Insert into … values ( SELECT … FROM … )

+975 votes
asked Aug 25, 2008 by claude-houle

I am trying to insert into a table using the input from another table. Although this is entirely feasible for many database engines, I always seem to struggle to remember the correct syntax for the SQL engine of the day (MySQL, Oracle, SQL Server, Informix, and DB2).

Is there a silver-bullet syntax coming from an SQL standard (for example, SQL-92) that would allow me to insert the values without worrying about the underlying database?

20 Answers

+1155 votes
answered Aug 25, 2008 by claude-houle

Try:

INSERT INTO table1 ( column1 )
SELECT  col1
FROM    table2  

This is standard ANSI SQL and should work on any DBMS

It definitely works for:

  • Oracle
  • MS SQL Server
  • MySQL
  • Postgres
  • SQLite v3
  • Teradata
  • DB2
  • Sybase
  • Vertica
  • HSQLDB
  • H2
  • AWS RedShift
  • SAP HANA
+726 votes
answered Aug 25, 2008 by travis

@Shadow_x99: That should work fine, and you can also have multiple columns and other data as well:

INSERT INTO table1 ( column1, column2, someInt, someVarChar )
SELECT  table2.column1, table2.column2, 8, 'some string etc.'
FROM    table2
WHERE   table2.ID = 7;

Edit: I should mention that I've only used this syntax with Access, SQL 2000/2005/Express, MySQL, and PostgreSQL, so those should be covered. A commenter has pointed out that it'll work with SQLite3.

+45 votes
answered Sep 28, 2008 by jonathan-leffler

Both the answers I see work fine in Informix specifically, and are basically standard SQL. That is, the notation:

INSERT INTO target_table[(<column-list>)] SELECT ... FROM ...;

works fine with Informix and, I would expect, all the DBMS. (Once upon 5 or more years ago, this is the sort of thing that MySQL did not always support; it now has decent support for this sort of standard SQL syntax and, AFAIK, it would work OK on this notation.) The column list is optional but indicates the target columns in sequence, so the first column of the result of the SELECT will go into the first listed column, etc. In the absence of the column list, the first column of the result of the SELECT goes into the first column of the target table.

What can be different between systems is the notation used to identify tables in different databases - the standard has nothing to say about inter-database (let alone inter-DBMS) operations. With Informix, you can use the following notation to identify a table:

[dbase[@server]:][owner.]table

That is, you may specify a database, optionally identifying the server that hosts that database if it is not in the current server, followed by an optional owner, dot, and finally the actual table name. The SQL standard uses the term schema for what Informix calls the owner. Thus, in Informix, any of the following notations could identify a table:

table
"owner".table
dbase:table
dbase:owner.table
dbase@server:table
dbase@server:owner.table

The owner in general does not need to be quoted; however, if you do use quotes, you need to get the owner name spelled correctly - it becomes case-sensitive. That is:

someone.table
"someone".table
SOMEONE.table

all identify the same table. With Informix, there's a mild complication with MODE ANSI databases, where owner names are generally converted to upper-case (informix is the exception). That is, in a MODE ANSI database (not commonly used), you could write:

CREATE TABLE someone.table ( ... )

and the owner name in the system catalog would be "SOMEONE", rather than 'someone'. If you enclose the owner name in double quotes, it acts like a delimited identifier. With standard SQL, delimited identifiers can be used many places. With Informix, you can use them only around owner names -- in other contexts, Informix treats both single-quoted and double-quoted strings as strings, rather than separating single-quoted strings as strings and double-quoted strings as delimited identifiers. (Of course, just for completeness, there is an environment variable, DELIMIDENT, that can be set - to any value, but Y is safest - to indicate that double quotes always surround delimited identifiers and single quotes always surround strings.)

Note that MS SQL Server manages to use [delimited identifiers] enclosed in square brackets. It looks weird to me, and is certainly not part of the SQL standard.

+7 votes
answered Jun 6, 2012 by swati-biswas

Here is another example where source is taken using more than one table:

INSERT INTO cesc_pf_stmt_ext_wrk( 
  PF_EMP_CODE    ,
  PF_DEPT_CODE   ,
  PF_SEC_CODE    ,
  PF_PROL_NO     ,
  PF_FM_SEQ      ,
  PF_SEQ_NO      ,
  PF_SEP_TAG     ,
  PF_SOURCE) 
SELECT
  PFl_EMP_CODE    ,
  PFl_DEPT_CODE   ,
  PFl_SEC         ,
  PFl_PROL_NO     ,
  PF_FM_SEQ       ,
  PF_SEQ_NO       ,
  PFl_SEP_TAG     ,
  PF_SOURCE
 FROM cesc_pf_stmt_ext,
      cesc_pfl_emp_master
 WHERE pfl_sep_tag LIKE '0'
   AND pfl_emp_code=pf_emp_code(+);

COMMIT;
+21 votes
answered Oct 16, 2012 by northben

This can be done without specifying the columns in the INSERT INTO part if you are supplying values for all columns in the SELECT part.

Let's say table1 has two columns. This query should work:

INSERT INTO table1
SELECT  col1, col2
FROM    table2

This WOULD NOT work (value for col2 is not specified):

INSERT INTO table1
SELECT  col1
FROM    table2

I'm using MS SQL Server. I don't know how other RDMS work.

+8 votes
answered Oct 17, 2012 by faiz

For Microsoft SQL Server, I will recommend learning to interpret the SYNTAX provided on MSDN. With Google it's easier than ever, to look for syntax.

For this particular case, try

Google: insert site:microsoft.com

The first result will be http://msdn.microsoft.com/en-us/library/ms174335.aspx

scroll down to the example ("Using the SELECT and EXECUTE options to insert data from other tables") if you find it difficult to interpret the syntax given at the top of the page.

[ WITH <common_table_expression> [ ,...n ] ]
INSERT 
{
        [ TOP ( expression ) [ PERCENT ] ] 
        [ INTO ] 
        { <object> | rowset_function_limited 
          [ WITH ( <Table_Hint_Limited> [ ...n ] ) ]
        }
    {
        [ ( column_list ) ] 
        [ <OUTPUT Clause> ]
        { VALUES ( { DEFAULT | NULL | expression } [ ,...n ] ) [ ,...n     ] 
        | derived_table       <<<<------- Look here ------------------------
        | execute_statement   <<<<------- Look here ------------------------
        | <dml_table_source>  <<<<------- Look here ------------------------
        | DEFAULT VALUES 
        }
    }
}
[;]

This should be applicable for any other RDBMS available there. There is no point in remembering all the syntax for all products IMO.

+7 votes
answered Mar 22, 2013 by grungondola

I actually prefer the following in SQL Server 2008:

SELECT Table1.Column1, Table1.Column2, Table2.Column1, Table2.Column2, 'Some String' AS SomeString, 8 AS SomeInt
INTO Table3
FROM Table1 INNER JOIN Table2 ON Table1.Column1 = Table2.Column3

It eliminates the step of adding the Insert () set, and you just select which values go in the table.

+23 votes
answered Apr 1, 2013 by santhosh

Most of the databases follow the basic syntax,

INSERT INTO TABLE_NAME
SELECT COL1, COL2 ...
FROM TABLE_YOU_NEED_TO_TAKE_FROM
;

Every database I have used follow this syntax namely, DB2, SQL Server, MY SQL, PostgresQL

+7 votes
answered Nov 20, 2013 by elijah7

This worked for me:

insert into table1 select * from table2

The sentence is a bit different from Oracle's.

+66 votes
answered Jan 10, 2014 by ianauld

To get only one value in a multi value INSERT from another table I did the following in SQLite3:

INSERT INTO column_1 ( val_1, val_from_other_table ) 
VALUES('val_1', (SELECT  val_2 FROM table_2 WHERE val_2 = something))
Welcome to Q&A, where you can ask questions and receive answers from other members of the community.
...