Bug#427499: Merge a few bugs and provide LSB header
severity 336258 normal merge 336258 400205 427499 user [EMAIL PROTECTED] usertags 336258 missing-dependency thanks Please make sure the new init.d script include dependency information using the LSB header. Here is a draft header based on the comments in bug #427499. ### BEGIN INIT INFO # Provides: clvmd # Required-Start:$remote_fs # Required-Stop: $remote_fs # Should-Start: fence # Should-Stop: fence # Default-Start: 2 3 4 5 # Default-Stop: 0 1 6 ### END INIT INFO -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of unsubscribe. Trouble? Contact [EMAIL PROTECTED]
Bug#427499: Merge a few bugs and provide LSB header
Hello, I suggest including the following LSB compliant headers at ALL the respective redhat cluster init.d scripts in order to maintain consistency between cluster service essential daemons. Please, review the ccs dependencies, as ssh must be running before it. Also, the strong dependency on $remote_fs or $local_fs may conflict with GFS cluster fs. Followed hints at http://wiki.debian.org/PkgSplit and its resource links [34] and [35]. Regards. Andre Felipe Machado ### BEGIN INIT INFO # Provides: clvmd # Required-Start:fence # Required-Stop: fence # Should-Start: $all # Should-Stop: $all # Default-Start: 2 3 4 5 # Default-Stop: 0 1 6 # Short-Description: Cluster aware LVM daemon. # Description: Coordinate LVM synchronization accross the #Red Hat Cluster. ### END INIT INFO ### BEGIN INIT INFO # Provides: fence # Required-Start:cman # Required-Stop: cman # Should-Start: $all # Should-Stop: $all # Default-Start: 2 3 4 5 # Default-Stop: 0 1 6 # Short-Description: Cluster device fencing daemon. # Description: Coordinate device fencing accross the #Red Hat Cluster. ### END INIT INFO ### BEGIN INIT INFO # Provides: cman # Required-Start:ccs # Required-Stop: ccs # Should-Start: $all # Should-Stop: $all # Default-Start: 2 3 4 5 # Default-Stop: 0 1 6 # Short-Description: Cluster manager daemon. # Description: Manage cluster synchronization accross the #Red Hat Cluster. ### END INIT INFO ### BEGIN INIT INFO # Provides: ccs # Required-Start:$network ssh # Required-Stop: $network ssh # Should-Start: $all # Should-Stop: $all # Default-Start: 2 3 4 5 # Default-Stop: 0 1 6 # Short-Description: Cluster aware LVM daemon. # Description: Cluster configuration system to manage the #cluster.conf file across the #Red Hat Cluster. ### END INIT INFO -- A Debian user never dies. Issues a last command: shutdown -h now http://www.techforce.com.br
Bug#427499: Merge a few bugs and provide LSB header
[Andre Felipe Machado] I suggest including the following LSB compliant headers at ALL the respective redhat cluster init.d scripts in order to maintain consistency between cluster service essential daemons. It seem mostly ok, but the dependency on '$all' is broken. They can not all start after all other packages. It will not work. In addition, all of them should depend on $remote_fs, to make sure /usr/ is mounted when the scripts are executed. Please, review the ccs dependencies, as ssh must be running before it. Also, the strong dependency on $remote_fs or $local_fs may conflict with GFS cluster fs. Unlikely. See http://wiki.debian.org/LSBInitScripts for information on how to specify the dependencies. Happy hacking, -- Petter Reinholdtsen -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of unsubscribe. Trouble? Contact [EMAIL PROTECTED]
Bug#427499: Merge a few bugs and provide LSB header
Hello, Petter Thanks for the fast answer. It seem mostly ok, but the dependency on '$all' is broken. They can not all start after all other packages. It will not work. The Should-Start: $all is supposed to be a weak dependency, a desired order that will not cause a failure if not satisfied. I imagined that when the init installer, at first pass, will try to move each script to the end of queue as per weak dependency spec, then, at second pass, will try to satisfy the strong dependencies, correctly ordering all. In addition, all of them should depend on $remote_fs, to make sure /usr/ is mounted when the scripts are executed. My first guess was to specify strong dependencies $remote_fs and $network for all packages. But then realized that each script will depend on previously started other and only the first (ccs) should have such strong dependencies. Specifying strong dependencies almost equal to all scripts, may confuse the installer order resolving. I am not sure if the installer could accomplish such task, yet. If the DD could make things easier to the installer, the better. As the redhat-cluster-suite should install all init.d scripts, these headers should be enough to guarantee consistency. Also, I am not sure about needing the non-standard Debian and Suse custom headers for reverse dependency, given that the package will install all scripts. X-Start-Before: boot_facility_1 [boot_facility_2...] X-Stop-After: boot_facility_1 [boot_facility_2...] Regards. Andre Felipe -- A Debian user never dies. Issues a last command: shutdown -h now http://www.techforce.com.br
Bug#427499: Merge a few bugs and provide LSB header
[Andre Felipe Machado] The Should-Start: $all is supposed to be a weak dependency, a desired order that will not cause a failure if not satisfied. I believe this is a misunderstanding on what a weak dependency is. A weak dependency mean that it will not fail to register/install this script if the dependency is missing. If it is present (all $all matches all/any script, thus always present), the script have to start after and stop before its weak dependency. If it can't, the ordering is not well defined. I imagined that when the init installer, at first pass, will try to move each script to the end of queue as per weak dependency spec, then, at second pass, will try to satisfy the strong dependencies, correctly ordering all. The current implementation of insserv will first order all scripts according to their dependencies ignoring the $all, and finally move all scripts with the $all dependency to the last order + 1. My first guess was to specify strong dependencies $remote_fs and $network for all packages. But then realized that each script will depend on previously started other and only the first (ccs) should have such strong dependencies. Specifying strong dependencies almost equal to all scripts, may confuse the installer order resolving. Yes, if the packages have hard dependencies on each other and can know that other scripts will be installed in time, it can drop the $remote_fs dependency. I would not recommend it, though, as it document the need for a mounted /usr/ and it is useful to include in case the script dependencies are changed in the future. I am not sure if the installer could accomplish such task, yet. The dependency based boot sequencing system insserv that is used in Debian handle this just fine. If the DD could make things easier to the installer, the better. As the redhat-cluster-suite should install all init.d scripts, these headers should be enough to guarantee consistency. I trust you there. :) Also, I am not sure about needing the non-standard Debian and Suse custom headers for reverse dependency, given that the package will install all scripts. They are probably not needed in this case. They have proven useful in other cases. Happy hacking, -- Petter Reinholdtsen -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of unsubscribe. Trouble? Contact [EMAIL PROTECTED]
Bug#427499: Merge a few bugs and provide LSB header
Hello, Petter Many thanks for the explanations. A weak dependency mean that it will not fail to register/install this script if the dependency is missing. If it is present (all $all matches all/any script, thus always present), the script have to start after and stop before its weak dependency. If it can't, the ordering is not well defined. I read [0] and tried to move the cluster daemons to the end of init.d queue as far as possible, in a defined order inside the group. As the Should-Start does not guarantee this, removed then at the new proposed headers below. The current implementation of insserv will first order all scripts according to their dependencies ignoring the $all, and finally move all scripts with the $all dependency to the last order + 1. Given that $all is composed of base services, should not the order be first honor the $all then the other strong dependencies in the header? Yes, if the packages have hard dependencies on each other and can know that other scripts will be installed in time, it can drop the $remote_fs dependency. I would not recommend it, though, as it document the need for a mounted /usr/ and it is useful to include in case the script dependencies are changed in the future. Well, I do not have experience with insserv yet, so the objective is to move the cluster daemons group to the very end of boot sequence in a enforced defined order. ($all) $network $remote_fs ssh ccs cman fence clvmd There are system services like $named and $time and $syslog that should start before any cluster daemon, if available. Their absence could lead to troubles in a production site, but they are not essential for a cluster start. I suggest that redhat-cluster-suite depends on ssh too, because when problems arise, without ssh is very difficult to solve them without it. (my own experience). I guess that the new proposed headers below will accomplish the objective and still LSB 3.1 compliant (did not carefully check against 3.2 [1] but seem ok at first sight). Please, review them. ### BEGIN INIT INFO # Provides: clvmd # Required-Start:$network $remote_fs fence # Required-Stop: fence $remote_fs $network # Default-Start: 2 3 4 5 # Default-Stop: 0 1 6 # Short-Description: Cluster aware LVM daemon. # Description: Coordinate LVM synchronization accross the #Red Hat Cluster. ### END INIT INFO ### BEGIN INIT INFO # Provides: fence # Required-Start:$network $remote_fs cman # Required-Stop: cman $remote_fs $network # Default-Start: 2 3 4 5 # Default-Stop: 0 1 6 # Short-Description: Cluster device fencing daemon. # Description: Coordinate device fencing accross the #Red Hat Cluster. ### END INIT INFO ### BEGIN INIT INFO # Provides: cman # Required-Start:$network $remote_fs ccs # Required-Stop: ccs $remote_fs $network # Default-Start: 2 3 4 5 # Default-Stop: 0 1 6 # Short-Description: Cluster manager daemon. # Description: Manage cluster synchronization accross the #Red Hat Cluster. ### END INIT INFO ### BEGIN INIT INFO # Provides: ccs # Required-Start:$network $remote_fs ssh # Required-Stop: ssh $remote_fs $network # Should-Start: $all # Should-Stop: $all # Default-Start: 2 3 4 5 # Default-Stop: 0 1 6 # Short-Description: Cluster aware LVM daemon. # Description: Cluster configuration system to manage the #cluster.conf file across the #Red Hat Cluster. ### END INIT INFO Regards. Andre Felipe Machado [0] http://refspecs.freestandards.org/LSB_3.1.0/LSB-Core-generic/LSB-Core-generic/initscrcomconv.html [1] http://refspecs.freestandards.org/LSB_3.2.0/LSB-Core-generic/LSB-Core-generic/initscrcomconv.html -- A Debian user never dies. Issues a last command: shutdown -h now http://www.techforce.com.br
Bug#427499: Merge a few bugs and provide LSB header
[Andre Felipe Machado] I read [0] and tried to move the cluster daemons to the end of init.d queue as far as possible, in a defined order inside the group. As the Should-Start does not guarantee this, removed then at the new proposed headers below. Why do you want to start the cluster daemon after unrelated services like xdm and munin-node? I believe it is better to not depend on '$all', and instead list the services it need to start after and stop before. Given that $all is composed of base services, should not the order be first honor the $all then the other strong dependencies in the header? It is not obvious, given the semantic of $all. And it is not how it is implemented in the only implementation I know about, insserv. Well, I do not have experience with insserv yet, so the objective is to move the cluster daemons group to the very end of boot sequence in a enforced defined order. ($all) $network $remote_fs ssh ccs cman fence clvmd There are system services like $named and $time and $syslog that should start before any cluster daemon, if available. Their absence could lead to troubles in a production site, but they are not essential for a cluster start. Then it is better to list $named, $time and $syslog in required-* or should-* I guess that the new proposed headers below will accomplish the objective and still LSB 3.1 compliant (did not carefully check against 3.2 [1] but seem ok at first sight). Look good, but I would drop the $all dependency. Also, there is no need to order the start and stop dependencies in opposite order. Instead of # Required-Start:$network $remote_fs fence # Required-Stop: fence $remote_fs $network I would use # Required-Start:$remote_fs $network fence # Required-Stop: $remote_fs $network fence It make it easier to see that the two lists are identical. The other look ok, except the ccs one. I propose this one: ### BEGIN INIT INFO # Provides: ccs # Required-Start:$remote_fs $network ssh # Required-Stop: $remote_fs $network ssh # Should-Start: $named $time $syslog # Should-Stop: $named $time $syslog # Default-Start: 2 3 4 5 # Default-Stop: 0 1 6 # Short-Description: Cluster aware LVM daemon. # Description: Cluster configuration system to manage the #cluster.conf file across the #Red Hat Cluster. ### END INIT INFO Happy hacking, -- Petter Reinholdtsen -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of unsubscribe. Trouble? Contact [EMAIL PROTECTED]
Bug#427499: Merge a few bugs and provide LSB header
Hello, Petter Many thanks for the explanations. The new corrected headers are below. Please, do not forget to make redhat-cluster-suite dependent on ssh if you implement the ccs header. It is really difficult to solve problems in a cluster without ssh. ### BEGIN INIT INFO # Provides: clvmd # Required-Start:$network $remote_fs fence # Required-Stop: $network $remote_fs fence # Default-Start: 2 3 4 5 # Default-Stop: 0 1 6 # Short-Description: Cluster aware LVM daemon. # Description: Coordinate LVM synchronization accross the #Red Hat Cluster. ### END INIT INFO ### BEGIN INIT INFO # Provides: fence # Required-Start:$network $remote_fs cman # Required-Stop: $network $remote_fs cman # Default-Start: 2 3 4 5 # Default-Stop: 0 1 6 # Short-Description: Cluster device fencing daemon. # Description: Coordinate device fencing accross the #Red Hat Cluster. ### END INIT INFO ### BEGIN INIT INFO # Provides: cman # Required-Start:$network $remote_fs ccs # Required-Stop: $network $remote_fs ccs # Default-Start: 2 3 4 5 # Default-Stop: 0 1 6 # Short-Description: Cluster manager daemon. # Description: Manage cluster synchronization accross the #Red Hat Cluster. ### END INIT INFO ### BEGIN INIT INFO # Provides: ccs # Required-Start:$network $remote_fs ssh # Required-Stop: $network $remote_fs ssh # Should-Start: $named $time $syslog # Should-Stop: $named $time $syslog # Default-Start: 2 3 4 5 # Default-Stop: 0 1 6 # Short-Description: Cluster configuration daemon. # Description: Cluster configuration system to manage and #sync the cluster.conf file across the #Red Hat Cluster. ### END INIT INFO Regards. Andre Felipe Machado -- A Debian user never dies. Issues a last command: shutdown -h now http://www.techforce.com.br
