Ajay Dalvi wrote:

>---------- Forwarded message ----------
>From: Ajay Dalvi <dalvi_ajay at extenprise.net>
>Date: 12 Jul 2005 12:39:46 +0530
>Subject: Problem related to slony
>To: cbbrowne at gmail.com
>
>
> 
> Hi Christopher,
> 
> I want to ask one problem that I am getting while using slony1.1.0
> 
> The scenario I am following is
> I am having one Master and two slave nodes.
> I have configured them properly. 
> Now I have written one automated script. In this script I first start
>"slon" process for master and for
> each slave. The I subscribe each slave for particular set one at a
>time. Now since i don't want the replication process to be run
>continuously, I am looking message "copy set [set no] done" in the
>logs. As soon as I get this message, I am unsubscribing that node.
>After that same process is followed for other slave node.
>           Now the problem I am getting is that. Sometimes the node is
>not getting unsubscribed from the set. So when I run my srcipt once,
>it does the replication and comes to an end. Now when next time I run
>that script, it is giving me following error.
> 
> -------------------------------------------------------------------------------------------------------
> "select "_cistera_cluster".setAddTable_int(1, 1,
>'"public"."component_properties"', 'name_pk', 'Table
>component_properties'); " PGRES_FATAL_ERROR ERROR:  Cannot insert a
>duplicate key into unique index sl_table-pkey"
> -------------------------------------------------------------------------------------------------------
> 
> I think this is just because the node is not getting unsubscribe when
>last time the script gets executed.
> 
> One more thing is that, to konw that replication is over and to stop
>the slon prcess, I am looking log entries for message "copy set [set
>no] done".  When i get this message, I am immediately unsubscribing
>the node. So is there any more message I should also look in to the
>loggs before unsubscribing the node to overcome above error that I am
>getting?
> 
> So if you can give me any suggessions, about why is this happening......
>  
>

In future, please forward messages to the Slony-I mailing list (see
http://gborg.postgresql.org/mailman/listinfo/slony1).

It seems likely that the problem you are running into is that the
unsubscribe event isn't propagating as quickly as you want it to.

I would think that better ways to detect that the set is *TRULY*
subscribed would include:

1.  Watching other nodes for the confirmation of the subscription event.

2.  Watching other nodes for confirmation of SYNC events on the new
subscriber node.

3.  Looking at sl_subscribe on the subscriber to see if the set is
marked as subscribed.  It is marked as NOT subscribed there until the
COPY_SET event completes.