> Darcy Buskermolen wrote:
> 
>>
>>> - sl_table and sl_sequence replication to client nodes is deferred until
>>> a subscription is started. IMHO both should be replicated immediately
>>> (just as sl_set), because their tab_id/seq_id must be unique across the
>>> cluster and in case of complicated scenarios it might well happen that a
>>> duplicate id is used which is not discovered until the subscription is
>>> started.
>>
>>
>>
>> Jan/Chris thought on this one ?
> 
> 
> 
> Having another thought about that, using sl_rowid_seq to get unique ids 
> seems sufficient. pgAdmin users will never need to enter an ID with that 
> too.

Trying this, it appears that the int8 rowid isn't too well suited for 
the int4 tab_id....

Any other idea? I'm currently using COALESCE(MAX(tab_id), 0)+1, which is 
ok if there's only one serving node. In all other cases, the user has to 
track the ids himself. Not too satisfying, since apparently the ID isn't 
of any use to the user, just for internal use.

Simply changing sl_tab.tab_id and sl_sequence.seq_id to int8 would give 
a new chance to the sl_rowid_seq idea.


Regards,
Andreas