OpenVZ and the Ubuntu Upstart init daemon [message #11810] |
Sat, 07 April 2007 07:46 |
Daniel Pittman
Messages: 26 Registered: January 2007
|
Junior Member |
|
|
G'day. I have been playing with the 'upstart' daemon that replaces the
traditional /sbin/init process in Ubuntu Edgy and Feisty.
I didn't have a huge degree of luck with this -- once I allocated a tty
for the console I ended up with the system failing during the bootstrap
process.
Before investing more time in this I was wondering if anyone else had
been working on getting upstart to cooperate in the OpenVZ boot process?
Regards,
Daniel
--
Digital Infrastructure Solutions -- making IT simple, stable and secure
Phone: 0401 155 707 email: contact@digital-infrastructure.com.au
http://digital-infrastructure.com.au/
|
|
|
Re: OpenVZ and the Ubuntu Upstart init daemon [message #11815 is a reply to message #11810] |
Sat, 07 April 2007 13:20 |
Daniel Pittman
Messages: 26 Registered: January 2007
|
Junior Member |
|
|
Daniel Pittman <daniel@rimspace.net> writes:
First, sorry for the cross-post. Since this is an issue with the
interaction between the two software packages -- and seems to be of
general interest on the OpenVZ users list where I first posted -- I felt
it was best to copy all three locations up front.
> G'day. I have been playing with the 'upstart' daemon that replaces
> the traditional /sbin/init process in Ubuntu Edgy and Feisty.
>
> I didn't have a huge degree of luck with this -- once I allocated a
> tty for the console I ended up with the system failing during the
> bootstrap process.
>
> Before investing more time in this I was wondering if anyone else had
> been working on getting upstart to cooperate in the OpenVZ boot
> process?
I spent some more time tracking this down and seem to be a conflict
between the expectations of the Upstart process and the environment
provided by OpenVZ when starting init in a new VE.
All the testing was with vzctl 3.0.16-1dso2 (from Debian/testing) and
Upstart 0.3.8 from Feisty.
At issue are the expectations about file description layout that are
made in init/main.c of upstart; on line 150 the 'control_open' method is
called.
That, eventually, binds a Unix domain socket for the Upstart control
connection, used to support 'telinit' among other options.
Later, on line 209, Upstart cleans up the environment:
/* Close any file descriptors we inherited, and open the console
* device instead. Normally we reset the console, unless we're
* inheriting one from another init process.
*/
for (int i = 0; i < 3; i++)
close (i);
During a normal boot sequence those three descriptors are bound to
STDIN, OUT and ERR just as expected. The result is that Upstart
correctly rebinds the console and everything "just works."
Under OpenVZ startup, however, things don't quite look the same.
OpenVZ will exec the init process with *no* open file descriptors.
This causes the control_open method to bind file description 0 to the
Unix domain socket -- and the later loop to carefully close the socket
again.
I was able to verify my belief that this was causing the problems by
replacing init with a shell script that simply did:
exec /sbin/init.real "$@" </dev/null >/dev/null 2>/dev/null
Once that was in place Upstart kept the control socket open and
proceeded considerably further through the boot sequence.
It isn't quite clear to me who is at fault here -- I presume that in the
real world the kernel and linuxrc processes never start upstart without
the first three file descriptors open, but I don't know that is assured
anywhere.
I am happy to help with further testing or to explain how to configure
an OpenVZ environment to reproduce this, if desired.
Regards,
Daniel
--
Digital Infrastructure Solutions -- making IT simple, stable and secure
Phone: 0401 155 707 email: contact@digital-infrastructure.com.au
http://digital-infrastructure.com.au/
|
|
|
[Devel] Re: OpenVZ and the Ubuntu Upstart init daemon [message #11821 is a reply to message #11810] |
Sun, 08 April 2007 03:41 |
Daniel Pittman
Messages: 26 Registered: January 2007
|
Junior Member |
|
|
Scott James Remnant <scott@ubuntu.com> writes:
> On Sat, 2007-04-07 at 23:20 +1000, Daniel Pittman wrote:
Please respect the reply-to -- the issues, if any, now seem to be
located in the Upstart area rather than OpenVZ specific.
>> > G'day. I have been playing with the 'upstart' daemon that replaces
>> > the traditional /sbin/init process in Ubuntu Edgy and Feisty.
[...]
>> I spent some more time tracking this down and seem to be a conflict
>> between the expectations of the Upstart process and the environment
>> provided by OpenVZ when starting init in a new VE.
>
> I appreciate your debugging efforts; I've been asking OpenVZ users to
> help with the problem for some time with little avail.
Well, hopefully I should be able to dig through the issues and get to
the bottom of the problems.
[...]
>> This causes the control_open method to bind file description 0 to the
>> Unix domain socket -- and the later loop to carefully close the socket
>> again.
>
> This was a bug introduced relatively recently while clearing out the
> main() function, and has already been fixed in bzr trunk:
Great. I have worked around that for the moment with my local shell
script and am working on getting the rest of the boot process up and
running.
Is it likely that this issue will be fixed before the Feisty release?
I know the freeze is in effect at the moment but this is definitely a
big incompatibility with the OpenVZ system.
> I've had reports of OpenVZ failing for some time before this though;
> can you confirm that with this patch, Upstart works correctly?
No, but it certainly gets a lot further. I am still working on
debugging the early "boot" process to find out where it fails.
I will follow up on this (on the Ubuntu list) once I have more results.
Hopefully this should lead to full compatibility between the two.
Once I have final results I will also copy of OpenVZ list, but I don't
think they need to know all the messy details of the debugging process.
Regards,
Daniel
--
Digital Infrastructure Solutions -- making IT simple, stable and secure
Phone: 0401 155 707 email: contact@digital-infrastructure.com.au
http://digital-infrastructure.com.au/
_______________________________________________
Devel mailing list
Devel@openvz.org
https://openvz.org/mailman/listinfo/devel
|
|
|
|
|
|
Re: OpenVZ and the Ubuntu Upstart init daemon [message #11827 is a reply to message #11825] |
Sun, 08 April 2007 06:45 |
Daniel Pittman
Messages: 26 Registered: January 2007
|
Junior Member |
|
|
"Stephen Fletcher" <fletch@cisco.com> writes:
> Ill try and make up for my lameness :)
Actually, the bug has been tracked down -- when upstart is run without
any file descriptors open (as it is under OpenVZ but not during normal
Ubuntu booting) it would accidentally close the control socket.
I am waiting to hear about the one remaining issue, a signal killing
upstart in the VE where it doesn't kill the default init, before I can
declare Feisty fully working ... but I do have a patched upstart and
init scripts set that let me run up a full VE now. :)
Thanks for the comment though. It looks like that effect is achieved by
OpenVZ -- at least in my version -- which claims the kernel command line
had *only* the word 'quiet' present.
Regards,
Daniel
--
Digital Infrastructure Solutions -- making IT simple, stable and secure
Phone: 0401 155 707 email: contact@digital-infrastructure.com.au
http://digital-infrastructure.com.au/
|
|
|
[Devel] Re: OpenVZ and the Ubuntu Upstart init daemon [message #11869 is a reply to message #11815] |
Sat, 07 April 2007 13:51 |
Matthias Urlichs
Messages: 1 Registered: April 2007
|
Junior Member |
|
|
Hi,
Daniel Pittman:
> Under OpenVZ startup, however, things don't quite look the same.
> OpenVZ will exec the init process with *no* open file descriptors.
>
Upstart/init is not the only program that breaks when stdin/out/err are
closed when you call it.
On the other hand, it's not too much to expect upstart to be a bit more
lenient in what it expects, given that there may or may not be a way to
have these file descriptors open when launching it.
So in this case I'd like to apply the "when in doubt, do both"
guideline and fix upstart *and* openVZ.
--
Matthias Urlichs | {M:U} IT Design @ m-u-it.de | smurf@smurf.noris.de
Disclaimer: The quote was selected randomly. Really. | http://smurf.noris.de
- -
To be thrown on one's own resources is to be cast in the very lap of
fortune; for our faculties undergo a development, and display an energy, of
which they were previously unsusceptible.
-- Benjamin Franklin
_______________________________________________
Devel mailing list
Devel@openvz.org
https://openvz.org/mailman/listinfo/devel
|
|
|
[Devel] Re: OpenVZ and the Ubuntu Upstart init daemon [message #11870 is a reply to message #11815] |
Sat, 07 April 2007 21:29 |
Scott James Remnant
Messages: 1 Registered: April 2007
|
Junior Member |
|
|
On Sat, 2007-04-07 at 23:20 +1000, Daniel Pittman wrote:
> > G'day. I have been playing with the 'upstart' daemon that replaces
> > the traditional /sbin/init process in Ubuntu Edgy and Feisty.
> >
> > I didn't have a huge degree of luck with this -- once I allocated a
> > tty for the console I ended up with the system failing during the
> > bootstrap process.
> >
> > Before investing more time in this I was wondering if anyone else had
> > been working on getting upstart to cooperate in the OpenVZ boot
> > process?
>
> I spent some more time tracking this down and seem to be a conflict
> between the expectations of the Upstart process and the environment
> provided by OpenVZ when starting init in a new VE.
>
I appreciate your debugging efforts; I've been asking OpenVZ users to
help with the problem for some time with little avail.
> All the testing was with vzctl 3.0.16-1dso2 (from Debian/testing) and
> Upstart 0.3.8 from Feisty.
>
> At issue are the expectations about file description layout that are
> made in init/main.c of upstart; on line 150 the 'control_open' method is
> called.
[...]
> This causes the control_open method to bind file description 0 to the
> Unix domain socket -- and the later loop to carefully close the socket
> again.
>
This was a bug introduced relatively recently while clearing out the
main() function, and has already been fixed in bzr trunk:
http://codebrowse.launchpad.net/~keybuk/upstart/main/revisio n/scott%
40netsplit.com-20070313191319-gztu8c0r0sjla0hp?start_revid=s cott%
40netsplit.com-20070316171800-scmrd6w9r22uf4me
I've had reports of OpenVZ failing for some time before this though; can
you confirm that with this patch, Upstart works correctly?
Scott
--
Scott James Remnant
Ubuntu Development Manager
scott@ubuntu.com
_______________________________________________
Devel mailing list
Devel@openvz.org
https://openvz.org/mailman/listinfo/devel
|
|
|