Re: "Your system clock just jumped" on Debian+VMware ESX

[Florian Reitmeir <florian@xxxxxxxxxxxx>]

On Fri, 29 Feb 2008, Roger Dingledine wrote:

> On Fri, Feb 29, 2008 at 09:01:48AM +0100, Florian Reitmeir wrote:
> > > I continue to see Tor report errors of jumps in the system time from
> > > 4397-4399 seconds. Strangely, this jump in time does not appear to be
> > > reflected in the time stamp that Tor assigns to the error message.
> > 
> > maybe it will help getting some more light on the issue if you just install a
> > cronjob which prints the output of "date" in a file, and then try to see
> > if the issue is a tor bug or a kernel bug. 4300 secounds are a bit more than
> > 60minutes.. so if its a kernel bug it should happen to other processes too.
> 
> The other hint here is that FreeBSD users used to see this problem,
> until a recent fix to FreeBSD made it go away. But last I checked,
> nobody has tracked down what the fix was.
> 
> What's happening inside Tor is that it goes to sleep until a new
> event happens or until one second elapses. When it wakes up, it asks
> gettimeofday() how much time passed, and in this case it finds that it
> has travelled back in time 4397 seconds since the last time it asked.
> 
> I'm guessing this is a kernel thing, so running "date" a lot will probably
> not help to notice it. Is gettimeofday() the wrong way to ask what time
> it is under vmware? :)

hm it's complicated :-) the depends on the os version.

http://www.vmware.com/pdf/vmware_timekeeping.pdf

-- 
Florian Reitmeir