Oh, its a sticky problem, to be sure. Have to get something working at
some point, though...
You're point about approximate solutions is well taken.
On Friday, September 19, 2003, at 01:11 AM, Tom Lane wrote:
> Andrew Rawnsley <ronz@ravensfield.com> writes:
>> On Friday, September 19, 2003, at 12:21 AM, Tom Lane wrote:
>>> What is the problem you really want to solve?
>
>> Hacking some semblance of sequence support into eRserver.
>
> Hmm. I don't see a lot of value in an approximate solution. Either
> the
> sequence is up to date at the slave, or it is not. What's the point of
> "almost up to date"? You'd still have to take some action along the
> lines of "select setval('seq', (select max(col) from tab))" during any
> failover. If you have to do that, it doesn't matter what the sequence
> value was.
>
> Perhaps sequence increments could be broadcast to slaves as-is, using
> some variant of the existing erserver protocol that understands that
> these things happen outside transaction control.
>
> regards, tom lane
>
--------------------
Andrew Rawnsley
President
The Ravensfield Digital Resource Group, Ltd.
(740) 587-0114
www.ravensfield.com