Bug#887736: stretch-pu: package openvswitch/2.6.2~pre+git20161223-3

[Adam D. Barratt]

On Wed, 2019-08-21 at 15:28 -0400, Felipe Sateler wrote:
> 
> 
> On Tue, Aug 20, 2019 at 6:23 PM Adam D. Barratt <
> a...@adam-barratt.org.uk> wrote:
> > Control: tags -1 + moreinfo
> > 
> > On Fri, 2018-01-19 at 15:21 +0100, Thomas Goirand wrote:
> > > I started maintaining OpenVSwitch long after the Stretch release,
> > and
> > > discovered #858418, which is very annoying for OpenVSwitch users.
> > > 
> > > tl;dr: #858418 prevent anyone that has a valid
> > > /etc/network/interfaces
> > > with OpenVSwitch directive from having a working network at boot.
> > The
> > > init script uses a non-documented, not-to-be-used systemd
> > internal,
> > > which is miserably failing.
> > > 
> > > After a long discussion with the bug reporter (which can be read
> > on
> > > the BTS), I came to the conclusion that he's right, and that the
> > most
> > > reasonable and safe way to fix the current situation is to apply
> > the
> > > patch he suggested (and which resulting debdiff I attached to
> > this
> > > bug).
> > 
> > As I understand things, that fix swaps use of one systemd internal
> > for another, which doesn't seem like a great plan.
> > 
> > When this was discussed (some time ago) on IRC, one of the systemd
> > maintainers essentially said "don't do that". With apologies for
> > the delay in doing so, I've CCed the maintainer list to see if we
> > can find a mutually acceptable solution.
> 
> Both `service` and `invoke-rc.d` wrappers have a few safeguards
> against running in unwanted contexts. Have you tried using one of
> them?

Ping? We're heading towards the final point release for stretch before
it moves to LTS, so need to decide what to do with this request.

Regards,

Adam

Bug#887736: stretch-pu: package openvswitch/2.6.2~pre+git20161223-3

[Adam D. Barratt]

On Wed, 2019-08-21 at 15:28 -0400, Felipe Sateler wrote:
> Both `service` and `invoke-rc.d` wrappers have a few safeguards
> against running in unwanted contexts. Have you tried using one of
> them?
> 

For avoidance of any possible doubt, "you" here is Thomas, not me.

Regards,

Adam

Bug#887736: stretch-pu: package openvswitch/2.6.2~pre+git20161223-3

[Felipe Sateler]

On Tue, Aug 20, 2019 at 6:23 PM Adam D. Barratt 
wrote:

> Control: tags -1 + moreinfo
>
> On Fri, 2018-01-19 at 15:21 +0100, Thomas Goirand wrote:
> > I started maintaining OpenVSwitch long after the Stretch release, and
> > discovered #858418, which is very annoying for OpenVSwitch users.
> >
> > tl;dr: #858418 prevent anyone that has a valid
> > /etc/network/interfaces
> > with OpenVSwitch directive from having a working network at boot. The
> > init script uses a non-documented, not-to-be-used systemd internal,
> > which is miserably failing.
> >
> > After a long discussion with the bug reporter (which can be read on
> > the BTS), I came to the conclusion that he's right, and that the most
> > reasonable and safe way to fix the current situation is to apply the
> > patch he suggested (and which resulting debdiff I attached to this
> > bug).
>
> As I understand things, that fix swaps use of one systemd internal for
> another, which doesn't seem like a great plan.
>
> When this was discussed (some time ago) on IRC, one of the systemd
> maintainers essentially said "don't do that". With apologies for the
> delay in doing so, I've CCed the maintainer list to see if we can find
> a mutually acceptable solution.
>

Both `service` and `invoke-rc.d` wrappers have a few safeguards against
running in unwanted contexts. Have you tried using one of them?

-- 

Saludos,
Felipe Sateler

Bug#887736: stretch-pu: package openvswitch/2.6.2~pre+git20161223-3

[Adam D. Barratt]

Control: tags -1 + moreinfo

On Fri, 2018-01-19 at 15:21 +0100, Thomas Goirand wrote:
> I started maintaining OpenVSwitch long after the Stretch release, and
> discovered #858418, which is very annoying for OpenVSwitch users.
> 
> tl;dr: #858418 prevent anyone that has a valid
> /etc/network/interfaces
> with OpenVSwitch directive from having a working network at boot. The
> init script uses a non-documented, not-to-be-used systemd internal,
> which is miserably failing.
> 
> After a long discussion with the bug reporter (which can be read on
> the BTS), I came to the conclusion that he's right, and that the most
> reasonable and safe way to fix the current situation is to apply the
> patch he suggested (and which resulting debdiff I attached to this
> bug).

As I understand things, that fix swaps use of one systemd internal for
another, which doesn't seem like a great plan.

When this was discussed (some time ago) on IRC, one of the systemd
maintainers essentially said "don't do that". With apologies for the
delay in doing so, I've CCed the maintainer list to see if we can find
a mutually acceptable solution.

Regards,

Adam

Bug#887736: stretch-pu: package openvswitch/2.6.2~pre+git20161223-3

[Thomas Goirand]

Package: release.debian.org
Severity: normal
Tags: stretch
User: release.debian@packages.debian.org
Usertags: pu

Dear release team,

I started maintaining OpenVSwitch long after the Stretch release, and
discovered #858418, which is very annoying for OpenVSwitch users.

tl;dr: #858418 prevent anyone that has a valid /etc/network/interfaces
with OpenVSwitch directive from having a working network at boot. The
init script uses a non-documented, not-to-be-used systemd internal,
which is miserably failing.

After a long discussion with the bug reporter (which can be read on the
BTS), I came to the conclusion that he's right, and that the most
reasonable and safe way to fix the current situation is to apply the
patch he suggested (and which resulting debdiff I attached to this bug).

The resulting sbuild binary package is available here:

http://sid.gplhost.com/stretch-proposed-updates/openvswitch/

Please allow me to upload it to Stretch. I'll try to come with a better
way to fix in Sid, probably importing the work from Ubuntu that they
did with an ifupdown hook.

Cheers,

Thomas Goirand (zigo)
diff -Nru openvswitch-2.6.2~pre+git20161223/debian/changelog 
openvswitch-2.6.2~pre+git20161223/debian/changelog
--- openvswitch-2.6.2~pre+git20161223/debian/changelog  2017-01-09 
23:46:48.0 +0100
+++ openvswitch-2.6.2~pre+git20161223/debian/changelog  2018-01-19 
13:46:44.0 +0100
@@ -1,3 +1,11 @@
+openvswitch (2.6.2~pre+git20161223-3+deb9u1) stretch; urgency=medium
+
+  * Added SYSTEMCTL_SKIP_REDIRECT=yes on top of existing
+_SYSTEMCTL_SKIP_REDIRECT=yes in the openvswitch-switch init script
+   (Closes: #858418).
+
+ -- Thomas Goirand   Fri, 19 Jan 2018 13:46:44 +0100
+
 openvswitch (2.6.2~pre+git20161223-3) unstable; urgency=medium
 
   * Avoid installing ovs-vswitchd.conf.db.5 manpage into directory for
diff -Nru openvswitch-2.6.2~pre+git20161223/debian/openvswitch-switch.init 
openvswitch-2.6.2~pre+git20161223/debian/openvswitch-switch.init
--- openvswitch-2.6.2~pre+git20161223/debian/openvswitch-switch.init
2016-09-30 17:27:55.0 +0200
+++ openvswitch-2.6.2~pre+git20161223/debian/openvswitch-switch.init
2018-01-19 13:46:44.0 +0100
@@ -28,6 +28,7 @@
 (test -x /usr/sbin/ovs-vswitchd && test -x /usr/sbin/ovsdb-server) || exit 0
 
 _SYSTEMCTL_SKIP_REDIRECT=yes
+SYSTEMCTL_SKIP_REDIRECT=yes
 
 . /usr/share/openvswitch/scripts/ovs-lib
 test -e /etc/default/openvswitch-switch && . /etc/default/openvswitch-switch