Slony1 PostgreSQL Replication - Use of WAIT FOR EVENT

Author

Use of WAIT FOR EVENT

Nuno Santos

2005-07-27, 7:27 am

Hello.

I'm trying to improve the time it takes to add a new table to an
existing set. With 3 replicas, all in the same host, it is taking close
to 10 seconds to add a new table. I suspect the problem is with all the
WAIT FOR EVENT commands I'm issuing during the creation script. The
documentation is not totally clear about when it is necessary to wait
for the previous event to be processed, so I don't know if I can remove
some waits. Could anyone give me some hints about which are really
necessary? For instance, is it necessary to wait between the subscribe
set commands?

My script is the following:

cluster name = cluster;
node 1 admin conninfo = 'host=localhost dbname=db1 user=postgres;
node 2 admin conninfo = 'host=localhost dbname=db2 user=postgres;
node 3 admin conninfo = 'host=localhost dbname=db3 user=postgres;
create set (id=4, origin=1, comment='Temp set');
WAIT FOR EVENT (ORIGIN = ALL, CONFIRMED = ALL, WAIT ON = 1);
set add table (set id=4, origin=1, id=6, fully qualified name =
'public.dir17', comment='dir17');
WAIT FOR EVENT (ORIGIN = ALL, CONFIRMED = ALL, WAIT ON = 1);
subscribe set ( id = 4, provider = 1, receiver = 2, forward = yes);
WAIT FOR EVENT (ORIGIN = ALL, CONFIRMED = ALL, WAIT ON = 1);
subscribe set ( id = 4, provider = 1, receiver = 3, forward = yes);
WAIT FOR EVENT (ORIGIN = ALL, CONFIRMED = ALL, WAIT ON = 1);
merge set ( ID = 1, ADD ID = 4, ORIGIN = 1 );

I have another related question. If I send two commands before waiting
for an event, is the wait guaranteed to return only when both events
were processed? For instance:

subscribe set ( id = 4, provider = 1, receiver = 2, forward = yes);
subscribe set ( id = 4, provider = 1, receiver = 3, forward = yes);
WAIT FOR EVENT (ORIGIN = ALL, CONFIRMED = ALL, WAIT ON = 1);
merge set ( ID = 1, ADD ID = 4, ORIGIN = 1 );

I know that Slonik only waits for the previous event, but since the
subscribe events are issued on the same node, aren't they ordered, so
that it is enough to wait for the last one?

Thank you in advance,
Nuno

Christopher Browne

2005-07-27, 9:24 am

Nuno Santos wrote:

>Hello.
>
>I'm trying to improve the time it takes to add a new table to an
>existing set. With 3 replicas, all in the same host, it is taking close
>to 10 seconds to add a new table. I suspect the problem is with all the
>WAIT FOR EVENT commands I'm issuing during the creation script. The
>documentation is not totally clear about when it is necessary to wait
>for the previous event to be processed, so I don't know if I can remove
>some waits. Could anyone give me some hints about which are really
>necessary? For instance, is it necessary to wait between the subscribe
>set commands?
>
>My script is the following:
>
>cluster name = cluster;
>node 1 admin conninfo = 'host=localhost dbname=db1 user=postgres;
>node 2 admin conninfo = 'host=localhost dbname=db2 user=postgres;
>node 3 admin conninfo = 'host=localhost dbname=db3 user=postgres;
>create set (id=4, origin=1, comment='Temp set');
>WAIT FOR EVENT (ORIGIN = ALL, CONFIRMED = ALL, WAIT ON = 1);
>set add table (set id=4, origin=1, id=6, fully qualified name =
>'public.dir17', comment='dir17');
>WAIT FOR EVENT (ORIGIN = ALL, CONFIRMED = ALL, WAIT ON = 1);
>subscribe set ( id = 4, provider = 1, receiver = 2, forward = yes);
>WAIT FOR EVENT (ORIGIN = ALL, CONFIRMED = ALL, WAIT ON = 1);
>subscribe set ( id = 4, provider = 1, receiver = 3, forward = yes);
>WAIT FOR EVENT (ORIGIN = ALL, CONFIRMED = ALL, WAIT ON = 1);
>merge set ( ID = 1, ADD ID = 4, ORIGIN = 1 );
>
>
>
>
I would be inclined to take out the WAIT FOR EVENTs that occur before
the SUBSCRIBE SET requests, but not any of the subsequent ones.

>I have another related question. If I send two commands before waiting
>for an event, is the wait guaranteed to return only when both events
>were processed? For instance:
>
>subscribe set ( id = 4, provider = 1, receiver = 2, forward = yes);
>subscribe set ( id = 4, provider = 1, receiver = 3, forward = yes);
>WAIT FOR EVENT (ORIGIN = ALL, CONFIRMED = ALL, WAIT ON = 1);
>merge set ( ID = 1, ADD ID = 4, ORIGIN = 1 );
>
>I know that Slonik only waits for the previous event, but since the
>subscribe events are issued on the same node, aren't they ordered, so
>that it is enough to wait for the last one?
>
>
That seems right.

The thing is, the SUBSCRIBE SET events likely run for long enough that
any time "wasted" by waits should be pretty immaterial there. Waiting a
few seconds for event propagation after a SUBSCRIBE SET that took 20
minutes doesn't strike me as a problem.

What I wouldn't do is to put WAIT FOR EVENT after each of the operations
that modify sets, thus, CREATE SET, SET ADD TABLE, SET ADD SEQUENCE, and
such...

Sponsored Links