A convenient Python interface to Postgres's LISTEN and NOTIFY features.
First you need to make a connection:
import pgpubsub pubsub = pgpubsub.connect(user='postgres', database='test')
The arguments accepted by the pgpubsub.connect() function are identical to those supported by the psycopg2.connect() function.
To send an event, use the pubsub.notify() method:
pubsub.notify('test_channel', 'some message')
To receive events, you must first subscribe to a specific channel with the pubsub.listen() method:
You can call pubsub.listen() multiple times to receive events from multiple channels:
Note that channel names must be a valid SQL identifiers. To quote from the Postgres docs:
SQL identifiers and key words must begin with a letter (a-z, but also letters with diacritical marks and non-Latin letters) or an underscore (_). Subsequent characters in an identifier or key word can be letters, underscores, digits (0-9), or dollar signs ($).
Because channels are SQL identifiers rather than strings, they can’t be quoted/escaped by Psycopg2 like strings can. It is not safe to build a channel name from untrusted user input.
DO NOT DO THIS:
channel = 'events_' + username pubsub.listen(channel)
If you do, then your whole database could be destroyed by someone with the username “; DROP TABLE users;”. Mandatory XKCD.
Once you have subscribed to one or more channels, you can choose to receive events either by iterating over pubsub.events(), or by repeatedly calling the pubsub.get_event() method.
This is a generator over the stream of events coming on the pubsub. It lets you loop over the events just as you would a list:
for e in pubsub.events(): print e.payload
Behind the scenes, the pubsub is blocking on the standard library’s select.select function. You can provide two additional arguments to pubsub.events() to control how timeouts are handled when waiting on select.select:
select_timeout: The number of seconds to wait on select.select before giving up and trying again. Defaults to 5.
yield_timeouts: This defaults to False. If set to True, then pubsub.events() will yield a None each time you go for select_timeout seconds before receiving an event. This is useful for things like WebSockets where you may want to send a keepalive message even if no new data has been received:
for e in pubsub.events(yield_timeouts=True): if e is None: send_websocket_ping() else: send_websocket_message(e.payload)
This method always returns immediately. If an event has been received, it will return that event. If no event has been received, it will return None.
If multiple events have been received and are waiting in the queue, then repeated get_event() calls will keep returning the next event until there aren’t any left and it returns None:
>>> pubsub.listen('test') >>> pubsub.get_event() # Nothing delivered yet, so returns None >>> pubsub.notify('test', 'message 1') >>> pubsub.notify('test', 'message 2') >>> pubsub.get_event() Notify(9425, 'test', 'message 1') >>> pubsub.get_event() Notify(9425, 'test', 'message 2') >>> pubsub.get_event() # No more messages, so returns None
The pubsub.get_event() method is intended for integration into event loops where blocking on pubsub.events() would cause problems.
If you want to stop receiving events on one of the channels you’re currently subscribed to, you can call pubsub.unlisten():
The event objects returned by pubsub.events() and pubsub.get_event() are instances of psycopg2’s Notify class. They have three possibly-interesting attributes:
payload: A string containing the actual message.
channel: The name of the channel to which the event was sent.
pid: The pid of the process on the Postgres server that’s handling the sender’s connection. This can be useful to prevent an endless loop in a program that both sends and receives events:
my_pid = pubsub.conn.get_backend_pid() pubsub.listen('echo') for e in pubsub.events(): sender_pid = e.pid if sender_pid != my_pid: pubsub.notify('echo', e.payload)
Q & A
Is it safe to pass pubsub objects between threads?
Why use the verbs ‘notify’ and ‘listen’ instead of ‘publish’ and ‘subscribe’?
The methods in pgpubsub are designed to look as much as possible like the actual SQL commands in Postgres, which are NOTIFY and LISTEN. The Postgres docs also refer to ‘notification events’ rather than ‘messages’, so pgpubsub uses the same term.
Why is there no callback-style interface?
Someday there might be, if there’s demand for it and a well-reasoned spec.