problem with debconf and start-stop-daemon
I'm currently working on adding debconf to quota. Well in fact Torsten Lnadschoff did that and send me the patch. However, during my tests I found a strange problem. In quota.postinst rpc.quotad is started using start-stop-daemon. This works as longs as I know the package. Now it doesn't anymore. That is the daemon is correctly started but quota.postinst does not return anymore. It remains defunc until I kill rpc.rquotad. As soon as I comment out the line . /usr/share/debconf/confmodule all works well again. Since I do not know much about debconf I'm not seeing the reason but I suppose it should be pretty simple or else other packages wouldn't work either. Any idea? Michael -- Michael Meskes Michael@Fam-Meskes.De Go SF 49ers! Go Rhein Fire! Use Debian GNU/Linux! Use PostgreSQL!
Re: problem with debconf and start-stop-daemon
On Fri, Jan 11, 2002 at 12:44:39PM +0100, Michael Meskes wrote: In quota.postinst rpc.quotad is started using start-stop-daemon. This works as longs as I know the package. Now it doesn't anymore. That is the daemon is correctly started but quota.postinst does not return anymore. It remains defunc until I kill rpc.rquotad. As soon as I comment out the line . /usr/share/debconf/confmodule all works well again. You need to explicitly end Debconf processing in the postinst by calling db_stop. debconf causes child processes to have an extra file descriptor open and waits for these to be closed before exiting and the daemon doesn't know it has this file open so doesn't close it. -- You grabbed my hand and we fell into it, like a daydream - or a fever. pgpTd7ITXX72V.pgp Description: PGP signature
Re: problem with debconf and start-stop-daemon
[Mark Brown] You need to explicitly end Debconf processing in the postinst by calling db_stop. debconf causes child processes to have an extra file descriptor open and waits for these to be closed before exiting and the daemon doesn't know it has this file open so doesn't close it. Is this the case for all postinst scripts? db_stop is not mentioned in my /usr/share/doc/debconf-doc/tutorial.html, which I use as my debconf reference.
Re: problem with debconf and start-stop-daemon
On Fri, Jan 11, 2002 at 11:58:47AM +, Mark Brown wrote: You need to explicitly end Debconf processing in the postinst by calling That's it. Thanks a lot. Michael -- Michael Meskes Michael@Fam-Meskes.De Go SF 49ers! Go Rhein Fire! Use Debian GNU/Linux! Use PostgreSQL! pgpfJMEeIzDdh.pgp Description: PGP signature
Re: problem with debconf and start-stop-daemon
On Fri, Jan 11, 2002 at 01:04:25PM +0100, Petter Reinholdtsen wrote: Is this the case for all postinst scripts? db_stop is not mentioned in my /usr/share/doc/debconf-doc/tutorial.html, which I use as my debconf reference. No. If your postinst does not leave processes running then there won't be any problem. -- You grabbed my hand and we fell into it, like a daydream - or a fever. pgpchpcTtjxc7.pgp Description: PGP signature
Re: problem with debconf and start-stop-daemon
On Fri, Jan 11, 2002 at 01:04:25PM +0100, Petter Reinholdtsen wrote: [Mark Brown] You need to explicitly end Debconf processing in the postinst by calling db_stop. debconf causes child processes to have an extra file descriptor open and waits for these to be closed before exiting and the daemon doesn't know it has this file open so doesn't close it. Is this the case for all postinst scripts? db_stop is not mentioned in my /usr/share/doc/debconf-doc/tutorial.html, which I use as my debconf reference. See the Troubleshooting section. The debconf documentation often doesn't use the db_ prefix when discussing commands, since that's only used by the shell confmodule and not the Perl confmodule. -- Colin Watson [EMAIL PROTECTED]
Re: problem with debconf and start-stop-daemon
On Fri, 11 Jan 2002, Michael Meskes wrote: In quota.postinst rpc.quotad is started using start-stop-daemon. This works Don't, please. This will be forbidden in the future (right now this is ok), you should use the /etc/init.d interface. If you need extra functionality, you should request the maintainer of the package with the /etc/init.d script to add it, and use that. as longs as I know the package. Now it doesn't anymore. That is the daemon is correctly started but quota.postinst does not return anymore. It remains defunc until I kill rpc.rquotad. As soon as I comment out the line . /usr/share/debconf/confmodule all works well again. Ah, it looks like you need to have a db_stop BEFORE you call start-stop-daemon or the initscript. Looks like the daemon is dumb and does not close all open descriptors, creates a new session id, etc... -- One disk to rule them all, One disk to find them. One disk to bring them all and in the darkness grind them. In the Land of Redmond where the shadows lie. -- The Silicon Valley Tarot Henrique Holschuh
Re: problem with debconf and start-stop-daemon
On Fri, Jan 11, 2002 at 11:40:28AM -0200, Henrique de Moraes Holschuh wrote: Ah, it looks like you need to have a db_stop BEFORE you call start-stop-daemon or the initscript. Looks like the daemon is dumb and does No, the ordering is unimportant. db_stop stops debconf no matter when you call it. not close all open descriptors, creates a new session id, etc... Or has called the standard daemon() function which doesn't close all the file descriptors. -- You grabbed my hand and we fell into it, like a daydream - or a fever. pgpX0u114EEbs.pgp Description: PGP signature
Re: problem with debconf and start-stop-daemon
On Fri, Jan 11, 2002 at 11:40:28AM -0200, Henrique de Moraes Holschuh wrote: Don't, please. This will be forbidden in the future (right now this is ok), you should use the /etc/init.d interface. If you need extra functionality, Sorry, I wasn't precies enough. It does call /etc/init.d/quotarpc which then calls start-stop-daemon. start-stop-daemon or the initscript. Looks like the daemon is dumb and does not close all open descriptors, creates a new session id, etc... Yup. Michael -- Michael Meskes Michael@Fam-Meskes.De Go SF 49ers! Go Rhein Fire! Use Debian GNU/Linux! Use PostgreSQL!
Re: problem with debconf and start-stop-daemon
Mark Brown [EMAIL PROTECTED] writes: On Fri, Jan 11, 2002 at 11:40:28AM -0200, Henrique de Moraes Holschuh wrote: Ah, it looks like you need to have a db_stop BEFORE you call start-stop-daemon or the initscript. Looks like the daemon is dumb and does No, the ordering is unimportant. db_stop stops debconf no matter when you call it. not close all open descriptors, creates a new session id, etc... Or has called the standard daemon() function which doesn't close all the file descriptors. Yeah, but it is really a bug that should be filed. The daemon will be killed by SAK otherwise (look at #92277 for further enlightenment). Ciao Racke -- For projects and other business stuff please refer to COBOLT NetServices (URL: http://www.cobolt.net; Email: [EMAIL PROTECTED]; Phone: 0041-1-3884400)
Re: problem with debconf and start-stop-daemon
On Fri, Jan 11, 2002 at 09:45:22AM -0500, Stefan Hornburg (Racke) wrote: Mark Brown [EMAIL PROTECTED] writes: Or has called the standard daemon() function which doesn't close all the file descriptors. Yeah, but it is really a bug that should be filed. The daemon will be killed by SAK otherwise (look at #92277 for further enlightenment). It does close the standard file descriptors. It just doesn't close anything else (like stuff from debconf, for example). -- You grabbed my hand and we fell into it, like a daydream - or a fever.
Re: problem with debconf and start-stop-daemon
Previously Henrique de Moraes Holschuh wrote: On Fri, 11 Jan 2002, Michael Meskes wrote: In quota.postinst rpc.quotad is started using start-stop-daemon. This works Don't, please. This will be forbidden in the future (right now this is ok), you should use the /etc/init.d interface. Why? That is silly, there are perfectly good reasons to use a different method for starting/stopping a daemon in the maintainer scripts then init scripts use. Wichert. -- _ /[EMAIL PROTECTED] This space intentionally left occupied \ | [EMAIL PROTECTED]http://www.liacs.nl/~wichert/ | | 1024D/2FA3BC2D 576E 100B 518D 2F16 36B0 2805 3CB8 9250 2FA3 BC2D |
Re: problem with debconf and start-stop-daemon
On Fri, 11 Jan 2002, Wichert Akkerman wrote: Previously Henrique de Moraes Holschuh wrote: On Fri, 11 Jan 2002, Michael Meskes wrote: In quota.postinst rpc.quotad is started using start-stop-daemon. This works Don't, please. This will be forbidden in the future (right now this is ok), you should use the /etc/init.d interface. Why? That is silly, there are perfectly good reasons to use a different method for starting/stopping a daemon in the maintainer scripts then init scripts use. Read the invoke-rc.d stuff, and policy-rc.d stuff in the BTS (debian-policy). I don't know the bug numbers offhand. It has been approved, the code is in woody and it works (try running invoke-rc.d sometime), and we are just waiting woody to go out of the door to start deploying the stuff. Basically, the idea is to finish once and for all with the mess of daemons starting and restarting (and stopping) when the admin does not them to, due to maintainer scripts playing around. The current default behaviour will NOT change unless you go out of your way to configure the system otherwise, btw. But people that, for example, don't want any daemons running before they are configured and reviewed will be able to do that. Please tell me one good reason not to use the init.d script interface to muck around with daemons _in maintainer scripts_? I mean, a good one that justfies bowling over any changes the local administrator might have done to the initscripts? -- One disk to rule them all, One disk to find them. One disk to bring them all and in the darkness grind them. In the Land of Redmond where the shadows lie. -- The Silicon Valley Tarot Henrique Holschuh
Re: problem with debconf and start-stop-daemon
On 11-Jan-02, 08:45 (CST), Stefan Hornburg (Racke) [EMAIL PROTECTED] wrote: Yeah, but it is really a bug that should be filed. The daemon will be killed by SAK otherwise (look at #92277 for further enlightenment). You can't, in general, close *all* open file descriptors. OPEN_MAX may not exist (and I would guess that it doesn't on the HURD). It's completely reasonable for a daemon to that doesn't open any extras to assume that only stdin, stdout, and stderr are open. Steve
Re: problem with debconf and start-stop-daemon
On Fri, Jan 11, 2002 at 10:34:18AM -0600, Steve Greenland wrote: You can't, in general, close *all* open file descriptors. OPEN_MAX may not exist (and I would guess that it doesn't on the HURD). It's completely reasonable for a daemon to that doesn't open any extras to assume that only stdin, stdout, and stderr are open. If OPEN_MAX is undefined you could always use INT_MAX :-) . -- You grabbed my hand and we fell into it, like a daydream - or a fever. pgp2B5Sy7URvY.pgp Description: PGP signature
Re: problem with debconf and start-stop-daemon
On Fri, Jan 11, 2002 at 04:49:19PM +, Mark Brown wrote: On Fri, Jan 11, 2002 at 10:34:18AM -0600, Steve Greenland wrote: You can't, in general, close *all* open file descriptors. OPEN_MAX may not exist (and I would guess that it doesn't on the HURD). It's completely reasonable for a daemon to that doesn't open any extras to assume that only stdin, stdout, and stderr are open. If OPEN_MAX is undefined you could always use INT_MAX :-) . iirc, there has been a problem with a program closing all descriptors before, because they were used by libc (?) for something. So no, you shouldn't use INT_MAX either :) -- Adam Olsen, aka Rhamphoryncus
Re: problem with debconf and start-stop-daemon
On Fri, Jan 11, 2002 at 04:49:19PM +, Mark Brown wrote: You can't, in general, close *all* open file descriptors. OPEN_MAX may not exist (and I would guess that it doesn't on the HURD). If OPEN_MAX is undefined you could always use INT_MAX :-) . I bet INT_MAX does not exist on Hurd, either... Michael -- -:- aj [EMAIL PROTECTED] has joined #debian-devel willy oh, crap, 13 RC bugs against base?
Re: problem with debconf and start-stop-daemon
Michael Banck [EMAIL PROTECTED] writes: On Fri, Jan 11, 2002 at 04:49:19PM +, Mark Brown wrote: You can't, in general, close *all* open file descriptors. OPEN_MAX may not exist (and I would guess that it doesn't on the HURD). If OPEN_MAX is undefined you could always use INT_MAX :-) . I bet INT_MAX does not exist on Hurd, either... I bet it does. The ANSI C standard requires that INT_MAX be defined. POSIX only requires that OPEN_MAX be defined if it has a fixed, uniform limit. -- While the Melissa license is a bit unclear, Melissa aggressively encourages free distribution of its source code. --Kevin Dalley [EMAIL PROTECTED]
Re: problem with debconf and start-stop-daemon
Previously Henrique de Moraes Holschuh wrote: Please tell me one good reason not to use the init.d script interface to muck around with daemons _in maintainer scripts_? The --exec option for start-stop-daemon. This option is very useful: it gives start-stop-daemon the ability to verify that it is killing the right process, and not another one with the same name or with the same (recycled) pid. This is definitely the preferred way of using start-stop-daemon. The only problem with the --exec option is that it does not work if you use it in the postinst after an upgrade to restart the daemon since the binary has been replaced and will now have a different inode. This gives you two options: 1. don't use --exec at all, not even in the init script. 2. don't use --exec in the postinst, or use it with a saved (hardlinked) copy of the original file Option 1 means we loose the advantage of being sure we are killing the right process. Option 2 is obviously better in that it does not have that problem at all of you use a saved copy, or only during upgrades if you only don't use --exec in the maintainer script. Wichert. -- _ /[EMAIL PROTECTED] This space intentionally left occupied \ | [EMAIL PROTECTED]http://www.liacs.nl/~wichert/ | | 1024D/2FA3BC2D 576E 100B 518D 2F16 36B0 2805 3CB8 9250 2FA3 BC2D |
Re: problem with debconf and start-stop-daemon
On Fri, 11 Jan 2002, Wichert Akkerman wrote: Previously Henrique de Moraes Holschuh wrote: Please tell me one good reason not to use the init.d script interface to muck around with daemons _in maintainer scripts_? The --exec option for start-stop-daemon. This option is very useful: it [...] The only problem with the --exec option is that it does not work if you use it in the postinst after an upgrade to restart the daemon since the binary has been replaced and will now have a different inode. Yes. When one both replace the daemon executable AND the initscript uses start-stop-daemon --exec, one has to stop the daemon in prerm. Lots of packages do this already. I believe it is the default behaviour for dh_initscript for example. Any package that is -replacing- the daemon executable should know very well what its initscript does, and how. In fact, in just about 99% of the cases, either the same binary package or source package will provide the initscript and the executable. So I will assume we are talking about a maintainer script maintained by the same person/crew that maintains the initscript. The prerm solution is a problem for those who need to shorten the downtime of the service. e.g, IDS daemons. For those, it is still possible to have a new target (say, stop-nocheck) in the init.d script that does not use --exec, exactly the way it would be done in the postinst. 1. don't use --exec at all, not even in the init script. 2. don't use --exec in the postinst, or use it with a saved (hardlinked) copy of the original file Actually, I was not aware of 2). I like it. While it would be possible (but ugly) to have such an stop-alike /usr/sbin/daemon.temp initscript target, it would look a bit kludgy (but perfectly legal, and supported). Well, policy will say should, not must for the use of invoke-rc.d and initscripts. There is no reason why we could not add a sentence to the effect of maintainer scripts that cannot use the initscript for some reason to execute an action on the service, must verify (using invoke-rc.d --query), if they would be allowed to execute a given initscript action (stop, start, restart, reload) before doing so directly. i.e., you could either live with a new action in your init.d script (using either method 1 or 2 to stop the daemon), stop it in prerm, or call invoke-rc.d to verify if the admin frowns upon what you're going to do and if it says its ok, do the start-stop-daemon thing directly. Would that be good enough? -- One disk to rule them all, One disk to find them. One disk to bring them all and in the darkness grind them. In the Land of Redmond where the shadows lie. -- The Silicon Valley Tarot Henrique Holschuh
