Good day. I renewed my server (ubuntu 9.04 to 9.10) via:

sudo apt-get install update-manager-core
sudo do-release-upgrade

After reboot I get:

mountall start / starting
init: hwclock main process (956) terminated with status 77
mountall: / proc: unable to mount: Device or resource busy
mountall: / proc / self / mountinfo: No such file or directory
mountall: root filesystem isn't mounted
init: mountall main process (957) terminated with status 1
General error mounting filesystems.
A maintenance shell will now be started.
CONTROL-D will terminate this shell and re-try.
Give root password for maintenance
(or type Control-D to continue):

Help me please:)

The version of udev used by Ubuntu 9.10 doesn't support old kernels like 2.6.18 (Xen) or 2.6.23 (UML). If you're on Xen, switch to the "Latest 2.6 paravirt" kernel (and install ntpd to keep the clock straight). If you're on UML, um, file a ticket to switch to Xen.

I do not know that I have: XEN and UML
Can I save my data, or they have already lost?

If you're in Newark or Atlanta, or created your node since about 2008-05, you're on Xen.

To confirm it, edit your configuration profile in the manager. If your disk images begin with "/dev/xvd" and the available kernels include 2.6.18 and 2.6.31, you're on Xen.



Nothing is lost, or damaged at all. Just edit your configuration profile and switch to the "Latest 2.6 paravirt" kernel, like I said.

thx

Aside from general principal, is the ntp suggestion really required, and if so because of the paravirt kernel or Ubuntu 9.10? I'm still on Ubuntu 8.04 but I haven't been seeing any clock slips on my nodes whether running the older 2.6.18 or the newer 2.6.31 paravirt. The clock source does differ between the two (jiffies for 2.6.18, xen for 2.6.31), but they both seem stable.

-- David

It's because the paravirt kernel has no access to the host clock, so the clock will slip. As such, ntp really is required on the paravirt kernel.

I'm not sure why it'd be a concern, though. All you have to do is run "sudo apt-get install ntp" and you're done. It installs in a pre-configured state.

How much you notice the need for ntp may also depend on the host. I have two linodes in different DCs, and one drifts at an average of about 34ppm (about 3sec per day), while the other drifts at only 0.5ppm (about 1 sec every three weeks).

But the paravirt's clock source (at least on my Linodes) appears to be xen, which I believe links me to the hypervisor/host's clock.


I wouldn't call it a concern (though one less daemon is one less daemon), more a curiosity to understand better. I always install ntp on all my standalone boxes, but was just wondering about the need in the VPS environment, given that the host is keeping track.

From various older postings, I take it that the pv_ops kernels were not always very good with timing - I wonder if that's improved in recent versions, since I'm not seeing any slip on my (small sample size) 2 Linodes.

Of course, it's not like running ntp is going to hurt if the clock is already in sync, it just feels a little wasteful (I know, it's a low overhead daemon) to have dozens of ntp clients on a single host if the host itself can keep those guests in sync.

-- David

FWIW, I've used 2.6.29, 2.6.30.5 and 2.6.31.5, and haven't had problems with timekeeping. (I do run ntpd; drift is about 31.1 ppm.)

Edit: As an NTP server operator, I'd like to thank you for not wanting to cause unnecessary NTP traffic.

I've been running a pv_ops kernel for several months now, without ntp, and my clock seems to be right on the money.

I just did a dist upgrade from Ubuntu Server 9.04 to 9.10 on a XENed x64 Linode. I used the "Latest 2.6 Paravirt (2.6.31.5-x86_64-linode9)" in order to boot successfully. Thanks mnordhoff