On Thu, Oct 08, 2009 at 09:15:52AM -0400, Anurag S. Maskey wrote: > >>> With this fini we are reverting our changes. We should find out why >>> bug 9431 was filed in the first place. >>> >> >> You filed and fixed the bug. Fill us in :) >> > The decision came out from one of the meetings over the summer summer > when we had those daily meetings. The description of the bug shows that > the decision was already made on what to do when the bug was filed. > Maybe Renee or Alan remembers more on this. I think when somebody did a > "svcadm refresh nwam", we do not want to tear down the network just > because of that. The user may have just changed one of the SMF > properties (like debug) and we destroy the network configuration).
I vaguely remember the discussion, though not specifics. In any case, I absolutely agree that the goal is for refresh to be as minimally invasive as possible; that is, if we can keep everything up, we should. I'm pretty sure we all agree with that goal. But we need to make sure the current code is in fact doing that. When I reviewed Michael's code, I had the same question Anurag did that started this thread: why are we doing the _fini on refresh? When I first glanced at the nwamd_init_ncus() code, I was concerned that we would be leaking objects, as Michael suggested; but now looking more closely, I see the part you're talking about, and I see that's not happening. I do also agree with Michael's assertion that the SIGTHAW and SIGHUP should be treated the same. Does it make sense to omit the nwamd_fini_ncus() call in both cases? -renee > > Anurag > > > > >> >>>> I also think that we would be leaking objects if we don't. But that >>>> could be fixed separately if we want. >>>> >>> Objects are not leaked if we don't fini them. The >>> _handle_init_event() code reads in new objects and frees the old >>> objects. If that's not happening, then bugs should be filed. >>> >> >> Okay. >> >> mph >> >> >>> Anurag >>> >>>
