Re: A method to asynchronously LISTEN ?

Hello,

This is not what was expected, when using the pgjdbc-ng driver there is no need to poll.
( [ https://github.com/impossibl/pgjdbc-ng | https://github.com/impossibl/pgjdbc-ng ] )

Do you think this feature could be added to the pgsql-jdbc driver ?
If not, our devs will have to use both drivers, the pgjdbc-ng for the asynchronous notifications without polling and the pgsql-jdbc for the dbcp.
Best regards,

Philippe Ebersohl

I got an example, and the declaration is much alike:

public void listen() throws SQLException {

PGNotificationListener listener = new PGNotificationListener() {

public void notification ( int processId , String channelName , String payload ) {

System. out .println( "Received Notification: " + processId + ", " + channelName + ", " + payload );

}

public void closed () {

// initiate reconnection & restart listening

}

};

Statement statement = conn .createStatement() ;

statement .execute( "LISTEN " + QNAME );

statement .close();

conn .addNotificationListener( listener );

System. out .println( "Wait for events" );

}

But there is no SELECT involved.
The main just does:

postgresNG .listen();

Object lock = new Object();

synchronized ( lock )

{

lock . wait ();

}

And when NOTIFY is issued from pgAdmin for example, the answer is printed immediately by the java program.

From: "Dave Cramer" <pg(at)fastcrypt(dot)com>
To: "Philippe Ebersohl" <philippe(dot)ebersohl(at)dalim(dot)com>
Cc: "List" <pgsql-jdbc(at)postgresql(dot)org>
Sent: Wednesday, 10 April, 2019 16:20:39
Subject: Re: A method to asynchronously LISTEN ?

Philippe,
Here's what's going on. The FE/BE protocol is somewhat synchronous and the NOTIFY is not being read until some other activity occurs on the channel.

If you modify your test to do the following it works fine.

public void testNotify() throws SQLException, InterruptedException {
AtomicBoolean flag = new AtomicBoolean( false );
Statement stmt = conn .createStatement();
((PGConnection) conn ).addNotificationListener( new PGNotificationListener() {
@Override
public void notification(PGNotification notification) {
System. out .println( "Notification Received" );
flag .set( true );
}
});
stmt.executeUpdate( "LISTEN mynotification" );
for ( int i= 0 ;i< 30 ; i++) {
stmt.execute( "select 1" );
Thread. sleep ( 1000 );
}
stmt.close();
Dave Cramer

[ mailto:davec(at)postgresintl(dot)com | davec(at)postgresintl(dot)com ]
[ http://www.postgresintl.com/ | www.postgresintl.com ]

On Wed, 13 Mar 2019 at 10:37, Dave Cramer < [ mailto:pg(at)fastcrypt(dot)com | pg(at)fastcrypt(dot)com ] > wrote:

On Mon, 11 Mar 2019 at 05:41, Philippe Ebersohl < [ mailto:philippe(dot)ebersohl(at)dalim(dot)com | philippe(dot)ebersohl(at)dalim(dot)com ] > wrote:

BQ_BEGIN

Hello,
could you pleas advise what is wrong in our testing ?

We retrieved the git sources, build them.
Modified the testNotify() as shown below.
When performing "NOTIFY mynotification" from an SQL client, we do not see the " Notification received" message as expected.

Thank you.

Regards,

Philippe Ebersohl

Hello Phillipe,

I will look at this as soon as possible. I am currently busy with other things.

Dave Cramer

[ mailto:davec(at)postgresintl(dot)com | davec(at)postgresintl(dot)com ]
[ http://www.postgresintl.com/ | www.postgresintl.com ]

BQ_BEGIN

From: "Philippe Ebersohl" < [ mailto:philippe(dot)ebersohl(at)dalim(dot)com | philippe(dot)ebersohl(at)dalim(dot)com ] >
To: "Dave Cramer" < [ mailto:pg(at)fastcrypt(dot)com | pg(at)fastcrypt(dot)com ] >
Cc: "David G. Johnston" < [ mailto:david(dot)g(dot)johnston(at)gmail(dot)com | david(dot)g(dot)johnston(at)gmail(dot)com ] >, "List" < [ mailto:pgsql-jdbc(at)postgresql(dot)org | pgsql-jdbc(at)postgresql(dot)org ] >
Sent: Thursday, 28 February, 2019 10:52:44
Subject: Re: A method to asynchronously LISTEN ?

Hello,
we retrieved the sources.

After starting the testNotify() below, we execute from pgadmin : notify mynotification

But the listener is not notified.

Did we miss something ?

public void testNotify() throws SQLException, InterruptedException {
final AtomicBoolean flag = new AtomicBoolean(false);
Statement stmt = conn.createStatement();
((org.postgresql.PGConnection)conn).addNotificationListener(new org.postgresql.PGNotificationListener() {
@Override
public void notification(org.postgresql.PGNotification notification) {
flag.set(true);

System.out.println("Notification received");

}
});
stmt.executeUpdate("LISTEN mynotification");
// stmt.executeUpdate("NOTIFY mynotification");

Thread.sleep(30000);

org.postgresql.PGNotification[] notifications = conn.unwrap(org.postgresql.PGConnection.class).getNotifications();
//assertTrue(flag.get());
//assertNotNull(notifications);
//assertEquals(1, notifications.length);
//assertEquals("mynotification", notifications[0].getName());
//long endMillis = System.currentTimeMillis();
//long runtime = endMillis - startMillis;
//assertNull("There have been notifications, although none have been expected.",notifications);
//assertTrue("We didn't wait long enough! runtime=" + runtime, runtime > 450);

stmt.close();
}

Regards.

Philippe Ebersohl

From: "Dave Cramer" < [ mailto:pg(at)fastcrypt(dot)com | pg(at)fastcrypt(dot)com ] >
To: "Philippe Ebersohl" < [ mailto:philippe(dot)ebersohl(at)dalim(dot)com | philippe(dot)ebersohl(at)dalim(dot)com ] >
Cc: "David G. Johnston" < [ mailto:david(dot)g(dot)johnston(at)gmail(dot)com | david(dot)g(dot)johnston(at)gmail(dot)com ] >, "List" < [ mailto:pgsql-jdbc(at)postgresql(dot)org | pgsql-jdbc(at)postgresql(dot)org ] >
Sent: Wednesday, 27 February, 2019 11:53:59
Subject: Re: A method to asynchronously LISTEN ?

On Wed, 27 Feb 2019 at 05:47, Philippe Ebersohl < [ mailto:philippe(dot)ebersohl(at)dalim(dot)com | philippe(dot)ebersohl(at)dalim(dot)com ] > wrote:

BQ_BEGIN

Hello,
I do not think there is a need for something like conn.listenSync().
We were just intrigued about the test.

In fact the current proposition should just work fine and we would like to test it.
How could we have access to a postgresql-*.jdbc.jar that includes this feature ?

Regards,
Philippe

BQ_END

I just rebased it over the current code [ https://github.com/pgjdbc/pgjdbc/pull/579 | https://github.com/pgjdbc/pgjdbc/pull/579 ]
You should be able to build it manually using the commandline instructions to get the pull

git checkout -b davecramer-notify master
git pull [ https://github.com/davecramer/pgjdbc.git | https://github.com/davecramer/pgjdbc.git ] notify

then use maven to build it

Dave Cramer

[ mailto:davec(at)postgresintl(dot)com | davec(at)postgresintl(dot)com ]
[ http://www.postgresintl.com/ | www.postgresintl.com ]

BQ_BEGIN

From: "David G. Johnston" < [ mailto:david(dot)g(dot)johnston(at)gmail(dot)com | david(dot)g(dot)johnston(at)gmail(dot)com ] >
To: "Dave Cramer" < [ mailto:pg(at)fastcrypt(dot)com | pg(at)fastcrypt(dot)com ] >
Cc: "Philippe Ebersohl" < [ mailto:philippe(dot)ebersohl(at)dalim(dot)com | philippe(dot)ebersohl(at)dalim(dot)com ] >, "List" < [ mailto:pgsql-jdbc(at)postgresql(dot)org | pgsql-jdbc(at)postgresql(dot)org ] >
Sent: Tuesday, 26 February, 2019 22:17:12
Subject: Re: A method to asynchronously LISTEN ?

On Tue, Feb 26, 2019 at 12:04 PM Dave Cramer < [ mailto:pg(at)fastcrypt(dot)com | pg(at)fastcrypt(dot)com ] > wrote:

BQ_BEGIN

On Tue, 26 Feb 2019 at 12:29, David G. Johnston < [ mailto:david(dot)g(dot)johnston(at)gmail(dot)com | david(dot)g(dot)johnston(at)gmail(dot)com ] > wrote:

BQ_BEGIN

The question being asked is whether executeUpdate method performs synchronous execution of the listener function queue during its execution.

This after sending the command to the server and the server completing execution of the NOTIFY. This assumes that the server places the notification on the channel for pick-up immediately (which will not be the case within a transaction).

While the client is, IIRC, synchronous, thus the local order of operations can be controlled, the server is asynchronous and thus this test is exposed to timing issues. Maybe we need something like conn.listenSync() that blocks until a notification payload is received on the connection...?

BQ_END

kind of defeats the asynchronous aspect, no ?

BQ_BEGIN

BQ_BEGIN

BQ_BEGIN

BQ_END

BQ_END

BQ_END

BQ_END

You'd still want an asynchronous API for people but JavaScript introduced the "async/await" feature for a reason. In this case making it "sync" instead of coding up wait loop seems desirable. That said its quite possible I'm missing some existing feature as I haven't used this API at all.

David J.

[ https://euroduo.dalim.com/ ]

BQ_END

[ https://euroduo.dalim.com/ ]

[ https://euroduo.dalim.com/ ]

BQ_END

BQ_END