Re: BUG #3948: date/time functions returning wrong value

On Sun, Feb 10, 2008 at 09:33:46PM -0500, Tom Lane wrote:
> Alvaro Herrera <alvherre(at)commandprompt(dot)com> writes:
> > I think what I conclude from this is that Windows TZ database is so
> > bogus that we should avoid trying to rely on it -- I say if the user
> > does not set "timezone" in postgresql.conf, refuse to start.
>
> Remember we're also relying on the OS for the time of day :-(.
> I'm not sure there's any point in trying to be a lot better than
> it is about timekeeping. In particular, even if the user has set
> the displayed local time correctly, what Windows will tell us the
> UTC time is depends entirely on its idea of the timezone offset.
> If we have a different idea of the timezone offset, all it will buy
> us is complaints from users that our time is wrong.
>
> (Now, this pessimistic view of things might be obsolete if Windows
> systems commonly get their UTC time from NTP, as is standard on
> Linux and Mac these days. I hadn't heard that Microsoft was up
> to speed on that, though --- if they did, I think it'd force them
> to be a whole lot more careful keeping their local timezone knowledge
> up to date ...)

Since Windows 2000, all domain members in a Windows domain member machines
will synchronize their time with NTP. It's a requirement of Kerberos that
the clocks don't drift. Recommended best practice is to have your domain
controller sync to either an external source on the net or to a GPS.

Since Windows 2000, there has been an NTP client included in Windows.

Since Windows XP, I think (could be 2000), all non-domain machines will
ntp-synchronize with servers at microsoft by default.

Since Windows NT, it's been possible and recommended to synchronize all
members of a domain to the domain controllers with the "net time" tools
available.

//Magnus