Bug#858418: [PKG-Openstack-devel] Bug#858418: Is it possible to release fixed packages now?

[Thomas Goirand]

On 01/18/2018 10:11 PM, Kepi wrote:
> Hi,

FYI:

https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=887736

Please do test the package that I built, so we can make sure it fixes
the issue for you, and report to this bug if it's a good fix or not, to
give yet-another insurance to the release team that we're moving to the
right direction.

Cheers,

Thomas Goirand (zigo)

Bug#858418: [PKG-Openstack-devel] Bug#858418: Is it possible to release fixed packages now?

[Thomas Goirand]

On 01/18/2018 10:11 PM, Kepi wrote:
> I'm really not sure what is best approach now and I understand your
> point of view. Problem is that currently Open vSwitch from stable
> distribution:
> 
> * can break your primary network with correct configuration
> * there is no info in README that you shouldn't use interfaces
> * IMHO there shouldn't be any such info :) as we want to use interfaces,
>   one reason why we love Debian :)
> * workaround exists, even when it is ugly
> 
> In my case issue is fixed using this workaround and I can use my servers
> again. I simply wanted to help others who will hit this bug and may not
> have serial console ready.

Hi,

You do have very convincing argument, and with them, I believe I can
attempt to get the stable release team to accept the "fix" in Stretch.
We'll have to get something better done for unstable though. Hopefully,
I'll find enough time to make the patch, build, and send the bug to the
release team for approval over this week-end.

> BTW: I'm truly sorry if my previous (or this) comment sounds offensive
> there is huge chance it is caused by my poor English skill.

No worries, this happened to me many times as well.

Cheers,

Thomas Goirand (zigo)

Bug#858418: [PKG-Openstack-devel] Bug#858418: Is it possible to release fixed packages now?

[Kepi]

Thomas Goirand  writes:

> On 01/09/2018 07:13 PM, Kepi wrote:
>> Is it possible to release fixed packages soon?
>> 
>> This bug is making remote systems unaccessible after reboot, IMHO
>> serverity should be critical. I don't think it is acceptable to have
>> such bug for more than half year on stable distribution.
>> 
>> I can confirm that upstream patch mentioned by Nicolas is working.
>> 
>> Thanks
>
> Hi,
>
> While upstream patch may indeed fix the issue, it still relies on the
> SYSTEMCTL_SKIP_REDIRECT systemd internal. Using any of these is in fact
> *very wrong* and should never be used in the packaging of any package in
> Debian. If you don't trust me, you can attempt asking the systemd
> maintainers about this.
>
> For this reason, unfortunately, it is very unlikely that the release
> team will accept such a change in the stable release, and they will
> likely ask for a better fix.
>
> So, we should come up with a better way to fix the issue, a way that
> will be accepted by the release team. According to my own experience, it
> is also very unlikely that the release team will accept some new
> .service files either, unfortunately.
>
> I'd very happily accept any contribution if it's good enough. Though I
> may very much dislike gratuitous comments on how much the work of
> current package maintainers is unacceptable, even if I only very
> recently started to take care of the OpenVSwitch package (ie: at least 6
> months after this bug was opened, and many months after Stretch was
> released). So please, Kepi, moderate yourself. FYI, I've been thinking
> for a long time about ways to fix this problem, but so far haven't
> figure out an acceptable way.
>
> BTW, I don't agree severity should be set to critical. Definition for
> critical is:
>
> "makes unrelated software on the system (or the whole system) break, or
> causes serious data loss, or introduces a security hole on systems where
> you install the package."
>
> Here, only the network as controlled by OpenVSwitch breaks (ie: only
> OpenVSwitch breaks), and there's no data loss or security hole. Besides
> this, it's not ideal, but it can be worked around by not using
> /etc/network/interfaces for configuring OpenVSwitch. There's multiple
> ways to achieve that. Yes, it's not ideal, but until we find a
> satisfying way (in the eyes of the release team, that is) to fix the
> issue, I don't think there's a choice, and that's still a workaround.
>
> Last, is Philippe Latu's solution working? It doesn't feel right to me,
> and didn't have time to test it out...

Hi,

I run into related issue when you discussed problem with internal
variable just couple minutes after I filled report.

I agree it should be fixed "right way", but:

a) isn't _SYSTEMCTL_SKIP_REDIRECT which is already present in init
script internal too? It if is so, I think this workaround should
be included before better solution is found.

b) This issue is critical at least in impact. If you use OvS for primary
interface it renders whole system unusable after restart. I simply
believe this is critical issues because such bug shouldn't be present in
stable distribution. Or if it is present, it should be fixed (worked
around) in matter of days, not weeks or months.

I'm really not sure what is best approach now and I understand your
point of view. Problem is that currently Open vSwitch from stable
distribution:

* can break your primary network with correct configuration
* there is no info in README that you shouldn't use interfaces
* IMHO there shouldn't be any such info :) as we want to use interfaces,
  one reason why we love Debian :)
* workaround exists, even when it is ugly

In my case issue is fixed using this workaround and I can use my servers
again. I simply wanted to help others who will hit this bug and may not
have serial console ready.

BTW: I'm truly sorry if my previous (or this) comment sounds offensive,
there is huge chance it is caused by my poor English skill. I have
great respect for all work done by maintainers and I wasn't trying to
insult anybody. I still think that impact of this bug is greater than
it seams and should be at least worked around until better fix is ready.

Kind Regards,

Kepi


signature.asc
Description: PGP signature

Bug#858418: [PKG-Openstack-devel] Bug#858418: Is it possible to release fixed packages now?

[Thomas Goirand]

On 01/09/2018 07:13 PM, Kepi wrote:
> Is it possible to release fixed packages soon?
> 
> This bug is making remote systems unaccessible after reboot, IMHO
> serverity should be critical. I don't think it is acceptable to have
> such bug for more than half year on stable distribution.
> 
> I can confirm that upstream patch mentioned by Nicolas is working.
> 
> Thanks

Hi,

While upstream patch may indeed fix the issue, it still relies on the
SYSTEMCTL_SKIP_REDIRECT systemd internal. Using any of these is in fact
*very wrong* and should never be used in the packaging of any package in
Debian. If you don't trust me, you can attempt asking the systemd
maintainers about this.

For this reason, unfortunately, it is very unlikely that the release
team will accept such a change in the stable release, and they will
likely ask for a better fix.

So, we should come up with a better way to fix the issue, a way that
will be accepted by the release team. According to my own experience, it
is also very unlikely that the release team will accept some new
.service files either, unfortunately.

I'd very happily accept any contribution if it's good enough. Though I
may very much dislike gratuitous comments on how much the work of
current package maintainers is unacceptable, even if I only very
recently started to take care of the OpenVSwitch package (ie: at least 6
months after this bug was opened, and many months after Stretch was
released). So please, Kepi, moderate yourself. FYI, I've been thinking
for a long time about ways to fix this problem, but so far haven't
figure out an acceptable way.

BTW, I don't agree severity should be set to critical. Definition for
critical is:

"makes unrelated software on the system (or the whole system) break, or
causes serious data loss, or introduces a security hole on systems where
you install the package."

Here, only the network as controlled by OpenVSwitch breaks (ie: only
OpenVSwitch breaks), and there's no data loss or security hole. Besides
this, it's not ideal, but it can be worked around by not using
/etc/network/interfaces for configuring OpenVSwitch. There's multiple
ways to achieve that. Yes, it's not ideal, but until we find a
satisfying way (in the eyes of the release team, that is) to fix the
issue, I don't think there's a choice, and that's still a workaround.

Last, is Philippe Latu's solution working? It doesn't feel right to me,
and didn't have time to test it out...

Cheers,

Thomas Goirand (zigo)