Ideas
The pre_prepare
module aims to prepare all your statements as soon as
possible and in a way that allows client queries not to bother at all and
just call EXECUTE
.
Setup
compile and install
The module is using the PG_XS build infrastructure, so just:
make
make install
psql -f `pg_config --sharedir`/contrib/pre_prepare.sql -U user dbname
Depending on the way you got PostgreSQL installed in the first place, you
may need to use sudo for the make install
step.
postgresql.conf
Add a custom class then the following settings:
preprepare.relation = 'preprepare.statements'
The pre_prepare.relation
is the name of the table where you’ll put all the
statements you want to module to prepare. The following SQL query against
the given relation must return the names and SQL statements to be prepared.
SELECT name, statement FROM <pre_prepare.relation>;
The statements won’t be edited, so must be the all PREPARE stuff.
When running PostgreSQL 9.1 or earlier, an additional setting is required:
custom_variable_classes = 'preprepare' # only for 9.1 and earlier
pgbouncer
When using pgbouncer, which is a good idea, consider setting up
connect_query
to prepare all the statements at server connection time.
[databases]
foo = port=5432 connect_query='SELECT prepare_all();'
Of course, if using pre_prepare
, you’ll want to avoid using DISCARD ALL
as your reset_query
…
local_preload_libraries
If you have 8.3
It’s unfortunately not possible to call SPI_connect()
from the module
initialization routine (_PG_init()
) when called via
local_preload_libraries
, it’s too much early. So fully transparent
preparing of user given statement is not possible with the pre_prepare
module, you still have to explicitely call SELECT prepare_all()
.
If running 8.4 or later
To call SPI_connect()
in fact what’s needed is an opened transaction and a
current active Snapshot. Both are possible to acquire from within a
dynamically loaded module, so preprepare
is supporting the case.
Update your postgresql.conf
to have the following:
custom_variable_classes = 'preprepare' # only for 9.1 and earlier
preprepare.at_init = on
preprepare.relation = 'preprepare.statements'
Then reload
PostgreSQL and enjoy: any new connection will have already
prepared your statements by the time you’re able to send queries, so you can
forget about PREPARE
and directly EXECUTE
. No need for extra software.
Please note that if your statements contain any error, PostgreSQL will
handle it as a FATAL
error and this will effectively prevent you from
connecting to your server. You’ll have to turn preprepare.at_init off
again then reload
, or remove pre_prepare
from local_preload_libraries
then restart
.
Usage
First create a table where to store your statements, e.g.:
create table pre_prepare.statements(name text primary key, statement text);
The statement stored is the complete statement including the PREPARE name AS
part.
insert into pre_prepare.statements values ('test', 'prepare test as select 1');
prepare_all()
Then connect to PostgreSQL (via pgbouncer
if not using
local_preload_libraries
) and run the EXECUTE
command matching with your
PREPARE
statements, they all are already prepared.
If not using pgbouncer
nor preprepare.at_init = on
, you’ll have to call
the function yourself to have your statements prepared:
SELECT prepare_all();
prepare_all(schema.table)
If you’re not in a position to always prepare the same set of queries, you
can use the second form of prepare_all
calling, which accepts a specific
statements table:
SELECT prepare_all('public.expensive_planning');
This can be automated in a special pgbouncer
fake database
where the
connect_query
will prepare specific queries when you know you need them in
some occasion, but they’re way to expensive to always prepare ahead of time.
discard()
The module also offers a discard()
function which does the same as
DISCARD ALL
except that it won’t call DEALLOCATE ALL
.