Learn how to quickly get started with Artemis.
Artemis is available on Clojars.
Include it in your project.clj
or deps.edn
dependencies.
Once you've installed Artemis, you'll need a GraphQL server to make use of it. We've set-up a publicly accessible Graphcool server that you can use if you'd like to follow along verbatim. You can, of course, also use your own server; the ideas will transfer over.
In your ClojureSript project, start off by creating a network-chain, store, and client. We'll cover each of these in more depth in later sections, including how to customize each, but for now you can use the built-in versions.
(require '[artemis.core :as a]
'[artemis.network-steps.http :as http]
'[artemis.stores.mapgraph.core :as mgs])
(def graphcool-url "https://api.graph.cool/simple/v1/cjjh9nmy118fs0127i5t71oxe")
(def network-chain (http/create-network-step graphcool-url))
(def store (mgs/create-store))
(def client (a/create-client :network-chain network-chain
:store store))
You can create a client without specifying either store or network chain. By
default, it'll use the same http-based network chain, but with /graphql
set
as the server endpoint. It'll also use the same Mapgraph-based store (we'll
touch on Mapgraph later).
You're now ready to start executing GraphQL queries. To fire off your first
query, create a GraphQL document and call query!
.
(require '[artemis.document :refer [parse-document]])
(def get-jerry
(parse-document
"query getJerry {
User(id: \"cjjh9x0q97x7r0111osr3t352\") {
id
name
}
}"))
(a/query! client get-jerry :fetch-policy :remote-only)
You might have noticed that we're passing a :fetch-policy
option. Don't
worry about that for now, it'll be covered in a later topic.
If you inspect network traffic using your browser's devtools, you should see a request to the GraphQL server and a response that looks something like this:
{
"data": {
"User": {
"id": "cjjh9x0q97x7r0111osr3t352",
"name": "Jerry Seinfeld"
}
}
}
At this point, you probably want to access that data, and that brings us to
the main design pattern that Artemis is based around: using channels to manage
the asynchronous nature of querying a GraphQL source. Every call to query!
will return a channel that will receive messages it with information about your
GraphQL query.
Let's bring in core.async, then use
go
to take off our channel.
(require-macros '[cljs.core.async.macros :refer [go go-loop]])
(require '[cljs.core.async :refer [<!]])
(go (let [x (<! (a/query! client get-jerry :fetch-policy :remote-only))]
(.log js/console x)))
You should see the following logged to your browser's console:
{:data nil,
:variables nil,
:in-flight? true,
:network-status :fetching}
Hmm, notice that data is nil
. That can't be right -- we saw that data come
through on our network tab. The reason the data appears as nil
is because our
channel actually receives multiple messages throughout the lifetime of the
query, each message providing information about what state the query is
currently in. To read multiple messages, let's use go-loop
instead.
(let [get-jerry-chan (a/query! client get-jerry :fetch-policy :remote-only)]
(go-loop []
(when-let [x (<! get-jerry-chan)]
(.log js/console x)
(recur))))
You should now see two log statemens, the first which is identical to what we previously saw, and the second which is something like:
{:in-flight? false,
:variables nil,
:network-status :ready,
:data {:User {:id "cjjh9x0q97x7r0111osr3t352",
:name "Jerry Seinfeld"}}}
If we inspect each of the messages, it should be apparent what each tells us about the state of our query at the time the message was put onto the channel.
The first message lets us know that we don't have any data available, and
that's because our network status is currently :fetching
, meaning we're
getting the data. You'll also notice that :in-flight?
is true
for the first
message, so we know our query is currently over the wire. The second message,
of course, shows different values for these keys.
We can use this kind of information per-message to easily render loading indicators, delay transitions, or make other query state-based decisions.
Channels, it turns out, are a nice, easy to understand way to track ordered updates on the state of our data fetching and manipulation. This simple design becomes especially useful as our application grows and we're executing multiple queries and potentially even doing so across multiple sources, such as a local cache and a server.
You might notice that we're using recur
within our go-loop
. We can safely
do this without ending up in an infinte loop because Artemis automatically
closes the channels returned by its core functions at the right time.
Now that we've learned how to query for data with Artemis, let's take a look at updating data.
Let's start off by creating a mutation document:
(def create-newman
(parse-document
"mutation createNewman {
createUser(name: \"Newman\") {
id
name
}
}"))
Now, just like with query!
, we pass our document to the mutate!
. Like its
counterpart function, mutate!
also returns a channel, so we'll want to take
messages off it.
(let [create-newman-chan (a/mutate! client create-newman)]
(go-loop []
(when-let [x (<! create-newman-chan)]
(.log js/console x)
(recur))))
Because GraphQL mutations also include the desired return data, the messages
that Artemis puts on your channel are pretty much the same as those you'll get
when a running a query -- you'll have access to the returned :data
, any
:errors
, as well as meta-data, like :in-flight?
and :network-status
.
That's about all it takes to execute GraphQL mutations with Artemis.
(ns app.core
(:require-macros [cljs.core.async.macros :refer [go go-loop]])
(:require [artemis.core :as a]
[artemis.document :refer [parse-document]]
[artemis.network-steps.http :as http]
[artemis.stores.mapgraph.core :as mgs]))
;;; Set-up
(def graphcool-url "https://api.graph.cool/simple/v1/cjjh9nmy118fs0127i5t71oxe")
(def network-chain (http/create-network-step graphcool-url))
(def store (mgs/create-store))
(def client (a/create-client :network-chain network-chain
:store store))
;;; Query
(def get-jerry
(parse-document
"query getJerry {
User(id: \"cjjh9x0q97x7r0111osr3t352\") {
id
name
}
}"))
(let [get-jerry-chan (a/query! client get-jerry :fetch-policy :remote-only)]
(go-loop []
(when-let [x (<! get-jerry-chan)]
(.log js/console x)
(recur))))
;;; Mutation
(def create-newman
(parse-document
"mutation createNewman {
createUser(name: \"Newman\") {
id
name
}
}"))
(let [create-newman-chan (a/mutate! client create-newman)]
(go-loop []
(when-let [x (<! create-newman-chan)]
(.log js/console x)
(recur))))
Can you improve this documentation?Edit on GitHub
cljdoc is a website building & hosting documentation for Clojure/Script libraries
× close