Paul Kraus wrote: > On 10/24/06, David Bustos <David.Bustos at sun.com> wrote: > >> svcadm refresh is meant to commit configuration changes and notify the >> service of those changes. For syslog, this happens to be the same >> operation as close-log-files-for-rotation, which is what we really want >> to do. But from an SMF perspective, they're not the same, so logadm >> shouldn't use svcadm refresh. In particular, if an administrator were >> in the middle of making changes, and logadm refreshed syslog, then the >> changes might be committed prematurely. > > According to the man page for syslogd, whenever it receives a > HP signal it re-reads its configuration file, so I don't see how a > kill -HUP is any different from a svcadm refresh ... unless svcadm > refresh does additional things. Am I missing something here ? > > I expect that if an admin were in the middle of making > changes, and had saved partial changes to the /etc/syslog.conf file, > either mechnism (kill -HUP or svcadm refresh) would have syslogd load > a partial configuration. If the admin does not save he changes to > /etc/syslog.conf until they are complete, then only the complete > config change will be used. >
What David was saying is if you were in the middle of making changes to the service configuration (using svccfg) not /etc/syslog.conf, then those changes could be committed prematurely when logadm refreshed syslog. -- Renaud
