David Bustos wrote: > Quoth Alan Maguire on Thu, Jun 26, 2008 at 04:48:35PM +0100: > >> David Bustos wrote: >> >>> Quoth Alan Maguire on Mon, Jun 23, 2008 at 05:05:13PM +0100: >>> >>>> http://cr.opensolaris.org/~amaguire/smf_ordered_values/ >>>> >>> cmd/svc/configd/backend.c >>> > ... > >>> 1697: I don't understand why we would need to check whether the >>> database has been upgraded without upgrading it. Please explain it >>> to me. >>> >> The logic here is that during boot, we want to determine >> as soon as possible if the repository is already upgraded, >> as if it is, we can use the value-ordered SELECT statement >> when retrieving values. By doing the upgrade check here, >> we catch the fact that the repository has been upgraded, >> set be_normal_upgraded to TRUE, and thus property >> retrieval will determine that we can use the value_order >> column in the SELECT statement used in fill_property_callback() >> (in file_object.c). I've elaborated a bit in the comment here and >> above backend_check_upgrade(), hopefully it's a bit >> clearer now. >> > > Ah, yes, we assume that the repository is not upgraded, so since we want > reads to be ordered if possible before the repository becomes writable, > we need to check on each start up. Do you know whether clients can get > any requests in between there and the zone-boot check? Doesn't look like it - backend_init() is called way before door_server_create() in configd.c. > If not, > I suspect you could consolidate the two calls into one. But I don't > think there's enough benefit to that to change your code now. > > okay. > By the way, did you notice the "Eventually, we'll support schema upgrade > here." comment at 1576? It seems like you're making this a bit stale. > I suppose it's not a big enough deal to change now, though. > > hmm, probably no harm to remove it, I suspect it might confuse people unnecessarily. Thanks!
Alan
