How to get a Xen domU to notice a change in the block size of one of its phy devices

LVM snapshots are meant to capture the filesystem in a frozen state. They are not meant to be a backup in and of themselves. They are, however, useful for obtaining backup images that are consistent because the frozen image cannot and will not change during the backup process. So while you won't use them directly to make long-term backups, they will be of great value in any backup process that you decide to use.

There are a few steps to implement a snapshot. The first is that a new logical volume has to be allocated. The purpose of this volume is to provide an area where deltas (changes) to the filesystem are recorded. This allows the original volume to continue on without disrupting any existing read/write access. The downside to this is that the snapshot area is of a finite size, which means on a system with busy writes, it can fill up rather quickly. For volumes that have significant write activity, you will want to increase the size of your snapshot to allow enough space for all changes to be recorded. If your snapshot overflows (fills up) both the snapshot will halt and be marked as unusable. Should this happen, you will want to release your snapshot so you can get the original volume back online. Once the release is complete, you'll be able to remount the volume as read/write and make the filesystem on it available.

The second thing that happens is that LVM now "swaps" the true purposes of the volumes in question. You would think that the newly allocated snapshot would be the place to look for any changes to the filesystem, after all, it's where all the writes are going to, right? No, it's the other way around. Filesystems are mounted to LVM volume names, so swapping out the name from underneath the rest of the system would be a no-no (because the snapshot uses a different name). So the solution here is simple: When you access the original volume name, it will continue to refer to the live (read/write) version of the volume you did the snapshot of. The snapshot volume you create will refer to the frozen (read-only) version of the volume you intend to back up. A little confusing at first, but it will make sense.

All of this happens in less than 2 seconds. The rest of the system doesn't even notice. Unless, of course, you don't release the snapshot before it overflows...

At some point you will want to release your snapshot to reclaim the space it occupies. Once the release is complete, the snapshot volume is released back into the volume, and the original remains.

I do not recommend pursuing this as a long-term backup strategy. You are still hosting data on the same physical drive that can fail, and recovery of your filesystem from a drive that has failed is no backup at all.

So, in a nutshell:

• Snapshots are good for assisting backups
• Snapshots are not, in and of themselves, a form of backup
• Snapshots do not last forever
• A full snapshot is not a good thing
• Snapshots need to be released at some point
• LVM is your friend, if you use it wisely.

I eventually solved this by using the following advice and settings from the open-iscsi documentation:

8.2 iSCSI settings for iSCSI root
---------------------------------

When accessing the root parition directly through a iSCSI disk, the
iSCSI timers should be set so that iSCSI layer has several chances to try to
re-establish a session and so that commands are not quickly requeued to
the SCSI layer. Basically you want the opposite of when using dm-multipath.

For this setup, you can turn off iSCSI pings by setting:

node.conn[0].timeo.noop_out_interval = 0
node.conn[0].timeo.noop_out_timeout = 0

And you can turn the replacement_timer to a very long value:

node.session.timeo.replacement_timeout = 86400

After setting up the connection to each LUN as described above, the failover works like a charm, even if it takes several minutes to happen.