[CRIU] [PATCH 1/2] timerfd: move the clockid check to dump time too

Cyrill Gorcunov gorcunov at gmail.com
Fri Oct 7 11:01:16 PDT 2016


On Fri, Oct 07, 2016 at 11:29:29AM -0600, Tycho Andersen wrote:
> On Fri, Oct 07, 2016 at 10:50:16AM +0300, Cyrill Gorcunov wrote:
> > On Thu, Oct 06, 2016 at 10:27:56PM +0000, Tycho Andersen wrote:
> > > but let's keep it during restore, since there might be images around that
> > > were generated before we did this check at dump.
> > > 
> > > Signed-off-by: Tycho Andersen <tycho.andersen at canonical.com>
> > > CC: Cyrill Gorcunov <gorcunov at openvz.org>
> > 
> > While I agree on the series in general, the question remains:
> > how should we restore abs time for bootclocks when container
> > is migrating? iirc in pcs6 we assume that when container
> > is offline the "sleep time" increasing. With this series
> > we need to investigate if we do the same (or if we should
> > do the same and remember total-sleep-time somewhere in
> > image).
> 
> So I don't know much about timerfds, but this stuff isn't namespaced,
> is it? So at least with the kernel API right now, there's probably no
> safe way for us to set the boot time, because it would set the boot
> time on the host too?

True, we can't set the clock but i don't remember from the scratch
if we keep only remaining time before clock trigger or we carry
abs time as well, need to revisit this code in criu. I'll ping
you later.

> 
> (I'm essentially totally ignorant here; I don't even know what you
> mean by PCS6 :)

virtuozzo 6 series, in-kernel containers


More information about the CRIU mailing list