[Xen-users] CentOS 5.4 jumps in time and shows 9999.99% CPU - anyone.

[Florian Heigl <florian.heigl@xxxxxxxxx>]

Hi,

(repost: the mailer daemon just ated the mail as described. I hope it was yummy)

I spend half an hour asking myself if this is for xen-users or -devel.
Since I can't decide I'll cross-post a little, hoping that the topic
catches only people who happend to have or fix this issue.:

I'm pretty amazed by one bug that I can't seem to pinpoint or get rid
of right now:
This is using old CentOS 5.4 with the 3.x Xen version that came with it.

Sometimes, without much reason, a domU will make a jump in time to the
future, by like 30 minutes to an hour.
This, as you could expect, breaks the OS quite badly. It stops most
processing until that point in time is reached and then
happily continues to work.
You can recognize it when all processes in top end up at 9999.99% cpu
usage, or by issueing date (examples follow)

There are NO related errors in dmesg or on the host. No others either.
reboot -f  inside the VM will not work any more once this happened, a
crash via sysrq does work.

There have been old reports of this issue, some, but not all in
combination with live migration, which we're not using.
In that case the restore functions overwrote the timer information, I
seem to be getting broken information instead. (wtf)
So live migration issue == fixed. Other issue == i don't know

If you look at this and ignore the comments from other people that
didn't have the same issue (i.e. they have a vmware timeacceleration
platform) then it makes me think this is something introduced with 5.4
https://www.centos.org/modules/newbb/viewtopic.php?topic_id=23402
(timmsteve and the thread starter)


I would like to fix the issue instead of switching to independent wallclock.
Reasons:
running independent wallclock uses up CPU time for nothing.
having one all VMs run using the time of one shared perfectly synced
clock just makes a lot more sense.

Backporting a patch to old Xen wouldn't worry me too much, I just need
to get a solution somehow.

I'm not sure what the bigger problem is right now:
Not finding a way to fix it, or not being able to reproduce it at will.
Either of this would make me glad right now.


OS CentOS 5.4
Upgrading the hosts would is possible, upgrading the VMs is *not*.

Real time:
Tue Nov 22 16:37:29  2011


Time / top displayed on the system, almost 20 minutes in the future.

top - 16:52:04 up 8 days,  3:41,  2 users,  load average: 0.25, 0.08, 0.02
Tasks:  87 total,   1 running,  86 sleeping,   0 stopped,   0 zombie
Cpu(s):  0.9%us,  0.7%sy,  0.0%ni, 97.7%id,  0.1%wa,  0.0%hi,  0.1%si,  0.5%st
Mem:    524464k total,   514608k used,     9856k free,    53556k buffers
Swap:   522104k total,       52k used,   522052k free,    60292k cached

 PID USER      PR  NI  VIRT  RES  SHR S %CPU %MEM    TIME+  COMMAND
 1380 xxx       15   0  194m  24m  12m S 9999.0  4.7  17:17.77 xxxx
   1 root      15   0  2088  732  628 S 9999.0  0.1   0:00.11 init
   2 root      RT  -5     0    0    0 S 9999.0  0.0   0:02.50
migration/0
   3 root      34  19     0    0    0 S 9999.0  0.0   0:00.00
ksoftirqd/0
   4 root      RT  -5     0    0    0 S 9999.0  0.0   0:00.00
watchdog/0
   5 root      10  -5     0    0    0 S 9999.0  0.0   0:00.05
events/0
   6 root      14  -5     0    0    0 S 9999.0  0.0   0:00.21 khelper
   7 root      11  -5     0    0    0 S 9999.0  0.0   0:00.00 kthread


Current Clocksource = jiffies
Independent wallclock = 0
permitted_clock_jitter = 10000000


Did anybody who had this issue *fix* it?
Is it *definitely* gone with indepent clocks?
Does CentOS / RedHat really have a QA team? :)

Going to read through the kernel-xen patches in 5.7 now :/

Thanks for reading / any help,
Florian

-- 
the purpose of libvirt is to provide an abstraction layer hiding all
xen features added since 2006 until they were finally understood and
copied by the kvm devs.

_______________________________________________
Xen-users mailing list
Xen-users@xxxxxxxxxxxxxxxxxxx
http://lists.xensource.com/xen-users