
            Select @ColumnName=ColumnName from @ColumnTable where id=@k

            Set @Query='Update '+@Table_Name+' Set '+@ColumnName+'=NULL where '+@ColumnName+'='''' ' 
            Exec(Query)

            Set @k=@k+1
        END



    Set @i=@i+1

END

来源：https://stackoverflow.com/questions/38459106/set-empty-strings-to-null-in-the-whole-database

标签

sql

postgresql

plpgsql

dynamic-sql

information-schema