Liane Praza wrote: > Tony Nguyen writes: > >> A transient enabling creates general_ovr/enabled property and sets the >> state to 'true (temporary)'. However, if the general/enabled property is >> missing, the instance can't come online and stays at disabled state. >> Adding the general/enabled property will make the instance become >> online. So it looks like general_ovr/enabled requires the existence of >> general/enabled. >> >> Two questions. >> Why does general/enabled property need to exist? Or I'm missing >> something here. >> > > Currently, svc.startd takes the existance of a general property group > on an instance as a key to take notice of the instance and start > doing something about it. > > There are 2 main reasons for this: > 1) startd doesn't do anything useful with a service that isn't enabled. > > 2) It was a quickly-implementable way to ensure that startd didn't > notice a service until we finished importing it. That is, during > svccfg import we set the general/enabled property *last* so that > startd doesn't try to start up a service that's half imported. > > I'm not proud of that particular bit of implementation, but it > solved the problem at hand when we were under quite a bit of time > pressure. > > The comments with libscf_get_basic_instance_data() and > dgraph_update_general() in the startd code give a clue to the > behaviour, but we should be more explicit in the header comment of > graph.c. > > (svccfg seems silent about this behaviour in its comments, though the > SCI_GENERALLAST flag is the setting which controls creating general > last.) > > I'm also not the biggest fan of the general_ovr implementation, and > hope that someday we will have temporary overrides of properties that > aren't required to use an alternate namespace defined only by convention. > > This make a lot more sense. I'll look closer into this. >> We can create general_ovr/enabled if it doesn't exist in >> set_inst_enabled_flags, can't we do the same for general/enabled if >> doesn't already exist when need to enable general_ovr/enabled? >> > > I infer from your statement that > smf_[enable|disable]_instance(*, SMF_TEMPORARY) doesn't create both > general/enabled and general_ovr/enabled (when they don't exist)? If so, > sounds like a bug. > > Feel free to file and fix. :) > > In set_inst_enabled, we only check for the existence of property group we're enabling(either general or general_ovr) and create that if it doesn't exist. We should probably always check for the existence of general/enabled property and create it if necessary. I'll file a bug and assign myself to it :^)
Thanks, -tony
