> Am 24.11.2016 um 15:05 schrieb Per Jessen <[email protected]>: > > Jan Beulich wrote: > >>>>> On 24.11.16 at 13:51, <[email protected]> wrote: >>> Jan Beulich wrote: >>> >>>>>>> On 22.11.16 at 07:50, <[email protected]> wrote: >>>>> i just installed our first Leap 42.2 DomU and as i wanted to adapt >>>>> our usual scriptset i noticed, that the settings for independent >>>>> wallclock seem to be missing. >>>>> While the doc is stating that independent wallclock settings should >>>>> apply for Leap 42.2, /proc/sys/xen is not showing the appropriate >>>>> file. >>>>> >>>>> /proc/sys/xen # ls >>>>> balloon >>>>> >>>>> Has something gone wrong during installation, or is the >>>>> documentation faulty? >>>> >>>> I don't think the pv-ops kernel has ever supported the concept of >>>> independent wallclock as configurable behavior. If you observe any >>>> misbehavior without it, please open a bug. >>> >>> We started switching to independent wallclock when systemd was >>> introduced - systemd kept complaining about the clock being adjusted >>> or something like that. >> >> Over time we had kernel problems in that area, so much would >> depend on when exactly you found it necessary to switch. > > Judging by my mailing list postings, I think we started in March 2015. > https://lists.opensuse.org/opensuse-virtual/2015-03/msg00020.html > >> Also I don't think your reply really demonstrates any observed >> misbehavior with the pv-ops kernel, yet that's the only criteria >> based upon which fixing that kernel or considering to re-introduce >> that control would become possible routes forward. > > We observed ntp-related poor behaviour by systemd, that's all. Hundreds > of message being written about "Time has been changed". Switching to > local ntp and independent wallclock solved the issue. > >>> After a lot of research, we switched to using >>> independent_wallclock for all guests. Afair, this was/is also the >>> SUSE/Novell recommendation. >> >> Not as far as I know. As long as there are no bugs, running e.g. NTP >> once in Dom0 would seem the more efficient model compared to >> running it in every DomU. But of course I don't know who gave this >> recommendation, and in which context that was. > > http://www.novell.com/documentation/vmserver/config_options/data/b9qzhq5.html > https://doc.opensuse.org/documentation/leap/virtualization/html/book.virt/sec.kvm.managing.clock.html > > "To effectively use an external time source, such as NTP, for time > synchronization on a virtual machine, the virtual machine's independent > wallclock setting must be enabled (set to 1)." > > Perhaps not a recommendation, but given that systemd didn't like the > domU just relying on the Dom0 clock, using running ntp in the DomU and > enabling independent wallclock was a solution. > > https://wiki.debian.org/Xen/Clocksource > https://lists.opensuse.org/opensuse-virtual/2015-03/msg00022.html > > I remember finding it very difficult to get any solid information on > this topic. > > > -- > Per Jessen, Zürich (9.9°C) >
Hello, we started using independent wallclock because the little timejump while live migrating was sufficient to disrupt some time dependent services. All our hosts (and VMs) use NTP, but the drift seems to be enough. Setting independent wallclock solved this behaviour. Because this is in allignment with the official documentation, we never considered this to be a problem. Norman, Offenbach (-7°C) :-)
