PostgreSQL – how to run VACUUM from code outside transaction block?

I don’t know psycopg2 and PostgreSQL, but only apsw and SQLite, so I think I can not give a “psycopg2” help.

But it seams to me, that PostgreSQL might work similar as SQLite does, it has two modes of operation:

• Outside a transaction block: This is semantically equivalent to have a transaction block around every single SQL operation
• Inside a transaction block, that is marked by “BEGIN TRANSACTION” and ended by “END TRANSACTION”

When this is the case, the problem could be inside the access layer psycopg2. When it does normally operate in a way that transactions are implicitely inserted until a commit is made, there could be no “standard way” to make a vacuum.

Of course it could be possible, that “psycopg2” has its special “vacuum” method, or a special operation mode, where no implicit transactions are started.

When no such possibilities exists, there stays one single option (without changing the access layer 😉 ):

Most databases have a shell programm to access the database. The program could run this shell program with a pipe (entering the vacuum-command into the shell), thus using the shell programm to make the vacuum. Since vacuum is a slow operation as such, the start of an external programm will be neglectible. Of course, the actual program should commit all uncommited work before, else there could be a dead-lock situation – the vacuum must wait until end of your last transaction.

Don’t do it – you don’t need VACUUM FULL. Actually if you run somewhat recent version of Postgres (let’s say > 8.1) you don’t even need to run plain VACUUM manually.

Additionally, you can also get the messages given by the Vacuum or Analyse using:

>> print conn.notices #conn is the connection object

this command print a list with the log message of queries like Vacuum and Analyse:

INFO:  "usuario": processados 1 de 1 páginas, contendo 7 registros vigentes e 0 registros não vigentes; 7 registros amostrados, 7 registros totais estimados   
INFO:  analisando "public.usuario"

This can be useful to the DBAs ^^

While vacuum full is questionable in current versions of postgresql, forcing a ‘vacuum analyze’ or ‘reindex’ after certain massive actions can improve performance, or clean up disk usage. This is postgresql specific, and needs to be cleaned up to do the right thing for other databases.

from django.db import connection
# Much of the proxy is not defined until this is done
force_proxy = connection.cursor()
realconn=connection.connection
old_isolation_level = realconn.isolation_level
realconn.set_isolation_level(0)
cursor = realconn.cursor()
cursor.execute('VACUUM ANALYZE')
realconn.set_isolation_level(old_isolation_level)

Unfortunately the connection proxy provided by django doesn’t provide access to set_isolation_level.

Note if you’re using Django with South to perform a migration you can use the following code to execute a VACUUM ANALYZE.

def forwards(self, orm):

    db.commit_transaction()
    db.execute("VACUUM ANALYZE <table>")

    #Optionally start another transaction to do some more work...
    db.start_transaction()

For anyone else that has tried all the suggestions around this problem without success, you may be suffering the same fate as me: I had 2 (or more) SQL statements in one execute() call. It turns out that Postgres itself resets any autocommit/isolation after the first statement (separated by a ;). I finally came across the solution here: https://github.com/psycopg/psycopg2/issues/1201

So don’t do something like this:

cursor.execute("SELECT 1; VACUUM FULL")

Instead do:

cursor.execute("SELECT 1")
cursor.execute("VACUUM FULL")

While it’s not the OP’s exact situation, I ran into this as well. It turns out it was related to an implicit transaction being opened when the connection is used as a context. For instance, if the OP’s code had been something like this, even with the correct isolation levels it would’ve hit the exact same error for different reasons:

class db(object):
    def __init__(dbname, host, port, user, password):
        self.conn = psycopg2.connect("dbname=%s host=%s port=%s 
                                      user=%s password=%s" 
                                      % (dbname, host, port, user, password))
        self.conn.set_isolation_level(ISOLATION_LEVEL_AUTOCOMMIT)

    def _doQuery(self, query):
        # THE FOLLOWING LINE OPENS AN IMPLICIT TRANSACTION
        # No matter the autocommit settings, this WILL create a transaction
        with self.conn:
            with conn.cursor() as cursor:
                cursor.execute(query)
                self.conn.commit()

    def vacuum(self):
        query = "VACUUM FULL"
        self._doQuery(query)

The solution is to just not use the connection as a context:

    def _doQuery(self, query):
        # Note no "with conn:" context anymore
        with conn.cursor() as cursor:
            cursor.execute(query)
            self.conn.commit()