-----BEGIN PGP SIGNED MESSAGE----- Hash: SHA1 Hi,
Thomas Hood wrote: > Package: pdns-recursor > Version: 2.9.17-10 > Severity: normal > Tags: patch > > First I report some problems, then I suggest a solution. :) > > * On stop, resolvconf is told to delete the resolvconf record after the > daemon is stopped. It would be better to do this before the daemon is > stopped so that no new queries are sent to the daemon just when it is > about to be killed. > That's a good point. Next time i'll investigate that stuff better. > * The record should be called 'lo.pdns-recursor' so that it does not > conflict with records created by other nameservers which could > conceivably by running on the same machine. (pdns-recursor does not > Conflict with any other packages, so I infer that this is possible. > Anyway, there is no reason not to give the record a unique name.) > Agreed. > * Currently in the stop method the initscript sends a signal and then > exits immediately and unconditionally. However, the script should not > exit before the daemon has really stopped. More precisely, it should > not exit before the daemon releases resources that could be needed by > other daemons that are starting in the same runlevel change or dpkg run. > Furthermore, if the daemon could not be stopped then the script should > exit with an error status. (Admittedly, many initscripts in Debian > have this shortcoming.) > Agreed. No problem, it seems that people are using the package and how more problems getting solved how better :) > * On restart, a "sleep 3" is used to ensure that the daemon is restarted > after it is stopped. This either wastes time or is unreliable. For > the same reason, the previous problem isn't best solved by adding a > sleep to the end of the stop method. > What do you mean exactly, now we wait for 3 seconds and then try to start. (Assuming that it is stopped after 3 seconds.) Checked your pdns-recursor script and i like it. > * The script does "test -x $DAEMON" runs the daemon using $NAME, which > is inconsistent. > Agreed. > * The daemon is started using the command > > $NAME --daemon 2>&1 /dev/null > > As the daemon does not take any non-option arguments (according to the > man page), I infer that you really meant to do: > > $NAME --daemon 2>&1 >/dev/null > > However, I don't see why you want to hide the daemon's rather brief > startup message. > > [EMAIL PROTECTED]:~$ sudo /usr/sbin/pdns_recursor --daemon > Apr 16 18:53:50 Incoming query source port: 53 > Apr 16 18:53:50 Done priming cache with root hints > > The cost is great: you also hide the daemon's error messages, if any. > If the startup messages are superfluous then the daemon source should > be patched in order to suppress them. > That means patching of the sources. I don't know if we can get it working but we will try to fix that. In first instance we show the messages when the init script will run. > * Neither the daemon nor the initscript deletes the pidfile on stop. > hmm.. > > * Solution > > I wrote the initscripts for dnsmasq and pdnsd which, although possibly > imperfect, are less imperfect. Each uses start-stop-daemon to test for > running instances of the daemon, properly handles various failure cases > and interfaces with resolvconf correctly. I attach an initscript like > theirs for pdns-recursor. > > I have implemented the "Don't really start unless a certain environment > variable is set to a certain value" feature even though I don't like it > when maintainers try to undermine the standard mechanism for enabling > and disabling services in runlevels. :/ > > If you find any problems with the initscript then please discuss them > with me. If the problem turns out to be real then I would like to fix > any similar problems in the dnsmasq and/or pdnsd initscripts. > > The following log shows that the initscript works. The daemon does > produce ugly messages, but as I said earlier, these should be dealt > with in the daemon source. One possibility would be to change the > daemon to print them on standard output rather than on standard > error and then divert only standard output to /dev/null. This would > eliminate the unwanted startup messages without hiding error messages. > > # # Note that in the following case START is unset > # ./pdns-recursor start > Not starting PowerDNS recursor -- disabled. > # START=yes ./pdns-recursor start > Starting PowerDNS recursor: pdns_recursorApr 16 19:26:55 Incoming query > source port: 53 > Apr 16 19:26:55 Done priming cache with root hints > . > # START=yes ./pdns-recursor start > Starting PowerDNS recursor: pdns_recursor (already running). > # ./pdns-recursor stop > Stopping PowerDNS recursor: pdns_recursor. > # ./pdns-recursor stop > Stopping PowerDNS recursor: pdns_recursor (not running). > # START=yes ./pdns-recursor restart > Restarting PowerDNS recursor: pdns_recursorApr 16 19:27:19 Incoming > query source port: 53 > Apr 16 19:27:19 Done priming cache with root hints > . > # START=yes ./pdns-recursor restart > Restarting PowerDNS recursor: pdns_recursorApr 16 19:27:21 Incoming > query source port: 53 > Apr 16 19:27:21 Done priming cache with root hints > . > # # Note that in the following cases START is unset > # ./pdns-recursor restart > Stopping PowerDNS recursor: pdns_recursor. > # ./pdns-recursor restart > Stopping PowerDNS recursor: pdns_recursor (not running). > > Very good. I like the way how you solved it, and i'll try to fix those problems also in pdns-server. One thing: Probably you can file a bug against the package initscripts and ask them to provide a better skeleton script. I have taken the skeleton script for pdns-recursor to make a new initscript. But yours seems much better. Also i think we need something about this in the debian-policy. So that we have good initscripts in debian. And at last maybe a bugreport against lintian/linda to check for this kind of changes if possible. Thanks for reporting and providing a solution. Regards, PowerDNS maintainers. -----BEGIN PGP SIGNATURE----- Version: GnuPG v1.4.0 (GNU/Linux) Comment: Using GnuPG with Thunderbird - http://enigmail.mozdev.org iD8DBQFCYV9h2n1ROIkXqbARAiuTAJ4h72UqtLEuOJQatV3PMdAdeMxdRQCggEBQ QPbehVazJ9E1SsH61SHe+g0= =5pNU -----END PGP SIGNATURE----- -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]
