On Sat, Feb 1, 2014 at 10:49 PM, Clemens Eisserer <linuxhippy@gmail.com> wrote:
> My question on the list was merely to make sure there are no
> side-effects when increasing this delay above what seems to be
> considered safe limits. However, I still wonder why this parameter is
> capped to 10s and whether this restriction could be lifted in future
> postgresql versions?
I don't think there's any practical reason, other than that it was
assumed that increasing it further was not useful. There is perhaps a
tendency to set GUC limits as high as seems reasonable without
consider niche use-cases such as yours. If you want to hack it to go
higher it should be fine, provided that WalWriterDelay *
HIBERNATE_FACTOR cannot ever overflow a 32-bit signed integer. But
since those are milliseconds and not microseconds, it seems pretty
safe. This applies to 9.2+ only. I didn't check what things look like
back when the delay was passed to pg_usleep(), which was the case in
9.1.
--
Regards,
Peter Geoghegan