On Thu, Nov 17, 2022 at 09:40:26PM +0100, Martin Wilck wrote: > On Thu, 2022-11-17 at 12:53 -0600, Benjamin Marzinski wrote: > > On Wed, Nov 09, 2022 at 10:10:07PM +0100, mwilck@xxxxxxxx wrote: > > > From: Martin Wilck <mwilck@xxxxxxxx> > > > > > With greedy, we expect that the blacklists must be correctly set up, > > so > > we're just slowing things down to deal with people not configuring > > multipath correctly. > > Only in theory. Because of the failed-wwids logic, "greedy" works quite > well actually, even if the blacklist is not correctly set up. > With this as a special exception. > > > But since I rarely see greedy configurations, I > > don't really have strong feelings about this trade-off. > > I've been wondering whether we could make this depend on a config > option (yes I know, I've said often that we have too many of them). > We could also have it depend on "greedy". But it might also be useful > with "smart" if we have a lot of LUNs. > > > > > More suggestions below. > > > > [...] <snip> Ooops. I signed my name a bit too high. I had one more suggestion, but it's basically the same as what you suggest above. > > > > -Ben > > <snip> > > > @@ -96,6 +346,9 @@ is_path_valid(const char *name, struct config > > > *conf, struct path *pp, > > > return PATH_IS_ERROR; > > > } > > > > > > + if (is_device_in_use(pp->udev) > 0) > > > + return PATH_IS_NOT_VALID; > > > + Here. > > > > Can we make this only apply to "greedy"? For "strict", "no" and "yes" > > this makes the common case slower (you are running multipath on a > > machine with multipath devices that you've seen before) with no real > > benefit. > > > > It might also be useful to run this check before we return "maybe" > > for > > find_multipaths "smart", perhaps as an alternative to the O_EXCL test > > we > > currently use. > > > > > if (conf->find_multipaths == FIND_MULTIPATHS_GREEDY) > > > return PATH_IS_VALID; > > > -- dm-devel mailing list dm-devel@xxxxxxxxxx https://listman.redhat.com/mailman/listinfo/dm-devel
