High Availability

Anbox Cloud comes with support for High Availability (HA) for both Core and the Streaming Stack.
Next to Juju’s support for high availability of the Juju controller, you can add HA for the Anbox Management Service (AMS) and the Anbox Stream Gateway to ensure fault tolerance and higher uptime.

Enabling High Availability (HA for short) is achieved by adding new units via juju.
This will allocate a new machine, run new instances of the scaled application and configure the cluster automatically.

Adding a unit is done with the following syntax:

$ juju add-unit <application name> -n <number of units to add>

For example, to go from 1 to 5 ams units, you would run the following:

$ juju add-unit ams -n 4

Hint: By default Juju allocates small machines to limit costs, but you can request better resources by enforcing constraints:

$ juju set-constraints anbox-stream-gateway cores=4 memory=8GB.

This is heavily recommended on production environments.

Anbox Cloud Core

Anbox Cloud Core HA requires additional AMS instances as well as a load balancer to spread out requests:

$ juju deploy cs:~anbox-charmers/ams-load-balancer
$ juju relate ams ams-load-balancer
$ juju add-unit ams -n 2

Note: If you are using the amc snap on your machine, you can tell it to use the load balancer instead of talking directly to ams:

$ amc remote add lb
$ amc remote set-default lb

The port to use is always 8444, the same AMS is listening on.

Anbox Streaming Stack

Warning: If you are upgrading from 1.4 or earlier, check the upgrade instructions

In the Streaming Stack, both the Agent and the Gateway can be run in HA.

$ juju add-unit anbox-stream-gateway -n 2
$ juju add-unit anbox-stream-agent -n 2
$ juju relate anbox-stream-gateway:api anbox-stream-gateway-lb:reverseproxy

This would give you 3 instances of both the Stream Gateway and the Stream Agent.

Note: We recommend a minimum of 3 machines for the Streaming Stack.

Checking status

When adding new units, Juju will create new machine so it may take a few minutes for
your cluster to be fully operational.
You can check juju status to see the current deployment status:

Model    Controller      Cloud/Region         Version  SLA          Timestamp
default  anbox-cloud     localhost/localhost  2.8.0    unsupported  19:18:10Z

App                      Version  Status   Scale  Charm                 Store       Rev  OS      Notes

anbox-stream-agent                active       3  anbox-stream-agent    jujucharms   80  ubuntu
anbox-stream-gateway              active       3  anbox-stream-gateway  jujucharms   90  ubuntu
anbox-stream-gateway-lb           active       1  haproxy               jujucharms   56  ubuntu  exposed

Unit                        Workload  Agent  Machine  Public address  Ports               Message
anbox-stream-agent/0*       active    idle   0
anbox-stream-agent/1        active    idle   6
anbox-stream-agent/2        active    idle   5
anbox-stream-gateway-lb/0*  active    idle   2  80/tcp,443/tcp      Unit is ready
anbox-stream-gateway/0*     active    idle   1  4000/tcp,7003/tcp
anbox-stream-gateway/1      active    idle   3  4000/tcp,7004/tcp
anbox-stream-gateway/2      active    idle   4  4000/tcp,7005/tcp

Notice the scale of each application indicating how much units an application has.

Scaling down

Scaling down can be done by removing units via Juju.
Here you have to specifically target the unit you want to remove:

$ juju remove-unit anbox-stream-agent/2

The cluster will reconfigure itself to work with the removed unit.