Re: [Gnumed-devel] database replication with bucardo

[Busser, Jim]

On 2012-07-04, at 11:24 PM, Karsten Hilbert wrote:

> Bucardo needs to contain code to either mediate or prevent
> such situations or else it is not fit for purpose.

I cannot see how the above can be enforced.

Imagine two distinct installations of GNUmed, which have been set up (in 
Bucardo) as master-master, and suppose there is a loss of connectivity between 
the two masters -- say an internet outage or even some lag in connectivity.

Is it being suggested that when clinician A is busy using master "A" and 
clinician B is busy using master "B" that in the event of any loss of 
connectivity (or responsiveness) the masters are to be rendered unable to add 
new rows, because if they do add rows each master will be assigning pks based 
on the local postgres counter and cannot know the state of the other master.

I can understand how each master might contribute notice (and the detail) of 
every row change to a Bucardo daemon which, in the case of loss of 
connectivity, would queue the notices and catch up once the connectivity was 
re-established. But it leaves unanswered how the pks can be syncd.

The overview at Bucardo does not clearly enough answer this for my simple brain

        http://bucardo.org/wiki/Bucardo/Documentation/Overview

whereas this page suggests that postgres 'sequences' may need to be adjusted:

        http://bucardo.org/wiki/Bucardo/Sequences

-- Jim