Bug#770627: dbus: Please (consider) switch(ing) to no-await triggers
Hi! On Sun, 2014-12-07 at 14:01:51 +0100, Niels Thykier wrote: > On 2014-12-05 05:20, Guillem Jover wrote: > > Given this, the only options seem to be to possibly change the > > --force-configure-any default to enabled (I'll try to test if this > > helps in the dbus situation) which I'd rather not do (but oh well), > > or disable the triggers dependency checks for jessie (and possibly > > later releases, if apt does not get fixed for jessie) and go back > > to the previous (broken, but known) behavior. :( > As I mentioned to David/the APT maintainers, I could be interested in > having the patch for #769609 in APT. @Guillem, do you think that will > deal with the issue? That would fix the issue with dangling trigger-pending packages, but I don't think it would fix the dbus issue. I've not tried to check a dpkg workaround for that yet, sorry. I've got a fix for the nvidia kernel module one, and asked for access to the ppc system with the pam module 100% cpu usage bug, to try to confirm that's been fixed in .22, which I think it does, but would like to make sure. In any case I think .22 would be an improvement over what's currently in testing. Thanks, Guillem -- To UNSUBSCRIBE, email to debian-bugs-dist-requ...@lists.debian.org with a subject of "unsubscribe". Trouble? Contact listmas...@lists.debian.org
Bug#770627: dbus: Please (consider) switch(ing) to no-await triggers
On 2014-12-05 05:20, Guillem Jover wrote: > Hi! > > [...] >> My question is, crucial for who? If dbus-daemon has not yet reloaded, >> then dbus-daemon will continue to work fine (it's in a consistent state >> either way), but the package that shipped the system service or security >> policy (in #771417 it's systemd) won't necessarily work properly until >> dbus-daemon has had a chance to reload. > > Crucial for the awaiting package (I'll be updating the docs for 1.18.x). > So, from your explanation, then in this case they are crucial for the > activating packages yes, and the bug report against dbus can be closed. > Hi, Thanks for the explanation. I have taken the liberty of closing the dbus issue then (BCC'ed -done for that bug and Simon to take them out of the loop with this mail). > Given this, the only options seem to be to possibly change the > --force-configure-any default to enabled (I'll try to test if this > helps in the dbus situation) which I'd rather not do (but oh well), > or disable the triggers dependency checks for jessie (and possibly > later releases, if apt does not get fixed for jessie) and go back > to the previous (broken, but known) behavior. :( > > Thanks, > Guillem > > As I mentioned to David/the APT maintainers, I could be interested in having the patch for #769609 in APT. @Guillem, do you think that will deal with the issue? ~Niels -- To UNSUBSCRIBE, email to debian-bugs-dist-requ...@lists.debian.org with a subject of "unsubscribe". Trouble? Contact listmas...@lists.debian.org
Bug#770627: dbus: Please (consider) switch(ing) to no-await triggers
Hi! On Thu, 2014-12-04 at 20:23:28 +, Simon McVittie wrote: > Control: tags 771989 + moreinfo > > On Thu, 04 Dec 2014 at 16:39:31 +0100, Niels Thykier wrote: > > On 2014-12-04 09:11, Simon McVittie wrote: > [see https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=770627#70 for full > analysis] > > > If I am correct in my interpretation, then it seems to me that normal > > > "interest" triggers are right here. However, if there is consensus that > > > using the wrong trigger is a less severe bug than #771417, I can switch > > > to "interest-noawait" and open a bug for "should go back to ordinary > > > triggers post-jessie"... as long as that new bug is not *also* going to > > > be considered RC. > > If your interpretation is correct, then I am certainly not convinced > > that changing the triggers to the "interest" variant is the correct > > thing to do for dbus. > > Sadly, I am no expert on these matters, so I will have to defer it to > > the APT or dpkg maintainers for whether your interpretation is correct. > > Thanks. apt/dpkg people: please send any feedback on the correct thing to do > to <771...@bugs.debian.org>. Sorry, I mixed up the various bugs and > didn't reply to that one initially; I have now bounced the missing > messages to that bug (hopefully). It seems the bounced mails didn't make it. In any case, from your other mail: > My question is, crucial for who? If dbus-daemon has not yet reloaded, > then dbus-daemon will continue to work fine (it's in a consistent state > either way), but the package that shipped the system service or security > policy (in #771417 it's systemd) won't necessarily work properly until > dbus-daemon has had a chance to reload. Crucial for the awaiting package (I'll be updating the docs for 1.18.x). So, from your explanation, then in this case they are crucial for the activating packages yes, and the bug report against dbus can be closed. Given this, the only options seem to be to possibly change the --force-configure-any default to enabled (I'll try to test if this helps in the dbus situation) which I'd rather not do (but oh well), or disable the triggers dependency checks for jessie (and possibly later releases, if apt does not get fixed for jessie) and go back to the previous (broken, but known) behavior. :( Thanks, Guillem -- To UNSUBSCRIBE, email to debian-bugs-dist-requ...@lists.debian.org with a subject of "unsubscribe". Trouble? Contact listmas...@lists.debian.org
Bug#770627: dbus: Please (consider) switch(ing) to no-await triggers
On Thu, 04 Dec 2014, Simon McVittie wrote: > On Thu, 04 Dec 2014 at 16:39:31 +0100, Niels Thykier wrote: > > On 2014-12-04 09:11, Simon McVittie wrote: > [see https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=770627#70 for full > analysis] > > If your interpretation is correct, then I am certainly not convinced > > that changing the triggers to the "interest" variant is the correct > > thing to do for dbus. > > Sadly, I am no expert on these matters, so I will have to defer it to > > the APT or dpkg maintainers for whether your interpretation is correct. > > Thanks. apt/dpkg people: please send any feedback on the correct thing to do > to <771...@bugs.debian.org>. Sorry, I mixed up the various bugs and > didn't reply to that one initially; I have now bounced the missing > messages to that bug (hopefully). I reviewed the log and I tend to agree with you that using the "interest" variant is the correct course of action. The cases where a postinst will require dbus to have loaded new policies provided by one of its dependencies are probably fairly rare (possibly even non-existent in Debian) but we should not rely on this being the case because your explanation of the actual behaviour makes sense. It would be nice if apt could be smarter and fallback to "dpkg --configure --pending" when the single "dpkg --configure foo" that it invokes fails... At the same time, dpkg could be nicer and accept to configure other packages when it's required to achieve the goal requested by the caller. Hopefully this would not confuse apt too much since apt tracks what dpkg does via the status-fd anyway. Is this realistic or only wishful thinking ? Only the latter solution seems to satisfy our requirements in terms of upgrades... Cheers, -- Raphaël Hertzog ◈ Debian Developer Support Debian LTS: http://www.freexian.com/services/debian-lts.html Learn to master Debian: http://debian-handbook.info/get/ -- To UNSUBSCRIBE, email to debian-bugs-dist-requ...@lists.debian.org with a subject of "unsubscribe". Trouble? Contact listmas...@lists.debian.org
Bug#770627: dbus: Please (consider) switch(ing) to no-await triggers
Control: tags 771989 + moreinfo On Thu, 04 Dec 2014 at 16:39:31 +0100, Niels Thykier wrote: > On 2014-12-04 09:11, Simon McVittie wrote: [see https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=770627#70 for full analysis] > > > > If I am correct in my interpretation, then it seems to me that normal > > "interest" triggers are right here. However, if there is consensus that > > using the wrong trigger is a less severe bug than #771417, I can switch > > to "interest-noawait" and open a bug for "should go back to ordinary > > triggers post-jessie"... as long as that new bug is not *also* going to > > be considered RC. > > Thanks for reviewing the situation. > > If your interpretation is correct, then I am certainly not convinced > that changing the triggers to the "interest" variant is the correct > thing to do for dbus. > Sadly, I am no expert on these matters, so I will have to defer it to > the APT or dpkg maintainers for whether your interpretation is correct. Thanks. apt/dpkg people: please send any feedback on the correct thing to do to <771...@bugs.debian.org>. Sorry, I mixed up the various bugs and didn't reply to that one initially; I have now bounced the missing messages to that bug (hopefully). S -- To UNSUBSCRIBE, email to debian-bugs-dist-requ...@lists.debian.org with a subject of "unsubscribe". Trouble? Contact listmas...@lists.debian.org
Bug#770627: dbus: Please (consider) switch(ing) to no-await triggers
On 2014-12-04 09:11, Simon McVittie wrote: > On 04/12/14 07:00, Niels Thykier wrote: >> On 2014-11-30 02:52, Guillem Jover wrote: >>> If this cannot be fixed in apt proper (due to requiring it to >>> self-upgrade), then it should be fixable by switching dbus to >>> noawaiting trigger directives (which I think would be correct >>> regardless). > > [..] > > If I am correct in my interpretation, then it seems to me that normal > "interest" triggers are right here. However, if there is consensus that > using the wrong trigger is a less severe bug than #771417, I can switch > to "interest-noawait" and open a bug for "should go back to ordinary > triggers post-jessie"... as long as that new bug is not *also* going to > be considered RC. > > S > Thanks for reviewing the situation. If your interpretation is correct, then I am certainly not convinced that changing the triggers to the "interest" variant is the correct thing to do for dbus. Sadly, I am no expert on these matters, so I will have to defer it to the APT or dpkg maintainers for whether your interpretation is correct. ~Niels -- To UNSUBSCRIBE, email to debian-bugs-dist-requ...@lists.debian.org with a subject of "unsubscribe". Trouble? Contact listmas...@lists.debian.org
Bug#770627: dbus: Please (consider) switch(ing) to no-await triggers
On 04/12/14 07:00, Niels Thykier wrote: > On 2014-11-30 02:52, Guillem Jover wrote: >> If this cannot be fixed in apt proper (due to requiring it to >> self-upgrade), then it should be fixable by switching dbus to >> noawaiting trigger directives (which I think would be correct >> regardless). So these triggers are: whenever a new D-Bus system service is installed in /usr/share/dbus-1/system-services or new security policies are installed in /etc/dbus-1/system.d, we tell the system dbus-daemon to reload, and block until it has done so. deb-triggers(5) says we should -noawait when "the functionality provided by the trigger is not crucial". My question is, crucial for who? If dbus-daemon has not yet reloaded, then dbus-daemon will continue to work fine (it's in a consistent state either way), but the package that shipped the system service or security policy (in #771417 it's systemd) won't necessarily work properly until dbus-daemon has had a chance to reload. >From reading deb-triggers(5) and /usr/share/doc/dpkg-dev/triggers.txt.gz it seems that the difference is that when systemd triggers dbus, with an "interest" trigger, systemd does not satisfy other packages' dependencies until dbus' trigger has run, forcing the trigger to be run before other packages try to use systemd's D-Bus API; but with an "interest-noawait" trigger, systemd is assumed to be fully working and satisfies dependencies, potentially resulting in failed calls to its D-Bus API because dbus-daemon's security policy does not allow them yet. Am I correct? If I am correct in my interpretation, then it seems to me that normal "interest" triggers are right here. However, if there is consensus that using the wrong trigger is a less severe bug than #771417, I can switch to "interest-noawait" and open a bug for "should go back to ordinary triggers post-jessie"... as long as that new bug is not *also* going to be considered RC. S -- To UNSUBSCRIBE, email to debian-bugs-dist-requ...@lists.debian.org with a subject of "unsubscribe". Trouble? Contact listmas...@lists.debian.org