The NOTIFY command sends a notify event to each frontend application that has previously executed LISTEN notifyname for the specified notify condition in the current database.
The information passed to the frontend for a notify event includes the notify condition name and the notifying backend process's PID. It is up to the database designer to define the condition names that will be used in a given database and what each one means.
Commonly, the notify condition name is the same as the name of some table in the database, and the notify event essentially means "I changed this table, take a look at it to see what's new". But no such association is enforced by the NOTIFY and LISTEN commands. For example, a database designer could use several different condition names to signal different sorts of changes to a single table.
NOTIFY provides a simple form of signal or IPC (interprocess communication) mechanism for a collection of processes accessing the same Postgres database. Higher-level mechanisms can be built by using tables in the database to pass additional data (beyond a mere condition name) from notifier to listener(s).
When NOTIFY is used to signal the occurrence of changes to a particular table, a useful programming technique is to put the NOTIFY in a rule that is triggered by table updates. In this way, notification happens automatically when the table is changed, and the application programmer can't accidentally forget to do it.
NOTIFY interacts with SQL transactions in some important ways. Firstly, if a NOTIFY is executed inside a transaction, the notify events are not delivered until and unless the transaction is committed. This is appropriate, since if the transaction is aborted we would like all the commands within it to have had no effect, including NOTIFY. But it can be disconcerting if one is expecting the notify events to be delivered immediately. Secondly, if a listening backend receives a notify signal while it is within a transaction, the notify event will not be delivered to its connected frontend until just after the transaction is completed (either committed or aborted). Again, the reasoning is that if a notify were delivered within a transaction that was later aborted, one would want the notification to be undone somehow---but the backend cannot "take back" a notify once it has sent it to the frontend. So notify events are only delivered between transactions. The upshot of this is that applications using NOTIFY for real-time signaling should try to keep their transactions short.
NOTIFY behaves like Unix signals in one important respect: if the same condition name is signaled multiple times in quick succession, recipients may get only one notify event for several executions of NOTIFY. So it is a bad idea to depend on the number of notifies received. Instead, use NOTIFY to wake up applications that need to pay attention to something, and use a database object (such as a sequence) to keep track of what happened or how many times it happened.
It is common for a frontend that sends NOTIFY to be listening on the same notify name itself. In that case it will get back a notify event, just like all the other listening frontends. Depending on the application logic, this could result in useless work---for example, re-reading a database table to find the same updates that that frontend just wrote out. In Postgres 6.4 and later, it is possible to avoid such extra work by noticing whether the notifying backend process's PID (supplied in the notify event message) is the same as one's own backend's PID (available from libpq). When they are the same, the notify event is one's own work bouncing back, and can be ignored. (Despite what was said in the preceding paragraph, this is a safe technique. Postgres keeps self-notifies separate from notifies arriving from other backends, so you cannot miss an outside notify by ignoring your own notifies.)
name can be any string valid as a name; it need not correspond to the name of any actual table. If name is enclosed in double-quotes, it need not even be a syntactically valid name, but can be any string up to 31 characters long.
In some previous releases of Postgres, name had to be enclosed in double-quotes when it did not correspond to any existing table name, even if syntactically valid as a name. That is no longer required.
In Postgres releases prior to 6.4, the backend PID delivered in a notify message was always the PID of the frontend's own backend. So it was not possible to distinguish one's own notifies from other clients' notifies in those earlier releases.