Re: [Gnumed-devel] database replication with bucardo

[Busser, Jim]

On 2012-07-04, at 1:37 PM, Karsten Hilbert wrote:

> What happens here is that GNUmed tries to write a provider
> inbox message which fails because the PK PostgreSQL gets
> from the associated PK sequence is already there.
> 
> Now, this is either due to Bucardo having screwed up the
> sync between table and sequence on either end or else
> something else having screwed up the sequence on either end.

Is the replicated record intended to exactly mirror every datum *except* the pk 
which gets newly-assigned by the contextually-child system?

For example. suppose GNUmed was cloned via Bucardo at the point when GNUmed had 
only 11 patients, and now in one system Kirk is created (being assigned pk 12) 
at the same moment as -- in the other system -- some other "new patient" is 
being created (also assigned pk = 12).

This results in two systems whose data content

- except for the pks, will always always the same

but pks will

- be different, in the situation of any latency (or lag-time or transient 
disconnection) between syncs, during which there had been table alterations at 
both ends

- at the end of a sync, achieve the same sequence number at both ends, with the 
possibility of remaining the same (in sync) for those records which alter a 
given table only at one end between syncs

and this is ok because one would never expect a cross-database search to have 
meaning?

??

-- Jim