Client-side cursors are what Psycopg uses in its normal querying process.
They are implemented by the
AsyncCursor classes. In such
querying pattern, after a cursor sends a query to the server (usually calling
execute()), the server replies transferring to the client the whole
set of results requested, which is stored in the state of the same cursor and
from where it can be read from Python code (using methods such as
fetchone() and siblings).
This querying process is very scalable because, after a query result has been transmitted to the client, the server doesn’t keep any state. Because the results are already in the client memory, iterating its rows is very quick.
The downside of this querying method is that the entire result has to be transmitted completely to the client (with a time proportional to its size) and the client needs enough memory to hold it, so it is only suitable for reasonably small result sets.
PostgreSQL has also its own concept of cursor (sometimes also called portal). When a database cursor is created, the query is not necessarily completely processed: the server might be able to produce results only as they are needed. Only the results requested are transmitted to the client: if the query result is very large but the client only needs the first few records it is possible to transmit only them.
The downside is that the server needs to keep track of the partially processed results, so it uses more memory and resources on the server.
Psycopg allows the use of server-side cursors using the classes
AsyncServerCursor. They are usually created by passing the name
parameter to the
cursor() method (reason for which, in
psycopg2, they are usually called named cursors). The use of these classes
is similar to their client-side counterparts: their interface is the same, but
behind the scene they send commands to control the state of the cursor on the
server (for instance when fetching new records or when moving using
Using a server-side cursor it is possible to process datasets larger than what would fit in the client memory. However for small queries they are less efficient because it takes more commands to receive their result, so you should use them only if you need to process huge results or if only a partial result is needed.
“Stealing” an existing cursor¶
For instance if you have a PL/pgSQL function returning a cursor:
CREATE FUNCTION reffunc(refcursor) RETURNS refcursor AS $$ BEGIN OPEN $1 FOR SELECT col FROM test; RETURN $1; END; $$ LANGUAGE plpgsql;
you can run a one-off command in the same connection to call it (e.g. using
Connection.execute()) in order to create the cursor on the server:
after which you can create a server-side cursor declared by the same name, and
call directly the fetch methods, skipping the
cur = conn.cursor('curname') # no cur.execute() for record in cur: # or cur.fetchone(), cur.fetchmany()... # do something with record