On Fri, Apr 13, 2018 at 12:19:12AM +0200, Martin Wilck wrote: > On Thu, 2018-04-12 at 13:46 -0500, Benjamin Marzinski wrote: > > On Wed, Apr 04, 2018 at 06:16:25PM +0200, Martin Wilck wrote: > > > With "find_multipaths smart", we accept paths as valid if they are > > > already part of a multipath map. This patch avoids doing a full > > > path > > > and device-mapper map scan for this case, speeding up "multipath > > > -u" > > > considerably. > > > > I feel like this supports my idea from 17/20. I really think that in > > smart mode, the only time we should be claiming a device as multipath > > is > > on an add event, if it has always been claimed (or pretend claimed) > > as a > > multipath path, or if it is currently a multipath path. Once we have > > let > > a uevent go by for a path without setting SYSTEMD_READY=0, anything > > else > > is free to use it, and we simply can't safely set SYSTEMD_READY=0, > > unless we know that multipathd has already grabbed the device. > > We have to remember that in the udev db then. That doesn't work for > coldplug aside, but we agree that's not an issue. By checking a > suitable udev variable, we can make sure that we never set > SYSTEMD_READY=0 after having released the device to the system. We can safely set SYSTEMD_READY=0 if we know that the device is now part of an existing multipath device, but yeah. > > I feel > > like checking the wwids file should give you confidence that a device > > either currently is a multipath path, or has always been claimed as > > one. > > However, this patch can fix any loop holes where a device isn't in > > the > > wwids file, but it multipathed (I don't know of any offhand). > > multipath -u can't assume that multipathd is running. The map may have > been set up in the initrd, and thus would not necessarily be in the > WWIDs file in the root FS. That's a loophole. And because I accidentally forgot to add it before Reviewed-by: Benjamin Marzinski <bmarzins@xxxxxxxxxx> > Currently multipath -u scans all paths and maps in this case, just to > figure out if the given path is maybe already part of a map. My patch > was just meant as a shortcut for this case. If we could rely on the > WWIDs file that would be even easier, but I don't think we can. > > Martin > > -- > Dr. Martin Wilck <mwilck@xxxxxxxx>, Tel. +49 (0)911 74053 2107 > SUSE Linux GmbH, GF: Felix Imendörffer, Jane Smithard, Graham Norton > HRB 21284 (AG Nürnberg) -- dm-devel mailing list dm-devel@xxxxxxxxxx https://www.redhat.com/mailman/listinfo/dm-devel
