On Sun, 6 Jun 1999, Bruce Momjian wrote:
> I liked unlinking because it allowed old backends to still see the
> segments if they still have open file descriptors, and new backends can
> see there is no file there. That seemed nice, but you clearly
> demostrated it caused major problems. Maybe truncation is the answer.
> I don't know, but we need to resolve this for 6.5. I can't imagine us
> focusing on this like we have in the past few weeks. Let's just figure
> out an answer. I am on IRC now if someone can get on to discuss this. I
> will even phone someone in US or Canada to discuss it.
Personally, I think the right thing is to unlink the unused segments. For
the most part keeping them around is not going to cause any problems, but
I can't really think of any good reasons to keep them around. Keeping the
database directories clean is a good thing in my opinion.
> What is it on the backend that causes some backend to think there is
> another segment. Does it just go off the end of the max segment size
> and try to open another, or do we store the number of segments
> somewhere. I thought it was the former in sgml() area. I honestly don't
> care if the segment files stay around if that is going to be a reliable
> solution.
The new patch from Hiroshi Inoue <Inoue@tpf.co.jp> works. I believe it is
a reliable solution, I just don't agree it's the right one. That is
probably just a matter of opinion however. As his patch doesn't have any
immediate problems, so I vote for that to be included in 6.5.
Thanks,
Ole Gjerde