Jesse Bellingham
Jesse's Blog

Follow

Jesse's Blog

Follow

This Week I Learned: Monitoring and Observability with Prometheus, Grafana, and fly.io

Jesse Bellingham's photo
Jesse Bellingham
·Mar 31, 2022·

5 min read

This Week I Learned: Monitoring and Observability with Prometheus, Grafana, and fly.io

Photo by Luke Chesser on Unsplash

This Week I Learned

How to instrument an Elixir/Phoenix app with Prometheus and Grafana, hosted in fly.io.

Build it in production

Recently I started building a new side project with Elixir and Phoenix, and I had one thing in my mind from a recent changelog episode, "build it in production". So I'm doing just that. I've decided to host the app with fly.io, and so far I am very happy with that choice.

Something that I've developed a keen interest for in the last year or so is monitoring and observability, and so I knew right away that I wanted to instrument this app as soon as possible. There are a bunch of options out there for observability stacks, but as this is a side project, cost is a big factor when considering those options. I ended up going with Prometheus and Grafana Cloud, because fly.io have a hosted Prometheus instance that I can easily publish to with almost no configuration, and Grafana Cloud because the free tier is free forever and pretty generous, and it plays nicely with my choice of instrumentation library.

How do observability?

First thing we need to have is a way to publish metrics from our running (or soon to be running) app. PromEx is an Elixir library that makes it very easy to publish metrics on all kinds of things relating to an app's operation. We can add it to our list of dependencies in mix.exs:

defp deps do
  [
      ...other deps
      {:prom_ex, "~> 1.7.1"},
      ...other deps
  ]
end

With the dependency added, run

mix deps.get

to fetch the dependency and make it available to our project. We can then run

mix prom_ex.gen.config --datasource prometheus

which will produce a new file lib/app_name/prom_ex.ex. Follow the documentation provided at the top of this file to make the other necessary code changes. PromEx comes pre-configured to surface metrics with:

def plugins do
    [
      # PromEx built in plugins
      Plugins.Application,
      Plugins.Beam,
      {Plugins.Phoenix, router: SocialNetworkWeb.Router, endpoint: SocialNetworkWeb.Endpoint},
      Plugins.Ecto,
      Plugins.PhoenixLiveView
    ]
  end

Enabling metrics from things like ecto, phoenix, and live view is as easy as un-commenting those lines, and the corresponding lines for their dashboards under

def dashboards do
    [
      # PromEx built in Grafana dashboards
      {:prom_ex, "application.json"},
      {:prom_ex, "beam.json"},
      {:prom_ex, "phoenix.json"},
      {:prom_ex, "ecto.json"},
      {:prom_ex, "phoenix_live_view.json"}
    ]

There is one other piece of configuration I needed, that isn't present in the current documentation. In the application config, when you are configuring PromEx in your application pipeline, to also specify the metrics_server attribute:

config :social_network, SocialNetwork.PromEx,
    manual_metrics_start_delay: :no_delay,
    grafana: [
      host: System.get_env("GRAFANA_HOST") || raise("GRAFANA_HOST is required"),
      auth_token: System.get_env("GRAFANA_TOKEN") || raise("GRAFANA_TOKEN is required"),
      upload_dashboards_on_start: true,
      folder_name: "Social Network App Dashboards",
      annotate_app_lifecycle: true
    ],
    metrics_server: [
      port: 4021,
      path: "/metrics", # This is an optional setting and will default to `"/metrics"`
      protocol: :http, # This is an optional setting and will default to `:http`
      pool_size: 5, # This is an optional setting and will default to `5`
      cowboy_opts: [], # This is an optional setting and will default to `[]`
      auth_strategy: :none # This is an optional and will default to `:none`
    ]

Discovered thanks to this blog post, pointing to this piece of documentation. Note that the PromEx config also needs a couple of environment variables to wire it up to grafana to push the dashboards. We don't have the values for these yet so we'll come back to it.

As I mentioned above, fly.io have a hosted Prometheus instance we can publish to with a few lines of config in our fly.toml:

...
[metrics]
port = 4021
path = "/metrics"
...

Prometheus will now poll for metrics on the above endpoint, and with that, we've made all the code changes we need.

In order to publish our dashboards to grafana, we need to create an API token for PromEx to use. In the grafana UI, navigate to /org/apikeys, and click "Add API key". Give it a name, I've called mine "PromEx", and give it the role "Editor". Specify a time to live for the API key, and click "Add". Make a note of the key that gets generated, as you won't be able to see it again and will have to generate a new one if you lose it. Using the fly.io command line tool, you'll need to add secrets to your app for the grafana host and api token, something like this:

fly secrets set GRAFANA_HOST=<your grafana instance url here>
fly secrets set GRAFANA_TOKEN=<your grafana api token here>

We also need to generate an fly api token for grafana to use, which we can do with:

flyctl auth token

Make note of the token produced by this command. Same as with the last API key, you won't get to see it again.


Pulling data into Grafana

In order for grafana to consume our metrics from prometheus, we need to configure a data source for it. In the grafana UI, navigate to /datasources, click "Add data source", and select Prometheus from the list. Give it the name "prometheus", specifically in all lower case, this property needs to match exactly the datasource_id from prom_ex.ex. Configure the rest of the data source as follows:

image.png

Replacing < ORG SLUG > with your fly.io org slug which can be found by running:

flyctl orgs list

We also need to add a custom HTTP header with the key "Authorization", and the value "Bearer: < FLY API TOKEN >". Click "Save & test" we should see something like this:

image.png

Getting the "Data source is working" message above means we're ready to deploy our fly app and start generating some data. After a successful deploy, if we head over to /dashboards on the grafana UI we should see a folder called "Social Network App Dashboards":

image.png

Expand the folder and we should see dashboards for all of the options we specified in prom_ex.ex: image.png

And if we open one up, we should see something like this:

image.png

If it looks like the above, then everything is working correctly 🎉🎉

If you see a whole lot of "No data" across all the dashboards, even after hitting the app with some traffic, something isn't configured quite right. Double check your data source in grafana, and that clicking "Save & test" produces the "Data source is working". Also double check your app secrets in fly, and the PromEx configuration in the app itself.

 
Share this