Realtime GraphQL Subscriptions

In this section, you’ll learn how you can bring realtime functionality into your app by implementing GraphQL subscriptions. The goal is to implement two subscriptions to be exposed by your GraphQL server:

  • Send realtime updates to subscribed clients when a new Link element is created
  • Send realtime updates to subscribed clients when an existing Link element is upvoted

What are GraphQL subscriptions?

Subscriptions are a GraphQL feature that allows a server to send data to its clients when a specific event happens. Subscriptions are usually implemented with WebSockets. In that setup, the server maintains a steady connection to its subscribed client. This also breaks the “Request-Response-Cycle” that were used for all previous interactions with the API.

Instead, the client initially opens up a long-lived connection to the server by sending a subscription query that specifies which event it is interested in. Every time this particular event happens, the server uses the connection to push the event data to the subscribed client(s).

Implementing GraphQL subscriptions

We will be using PubSub from the apollo-server library that we have already been using for our GraphQL server to implement subscriptions to the following events:

  • A new model is created
  • An existing model updated
  • An existing model is deleted

You will do this by first adding an instance of PubSub to the context, just as we did with PrismaClient, and then calling its methods in the resolvers that handle each of the above events.

Setting up PubSub

Here, you’re creating an instance of PubSub and storing it in the variable pubsub, just as we stored an instance of PrismaClient in the variable prisma.

Great! Now we can access the methods we need to implement our subscriptions from inside our resolvers via context.pubsub!

Subscribing to new Link elements

Alright – let’s go ahead and implement the subscription that allows your clients to subscribe to newly created Link elements.

Just like with queries and mutations, the first step to implement a subscription is to extend your GraphQL schema definition.

Next, go ahead and implement the resolver for the newLink field. Resolvers for subscriptions are slightly different than the ones for queries and mutations:

  1. Rather than returning any data directly, they return an AsyncIterator which subsequently is used by the GraphQL server to push the event data to the client.
  2. Subscription resolvers are wrapped inside an object and need to be provided as the value for a subscribe field. You also need to provide another field called resolve that actually returns the data from the data emitted by the AsyncIterator.

The code seems pretty straightforward. As mentioned before, the subscription resolver is provided as the value for a subscribe field inside a plain JavaScript object.

Now you can see how we access pubsub on the context and call asyncIterator() passing the string "NEW_LINK" into it. This function is used to resolve subscriptions and push the event data.

Adding subscriptions to your resolvers

The last thing we need to do for our subscription implementation itself is to actually call this function inside of a resolver!

Now you can see how we pass the same string to the publish method as you added in your newLinkSubscribe function just above, along with passing in the newLink as a second argument!

Ok, I’m sure you’re dying to test out your brand-spanking new Subscription! All we need to do now is make sure your GraphQL server knows about your changes.

Testing subscriptions

With all the code in place, it’s time to test your realtime API ⚡️ You can do so by using two instances (i.e. tabs or windows) of the GraphQL Playground at once.

As you might guess, you’ll use one Playground to send a subscription query and thereby create a permanent websocket connection to the server. The second one will be used to send a post mutation which triggers the subscription.

In contrast to what happens when sending queries and mutations, you’ll not immediately see the result of the operation. Instead, there’s a loading spinner indicating that it’s waiting for an event to happen.

loading spinner

Time to trigger a subscription event.

Now observe the Playground where the subscription was running:

subscription running

Adding a voting feature

Implementing a vote mutation

The next feature to be added is a voting feature which lets users upvote certain links. The very first step here is to extend your Prisma data model to represent votes in the database.

As you can see, you added a new Vote type to the data model. It has one-to-many relationships to the User and the Link type.

To apply the changes and update your Prisma Client API so it exposes CRUD queries for the new Vote model, regenerate PrismaClient.

Now, with the process of schema-driven development in mind, go ahead and extend the schema definition of your application schema so that your GraphQL server also exposes a vote mutation:

.../hackernews-node/src/schema.graphql
type Mutation {
  post(url: String!, description: String!): Link!
  signup(email: String!, password: String!, name: String!): AuthPayload
  login(email: String!, password: String!): AuthPayload
  vote(linkId: ID!): Vote
}

It should also be possible to query all the votes from a Link, so you need to adjust the Link type in schema.graphql as well.

You know what’s next: Implementing the corresponding resolver functions.

Here is what’s going on:

  1. Similar to what you’re doing in the post resolver, the first step is to validate the incoming JWT with the getUserId helper function. If it’s valid, the function will return the userId of the User who is making the request. If the JWT is not valid, the function will throw an exception.
  2. To protect against those pesky “double voters” (or honest folks who accidentally click twice), you need to check if the vote already exists or not. First, you try to fetch a vote with the same linkId and userId. If the vote exists, it will be stored in the vote variable, resulting in the boolean true from your call to Boolean(vote) — throwing an error kindly telling the user that they already voted.
  3. If that Boolean(vote) call returns false, the vote.create method will be used to create a new Vote that’s connected to the User and the Link.

You also need to account for the new relations in your GraphQL schema:

  • votes on Link
  • user on Vote
  • link on Vote

Similar to before, you need to implement resolvers for these.

Finally you need to resolve the relations from the Vote type.

Great job!

Finally the Vote resolver needs to be included in the main resolvers object in index.js.

Your GraphQL API now accepts vote mutations! 👏

Subscribing to new votes

The last task in this chapter is to add a subscription that fires when new Votes are being created. You’ll use an analogous approach as for the newLink query for that.

Next, you need to add the subscription resolver function.

All right, that’s it! You can now test the implementation of your newVote subscription!

You can use the following subscription for that:

subscription {
  newVote {
    id
    link {
      url
      description
    }
    user {
      name
      email
    }
  }
}

If you’re unsure about writing one yourself, here’s a sample vote mutation you can use. You’ll need to replace the __LINK_ID__ placeholder with the id of an actual Link from your database. Also, make sure that you’re authenticated when sending the mutation.

mutation {
  vote(linkId: "__LINK_ID__") {
    link {
      url
      description
    }
    user {
      name
      email
    }
  }
}

authenticated when sending the mutation

Unlock the next chapter
Which of the following statements is true?
Subscriptions follow a request-response-cycle
Subscriptions are best implemented with MailChimp
Subscriptions are typically implemented via WebSockets
Subscriptions are defined on the 'Query' type and annotated with the @realtime-directive