James Carlson wrote: > Liane Praza writes: >> [1] I have a concern about a 'run-last' mechanism, given that there were >> two posited consumers for it within 24 hours of it being proposed, which >> is the same thing that happens every time such a mechanism is proposed. > > Really? I saw exactly one consumer -- the UPS shutdown case. I must > have missed the other.
Jordan's patching consumer, right? >> The specific request in this thread may or may not have been sound, >> but the time-to-second-consumer is standard, which means that I worry >> that there will never really be a single run-last service on any system, >> and that those run-last services will prove to be buggy in the face of >> their expectation being violated. > > The semantics I was expecting was that all of the services with this > flag would go into maintenance (and wouldn't work at all) if there > were more than one installed on the system. It would intentionally > work with at most one such service. > > Alternatively, I suppose a special guaranteed-to-be-last FMRI could be > reserved ... and then only one user could possibly install that way. Yes, that'd resolve that concern. But, what's an admin to do when they have installed two services with the run-last expectation? I admit I haven't spent a lot of time this week thinking about which alternative amongst those proposed in this thread is most architecturally sound. > In any event, I don't think the :true thing would work. That just > means it has no dependencies. It doesn't necessarily guarantee that > the service goes last on shutdown. Eh? As I suggested, network/loopback would depend on it. Given a reverse-dependency-order shutdown, what's your concern? liane
