On 27 December 2012 18:06, Heikki Linnakangas <hlinnakangas@vmware.com> wrote:
> This would have the advantage that MaxBackends would be kept set at zero,
> until we know the final value. That way it's obvious that you cannot trust
> the value of MaxBackends in a contrib module preload-function, for example,
> which would reduce the chance of programmer mistakes.
I admire your forward thinking on that; yes, that could cause
problems. But even then, we would be admitting that nobody now gets a
valid value of MaxBackends, which sounds like it might be a problem in
itself.
Perhaps we should try to solve that a different way? Can we ask for
reservations of bgworkers ahead of running their _init functions,
using an additional API call?
That way we'd know the final value and everybody would have the
correct value at init time.
-- Simon Riggs http://www.2ndQuadrant.com/PostgreSQL Development, 24x7 Support, Training & Services