Live storage migration using libvirt blockcopy and blockcommit
===============================================================

[Notes, from a discussion with Eric Blake.]


[FIXME] Address, feedback from Eric Blake on 17MAR2015:
-----------------------------------------------------------------------
 - If your end goal is to commit back into the migrated image, then
   --quiesce doesn't add any value. --quiesce is only useful if you plan
   to revert to a known point in time and want the filesystem stable at
   that point in time. That, and snapshot-create-as doesn't yet support
   --quiesce, although we already have an open BZ to add it.

 - Next, when you do the blockcopy, you probably want to pre-create the
   destination /path/to/sn2 with the backing information correctly
   pointing to /path/to/base

 - And use the blockcopy --reuse-external flag, because if you don't,
   then libvirt/QEMU will create /path/to/sn2 with /path1/to/base as its
   backing file, which is not what you want.
   
   If you want, you can use 'blockcopy ... --pivot', instead of 'blockcopy
   ...  --finish' + 'blockjob --pivot' (one command instead of two).

 - Might also be worth mentioning that there might be more efficient
   copies than `rsync`. For example, if you have BTRFS or ISCSI or other
   storage with a super-fast copy-on-write cloning operation, 10:42:42>
   so the step that copies /path1/to/base to /path2/to/base can take
   advantage of these faster storage-specific tricks.

  But otherwise, it looks like your page is mostly there, in describing
  a live storage migration 
-----------------------------------------------------------------------

Scenario
--------

Start with:

    [/path1/to/base] (qemu active)


then snapshot:

    [/path1/to/base] <- [sn1] (qemu active)


then background copy (using `rsync` or `cp`):

    [/path1/to/base] <- [sn1] (qemu active)
    |                                                                                                                                                   
    [/path2/to/base]


then shallow blockcopy:

    [/path1/to/base] <- [sn1] (qemu active)
      |                   |                                                                                                                               
      [/path2/to/base] <- [sn2]


then pivot:

    [/path2/to/base] <- [sn2] (qemu active)


then live commit:

    [/path2/to/base]


Procedure
---------

[Need to test this thoroughly.]

    $ virsh domblklist vm1

    # Create snapshot ( use --quiesce 
    $ virsh snapshot-create-as --domain vm1 sn1 \
        --diskspec vda,file=/export/images/sn1.qcow2 \
        --disk-only --atomic --no-metadata

    NOTE: Above, if you have QEMU guest agent installed in your virtual
    machine, try '--quiesce' option with 'snapshot-create-as' to ensure
    you have a consistent disk state.

    # Take a backup of disk
    $ rsync -avh --progress /export/images/base.img \
        /export/images/copy.img

    # Shallow blockcopy (to path 2)
    $ virsh blockcopy \
        --domain vm1 vda /export/images/copy.qcow2 \
        --wait --verbose --shallow --finish

    # Then, pivot to the copy (FIXME: test)
    $ virsh blockjob vda --pivot

    # Then, live blockcommit (to path 2):
    $ virsh blockcommit vm1 vda --active --verbose --pivot