Rails Health Endpoint Setup

[schmijos]

Quick guidance on how to setup a custom health probe would be nice, because otherwise the uptime monitoring is not continuous. This is measurement cosmetics.

Default behaviour without custom health probe

Reason explained by Nick:

I checked the raw underlying metrics and it seems it happens due to the query. For memory based metrics you don’t need any fancy query functions actually. One basically requests the underlying metric container_memory_working_set_bytes and the reporting of this metric stops as soon as the container/pod was marked as ‘Terminating’. That is why there is no overlapping in the graph.

Fixed behaviour with custom health probe

Change

Make code changes

# app/controllers/health_controller.rb
class HealthController < Rails::HealthController
  def up
    ActiveRecord::Base.connection.execute("SELECT 1") # Test if database is here
    render json: { status: "ok" }, status: :ok
  rescue StandardError => e
    render json: { status: "error", error: e.message }, status: :internal_server_error
  end
end

# config/routes.rb
get "/up" => "health#up"

# config/environments/production.rb
# Opt out of SSL redirection for health endpoint
config.ssl_options = { redirect: { exclude: ->(request) { request.path == "/up" } } }

# spec/requests/health_spec.rb
require "rails_helper"

RSpec.describe "Health" do
  describe "GET /up" do
    context "when the database is healthy" do
      before do
        expect(ActiveRecord::Base.connection).to receive(:execute).and_return(nil)
      end

      it "returns http success" do
        get up_path
        expect(response).to have_http_status(:ok)
        expect(response.parsed_body).to eq({ "status" => "ok" })
      end
    end

    context "when the database is unhealthy" do
      before do
        allow(ActiveRecord::Base.connection).to receive(:execute).and_raise(ActiveRecord::ConnectionNotEstablished)
      end

      it "returns http internal server error" do
        get up_path
        expect(response).to have_http_status(:internal_server_error)
        expect(response.parsed_body).to include({ "status" => "error" })
      end
    end
  end
end

Test

Deploy and visit page: example.com/up

Configure Deploio

Then setup the healthprobe (like described in https://docs.nine.ch/de/docs/deplo-io/configuration/deploio-health-probe/ )

nctl update app APP_NAME --health-probe-path="/up" --health-probe-period-seconds=5

[schmijos]

The redirect exception is not yet ideal, I think.

[schmijos]

It makes sense. As long as the app itself doesn't show a TLS certificate, we can only talk HTTP. We need the redirection exception.

[schmijos]

There is a downside to be mentioned: If you test for database uptime, then your whole app won't be available if the database is down (which might exactly be what you want or not).

[schmijos]

You might see errors like this if the health checkpoint is not properly setup on app side:

2026-06-02T07:38:53.405Z: [ce43450a-25e7-49e1-b231-03add9aa4c8c] Started GET "/" for 10.42.82.23 at 2026-06-02 07:38:53 +0000
2026-06-02T07:38:53.405Z: [ce43450a-25e7-49e1-b231-03add9aa4c8c] Redirected to https://10.42.82.220:8080/blitz
2026-06-02T07:38:53.405Z: [ce43450a-25e7-49e1-b231-03add9aa4c8c] Completed 301 Moved Permanently in 0ms
2026-06-02T07:38:53.407Z: 2026-06-02 07:38:53 +0000 HTTP parse error, malformed request: #

We fix redirection issues in Rails by excluding the health endpoint from redirection:

# config/environments/production.rb

  # Opt out of SSL redirection for health endpoint
  config.ssl_options = { redirect: { exclude: ->(request) { request.path == "/up" } } }