[rhelv6-beta-list] My first experiences with RHEL6 beta

[Lamar Owen]

On Tuesday, June 15, 2010 03:39:12 pm Chris Adams wrote:
> However, my point is that moving PEs around is fun (but not terribly
> useful in lots of cases), and changing which LVs they are assigned to is
> great and all, but it is of reduced usefulness if the things on top of
> those PEs can't handle the LVs changing sizes on the fly.

I actually put the ability to move PE's around to good use, on a Sun Enterprise 6500 (17 CPU, 20GB RAM) the other day.  The F12 Beta for SPARC is pretty fragile, so I set up the default LVM system on the first disk of a D1000 storedge shelf hanging off a pair of Qlogic ISP diffSCSI controllers.  After the system was installed, I created a RAID6 MD set with the other 8 drives I have in the D1000, added the DM device to the volume group, and pvmoved things over to the RAID6 MD set from the single drive.  Yep, pvmoved the PE's for the root filesystem while it was mounted and running, which is sweetness, and something much like the LUN migration our Clariions can do (other than boot records; booting from an MD isn't supported at all on F12 SPARC, so the boot is still on the single drive).  

I can see doing live disk or array migration, using hotswap disk caddies, of the root or any other non-swap filesystem on a running server with no downtime.  With SATA, SAS, and even SCSI hotplug, this gives great utility to servers that need to be highly available.  The only snag will be moving the boot sectors, but a competent sysadmin should be able to handle that without a lot of fuss.

The one thing to watch out for is moving PE's that contain swap; the F12 SPARC kernel, at least, panics when swap's PE's are moved out from under it (happens at the very last, when the move is fully committed).

At least ext[234] can be shrunk; you can't shrink even an UNmounted XFS filesystem.  XFS is grow-only.  Hit that a few weeks ago; had a 32-bit CentOS 4 VMware guest with a greater than 16TB XFS filesystem hanging off of it; worked just fine until the amount of actual data on the filesystem rolled 16TB, and then it quit working.  Had to upgrade to a 64-bit guest (rolled in CentOS 5) to get the filesystem to mount again, rather than the desired 'boot a 64-bit rescue CD, move a few of the last files to another filesystem, then shrink the filesystem back under 16TB' process that I would have preferred to do.