Autoscale Machines

Illustration by Annie Ruygt of two machines dancing and one standing still

You have an app with services that’s configured to automatically start and stop Machines based on traffic demand. But the traffic to your app changes significantly during the day and you don’t want to keep a lot of stopped Machines during the period of low traffic.

This blueprint will guide you through the process of configuring the fly-autoscaler app in conjunction with Fly Proxy autostop/autostart to always keep a fixed number of Machines ready to be quickly started by Fly Proxy.

Configure autostop/autostart

First, if you haven’t already done so, configure the app to allow Fly Proxy to automatically start and stop or suspend Machines based on traffic demand. The autostop/autostart settings apply per service, so you set them within the [[services]] or [http_service] sections of fly.toml:

...
[[services]]
  ...
  auto_stop_machines = "stop"
  auto_start_machines = true
  min_machines_running = 0
...

With these settings Fly Proxy will start an additional Machine if all the running Machines are above their concurrency soft_limit and stop running Machines when the traffic decreases. You can set Machines to "suspend" rather than "stop", for even faster start-up, but with some limitations on the type of Machine.

In the next section you’ll configure and deploy fly-autoscaler to ensure that the app always has a spare stopped Machine for Fly Proxy to start.

Configuring and deploying fly-autoscaler

fly-autoscaler is a metrics-based autoscaler that scales an app’s Machines based on any metric. You can configure it to ensure that there is always additional Machine available for Fly Proxy to start if the traffic increases.

First, create a new Fly.io app that will run the autoscaler.

$ fly apps create my-autoscaler

Create a deploy token so that the autoscaler app has permissions to scale your target app up and down:

$ fly tokens create deploy -a my-target-app
$ fly secrets set -a my-autoscaler --stage FAS_API_TOKEN="FlyV1 ..."

Create a read-only token so that the autoscaler app has access to a Prometheus instance:

$ fly tokens create readonly
$ fly secrets set -a my-autoscaler --stage FAS_PROMETHEUS_TOKEN="FlyV1 ..."

Configure your autoscaler fly.toml like this:

app = "my-autoscaler"

[build]
image = "flyio/fly-autoscaler:0.3.1"

[env]
FAS_PROMETHEUS_ADDRESS = "https://api.fly.io/prometheus/my-org"
FAS_PROMETHEUS_METRIC_NAME = "running_machines"
FAS_PROMETHEUS_QUERY = "count(fly_instance_up{app='$APP_NAME'})"

FAS_APP_NAME = "my-target-app"
FAS_CREATED_MACHINE_COUNT = "min(running_machines + 1, 10)"
FAS_INITIAL_MACHINE_STATE = "stopped"

[metrics]
port = 9090
path = "/metrics"

With this configuration, the autoscaler will create a new stopped Machine as soon as all available Machines are running (but never more than 10), and will destroy extra stopped Machines if more than one Machine is stopped.

Make sure you are using autoscaler version 0.3.1 or newer for FAS_INITIAL_MACHINE_STATE configuration option to work.

And finally, deploy the autoscaler, using the --ha option to deploy only one Machine:

$ fly deploy --ha=false

Read more

Open in ChatGPT
Report an issue or edit this page on GitHub