Hi. If I create a zpool with the following command: zpool create tank raidz2 da0 da1 da2 da3 da4 da5 da6 da7 and after a reboot the device names for some reason are changed so da2 and da5 are swapped, either by altering the LUN setting on the storage or by switching cables/swapping disks etc.? How will zfs handle that? Will it simply acknowledge that all devices are present and the pool is intact and report online or will it complain? The reason I ask is that six months ago I had two seperate storage systems on veritas volume manager (vxfs) attached to my solaris-9-nfs-server via a fiber-switch. I migrated all volumes from the smaller storage-system to the larger one and disconnected it. All went fine and the system was humming along. The fiber-switch rebooted three weeks ago and some of the devices (LUN''s) went into state failing as seen by veritas so I had to online the devices and fsck the volumes that went offline. The tape-drive also took a hit and so I did a devfsadm to have the devices (re)created. But then a device changed name from c3t1d0s2 to c3t0d0s2 and now veritas report the disk as missing and one volume as unavailable. regards Claus
Claus Guttesen wrote:> Hi. > > If I create a zpool with the following command: > > zpool create tank raidz2 da0 da1 da2 da3 da4 da5 da6 da7 > > and after a reboot the device names for some reason are changed so da2 > and da5 are swapped, either by altering the LUN setting on the storage > or by switching cables/swapping disks etc.? > > How will zfs handle that? Will it simply acknowledge that all devices > are present and the pool is intact and report online or will it > complain? >It will just work. The best way I have found to experiment with ZFS configureations is to get a bunch of USB sticks and have a play! Ian
On Fri, 2007-04-27 at 09:25 +1200, Ian Collins wrote:> Claus Guttesen wrote: > > > Hi. > > > > If I create a zpool with the following command: > > > > zpool create tank raidz2 da0 da1 da2 da3 da4 da5 da6 da7 > > > > and after a reboot the device names for some reason are changed so da2 > > and da5 are swapped, either by altering the LUN setting on the storage > > or by switching cables/swapping disks etc.? > > > > How will zfs handle that? Will it simply acknowledge that all devices > > are present and the pool is intact and report online or will it > > complain? > > > It will just work.one thing i would guess is that the device id will keep the same and be used as final proof while path name is only used as a clue. can somebody confirm this? thx.> > The best way I have found to experiment with ZFS configureations is to > get a bunch of USB sticks and have a play! > > Ian > > _______________________________________________ > zfs-discuss mailing list > zfs-discuss at opensolaris.org > http://mail.opensolaris.org/mailman/listinfo/zfs-discuss
On Thu, Apr 26, 2007 at 05:46:36PM -0400, Ming Zhang wrote:> > one thing i would guess is that the device id will keep the same and be > used as final proof while path name is only used as a clue. >The devid is the preferred method of finding devices. The path is used as a secondary measure. In all cases, the on-disk data is the only thing that is used as final proof. - Eric -- Eric Schrock, Solaris Kernel Development http://blogs.sun.com/eschrock
> zpool create tank raidz2 da0 da1 da2 da3 da4 da5 da6 da7 > > and after a reboot the device names for some reason are changed so da2 > and da5 are swapped, either by altering the LUN setting on the storage > or by switching cables/swapping disks etc.? > > How will zfs handle that? Will it simply acknowledge that all devices > are present and the pool is intact and report online or will it > complain?It should simply work.> The reason I ask is that six months ago I had two seperate storage > systems on veritas volume manager (vxfs) attached to my > solaris-9-nfs-server via a fiber-switch. I migrated all volumes from > the smaller storage-system to the larger one and disconnected it. All > went fine and the system was humming along. > > The fiber-switch rebooted three weeks ago and some of the devices > (LUN''s) went into state failing as seen by veritas so I had to online > the devices and fsck the volumes that went offline. The tape-drive > also took a hit and so I did a devfsadm to have the devices > (re)created. But then a device changed name from c3t1d0s2 to c3t0d0s2 > and now veritas report the disk as missing and one volume as > unavailable.Both VxVM and ZFS use similar schemes. They can make use of a local file that enumerates non-disk devices as storage spaces, and they scan for attached disks looking for a signature that identifies it as a ZFS/VxVM component. Data within the storage identifies its place in the pool or diskgroup. As long as the disk can be "seen" by the OS, the specific access name shouldn''t matter. Without more details, it''s impossible to guess what happened with your previous setup, but the general case of renaming (offline!) storage should not affect either ZFS or VxVM. -- Darren Dunham ddunham at taos.com Senior Technical Consultant TAOS http://www.taos.com/ Got some Dr Pepper? San Francisco, CA bay area < This line left intentionally blank to confuse you. >
Ming Zhang wrote:> On Fri, 2007-04-27 at 09:25 +1200, Ian Collins wrote: >> Claus Guttesen wrote: >> >>> Hi. >>> >>> If I create a zpool with the following command: >>> >>> zpool create tank raidz2 da0 da1 da2 da3 da4 da5 da6 da7 >>> >>> and after a reboot the device names for some reason are changed so da2 >>> and da5 are swapped, either by altering the LUN setting on the storage >>> or by switching cables/swapping disks etc.? >>> >>> How will zfs handle that? Will it simply acknowledge that all devices >>> are present and the pool is intact and report online or will it >>> complain? >>> >> It will just work. > > one thing i would guess is that the device id will keep the same and be > used as final proof while path name is only used as a clue. > > can somebody confirm this? thx.Yes, this works. We make use of the device id (which doesn''t change, for almost all circumstances) to identify luns and route read and write ops appropriately. The only circumstance that I am aware of where a devid might change is if a disk or tape''s firmware is changed such that its SCSI INQUIRY page83 response changes. This is a very rare thing and you the user would have to knowingly kick this off. cheers, James C. McPherson -- Solaris kernel software engineer, system admin and troubleshooter http://www.jmcp.homeunix.com/blog Find me on LinkedIn @ http://www.linkedin.com/in/jamescmcpherson
On Thu, 2007-04-26 at 14:50 -0700, Eric Schrock wrote:> On Thu, Apr 26, 2007 at 05:46:36PM -0400, Ming Zhang wrote: > > > > one thing i would guess is that the device id will keep the same and be > > used as final proof while path name is only used as a clue. > > > > The devid is the preferred method of finding devices. The path is used > as a secondary measure. In all cases, the on-disk data is the only > thing that is used as final proof.thanks. i am reading the on disk spec doc and can be validated there.> > - Eric > > -- > Eric Schrock, Solaris Kernel Development http://blogs.sun.com/eschrock
On Fri, 2007-04-27 at 07:53 +1000, James C. McPherson wrote:> Ming Zhang wrote: > > On Fri, 2007-04-27 at 09:25 +1200, Ian Collins wrote: > >> Claus Guttesen wrote: > >> > >>> Hi. > >>> > >>> If I create a zpool with the following command: > >>> > >>> zpool create tank raidz2 da0 da1 da2 da3 da4 da5 da6 da7 > >>> > >>> and after a reboot the device names for some reason are changed so da2 > >>> and da5 are swapped, either by altering the LUN setting on the storage > >>> or by switching cables/swapping disks etc.? > >>> > >>> How will zfs handle that? Will it simply acknowledge that all devices > >>> are present and the pool is intact and report online or will it > >>> complain? > >>> > >> It will just work. > > > > one thing i would guess is that the device id will keep the same and be > > used as final proof while path name is only used as a clue. > > > > can somebody confirm this? thx. > > > Yes, this works. We make use of the device id (which > doesn''t change, for almost all circumstances) to identify > luns and route read and write ops appropriately. > > > The only circumstance that I am aware of where a devid might > change is if a disk or tape''s firmware is changed such that > its SCSI INQUIRY page83 response changes. This is a very rare > thing and you the user would have to knowingly kick this off. >as Eric pointed out, i think at last the on disk data will help to validate them. thx.> > cheers, > James C. McPherson > -- > Solaris kernel software engineer, system admin and troubleshooter > http://www.jmcp.homeunix.com/blog > Find me on LinkedIn @ http://www.linkedin.com/in/jamescmcpherson