[OPEN-ILS-GENERAL] Re: Fast update => two copies of database

[LaJeunesse, Brad]

Hi Douglas,

I think you're misreading that. It's referring to the speed at which 
software features/enhancements are developed by our team and placed into 
production, not data propagation.

In our production environment, we currently have 3 database servers 
(each running an instance of Postgres). We're using slony [1] for 
database replication and failover.

--Brad

[1] http://gborg.postgresql.org/project/slony1/projdisplay.php

Douglas G. Danforth wrote:
> Hello,
> I learned of your efforts from a post on SlashDot.  They
> say
> "/Previously, if users wanted changes to their systems, they'd be put 
> into an 'enhancement queue'. Now, some features are implemented 
> overnight./"
>
> I think you can do better than "overnight".
> I worked at Tymshare in the '70s and we had the problem of updating our
> database of users for a e-mail system which would happen at off hours
> once a day.  Two of us said "We can do better than that" and came up
> with the method of having two identical copies of the database (in 
> your case
> this would be a good thing to have for redundancy and reliability).  One
> copy would be active with user processes running use that database.  The
> second "passive" database would be available for updating whenever a new
> account was added, removed, or changed.  When the change had been made
> then in one instruction (or a few) the pointer to the old database 
> would be
> switched to the new database.  Those users still accessing the old 
> database
> would slowly trickle off until the old database was empty of 
> processes.  It would
> then become the passive database ready for new additions, deletions, 
> or changes.
>
> Management at Tymshare was ecstatic at the speed and the fact that 
> changes to the
> database could take place at any time day or night.
>
> Just thought you might like to know.
>
> Best regards,
> Douglas G. Danforth, Ph.D.
>