Thread: NOTIFY/LISTEN in Postgresql

On 10/15/2012 08:54 PM, P. Broennimann wrote:
> Hi there
>
> 1) Can a Pg/SQL function "listen" for a notification sent from an
> external instance?

No, it's the other way around. A client can `LISTEN` for a `NOTIFY` sent
by another client, either directly or via a PL/PgSQL function.

What you want is a NOTIFY callback or NOTIFY trigger, something that
invokes a function without any client action when a NOTIFY comes in. No
such feature exists.

> I would like my stored function to pause/wait and continue its execution
> once an external event (NOTIFY event) occurs.

Use an advisory lock, they're ideal for that job:

http://www.postgresql.org/docs/current/static/explicit-locking.html#ADVISORY-LOCKS

> 2) In Pg/SQL I can implement a loop (until something happens) to
> pause/wait. This costs CPU time -> Is there another solution?

Depends on what you're waiting for. Details?

Again, an advisory lock may be a candidate.

--
Craig Ringer

<div class="moz-cite-prefix">On 10/15/2012 02:54 PM, P. Broennimann wrote:<br /></div><blockquote
cite="mid:CACm4aU-eFNaDiYw9uH1n9VPgHN5SqseJ3HqsbgXd7-Cw++NF-g@mail.gmail.com"type="cite"><font face="courier new,
monospace">Hithere</font><div><font face="courier new, monospace"><br /></font></div><div><font face="courier new,
monospace">1)Can a Pg/SQL function "listen" for a notification sent from an external instance?</font></div><div><span
style="font-family:'couriernew',monospace">I would like my stored function to pause/wait and continue its execution
oncean external event (NOTIFY event) occurs.</span></div></blockquote><blockquote
cite="mid:CACm4aU-eFNaDiYw9uH1n9VPgHN5SqseJ3HqsbgXd7-Cw++NF-g@mail.gmail.com"type="cite"><br /><div><font face="courier
new,monospace">2) In </font><span style="font-family:'courier new',monospace">Pg/SQL I can implement a loop (until
somethinghappens) to pause/wait. This costs CPU time -> Is there another solution? Actually I would need something
likea trigger to give my stored function the signal to continue.</span></div></blockquote><br /> Yes. I played around
withthis a while ago, and thought the possibility was beyond cool.<br /> However, after playing around with it and
gettingsome constructive criticism, I decided that it would be better to keep the different application layers
completelyseparated, so that the database layer would not be talking directly to the GUI.<br /><br /> Here is an
examplefunction that I sent to the list a year+ ago.<br /><a
href="http://archives.postgresql.org/pgsql-general/2011-06/msg00322.php">http://archives.postgresql.org/pgsql-general/2011-06/msg00322.php</a><br
/><br/> On the client end, I had something like this:<br /> class ListenThread(threading.Thread):     <br />     def
__init__(self,frame):<br/>         threading.Thread.__init__(self)<br />         self.frame=frame<br />         <br />
   def run(self):<br />             HOST, PORT = "192.168.1.207", 8080<br />         <br />             # Create the
server,binding to localhost on port 8080<br />             server = SocketServer.TCPServer((HOST, PORT),
MyTCPHandler)<br/>             server.frame=self.frame<br />             # Activate the server; this will keep running
untilyou<br />             # interrupt the program with Ctrl-C<br />             server.serve_forever()   <br /><br /> 

On Mon, Oct 15, 2012 at 8:44 AM, P. Broennimann
<peter.broennimann@gmail.com> wrote:
> Thx for the feedback I will take a look.
>
> Here some details. Basically what I'd like to achieve:
>
>   Internet <-> AppliA <-> PostgreSQL <-> AppliB
>
> 1) AppliA receives a request from the internet and calls a Pg/SQL function.
> 2) The Pg/SQL function informs AppliB that there is some work waiting
> (NOTIFY).
> 3) AppliB does the work and posts the result back to PostgreSQL.
>
> ... and now here's the problem -> The initial Pg/SQL function should give
> back the result computed by AppliB -> It has to wait somehow and get
> informed when the work from AppliB is done.

I have done this exact workload many times and it's 100% doable.  It
relies on dblink and read committed hacks but it's very robust.

In step step 2, you write out a record that is going to hold the
response from AppliB.  It can be a simple text field or whatever you
need.  Then you spawn a dblink call and issue the notify (it has to be
dblink, because notifications are not delivered till end of
transaction.  After you spawn, you start sleep looping, waiting for
AppliB to set the response field.  You can see it because AppliB is
writing it in a different transaction, and in ReadCommitted mode you
are allowed to see other transactions that complete.  Once the
response is set (make sure you have a timeout), you can return the
reponse or optionally grab more data.

merlin