Gordan Bobic | 1 Jan 09:30 2010
Picon

Host shutdown hangs on vz service stop

I'm finding that when the host node is being shut down, the vz service 
often hangs on stopping. I only have one test VM running. If I vzctl 
enter the vm and issue halt, it stops cleanly. Doing shutdown -h seems 
more unreliable.

Unless each VM is stopped from inside it (usually using halt), stopping 
the container doesn't manage to stop it most of the time, it just blocks.

Thus in turn makes the vz service block on shutdown. Is this a known 
issue? Are there any configuration options that could be affecting this? 
As it is, the teardown process doesn't seem to be particularly robust.

I'm using CentOS 5.4 guests and host if that makes any difference.

On a separate-ish note, it might be worth fixing the CentOS 5.x 
template's rc.sysinit - it tries to do an awful lot of things that don't 
work in an OpenVZ container (e.g. udev, initializing random seed, 
accessing hwclock, and trying to access /proc/misc with doesn't exist in 
the guest). It's only a case of commenting out a few lines, but it would 
make a big difference in terms of appearing as a more stable, polished 
and mature solution, especially for those of us running with the 
(incredibly useful) init-logger patch that allow the output of init to 
be captured.

Thanks.

Gordan
Gregor at HostGIS | 1 Jan 21:16 2010

Re: Host shutdown hangs on vz service stop

> Unless each VM is stopped from inside it (usually using halt), stopping 
> the container doesn't manage to stop it most of the time, it just blocks.

Does it give a timeout? I had an issue before, where "vzctl stop XXX" 
would eventually time out, leaving the VE in an unpleasant state: no 
processes left, but not stopped so it can't be restarted. Does that 
sound familiar?

For that one, "vzcfgvalidate" showed that I had mistakes in some limits 
and correcting the VE config corrected the issue. We never knew why 
these mistakes would cause stops to timeout, but it did work.

--

-- 
HostGIS, Open Source solutions for the global GIS community
Greg Allensworth - SysAdmin, Programmer, GIS Person, Security
    Network+   Server+   A+   Security+   Linux+
    PHP   PostgreSQL   MySQL   DHTML/JavaScript/AJAX

"No one cares if you can back up — only if you can recover."
Gordan Bobic | 2 Jan 00:30 2010
Picon

Re: Host shutdown hangs on vz service stop

Gregor at HostGIS wrote:
>> Unless each VM is stopped from inside it (usually using halt), 
>> stopping the container doesn't manage to stop it most of the time, it 
>> just blocks.
> 
> Does it give a timeout? I had an issue before, where "vzctl stop XXX" 
> would eventually time out, leaving the VE in an unpleasant state: no 
> processes left, but not stopped so it can't be restarted. Does that 
> sound familiar?

That does, indeed, sound very familiar.

> For that one, "vzcfgvalidate" showed that I had mistakes in some limits 
> and correcting the VE config corrected the issue. We never knew why 
> these mistakes would cause stops to timeout, but it did work.

I just vzcfgvalidate-d my config files and it threw up no problems or 
errors. :-/

Gordan
Gregor at HostGIS | 2 Jan 02:47 2010

Re: Host shutdown hangs on vz service stop

> That does, indeed, sound very familiar.
> I just vzcfgvalidate-d my config files and it threw up no problems or 
> errors. :-/

Aww, darn. My thread is here:
http://bugzilla.openvz.org/show_bug.cgi?id=1345

It's embarrassing, though a relief, that by the time I had opened the 
ticket the problem never happened again. We checked vzcfgvalidate, 
checked vzctl-libs versions, and changed to other kernels but it never 
happened again. So, we never positively identified a cause and solution.

The only changes I had made in previous months were to upgrade the 
kernel (we use 2.6.24, now at ovz009.1) and to fix the config flaws. So 
if you're not seeing config issues, maybe a kernel upgrade?

They had also suggested updating vzctl and vzctl-libs, and verifying 
that both packages' versions matched. Though this wasn't our issue, it 
was one thing they suggested during hunting this bug.

--

-- 
HostGIS, Open Source solutions for the global GIS community
Greg Allensworth - SysAdmin, Programmer, GIS Person, Security
    Network+   Server+   A+   Security+   Linux+
    PHP   PostgreSQL   MySQL   DHTML/JavaScript/AJAX

"No one cares if you can back up — only if you can recover."
Gordan Bobic | 2 Jan 11:41 2010
Picon

Re: Host shutdown hangs on vz service stop

Gregor at HostGIS wrote:
>> That does, indeed, sound very familiar.
>> I just vzcfgvalidate-d my config files and it threw up no problems or 
>> errors. :-/
> 
> Aww, darn. My thread is here:
> http://bugzilla.openvz.org/show_bug.cgi?id=1345
> 
> It's embarrassing, though a relief, that by the time I had opened the 
> ticket the problem never happened again. We checked vzcfgvalidate, 
> checked vzctl-libs versions, and changed to other kernels but it never 
> happened again. So, we never positively identified a cause and solution.
> 
> The only changes I had made in previous months were to upgrade the 
> kernel (we use 2.6.24, now at ovz009.1) and to fix the config flaws. So 
> if you're not seeing config issues, maybe a kernel upgrade?
> 
> They had also suggested updating vzctl and vzctl-libs, and verifying 
> that both packages' versions matched. Though this wasn't our issue, it 
> was one thing they suggested during hunting this bug.

I just a bit more testing, and it doesn't seem to end with a time-out
message at service shutdown. The service _eventually_ stops but it takes
a minute or so. Issuing halt in the CT takes about 5 seconds, after
which issuing vzctl stop takes about 2, and after that service vz stop
is instantaneous. I'd be curious to learn how the shutdown process works
when initialized from the HN. I'm guessing this is somehow related to
the difference between issuing "halt" and "shutdown -h now" in the CT
("shutdown -h now" seems to take a lot longer).

(Continue reading)

Aleksandar Ivanisevic | 4 Jan 10:05 2010
X-Face
Picon

Re: kernel panic 2.6.18-164.2.1.el5.028stab066.10 drbd_connector_callback

Dietmar Maurer <dietmar@...>
writes:

Would you mind sharing which exact version did you compile and how
exactly?

I've managed to compile the latest 8.3.7rc1 and it kinda works, but
it involved a lot of manual fiddling, and I'm not really sure its
right.

> We also had problem with DRDB on that kernel. We ended up compiling 
> the DRBD module ourselves.
>
> - Dietmar
>
>> -----Original Message-----
>> From: users-bounces@...
[mailto:users-bounces@...] On
>> Behalf Of Aleksandar Ivanisevic
>> Sent: Donnerstag, 24. Dezember 2009 13:07
>> To: users@...
>> Subject: [Users] kernel panic 2.6.18-164.2.1.el5.028stab066.10
>> drbd_connector_callback
>> 
>> 
>> Hi,
>> 
>> I'm trying to use drbd in the latest rhel5 based openvz kernel but i'm
>> getting the following oops when starting DRBD. I'm using drbd83
>> package from centos extras for userland.
(Continue reading)

Dietmar Maurer | 4 Jan 17:57 2010

RE: Re: kernel panic 2.6.18-164.2.1.el5.028stab066.10 drbd_connector_callback

> Would you mind sharing which exact version did you compile

I compiled 8.3.4

> and how exactly?

AFAIR I removed the drbd headers from the rh kernel source - ugly hack.

May I suggest you post that question to the drbd list - I guess they know the solution.

- Dietmar

> 
> I've managed to compile the latest 8.3.7rc1 and it kinda works, but
> it involved a lot of manual fiddling, and I'm not really sure its
> right.
> 
> > We also had problem with DRDB on that kernel. We ended up compiling
> > the DRBD module ourselves.
Grégory Bulot | 8 Jan 10:26 2010

Nfs In VE

[be care, my english's awfull, and i hope my question is not too stupid]

Hi,

In Lenny, how i can use an nfs server in VE ?

In Host :
modprobe nfsd works (/proc/net/rpc created)

But in VE /proc/net/rpc not (mount --bind from host to VE don't work,
did i miss anything ?)

# aptitude search vz | grep  ^i | awk -F ' - ' ' { print $1 } '
i A linux-headers-2.6.26-2-common-o
i   linux-headers-2.6.26-2-openvz-6
i   linux-image-2.6-openvz-686     
i A linux-image-2.6.26-2-openvz-686
i A vzctl                          
i A vzquota 

>From Host :
# lsmod | grep nfs
nfsd                  188752  0 
lockd                  56456  1 nfsd
nfs_acl                 3680  1 nfsd
auth_rpcgss            34752  1 nfsd
sunrpc                166880  7 nfsd,lockd,nfs_acl,auth_rpcgss
exportfs                4832  1 nfsd

>From VE: (just crazy test, because there no module capacity in VE)
(Continue reading)

frank | 8 Jan 11:08 2010

gfs and localflocks

Hi all,
we have a couple of servers with OpenVZ RHEL 5.4 kernel, and sharing a 
GFS filesystem.
We have done some locking tests with "ping_pong" over the GFS 
filesystem, and we have realized that GFS is doing local flocks, not 
distributed ones.
It is not due to mount options:

# mount -t gfs -o rw /dev/volCluster/lvol0 /mnt/gfs

but we got

# grep /mnt/gfs /proc/mounts
/dev/mapper/volCluster-lvol0 /mnt/gfs gfs 
rw,hostdata=jid=1:id=65537:first=0,localflocks 0 0

"localflocks" flag is on! We don't know why. We have the same 
configuration in another pair of servers without OpenVZ, and there it 
works doing distributed locking without problem; so it seems it is 
related to OpenVZ patched kernel.

Can this be a bug? Is there any explanation?

Regards.

Frank

--

-- 
Aquest missatge ha estat analitzat per MailScanner
a la cerca de virus i d'altres continguts perillosos,
(Continue reading)

Robert B | 8 Jan 15:51 2010

Setting defaults for templates...

I have a template that I use which needs a certain capability enabled each time I deploy it. Is there an easy way to set this capability inside the template itself?

I am sick of having to use this every time:

vzctl set # --capa sys_admin:on --save

If there's a way to script this, that would be even better because I could automate a few more things...

Thanks!

-- Robert

_______________________________________________
Users mailing list
Users@...
https://openvz.org/mailman/listinfo/users

Gmane