Description of problem: These RPMs cannot be used in production, as they touch running services, without consulting the systems admin. 1) There is not enough logic to know everything that must be done for a clean restart. (rc script/upstart/systemd/puppet/chef/cfengine/... controlled) 2) Upgrading services is a two stage process (only stage "b" causes an outage) a) update files (standard RPM install) b) restart services (Unknown environmental dependencies) The only rule is that "a" comes before "b". There could be days between the two, depending on environmental requirements (including political ones.) So, for everyone that is running their production cluster properly, they must rebuild the RPMs to omit the pre-uninstall scriptlet. This make the current RPM set unusable to anyone but experimenters, or those that don't care about end user experiences or production outages. Unix design philosophy: Do one thing and do it well. Software Package: Install software only Please remove pre-uninstall scriptlet. Version-Release number of selected component (if applicable): 3.5.0-2 How reproducible: Very Steps to Reproduce: 1. Upgrade RPM on any running node 2. 3. Actual results: Services are restarted Expected results: Services should not be started/stopped or restarted. Additional info:
Some users expect the services to be restarted, others dont. This is a conflict that we can only resolve by making the restart a configurable option. It would be good to have an option in /etc/sysconfig/gluster that influences the RPM scriptlets and does not restart when the option in enabled. Fedora packaging requires restarting of the services when updates are installed. This should probably stay the default. (Also, most sysadmins that I know only install updates during maintenance windows where systems are temporarily taken offline.)
If Fedora packaging allow it, I assume we should keep it.