Juju can manage the entire life cycle of an application. When deploying a Charm, Juju will start and provision machines. Likewise, when expanding capacity for applications, Juju can automatically utilize your provider's elastic computing capabilities and add more machines with the applications provisioned.
Should you ever need to destroy machines or applications, Juju can help with that too. Juju keeps a model of what it thinks the environment looks like, and based on that model, can harvest machines which it deems are no longer required. This can help keep your costs low, and keep you out of web consoles.
So how does harvesting in Juju work?
To discuss how Juju manages machines, it's important to first understand how Juju perceives machines in the environment.
Juju slots machines into 4 different states:
The machine is up and being utilized.
The machine is in the process of being terminated by Juju, but hasn't yet finished.
The machine has been successfully brought down by Juju, but is still being tracked for removal.
The machine is alive, but Juju knows nothing about it.
Using this method, Juju won't harvest any machines. This is the most conservative, and a good choice if you manage your machines utilizing a separate process outside of Juju.
This is Juju's default. Using this methodology, Juju will harvest only machine instances that are dead, and that Juju knows about. Unknown instances will not be harvested.
Utilizing this method, Juju will harvest only instances that Juju doesn't know about.
This is the most aggressive setting. Utilizing this, Juju will terminate all instances – destroyed or unknown – that it finds. This is a good option if you are only utilizing Juju for your environment.
Juju's harvesting behaviour is set through the environments.yaml file. To change the default behaviour, edit this file and set the "provisioner-harvest-mode" key to the desired harvesting mode.