04: Environment Configuration

ℹ️
You can view all source for examples on GitHub

Most applications need to read configuration values from their environment. Such configuration typically includes connection parameters for databases and API services. What’s more, configuration values vary per environment (local, qa, prod, etc), so you need some way of specifying what environment your system is running in. Below is an example of how to handle this with donut.system:

dev/donut/examples/tutorial/04_environment_configuration.clj
 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
(ns donut.examples.tutorial.04-environment-configuration
  (:require
   [aero.core :as aero]
   [clojure.java.io :as io]
   [donut.system :as ds]))

(defn env-config [& [profile]]
  (aero/read-config (io/resource "config/env.edn")
                    (when profile {:profile profile})))

(def DataStoreComponent
  {::ds/start (fn [_] (atom nil))})

(def APIPollerComponent
  {::ds/start  (fn [{:keys [::ds/config]}]
                 (let [{:keys [data-store source]} config]
                   (future (loop [i 0]
                             (println (str "polling " source))
                             (reset! data-store i)
                             (Thread/sleep (:interval config))
                             (recur (inc i))))))
   ::ds/stop   (fn [{:keys [::ds/instance]}]
                 (future-cancel instance))
   ::ds/config {:interval   (ds/ref [:env :api-poller :interval])
                :source     (ds/ref [:env :api-poller :source])
                :data-store (ds/ref [:services :data-store])}})

(def base-system
  {::ds/defs
   {:env      {}
    :services {:api-poller APIPollerComponent
               :data-store DataStoreComponent}}})

(defmethod ds/named-system :base
  [_]
  base-system)

(defmethod ds/named-system :dev
  [_]
  (ds/system :base {[:env] (env-config :dev)}))


(defmethod ds/named-system :prod
  [_]
  (ds/system :base {[:env] (env-config :prod)}))

The strategy here is:

  1. Use another library, aero, to transform config files into Clojure data structures for your application. aero is EDN, but with some enhancements, including a little syntax sugar for incorporating environment variables and for producing different values based on the :profile you pass in. env-config uses aero on line 8; see the aero docs for more info.
  2. Create named systems that introduce per-environment configuration by modifying a base system. The :dev and :prod named systems use the env-config function to read environment-specific values and place them in the :env component group.
  3. Create components that pull their environment configuration from the :env component group. APIPollerComponent has refs for [:env :api-poller :interval] and [:env :api-poller :source] on lines 24 and 25. These are deep refs and I cover them below.

To fully understand this strategy, we need to understand:

  • Constant instances
  • Deep refs
  • Named systems

Constant instances

If a component definition is anything other than a map that includes signal handler keys, then it’s treated as a “constant instance.” Observe:

(ns donut.examples.tutorial.04-constant-instance
  (:require
   [donut.system :as ds]))

(def system
  {::ds/defs
   {:env      {:db-conn-string "//localhost:5032etcetc"}
    :services {:db {::ds/start  (fn [{:keys [::ds/config]}]
                                  (prn "db-conn-string" (:db-conn-string config)))
                    ::ds/config {:db-conn-string (ds/ref [:env :db-conn-string])}}}}})

(ds/start system)

If you evaluate all of this at your REPL, you’ll see the following get printed:

"db-conn-string" "//localhost:5032etcetc"

What’s happening here is the [:services :db] component is referencing [:env :db-conn-string]. Recall that references are our means of conveying component instances into the signal handlers of another component.

That means that the component instance for [:env :db-conn-string] is the string "//localhost:5032etcetc". However, the component definition for [:env :db-conn-string] is not a map with signal handlers like we’ve seen so far; it’s the string "//localhost:5032etcetc".

This violates our understand of how component definitions work. So far, learned that component definitions are maps of signal handlers, and that signal handler return values become component instances. You would expect that for this to work, you would have to use the following component definition for [:env :db-conn-string]:

{::ds/start (constantly "//localhost:5032etcetc")}

However, because it’s such a common use case to want to include such constant instances, donut.system was designed to support you in including the value directly, rather than having to wrap it in a signal handler.

In practice, this means that components and component groups can essentially be paths in your system that house configuration. The full system at the top of this page takes this approach: it uses aero to read a config file, generating a Clojure map in the process. That Clojure map then gets placed under the :env component group. Other components can then access that configuration via refs.

Deep refs

APIPollerComponent at the top of the page includes these two refs:

(ds/ref [:env :api-poller :interval])
(ds/ref [:env :api-poller :source])

So far, we’ve only seen refs that take two-element vectors of the form [component-group-name component-name]. But in the code, the vectors have three elements.

Refs can actually take any number of elements; you can think of the vector you pass in as being used to perform a get-in on your system’s instances:

(get-in (::ds/instances system) [:env :api-poller :interval])

If a component instance is a deeply-nested map, you can use refs to refer to any path within that map. This also works with vectors, as you can use get-in on vectors:

(get-in [[:a :b] [:c :d]] [1 0])
;; =>
:c

Named systems

The multimethod ds/named-system serves as a system definition registry. We can then use the function ds/system to retrieve a system definition and optionally override component definitions. We see this in the example at the top of the page:

dev/donut/examples/tutorial/04_environment_configuration.clj
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
(def base-system
  {::ds/defs
   {:env      {}
    :services {:api-poller APIPollerComponent
               :data-store DataStoreComponent}}})

(defmethod ds/named-system :base
  [_]
  base-system)

(defmethod ds/named-system :dev
  [_]
  (ds/system :base {[:env] (env-config :dev)}))


(defmethod ds/named-system :prod
  [_]
  (ds/system :base {[:env] (env-config :prod)}))

We register a system definition named :base, then we build on that system in the :dev and :prod named systems, overriding the :env component groups with environment-specific configurations produced by the env-config function.

Together, all of these pieces allow you to define the main structure of your system – the components that produce the behavior your care about – while giving you the flexibility to configure these components for different environments.

Where do you actually put all this? My recommendation is to create a your-project.system namespace and put your system definitions there.

And with that, you now have all the basics you need for effectively using donut.system in a real project! Woo!

Summary

  • Any component definition that isn’t a map of signal handlers is treated as a constant instance
  • You can use ds/named-system to register different system definitions
  • Refs can actually refer to any part of a system’s ::ds/instances that’s reachable via get-in