Re: [Evolution] Evo just downloaded all my mail again

[Ángel <angel 16bits net>]

Adam Tauno Williams wrote:
> huw <huw synapticsilence net> wrote:
> > On Sat, 2013-06-29 at 07:55 +0100, huw wrote:
> > > I'd restarted my PC, then put it into suspend.  This morning I woke
> > it
> > > up, loaded Evolution, and it promptly downloaded all the mail in my
> > > inbox again, as if it had never been there (via an IMAP account).  I
> > > hadn't fiddled with any settings or deleted anything beforehand.
> > > Why did it do this?
> > This has just happened again, after booting into Linux after using
> > Windows.  Can nobody suggest why it's happening?
>
> Possibly dual-booting between different operating systems is trashing your clock. 
>  Windows stashes the system time in your local time zone and LINUX stashes 
> the system time in UTC.  So you are probably time-warping forward and back 
> every time you reboot; and you can't expect coherent behavior of *anything* 
> if you are now logging in four hours *prior* to the last time you logged in.

That's very easy to check. Just look at the clock on Windows and Linux
and see if they report different hours.

However, there's no need to virtualize for that:
> Dual-booting is evil; if possible just use virtualization.  Then many problems 'mysteriously' disappear.

Simply make Windows store the time in UTC.
(Set a DWORD named HKEY_LOCAL_MACHINE\SYSTEM\CurrentControlSet\Control
\TimeZoneInformation\RealTimeIsUniversal to 1) This is supported since
Windows Vista SP2 (Windows XP if you don't suspend/hibernate).

You could also put Linux in localtime instead, but that's uglier and you
would need to manage DST transitions yourself.


Tom Davies wrote:
> I think Windows Server has a different way of counting time that
> gradually gets it out-of-sync with the rest of the world.  On our
> network we have to reset the Windows Server clock about 1/year.  Just
> after it becomes around 10mins out of sync with reality the various
> desktops authentification suddenly starts to fail.
I would blame the clock of your server, not the OS. I have seen Linux
time drifting, too.

Running ntp periodically should "fix" your server. From the above data:
10 minutes in 6 months = 100 seconds/month ~= 3s/day


> > I don't think the desktops themselves suffer at all from having
> > different times in different OSes.  It's an interesting idea.  

Most programs won't care about the time, unless they are related to
security or authentication. Your authentication is probably based on
Kerberos tickets. If you put your OS date a few years back (eg. the
battery isn't working, and time is reset on each reboot) you will start
having problems on all https pages due to certificates still not issued.
For a few seconds or even minutes, nobody will care.

You would have inconsistent data, but the OSes themselves would work
fine. For programs expecting a monotonic clock such as for checking if
the summary file mtime is >= data file mtime, you would get some
pointless rebuilds (or non-updates) if accessing it from both OSes.

By the way, I have seen more time glitches with virtualization than
without. The clocks don't tick as accurately as expected by the guest OS
when inside a VM (but again, you usually don't care).

Regards