> But because there's no enforcement of "every transaction should go
> through pgpool", it's not enough for the managers who are ultimately
> responsible for deciding on system design. In the hypothetical case,
> we're aiming at multimaster systems that are there for reliability,
> not performance. Decreasing the reliance on fault-tolerant hardware
> by increasing the potential for human error does not solve that
> problem.
Enforcement? There would be plenty of ways to achieve that. For
example, you could set pg_hba.conf so that on ly the host where pgpool
is running on could connect to the host where postmaster is running
on.
> We have been (my colleague Brad is the one who's been working on
> this). But for something to qualify for real production-grade use,
> it needs to be rock solid stable in heavy use for a considerable
> period of time. We're not there yet, is all I'm suggesting. (This
> principle is why it's also a good thing that Red Hat Enterprise isn't
> always completely up to date with the community sources.)
Right. It's your freedom that you do not use pgpool until you think
it's solid enough.
--
SRA OSS, Inc. Japan
Tatsuo Ishii