Configuring models

A model influences all the machines that Juju creates within it and, in turn, the applications that get deployed onto those machines. It is therefore a very powerful feature to be able to configure at the model level.

Model configuration consists of a collection of keys and their respective values. An explanation of how to both view and set these key:value pairs is provided below. Notable examples are provided at the end.

Getting and setting values

You can display the current model settings by running the command:

juju model-config

This will include all the currently set key values - whether they were set by you, inherited as a default value or dynamically set by Juju.

A key's value may be set for the current model using the same command:

juju model-config no-proxy=jujucharms.com

It is also possible to specify a list of key-value pairs:

juju model-config test-mode=true enable-os-upgrade=false

Note: Juju does not currently check that the provided key is a valid setting, so make sure you spell it correctly.

To return a value to the default setting the --reset flag is used, along with the key name:

juju model-config --reset test-mode

After deployment, the model-defaults command allows a user to display the configuration values for a model as well as set default values that all new models will use. These values can even be specified for each cloud region instead of just the controller.

To set a default value for 'ftp-proxy', for instance, you would enter the following:

juju model-defaults ftp-proxy=10.0.0.1:8000

To see both the default values and what they've been changed to, you would use:

juju model-defaults

To set default values for all new models in a specific controller region, state the region in the command, shown here using the same example settings as our previous model-config key-value pairs example above:

juju model-defaults us-east-1 test-mode=true enable-os-upgrade=false

Model settings can also be made when creating a new controller via either the --config or --model-default options. The difference being that --config affects just the 'controller' and 'default' models while --model-default affects all models, including any future ones. Below we use the --config option:

juju bootstrap --config image-stream=daily lxd lxd-daily

See Creating a controller for in-depth coverage on how to create a controller.

Note that these defaults can be overridden, on a per-model basis, during the invocation of the add-model command (option --config) as well as by resetting specific options to their original defaults through the use of the model-config command (option --reset).

List of model keys

The table below lists all the model keys which may be assigned a value. Some of these keys deserve further explanation. These are explored in the sections below the table.

Apt mirror

The APT packaging system is used to install and upgrade software on machines provisioned in the model, and many charms also use APT to install software for the applications they deploy. It is possible to set a specific mirror for the APT packages to use, by setting 'apt-mirror':

juju model-config apt-mirror=http://archive.ubuntu.com/ubuntu/

It is also possible to set this to a local mirror if desired.

You may also run:

juju model-config --reset apt-mirror

to restore the default behaviour in a running model.

APT updates and upgrades with faster machine provisioning

When Juju provisions a machine, its default behaviour is to upgrade existing packages to their latest version. If your OS images are fresh and/or your deployed applications do not require the latest package versions, you can disable upgrades in order to provision machines faster.

Two Boolean configuration options are available to disable APT updates and upgrades: enable-os-refresh-update (apt-get update) and enable-os-upgrade (apt-get upgrade), respectively.

enable-os-refresh-update: false
enable-os-upgrade: false

You may also want to just update the package list to ensure a charm has the latest software available to it by disabling upgrades but enabling updates.

Disable network management

This can only be used with MAAS models and should otherwise be set to 'false' (default) unless you want to take over network control from Juju because you have unique and well-defined needs. Setting this to 'true' with MAAS gives you the same behaviour with containers as you already have with other providers: one machine-local address on a single network interface, bridged to the default bridge.

Firewall mode

Modes available include: - instance: Requests the use of an individual firewall per instance. - global: Uses a single firewall for all instances (access for a network port is enabled to one instance if any instance requires that port). - none: Requests that no firewalling should be performed inside the model, which is useful for clouds without support for either global or per instance security groups.

Juju lifecycle and harvesting

Juju keeps state on the running model and it can harvest (remove) machines which it deems are no longer required. This can help reduce running costs and keep the model tidy. Harvesting is guided by what "harvesting mode" has been set.

A Juju machine can be in one of four states:

• Alive: The machine is running and being used.
• Dying: The machine is in the process of being terminated by Juju, but hasn't yet finished.
• Dead: The machine has been successfully brought down by Juju, but is still being tracked for removal.
• Unknown: The machine exists, but Juju knows nothing about it.

Juju can be in one of several harvesting modes, in order of most conservative to most aggressive:

• none: Machines will never be harvested. This is a good choice if machines are managed via a process outside of Juju.
• destroyed: Machines will be harvested if i) Juju "knows" about them and ii) they are 'Dead'.
• unknown: Machines will be harvested if Juju does not "know" about them ('Unknown' state). Use with caution in a mixed environment or one which may contain multiple instances of Juju.
• all: Machines will be harvested if Juju considers them to be 'destroyed' or 'unknown'.

The default mode is destroyed.

Below, the harvest mode key for the current model is set to 'none':

juju model-config provisioner-harvest-mode=none

Retrying failed hooks

Prior to version 2.0, hooks returning an error would block until the user ran a command to retry them manually:

juju resolved unit-name/#

From version 2.0, Juju will automatically retry hooks periodically - there is an exponential backoff, so hooks will be retried after 5, 10, 20, 40 seconds up to a period of 5 minutes, and then every 5 minutes. The logic behind this is that some hook errors are caused by timing issues or the temporary unavailability of other applications - automatic retry enables the Juju model to heal itself without troubling the user.

However, in some circumstances, such as debugging charms, this behaviour can be distracting and unwelcome. For this reason, it is possible to set the automatically-retry-hooks option to 'false' to disable this behaviour. In this case, users will have to manually retry any hook which fails, using the command above, as with earlier versions of Juju.

Note: Even with the automatic retry enabled, it is still possible to use the juju resolved unit-name/# command to retry manually.

Versions and streams

The agent-stream option selects the versions of Juju which a model can deploy and upgrade to. This defaults to 'released', indicating that only the latest stable versions of Juju should be used, which is the recommended setting.

To run the upcoming stable release (before it has passed the normal QA process) you can set:

agent-stream: proposed

Alternatively, for testing purposes, you can use the latest unstable version of Juju by setting:

agent-stream: devel

The agent-version option selects a specific client version to be used, with some constraints. It is used as a parameter during bootstrap and permits you to tell Juju to bootstrap a new controller using the same major and minor version already in use, but with a different patch number. For example, Juju uses the major.minor.patch numbering scheme, so Juju 2.1.3 means major version 2, minor version 1, and patch version 3. On a system with this release of Juju installed, you can bootstrap a controller on AWS using a different patch release, like this:

juju bootstrap aws aws --agent-version='2.1.2'

You cannot bootstrap a controller on this system using Juju 1.x, Juju 2.2, and so on. Only different patch numbers may be used with agent-version.