Re: [Openais] init script runlevel at 20/20 instead of 20/80 debate
On 9/2/2010 9:09 PM, Steven Dake wrote: On 08/31/2010 08:45 PM, Vladislav Bogdanov wrote: Hi, 31.08.2010 22:22, Steven Dake wrote: I am pleased to announce Corosync 1.2.8 is available for immediate download from our website. Initscript doesn't seem to be fixed yet. http://marc.info/?l=openaism=128271460429681w=2 http://www.mail-archive.com/pacema...@oss.clusterlabs.org/msg05833.html Best, Vladislav Fabio, Any chance you can provide feedback on this topic? I'm at a loss in this area of distro integration. so the current init script has: # chkconfig: - 20 20 and that is definitely wrong. It must have slept through the crack when we re-did the init script a while ago. Kudos to Vladislav for noticing it. (making a bunch of assumptions here) the general rule is: stop-priority-value = 100 - start-priority-value. to guarantee the service start/stop symmetry that is pretty much the case for corosync. So a value of 20/80 would be correct. Now this should address the first concern reported. The other issue with start-early/late, that´s very debatable and dependent on use cases and distribution init process implementation (and the way it calculates init dependencies). My general approach is that the generic/default init script should satisfy the majority of the cases out there (it´s called generic for a good reason) and the current values do (modulo the error in the stop sequence that needs to be addressed). Every sysadmin has the power to tailor those values with system command tools, without the need to rebuild the rpm or editing the init script itself (chkconfig or whatever they use nowadays). The idea of having an early/late script is something I really don´t like. It involves a chain of changes to be done properly that adds an unnecessary complexity specially when there are already tools to move scripts priority around that are tailored to the OS/Distribution behavior. Fabio ___ Openais mailing list Openais@lists.linux-foundation.org https://lists.linux-foundation.org/mailman/listinfo/openais
Re: [Openais] init script runlevel at 20/20 instead of 20/80 debate
2010/9/3 Fabio M. Di Nitto fabbi...@fabbione.net: so the current init script has: # chkconfig: - 20 20 and that is definitely wrong. It must have slept through the crack when we re-did the init script a while ago. Kudos to Vladislav for noticing it. (making a bunch of assumptions here) the general rule is: stop-priority-value = 100 - start-priority-value. to guarantee the service start/stop symmetry that is pretty much the case for corosync. So a value of 20/80 would be correct. Now this should address the first concern reported. As for the starting order, I would be grad if you could also consider the attached patch. It will adjust the dependency with syslog correctly so that can prevent a problem when you use rsyslog as I reported before: https://lists.linux-foundation.org/pipermail/openais/2010-July/014946.html Thanks, -- Keisuke MORI --- corosync-1.2.8/init/generic.in.orig 2010-09-03 16:42:39.0 +0900 +++ corosync-1.2.8/init/generic.in 2010-09-03 16:46:34.0 +0900 @@ -12,9 +12,8 @@ # ### BEGIN INIT INFO # Provides: corosync -# Required-Start: $network -# Should-Start: $syslog -# Required-Stop: $network +# Required-Start: $network $syslog +# Required-Stop: $network $syslog # Default-Start: # Default-Stop: # Short-Description: Starts and stops Corosync Cluster Engine. ___ Openais mailing list Openais@lists.linux-foundation.org https://lists.linux-foundation.org/mailman/listinfo/openais
Re: [Openais] init script runlevel at 20/20 instead of 20/80 debate
On 9/3/2010 10:00 AM, Keisuke MORI wrote: 2010/9/3 Fabio M. Di Nitto fabbi...@fabbione.net: so the current init script has: # chkconfig: - 20 20 and that is definitely wrong. It must have slept through the crack when we re-did the init script a while ago. Kudos to Vladislav for noticing it. (making a bunch of assumptions here) the general rule is: stop-priority-value = 100 - start-priority-value. to guarantee the service start/stop symmetry that is pretty much the case for corosync. So a value of 20/80 would be correct. Now this should address the first concern reported. As for the starting order, I would be grad if you could also consider the attached patch. It will adjust the dependency with syslog correctly so that can prevent a problem when you use rsyslog as I reported before: https://lists.linux-foundation.org/pipermail/openais/2010-July/014946.html I think this is a sane patch and should go in. The only thing that puzzles me about this requirement is that many daemons do use syslog before the daemon is available and don´t fail. All C calls to syslog returns void so there is no way to know if they succeeded or not (as sign that the daemon is running), and glibc should make that transparent. So I wonder if the issue is really corosync that needs to start after syslog or we are masking another bug somewhere else (most likely glibc). Fabio ___ Openais mailing list Openais@lists.linux-foundation.org https://lists.linux-foundation.org/mailman/listinfo/openais
Re: [Openais] init script runlevel at 20/20 instead of 20/80 debate
03.09.2010 11:16, Fabio M. Di Nitto wrote: On 9/3/2010 10:00 AM, Keisuke MORI wrote: 2010/9/3 Fabio M. Di Nitto fabbi...@fabbione.net: so the current init script has: # chkconfig: - 20 20 and that is definitely wrong. It must have slept through the crack when we re-did the init script a while ago. Kudos to Vladislav for noticing it. (making a bunch of assumptions here) the general rule is: stop-priority-value = 100 - start-priority-value. to guarantee the service start/stop symmetry that is pretty much the case for corosync. So a value of 20/80 would be correct. Now this should address the first concern reported. As for the starting order, I would be grad if you could also consider the attached patch. It will adjust the dependency with syslog correctly so that can prevent a problem when you use rsyslog as I reported before: https://lists.linux-foundation.org/pipermail/openais/2010-July/014946.html I think this is a sane patch and should go in. BTW shouldn't openais initscript also include LSB stanza # Provides: corosync So f.e. pacemaker MCP will be satisfied at LSB level too, not only on chkconfig legacy one (LSB has priority BTW). Pacemaker MCP initscript already has # Required-Start: $network corosync And personally I start openais instead of corosync to have support for OCFS2/GFS2. Best, Vladislav ___ Openais mailing list Openais@lists.linux-foundation.org https://lists.linux-foundation.org/mailman/listinfo/openais
Re: [Openais] init script runlevel at 20/20 instead of 20/80 debate
On 9/3/2010 10:26 AM, Vladislav Bogdanov wrote: 03.09.2010 11:16, Fabio M. Di Nitto wrote: On 9/3/2010 10:00 AM, Keisuke MORI wrote: 2010/9/3 Fabio M. Di Nitto fabbi...@fabbione.net: so the current init script has: # chkconfig: - 20 20 and that is definitely wrong. It must have slept through the crack when we re-did the init script a while ago. Kudos to Vladislav for noticing it. (making a bunch of assumptions here) the general rule is: stop-priority-value = 100 - start-priority-value. to guarantee the service start/stop symmetry that is pretty much the case for corosync. So a value of 20/80 would be correct. Now this should address the first concern reported. As for the starting order, I would be grad if you could also consider the attached patch. It will adjust the dependency with syslog correctly so that can prevent a problem when you use rsyslog as I reported before: https://lists.linux-foundation.org/pipermail/openais/2010-July/014946.html I think this is a sane patch and should go in. BTW shouldn't openais initscript also include LSB stanza # Provides: corosync Yes, ACK. Fabio ___ Openais mailing list Openais@lists.linux-foundation.org https://lists.linux-foundation.org/mailman/listinfo/openais
Re: [Openais] init script runlevel at 20/20 instead of 20/80 debate
On 09/03/2010 01:00 AM, Keisuke MORI wrote: 2010/9/3 Fabio M. Di Nittofabbi...@fabbione.net: so the current init script has: # chkconfig: - 20 20 and that is definitely wrong. It must have slept through the crack when we re-did the init script a while ago. Kudos to Vladislav for noticing it. (making a bunch of assumptions here) the general rule is: stop-priority-value = 100 - start-priority-value. to guarantee the service start/stop symmetry that is pretty much the case for corosync. So a value of 20/80 would be correct. Now this should address the first concern reported. As for the starting order, I would be grad if you could also consider the attached patch. It will adjust the dependency with syslog correctly so that can prevent a problem when you use rsyslog as I reported before: https://lists.linux-foundation.org/pipermail/openais/2010-July/014946.html Thanks, I've attached a patch which applies against the current corosync trunk and merged it into trunk based upon the patch in that link. Regards -steve Index: init/generic.in === --- init/generic.in (revision 3035) +++ init/generic.in (working copy) @@ -12,9 +12,8 @@ # ### BEGIN INIT INFO # Provides:corosync -# Required-Start: $network -# Should-Start:$syslog -# Required-Stop: $network +# Required-Start: $network $syslog +# Required-Stop: $network $syslog # Default-Start: # Default-Stop: # Short-Description: Starts and stops Corosync Cluster Engine. ___ Openais mailing list Openais@lists.linux-foundation.org https://lists.linux-foundation.org/mailman/listinfo/openais
Re: [Openais] init script runlevel at 20/20 instead of 20/80 debate
On 09/02/2010 09:15 PM, Vladislav Bogdanov wrote: 03.09.2010 00:04, Steven Dake wrote: On 09/02/2010 01:28 PM, Vadym Chepkov wrote: On Sep 2, 2010, at 3:09 PM, Steven Dake wrote: On 08/31/2010 08:45 PM, Vladislav Bogdanov wrote: Hi, 31.08.2010 22:22, Steven Dake wrote: I am pleased to announce Corosync 1.2.8 is available for immediate download from our website. Initscript doesn't seem to be fixed yet. http://marc.info/?l=openaism=128271460429681w=2 http://www.mail-archive.com/pacema...@oss.clusterlabs.org/msg05833.html Best, Vladislav Fabio, Any chance you can provide feedback on this topic? I'm at a loss in this area of distro integration. I have complained about it some time ago, but was dismissed :) Vadym, My general take is it is the responsibility of individual distros to tailor the init scripts to their particular startup order. But rather then you being dismissed, I'd like to hear Fabio's reasoning so we can all come to some agreement. Unfortunately I don't know enough about the particulars of the init script dependencies to make that judgement. Fabio should be able to provide appropriate reasoning. One option to make everybody happy would be the following: Provide two init scripts, say 'corosync-early' and 'corosync-late' (the same is for openais), former has 20 80 - for CMAN/PCMKv1/whatever else, later has 98 02 - for PCMKv0. And control which one should be started by variable in /etc/sysconfig/corosync which should be mandatory and checked very early in that scripts. Redhat already has same suffixes for readehead. Agree with Fabio, this sounds too complicated and works around a problem which can be addressed with distro tools. Also pcmkv0 is legacy, so we would be working around a problem which doesn't exist upstream. I don't know enough about the LSB stanzas, but shouldn't these dependent init scripts all have a start-requires of corosync? Would that solve all these problems? Regards -steve Best, Vladislav ___ Openais mailing list Openais@lists.linux-foundation.org https://lists.linux-foundation.org/mailman/listinfo/openais
Re: [Openais] init script runlevel at 20/20 instead of 20/80 debate
03.09.2010 19:01, Steven Dake wrote: Agree with Fabio, this sounds too complicated and works around a problem which can be addressed with distro tools. Also pcmkv0 is legacy, so we would be working around a problem which doesn't exist upstream. I don't know enough about the LSB stanzas, but shouldn't these dependent init scripts all have a start-requires of corosync? Would that solve all these problems? They should and MCP has. I can't check right now, but I think it should also include stop-requires: corosync (if I understand LSB stanzas correctly). Best, Vladislav ___ Openais mailing list Openais@lists.linux-foundation.org https://lists.linux-foundation.org/mailman/listinfo/openais
[Openais] init script runlevel at 20/20 instead of 20/80 debate
On 08/31/2010 08:45 PM, Vladislav Bogdanov wrote: Hi, 31.08.2010 22:22, Steven Dake wrote: I am pleased to announce Corosync 1.2.8 is available for immediate download from our website. Initscript doesn't seem to be fixed yet. http://marc.info/?l=openaism=128271460429681w=2 http://www.mail-archive.com/pacema...@oss.clusterlabs.org/msg05833.html Best, Vladislav Fabio, Any chance you can provide feedback on this topic? I'm at a loss in this area of distro integration. Thanks -steve ___ Openais mailing list Openais@lists.linux-foundation.org https://lists.linux-foundation.org/mailman/listinfo/openais ___ Openais mailing list Openais@lists.linux-foundation.org https://lists.linux-foundation.org/mailman/listinfo/openais
Re: [Openais] init script runlevel at 20/20 instead of 20/80 debate
On 09/02/2010 09:09 PM, Steven Dake wrote: On 08/31/2010 08:45 PM, Vladislav Bogdanov wrote: Hi, 31.08.2010 22:22, Steven Dake wrote: I am pleased to announce Corosync 1.2.8 is available for immediate download from our website. Initscript doesn't seem to be fixed yet. http://marc.info/?l=openaism=128271460429681w=2 http://www.mail-archive.com/pacema...@oss.clusterlabs.org/msg05833.html Best, Vladislav Fabio, Any chance you can provide feedback on this topic? I'm at a loss in this area of distro integration. Sure, but it will have to wait tomorrow morning once I get in the office. Fabio ___ Openais mailing list Openais@lists.linux-foundation.org https://lists.linux-foundation.org/mailman/listinfo/openais
Re: [Openais] init script runlevel at 20/20 instead of 20/80 debate
On Sep 2, 2010, at 3:09 PM, Steven Dake wrote: On 08/31/2010 08:45 PM, Vladislav Bogdanov wrote: Hi, 31.08.2010 22:22, Steven Dake wrote: I am pleased to announce Corosync 1.2.8 is available for immediate download from our website. Initscript doesn't seem to be fixed yet. http://marc.info/?l=openaism=128271460429681w=2 http://www.mail-archive.com/pacema...@oss.clusterlabs.org/msg05833.html Best, Vladislav Fabio, Any chance you can provide feedback on this topic? I'm at a loss in this area of distro integration. I have complained about it some time ago, but was dismissed :) So, I build my own rpm diff -ur corosync-1.2.4.orig/init/generic.in corosync-1.2.4/init/generic.in --- corosync-1.2.4.orig/init/generic.in 2010-03-22 12:08:57.0 -0400 +++ corosync-1.2.4/init/generic.in 2010-07-28 10:00:00.0 -0400 @@ -6,7 +6,7 @@ # # License: Revised BSD -# chkconfig: - 20 20 +# chkconfig: - 99 01 # description: Corosync Cluster Engine # processname: corosync # ___ Openais mailing list Openais@lists.linux-foundation.org https://lists.linux-foundation.org/mailman/listinfo/openais
Re: [Openais] init script runlevel at 20/20 instead of 20/80 debate
On 09/02/2010 01:28 PM, Vadym Chepkov wrote: On Sep 2, 2010, at 3:09 PM, Steven Dake wrote: On 08/31/2010 08:45 PM, Vladislav Bogdanov wrote: Hi, 31.08.2010 22:22, Steven Dake wrote: I am pleased to announce Corosync 1.2.8 is available for immediate download from our website. Initscript doesn't seem to be fixed yet. http://marc.info/?l=openaism=128271460429681w=2 http://www.mail-archive.com/pacema...@oss.clusterlabs.org/msg05833.html Best, Vladislav Fabio, Any chance you can provide feedback on this topic? I'm at a loss in this area of distro integration. I have complained about it some time ago, but was dismissed :) Vadym, My general take is it is the responsibility of individual distros to tailor the init scripts to their particular startup order. But rather then you being dismissed, I'd like to hear Fabio's reasoning so we can all come to some agreement. Unfortunately I don't know enough about the particulars of the init script dependencies to make that judgement. Fabio should be able to provide appropriate reasoning. Regards -steve So, I build my own rpm diff -ur corosync-1.2.4.orig/init/generic.in corosync-1.2.4/init/generic.in --- corosync-1.2.4.orig/init/generic.in 2010-03-22 12:08:57.0 -0400 +++ corosync-1.2.4/init/generic.in2010-07-28 10:00:00.0 -0400 @@ -6,7 +6,7 @@ # # License: Revised BSD -# chkconfig: - 20 20 +# chkconfig: - 99 01 # description: Corosync Cluster Engine # processname: corosync # ___ Openais mailing list Openais@lists.linux-foundation.org https://lists.linux-foundation.org/mailman/listinfo/openais
Re: [Openais] init script runlevel at 20/20 instead of 20/80 debate
On Sep 2, 2010, at 5:04 PM, Steven Dake wrote: On 09/02/2010 01:28 PM, Vadym Chepkov wrote: On Sep 2, 2010, at 3:09 PM, Steven Dake wrote: On 08/31/2010 08:45 PM, Vladislav Bogdanov wrote: Hi, 31.08.2010 22:22, Steven Dake wrote: I am pleased to announce Corosync 1.2.8 is available for immediate download from our website. Initscript doesn't seem to be fixed yet. http://marc.info/?l=openaism=128271460429681w=2 http://www.mail-archive.com/pacema...@oss.clusterlabs.org/msg05833.html Best, Vladislav Fabio, Any chance you can provide feedback on this topic? I'm at a loss in this area of distro integration. I have complained about it some time ago, but was dismissed :) Vadym, My general take is it is the responsibility of individual distros to tailor the init scripts to their particular startup order. But rather then you being dismissed, I'd like to hear Fabio's reasoning so we can all come to some agreement. Unfortunately I don't know enough about the particulars of the init script dependencies to make that judgement. Fabio should be able to provide appropriate reasoning. Regards -steve So, I build my own rpm diff -ur corosync-1.2.4.orig/init/generic.in corosync-1.2.4/init/generic.in --- corosync-1.2.4.orig/init/generic.in 2010-03-22 12:08:57.0 -0400 +++ corosync-1.2.4/init/generic.in 2010-07-28 10:00:00.0 -0400 @@ -6,7 +6,7 @@ # # License: Revised BSD -# chkconfig: - 20 20 +# chkconfig: - 99 01 # description: Corosync Cluster Engine # processname: corosync # This is one of those cases where you can't make everybody happy. In some cases corosync needs to start early to support cman/gfs, in other case it needs to start later to support XEN with drbd block device. so 20 vs 99 is pretty much subjective and since you added postinstall scriptlet (using /bin/sh): if [ $1 -eq 1 ]; then /sbin/chkconfig --add corosync || : fi it doesn't destroy local preferences during upgrade, so it is just something that administrator has to decide for himself. I need to have corosync to start after ntpd, after sendmail, after winbind, after xend and this pretty much leaves me with 99 Vadym ___ Openais mailing list Openais@lists.linux-foundation.org https://lists.linux-foundation.org/mailman/listinfo/openais