Skip to content

Latest commit

 

History

History
118 lines (84 loc) · 5.61 KB

NUTSNBOLTS.md

File metadata and controls

118 lines (84 loc) · 5.61 KB
Raw

Forward incremental Backup

Initial configuration. "Current image" pointer is an abstraction for the image currently used by qemu to run a given domain.

  Current image
        |
        v
+-----------------+
|   disk.qcow2    |
+-----------(R/W)-+

Run fi-backup.

                             Current image
                                   |
                                   v
+------------------+   +------------------------+
|    disk.qcow2    |<--| disk.bimg-<timestamp1> |
+------------(R/O)-+   +------------------(R/W)-+

After the first run, disk.qcow2 is the backing file of disk.bimg-<timestamp1>. That means that disk.qcow2 is used in read-only mode by qemu. Thus, it can be safely copied to perform our backup.

                                                           Current image
                                                                 |
                                                                 v
+------------------+   +------------------------+   +------------------------+
|    disk.qcow2    |<--| disk.bimg-<timestamp1> |<--| disk.bimg-<timestamp2> |
+------------(R/O)-+   +------------------(R/O)-+   +------------------(R/W)-+

After the second run, disk.qcow2 has not changed, so we can just copy disk.bimg-<timestamp1> to our backup storage. This is the "key" of forward incremental backup: disk.bimg-<timestamp1> contains just the deltas since the first fi-backup run.

Consolidation

Consolidation reduces the number of backing files. This can be done one of two ways. (1) BlockCommit or (2) BlockPull BlockCommit copies the content of backing files into the current image, and then removing the backing images. BlockPull copies the state of the current image into the last backing image and then removing the original image and child images up to the last backing image.

This reduces the number of image files. See README.md for more information about the advantages of BlockCommit vs BlockPull. Like time to consolidate vs image size.

Here is an example of consolidation via BlockPull

                                                           Current image
                                                                 |
                                                                 v
+------------------+   +------------------------+   +------------------------+
|    disk.qcow2    |<--| disk.bimg-<timestamp1> |<--| disk.bimg-<timestamp2> |
+------------(R/O)-+   +------------------(R/O)-+   +------------------(R/W)-+

Running in consolidation BlockPull mode will "copy" backing images content into the current image.

                                                          Current image
               +--block-pull->+         +--block-pull->+        |
               |              |         |              |        v
+------------------+   +------v---------+-------+   +--v---------------------+
|    disk.qcow2    | X | disk.bimg-<timestamp1> | X | disk.bimg-<timestamp2> |
+-----------(R/O)--+   +------------------(R/O)-+   +------------------(R/W)-+

After that, the original image and old backing files can be deleted. fi-backup.sh will automatically delete every old backing file. :

       Current image
             |
             v
+------------------------+
| disk.bimg-<timestamp2> |
+------------------(R/W)-+

Here is an example of consolidation via BlockCommit

                                                           Current image
                                                                 |
                                                                 v
+------------------+   +------------------------+   +------------------------+
|    disk.qcow2    |<--| disk.bimg-<timestamp1> |<--| disk.bimg-<timestamp2> |
+------------(R/O)-+   +------------------(R/O)-+   +------------------(R/W)-+

Running in consolidation BlockCommit mode will "copy" backing images content into the current image.

                                                          Current image
               +<-block-pull--+         +<-block-pull--+        |
               |              |         |              |        v
+--------------V---+   +------+---------V-------+   +--+---------------------+
|    disk.qcow2    |<--| disk.bimg-<timestamp1> | X | disk.bimg-<timestamp2> |
+-----------(R/O)--+   +------------------(R/O)-+   +------------------(R/W)-+

After that, the unused files can be deleted.fi-backup.sh will automatically delete every unused file. :

       Current image
             |
             v
+------------------------+
|       disk.qcow2       |
+------------------(R/W)-+

Restore

Restore procedure for a domain A, with images in directory B, backup directory C is the following. Thanks to svorobyov for these steps.

  1. clean up the directory B from all images for domain A;
  2. copy the desired chains of the backing images for domain A from C to B (although it is not recommended, chains can be of different lengths, e.g., recover disk 1 to the state two days back and disk 2 to the state three days);
  3. specify (using virsh, virt-manager, virt-install, ...) that the disks for A are the last images of the chains in 2 (this is important, otherwise the VM state description may be inconsistent)
  4. optionally, consolidate for domain A (which will use chains of images in B)

Restore of backup with domain state dump can be performed

TBW <qemu-command-line> -incoming "exec: gzip -c -d <path/to/state/file>"