A backup of a controller enables one to re-establish the configuration and state of a controller. It does not influence workload instances on the backing cloud. That is, if such an instance is terminated directly in the cloud then a controller restore cannot re-create it.
This page will cover the following topics:
- Creating a backup
- Managing backups
- Restoring from a backup
- High availability considerations
Note: Data backups can also be made of the Juju client. See the Juju client page for guidance.
Juju provides commands for recovering a controller in case of breakage or in the case where the controller no longer exists.
The current state is held within the 'controller' model. Therefore, all backup
commands need to operate within that model explicitly or by ensuring the
current model is the 'controller' model. In the examples provided below, both
the controller name and the model name are expressed explicitly (e.g.
-m aws:controller). Due to the delicate nature of data backups, this method
is highly recommended.
create-backup command is used to create a backup. It does so by
generating an archive and downloading it to the Juju client system as a
'tar.gz' file (a local backup). If the
--keep-copy option is used then a
copy of the archive will also remain on the controller (a remote backup).
With the aid of the
--no-download option a local backup can be prevented, but
since the archive must be kept somewhere, this option implies
The name of the backup is composed of the creation time (in UTC) and a unique identifier.
The below examples assume the existence of the following controllers (output to
Controller Model User Access Cloud/Region Models Machines HA Version aws default admin superuser aws/us-east-1 2 1 none 2.3.7 lxd* default admin superuser localhost/localhost 2 1 none 2.3.7
To create a backup of the 'aws' controller:
juju create-backup -m aws:controller
20180515-191942.7e45250b-637a-4dc9-8389-c6aa70100cd6 downloading to juju-backup-20180515-191942.tar.gz
From the name of the archive we see that the backup was made on May 15, 2018 at 19:19:42 UTC.
Warning: Archive filenames do not include the associated controller name. Care should therefore be taken when archiving from multiple controllers. To specify a custom name use the
--filename option. This option does not affect the remote archive name.
To create a backup of the 'lxd' controller while both using a custom filename and adding an optional note:
juju create-backup -m lxd:controller --filename juju-backup-lxd-20180515-193724.tar.gz "fresh lxd controller"
The optional note is exposed via the
show-backup command detailed below.
Note: A backup of a fresh (empty) environment, regardless of cloud type, is approximately 56 MiB in size.
The following commands are available for managing backups (apart from restoring):
backups command displays the names of all remote backups for a given
controller. For instance, to see all remotely stored backups for the 'lxd'
juju backups -m lxd:controller
show-backup command provides a metadata record for a specific remote
backup (identified via the
backups command). For example, to query a backup
stored on the 'lxd' controller:
juju show-backup -m lxd:controller 20180515-195557.9c6a3650-2957-489a-8f0c-6c3b5ce2e055
backup ID: "20180515-193724.9c6a3650-2957-489a-8f0c-6c3b5ce2e055" checksum: "pmxx7bCwtZVV+KM48YKz5w6Boc0=" checksum format: "SHA-1, base64 encoded" size (B): 58605244 stored: 2018-05-15 19:40:28 +0000 UTC started: 2018-05-15 19:37:24 +0000 UTC finished: 2018-05-15 19:37:41 +0000 UTC notes: "fresh lxd controller" model ID: "9c6a3650-2957-489a-8f0c-6c3b5ce2e055" machine ID: "0" created on host: "juju-e2e055-0" juju version: 2.3.7
started time is the most pertinent of the various timestamps. It refers
to the time the backup was created.
download-backup command downloads a specific remote backup (again,
identified via the
backups command). Here, we download a backup that is
stored on the 'aws' controller:
juju download-backup -m aws:controller 20180515-191942.7e45250b-637a-4dc9-8389-c6aa70100cd6
upload-backup command uploads a specific local backup to a controller.
juju upload-backup -m lxd:controller juju-backup-20180515-193724.tar.gz
Note: It is not possible to upload a file that is equivalent to a backup stored remotely. The process will be cancelled and an error message will be printed.
remove-backup command removes a specific remote backup from a controller.
juju remove-backup -m aws:controller 20180515-191942.7e45250b-637a-4dc9-8389-c6aa70100cd6
To clean up any possible remote backups the
--keep-latest option can be used.
This option instructs Juju to remove all remote backups with the exception of
the most recently created one:
juju remove-backup -m aws:controller --keep-latest
To revert the state of an environment to a previous time the
command is used.
Warning: The restore process does not validate that a backup archive corresponds to the controller it was created from. Make sure you do not overwrite a controller with the wrong backup.
This command requires the use of the
--id option when referring to a remote
juju restore-backup -m lxd:controller --id 20180515-193724.9c6a3650-2957-489a-8f0c-6c3b5ce2e055
To apply a local backup the
--file option must be used:
juju restore-backup -m lxd:controller --file juju-backup-lxd-20180515-193724.tar.gz
Note: It is not possible to restore using a local backup that is equivalent to a remote backup. The process will be cancelled and an error message will be printed. The remote backup should just be used instead.
Although Controller high availability makes for a more robust
(and load balanced) Juju infrastructure, it should not replace the need for
data backups. It does, however, make the prospect of restoring from backup less
likely, since as long as one controller cluster member remains operational, the
others can be replaced via the
enable-ha command. A restore in an HA scenario
therefore only becomes necessary when all controllers have failed. However,
if a restore is applied to a cluster with active members all reachable
controllers will naturally have their data overwritten.
It is not possible at this time to restore while HA is enabled.
To restore to an HA cluster one needs to first remove HA (by removing all but one of the controller machines) and then perform a restore operation. HA may then be re-enabled afterwards.
For example, consider a three-member cluster with machines '0', '1', and '2' in
the 'controller' model and where a backup of the cluster was previously made
Here, we begin by removing machines '1' and '2' but you can remove any two:
juju remove-machine -m aws:controller 1 2
Now wait until Juju reports that it is in a non-HA state. This is indicated by
the text 'none' under the 'HA' column in the output to the
Controller Model User Access Cloud/Region Models Machines HA Version aws* default admin superuser aws/us-east-1 3 2 none 2.4-rc1
There should now be only a single machine listed in the output to
juju machines -m aws:controller.
We can now restore:
juju restore-backup -m aws:controller --file aws-ha3.tar.gz
After a while the two removed machines will reappear but in a 'down' state:
Machine State DNS Inst id Series AZ Message 0 started 22.214.171.124 i-0095fa21cda2b3b9c xenial us-east-1a running 1 down 126.96.36.199 i-08105aeb4e04a26e2 xenial us-east-1a running 2 down 188.8.131.52 i-0e6417bf06d36498b xenial us-east-1c running
Remove them by force:
juju remove-machine -m aws:controller 1 2 --force
You can now re-enable HA if desired:
juju enable-ha -c aws
See Controller high availability for guidance on using the
In the advent that all controllers are unresponsive the following steps should be taken:
- Remove the cluster
- Create a pristine controller
- Perform a data restore
- Enable HA
To demonstrate this, consider an initial AWS-based controller named 'aws-ha3-1' with three cluster members. The new controller will be called 'aws-ha3-2':
juju kill-controller aws-ha3-1 juju bootstrap aws aws-ha3-2 juju restore-backup -m aws-ha3-2:controller --file backup.tar.gz juju enable-ha -m aws-ha3-2:controller -n 3
Note: Section Recovering from controller failure details how to deal with a partially degraded cluster.