(aero is edn really, ok?)
A small library for explicit, intentful configuration.
Add the following dependency to your project.clj
file
Create a file called config.edn
containing the following
{:greeting "World!"}
In your code, read the configuration like this
(require '[aero.core :refer [read-config]])
(read-config "config.edn")
or to read from the classpath, like this
(read-config (clojure.java.io/resource "config.edn"))
Keep in mind that even though (read-config "config.edn")
will work in your Repl and when running tests, it's very likely to catastrophically fail if you run your application from the generated .jar
file.
So to avoid surprises it's better to always use io/resource
which works in all scenarios.
Configuration should be explicit, intentful, obvious, but not clever. It should be easy to understand what the config is, and where it is declared.
Determining config in stressful situations, for example, while diagnosing the cause of a production issue, should not be a wild goose chase.
Config files are often duplicated on a per-environment basis, attracting all the problems associated with duplication.
When looking at a config file, a reader will usually ask: "Does the value differ from the default, and if so how?". It's clearly better to answer that question in-place.
When config is left out of source code control it festers and diverges from the code base. Better to keep a single config file in source code control.
While it is good to keep config in source code control, it is important to ensure passwords and other sensitive information remain hidden.
While it can be very flexible to have 'clever' configuration 'programs', it can be unsafe, lead to exploits and compromise security. Configuration is a key input to a program. Always use data for configuration and avoid turing-complete languages!
We suggest using environment variables judiciously and sparingly, the way Unix intends, and not go mad. After all, we want to keep configuration explicit and intentional.
Also, see these arguments against.
Fortunately for Clojure developers like us, most of the tech to read configuration in a safe, secure and extensible way already exists in the Clojure core library (EDN).
Aero provides a small library of tag literals.
Use #env
to reference an environment variable.
{:database-uri #env DATABASE_URI}
It is considered bad practice to use environment variables for passwords and other confidential information. This is because it is very easy to leak a process's environment (e.g. via ps e -f
or to your application monitoring tool). Instead you should use #include
- see here.
Use #envf
to insert environment variables into a formatted string.
{:database #envf ["protocol://%s:%s" DATABASE_HOST DATABASE_NAME]}
Use #or
when you want to provide a list of possibilities, perhaps with a default at the end.
{:port #or [#env PORT 8080]}
#join
is used as a string builder, useful in a variety of situations such as building up connection strings.
{:url #join ["jdbc:postgresql://psq-prod/prod?user="
#env PROD_USER
"&password="
#env PROD_PASSWD]}
Use profile as a kind of reader conditional.
#profile
expects a map, from which it extracts the entry corresponding to the profile.
{:webserver
{:port #profile {:default 8000
:dev 8001
:test 8002}}}
You can specify the value of profile when you read the config.
(read-config "config.edn" {:profile :dev})
which will return
{:webserver
{:port 8001}}
(#profile
replaces the now deprecated #cond
, found in previous versions of Aero)
Use when config has to differ from host to host, using the hostname. You can specify multiple hostnames in a set.
{:webserver
{:port #hostname {"stone" 8080
#{"emerald" "diamond"} 8081
:default 8082}}}
Use to parse a String
value into a Long
, Double
, keyword or boolean.
{:debug #boolean #or [#env DEBUG "true"]
:webserver
{:port #long #or [#env PORT 8080]
:factor #double #env FACTOR
:mode #keyword #env MODE}}
#user
is like #hostname
, but switches on the user.
Use to include another config file. This allows you to split your config files to prevent them from getting too large.
{:webserver #include "webserver.edn"
:analytics #include "analytics.edn"}
NOTE: By default #include
will attempt to resolve the file to be included relative to the
config file it's being included from. (this won't work for jars)
You can provide your own custom resolver to replace the default behaviour or use one that
aero provides (resource-resolver
, root-resolver
). For example
(require '[aero.core :refer (read-config resource-resolver)])
(read-config "config.edn" {:resolver resource-resolver})
You can also provide a map as a resolver. For example
(read-config "config.edn" {:resolver {"webserver.edn" "resources/webserver/config.edn"}})
Merge multiple maps together
#merge [{:foo :bar} {:foo :zip}]
To avoid duplication you can refer to other parts of your configuration file using the #ref
tag.
The #ref
value should be a vector resolveable by get-in
. Take the following config map for example:
{:db-connection "datomic:dynamo://dynamodb"
:webserver
{:db #ref [:db-connection]}
:analytics
{:db #ref [:db-connection]}}
Both :analytics
and :webserver
will have their :db
keys resolved
to "datomic:dynamo://dynamodb"
References are recursive. They can be used in #include
files.
Aero supports user-defined tag literals. Just extend the reader
multimethod.
(defmethod reader 'mytag
[{:keys [profile] :as opts} tag value]
(if (= value :favorite)
:chocolate
:vanilla))
Passwords and other confidential information should not be stored in version control, nor be specified in environment variables. One alternative option is to create a private file in the HOME directory that contains only the information that must be kept outside version control (it is good advice that everything else be subject to configuration management via version control).
Here is how this can be achieved:
{:secrets #include #join [#env HOME "/.secrets.edn"]
:aws-secret-access-key
#profile {:test #ref [:secrets :aws-test-key]
:prod #ref [:secrets :aws-prod-key]}}
Here's some good advice on using Aero in your own programs.
Define a dedicated namespace for config that reads the config and provides functions to access it.
(ns myproj.config
(:require [aero.core :as aero]))
(defn config [profile]
(aero/read-config "dev/config.edn" {:profile profile}))
(defn webserver-port [config]
(get-in config [:webserver :port]))
This way, you build a simple layer of indirection to insulate the parts of your program that access configuration from the evolving structure of the configuration file. If your configuration structure changes, you only have to change the wrappers, rather than locate and update all the places in your code where configuration is accessed.
Your program should call the config
function, usually with an argument specifying the configuration profile. It then returned value passes the returned value through functions or via lexical scope (possibly components).
Aero has frictionless integration with Plumatic Schema. If you wish, specify your configuration schemas and run check
or validate
against the data returned from read-config
.
If you are using Stuart Sierra's component library, here's how you might integrate Aero.
(ns myproj.server
(:require [myproj.config :as config]))
(defrecord MyServer [config]
Lifecycle
(start [component]
(assoc component :server (start-server :port (config/webserver-port config))))
(stop [component]
(when-let [server (:server component)] (stop-server server))))
(defn new-server [config]
(->MyServer config))
(ns myproj.system
[com.stuartsierra.component :as component]
[myproj.server :refer [new-server]])
(defn new-production-system []
(let [config (config/config :prod)]
(system-using
(component/system-map :server (new-server config))
{})))
However, another useful pattern you might consider is to keep your system map and configuration map aligned.
For example, imagine you have a config file:
{:listener {:port 8080}
:database {:uri "datomic:mem://myapp/dev"}}
Here we create a system as normal but with the key difference that we configure the system map after we have created using merge-with merge
. This avoids all the boilerplate required in passing config around the various component constructors.
(defrecord Listener [database port]
Lifecycle …)
(defn new-listener []
(using (map->Listener {}) [:database])
(defrecord Database [uri]
Lifecycle …)
(defn new-database []
(map->Database {}))
(defn new-system-map
"Create a configuration-free system"
[]
(system-map
:listener (new-listener)
:database (new-database)))
(defn configure [system profile]
(let [config (aero/read-config "config.edn" {:profile profile})]
(merge-with merge system config)))
(defn new-dependency-map [] {})
(defn new-system
"Create the production system"
[profile]
(-> (new-system-map)
(configure profile)
(system-using (new-dependency-map))))
Also, if you follow the pattern described here you can also ensure accurate configuration is given to each component without having to maintain explicit schemas. This way, you only verify the config that you are actually using.
Aero is a great way to implement feature toggles.
If at all possible, try to avoid having lots of configuration files and stick with a single file. That way, you're encouraged to keep configuration down to a minimum. Having a single file is also useful because it can be more easily edited, published, emailed, watched for changes. It is generally better to surface complexity than hide it away.
aero.alpha.core
defines a new experimental API for tagged literals.
This API allows you to define tagged literal "macros" similar to macros in Clojure.
It is intended for use in creating your own conditional constructs like #profile
and #or
.
The easiest kind of tagged literal to create is a case-like one.
A case-like tagged literal is one which takes a map of possible paths to take.
An example of this in Aero is #profile
.
Here's how you can define your own version of #profile
:
(ns myns
(:require [aero.alpha.core :as aero.alpha]))
(defmethod aero.alpha/eval-tagged-literal 'profile
[tagged-literal opts env ks]
(aero.alpha/expand-case (:profile opts) tagged-literal opts env ks))
eval-tagged-literal
allows you to define macro tagged literals.
expand-case
is a function which forms the common behaviour beneath #user
, #profile
, etc.
#or
is very different from #profile
in implementation, and doesn't have a convenience function.
The source for #or
in aero.core
is a good example of doing custom partial expansion from a tagged literal.
The primitives you will need to understand are: aero.alpha.core/expand
, aero.alpha.core/expand-coll
, aero.alpha.core/expand-scalar
.
And helpers: aero.alpha.core/expand-scalar-repeatedly
.
These vars have docstrings which explain their specific purpose.
All expand-* functions take parameters opts
, env
, and ks
.
opts
are the same opts
that are passed to aero.core/read-config
.
env
is a map of ks
to their resolved values in the config, being absent from this map means the value is not yet resolved.
ks
is a vector representing the current key path into the location of this tagged literal.
Your implementation of eval-tagged-literal must assoc
the ks
into env
if it is successfully resolved.
Aero is built on Clojure's edn.
Aero is influenced by nomad, but purposely avoids instance, environment and private config.
Thanks to the following people for inspiration, contributions, feedback and suggestions.
The MIT License (MIT)
Copyright © 2015 JUXT LTD.
Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated documentation files (the "Software"), to deal in the Software without restriction, including without limitation the rights to use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of the Software, and to permit persons to whom the Software is furnished to do so, subject to the following conditions:
The above copyright notice and this permission notice shall be included in all copies or substantial portions of the Software.
THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.