[sql] How do you use script variables in psql?

In MS SQL Server, I create my scripts to use customizable variables:

DECLARE @somevariable int  
SELECT @somevariable = -1

INSERT INTO foo VALUES ( @somevariable )

I'll then change the value of @somevariable at runtime, depending on the value that I want in the particular situation. Since it's at the top of the script it's easy to see and remember.

How do I do the same with the PostgreSQL client psql?

One final word on PSQL variables:

1. They don't expand if you enclose them in single quotes in the SQL statement. Thus this doesn't work:

   SELECT * FROM foo WHERE bar = ':myvariable'
   

2. To expand to a string literal in a SQL statement, you have to include the quotes in the variable set. However, the variable value already has to be enclosed in quotes, which means that you need a second set of quotes, and the inner set has to be escaped. Thus you need:

   \set myvariable '\'somestring\''  
   SELECT * FROM foo WHERE bar = :myvariable
   

   EDIT: starting with PostgreSQL 9.1, you may write instead:

   \set myvariable somestring
   SELECT * FROM foo WHERE bar = :'myvariable'
   

Another approach is to (ab)use the PostgreSQL GUC mechanism to create variables. See this prior answer for details and examples.

You declare the GUC in postgresql.conf, then change its value at runtime with SET commands and get its value with current_setting(...).

I don't recommend this for general use, but it could be useful in narrow cases like the one mentioned in the linked question, where the poster wanted a way to provide the application-level username to triggers and functions.

FWIW, the real problem was that I had included a semicolon at the end of my \set command:

\set owner_password 'thepassword';

The semicolon was interpreted as an actual character in the variable:

\echo :owner_password thepassword;

So when I tried to use it:

CREATE ROLE myrole LOGIN UNENCRYPTED PASSWORD :owner_password NOINHERIT CREATEDB CREATEROLE VALID UNTIL 'infinity';

...I got this:

CREATE ROLE myrole LOGIN UNENCRYPTED PASSWORD thepassword; NOINHERIT CREATEDB CREATEROLE VALID UNTIL 'infinity';

That not only failed to set the quotes around the literal, but split the command into 2 parts (the second of which was invalid as it started with "NOINHERIT").

The moral of this story: PostgreSQL "variables" are really macros used in text expansion, not true values. I'm sure that comes in handy, but it's tricky at first.

postgres (since version 9.0) allows anonymous blocks in any of the supported server-side scripting languages

DO '
DECLARE somevariable int = -1;
BEGIN
INSERT INTO foo VALUES ( somevariable );
END
' ;

http://www.postgresql.org/docs/current/static/sql-do.html

As everything is inside a string, external string variables being substituted in will need to be escaped and quoted twice. Using dollar quoting instead will not give full protection against SQL injection.

I really miss that feature. Only way to achieve something similar is to use functions.

I have used it in two ways:

• perl functions that use $_SHARED variable
• store your variables in table

Perl version:

   CREATE FUNCTION var(name text, val text) RETURNS void AS $$
        $_SHARED{$_[0]} = $_[1];
   $$ LANGUAGE plperl;
   CREATE FUNCTION var(name text) RETURNS text AS $$
        return $_SHARED{$_[0]};
   $$ LANGUAGE plperl;

Table version:

CREATE TABLE var (
  sess bigint NOT NULL,
  key varchar NOT NULL,
  val varchar,
  CONSTRAINT var_pkey PRIMARY KEY (sess, key)
);
CREATE FUNCTION var(key varchar, val anyelement) RETURNS void AS $$
  DELETE FROM var WHERE sess = pg_backend_pid() AND key = $1;
  INSERT INTO var (sess, key, val) VALUES (sessid(), $1, $2::varchar);
$$ LANGUAGE 'sql';

CREATE FUNCTION var(varname varchar) RETURNS varchar AS $$
  SELECT val FROM var WHERE sess = pg_backend_pid() AND key = $1;
$$ LANGUAGE 'sql';

Notes:

• plperlu is faster than perl
• pg_backend_pid is not best session identification, consider using pid combined with backend_start from pg_stat_activity
• this table version is also bad because you have to clear this is up occasionally (and not delete currently working session variables)

Specifically for psql, you can pass psql variables from the command line too; you can pass them with -v. Here's a usage example:

$ psql -v filepath=/path/to/my/directory/mydatafile.data regress
regress=> SELECT :'filepath';
               ?column?                
---------------------------------------
 /path/to/my/directory/mydatafile.data
(1 row)

Note that the colon is unquoted, then the variable name its self is quoted. Odd syntax, I know. This only works in psql; it won't work in (say) PgAdmin-III.

This substitution happens during input processing in psql, so you can't (say) define a function that uses :'filepath' and expect the value of :'filepath' to change from session to session. It'll be substituted once, when the function is defined, and then will be a constant after that. It's useful for scripting but not runtime use.

I solved it with a temp table.

CREATE TEMP TABLE temp_session_variables (
    "sessionSalt" TEXT
);
INSERT INTO temp_session_variables ("sessionSalt") VALUES (current_timestamp || RANDOM()::TEXT);

This way, I had a "variable" I could use over multiple queries, that is unique for the session. I needed it to generate unique "usernames" while still not having collisions if importing users with the same user name.

I've posted a new solution for this on another thread.

It uses a table to store variables, and can be updated at any time. A static immutable getter function is dynamically created (by another function), triggered by update to your table. You get nice table storage, plus the blazing fast speeds of an immutable getter.

Variables in psql suck. If you want to declare an integer, you have to enter the integer, then do a carriage return, then end the statement in a semicolon. Observe:

Let's say I want to declare an integer variable my_var and insert it into a table test:

Example table test:

thedatabase=# \d test;
                         Table "public.test"
 Column |  Type   |                     Modifiers                     
--------+---------+---------------------------------------------------
 id     | integer | not null default nextval('test_id_seq'::regclass)
Indexes:
    "test_pkey" PRIMARY KEY, btree (id)

Clearly, nothing in this table yet:

thedatabase=# select * from test;
 id 
----
(0 rows)

We declare a variable. Notice how the semicolon is on the next line!

thedatabase=# \set my_var 999
thedatabase=# ;

Now we can insert. We have to use this weird ":''" looking syntax:

thedatabase=# insert into test(id) values (:'my_var');
INSERT 0 1

It worked!

thedatabase=# select * from test;
 id  
-----
 999
(1 row)

Explanation:

So... what happens if we don't have the semicolon on the next line? The variable? Have a look:

We declare my_var without the new line.

thedatabase=# \set my_var 999;

Let's select my_var.

thedatabase=# select :'my_var';
 ?column? 
----------
 999;
(1 row)

WTF is that? It's not an integer, it's a string 999;!

thedatabase=# select 999;
 ?column? 
----------
      999
(1 row)

You need to use one of the procedural languages such as PL/pgSQL not the SQL proc language. In PL/pgSQL you can use vars right in SQL statements. For single quotes you can use the quote literal function.

You can try to use a WITH clause.

WITH vars AS (SELECT 42 AS answer, 3.14 AS appr_pi)
SELECT t.*, vars.answer, t.radius*vars.appr_pi
FROM table AS t, vars;

I've found this question and the answers extremely useful, but also confusing. I had lots of trouble getting quoted variables to work, so here is the way I got it working:

\set deployment_user username    -- username
\set deployment_pass '\'string_password\''
ALTER USER :deployment_user WITH PASSWORD :deployment_pass;

This way you can define the variable in one statement. When you use it, single quotes will be embedded into the variable.

NOTE! When I put a comment after the quoted variable it got sucked in as part of the variable when I tried some of the methods in other answers. That was really screwing me up for a while. With this method comments appear to be treated as you'd expect.