HTTP callback protocol for GraphQL subscriptions

This reference describes a protocol for GraphQL servers (or subgraphs) to send subscription data to a subscribing graph router (such as the Apollo Router) via HTTP callbacks.

For routers with many simultaneous open subscriptions, this protocol scales better than WebSocket-based protocols, which require long-lasting open connections.

The Apollo Router provides preview support for this protocol as part of its support for federated subscriptions:

Protocol flow

All actions described in these steps are performed by one of Router, Subgraph, or Emitter.

Initialization

1. Before it executes a subscription operation, Router generates a unique ID (version 4 UUID recommended) to represent that operation.

2. Router sends a GraphQL subscription operation to Subgraph via HTTP POST using the standard GraphQL request payload:

   JSON

   1
   {
   2
     "query": "subscription { userWasCreated { name reviews { body } } }",
   3
     "extensions": {
   4
       "subscription": {
   5
         "callback_url": "http://localhost:4000/callback/c4a9d1b8-dc57-44ab-9e5a-6e6189b2b945",
   6
         "subscription_id": "c4a9d1b8-dc57-44ab-9e5a-6e6189b2b945",
   7
         "verifier": "XXX"
   8
       }
   9
     }
   10
   }

   Copy

   The extensions property of this payload includes a subscription object with the following:

   • callback_url: The URL that Emitter will send subscription data to
   • subscription_id: The generated unique ID for the subscription operation
   • verifier: A string that Emitter will include in all HTTP callback requests to verify its identity

3. Before Subgraph responds to Router's request, it sends a check message to the provided callback_url for confirmation from Router:

   JSON

   1
   {
   2
     "kind": "subscription",
   3
     "action": "check",
   4
     "id": "c4a9d1b8-dc57-44ab-9e5a-6e6189b2b945",
   5
     "verifier": "XXX"
   6
   }

   Copy

   All messages sent to callback_url are HTTP POST requests. Message types are documented below.

   This helps ensure that Subgraph is able to send callbacks successfully, and that the id and verifier fields are correct.

4. Router validates the check message using its id and verifier fields.

   Again, Subgraph has not yet responded to Router's original request!

   • If the check message is valid, Router responds with the following details:
     • A 204 HTTP status code
     • An empty response body
     • The response header subscription-protocol: callback
   • If the check message is invalid, Router responds with a status code besides 204 (400-level recommended).
     • If this occurs, Subgraph then responds to Router's request with a 400-level status code, and the subscription is canceled.

5. If validation succeeds, Subgraph spawns a background process or notifies a separate system (Emitter) to begin listening for subscription events.

6. Subgraph finally responds to Router with a 200-level status code. This indicates to Router that the subscription has been initialized.

With initialization complete, the protocol commences its main loop.

Main loop

The protocol's main loop remains active for the duration of the subscription. During the main loop, all of the following occur:

• Every five seconds, Emitter must send a heartbeat message to Router to confirm that Router is still listening.
• Whenever new subscription data is available, Emitter sends a next message to Router containing the new data.
• If an error occurs and the subscription must be terminated, Emitter sends a complete message to Router and includes the errors field.
• If the subscription reaches the end of its stream and no new data is forthcoming, Emitter sends a complete message to Router and omits the errors field.
• If Router terminates a particular subscription, it should return a 404 status code for all future HTTP callbacks sent for that subscription.
  • Relatedly, if Emitter receives a 404 status code from Router for an HTTP callback, it should consider the associated subscription terminated.

Message types

During the protocol flow, Subgraph and Emitter send various messages to Router's callback URL via HTTP POST requests.

All of these messages include the following base properties in their JSON body:

1
{
2
  "kind": "subscription",
3
  "action": "check",
4
  "id": "c4a9d1b8-dc57-44ab-9e5a-6e6189b2b945",
5
  "verifier": "XXX"
6
}

Fields and behaviors specific to individual message types are documented below.

check

During protocol initialization, Subgraph sends a synchronous check message to Router to help ensure that it can send callbacks successfully, and that the id and verifier fields provided by Router are correct.

A check message includes only base message fields:

1
{
2
  "kind": "subscription",
3
  "action": "check",
4
  "id": "c4a9d1b8-dc57-44ab-9e5a-6e6189b2b945",
5
  "verifier": "XXX"
6
}

If id and verifier both match Router's provided values, Router should respond with the following details:

• A 204 HTTP status code
• An empty response body
• The response header subscription-protocol: callback

Otherwise, Router should respond with an error.

heartbeat

As long as at least one subscription is active, Emitter must send a heartbeat message to Router every five seconds. This enables Emitter to confirm both that it can still reach Router's callback endpoint, and that all known subscriptions are still active.

The heartbeat message includes an ids field, which is an array of Emitter's active subscription IDs:

1
{
2
    "kind": "subscription",
3
    "action": "heartbeat",
4
    "id": "c4a9d1b8-dc57-44ab-9e5a-6e6189b2b945",
5
    "verifier": "XXX",
6
    "ids": ["c4a9d1b8-dc57-44ab-9e5a-6e6189b2b945", "c4a9d1b8-dc57-44ab-9e5a-6e6189b2b254"] 
7
}

On receiving a heartbeat message, Router checks whether all subscriptions listed in ids are indeed still active on its end.

• If all listed subscriptions are active, Router should respond with the following details:
  • A 204 HTTP status code
  • An empty response body
• If the list includes both active and inactive subscriptions, Router should respond with the following details:
  • A 400 HTTP status code
  • A JSON response body containing:
    • An invalid_ids field. This is a list containing all IDs of inactive subscriptions.
    • A verifier field. This value replaces the value of verifier provided by Emitter.
• If all listed subscriptions are inactive, Router should respond with the following details:
  • A 404 HTTP status code
  • An empty response body

Here's an example payload sent with a 400 HTTP status code if ids contains both active and inactive subscription IDs:

1
{
2
    "id": "c4a9d1b8-dc57-44ab-9e5a-6e6189b2b945",
3
    "invalid_ids": ["c4a9d1b8-dc57-44ab-9e5a-6e6189b2b254"],
4
    "verifier": "XXX"
5
}

next

Whenever a new subscription event occurs, Emitter sends the associated data to Router in a next message.

The next message includes a payload field, which contains the subscription data in standard GraphQL JSON response format:

1
{
2
  "kind": "subscription",
3
  "action": "next",
4
  "id": "c4a9d1b8-dc57-44ab-9e5a-6e6189b2b945",
5
  "verifier": "XXX",
6
  "payload": {
7
    "data": {
8
      "numberIncremented": 5
9
    }
10
  }
11
}

complete

Emitter sends a complete message to Router to terminate an active subscription. Emitter might terminate a subscription for the following reasons:

• The subscription has reached the end of its stream and no new data is forthcoming.
• An Emitter error occurred that caused the subscription to fail.

A complete message can include an errors field containing an array of GraphQL errors. This field is required if the subscription failed and optional if it completed successfully (it's usually an empty list in this case):

1
{
2
    "kind": "subscription",
3
    "action": "complete",
4
    "id": "c4a9d1b8-dc57-44ab-9e5a-6e6189b2b945",
5
    "verifier": "XXX",
6
    "errors": [{ // Optional if subscription completed successfully
7
      "message": "Something went wrong"
8
    }]
9
}

On receiving a complete message, Router terminates the associated subscription.

Error states

The following are common error states that can occur with this protcol:

• Emitter can't communicate with Router's callback endpoint, either because the endpoint isn't available or because its provided credentials (id and/or verifier) are invalid.
• Emitter receives an error HTTP status code from Router's callback endpoint.
  • If the error code is 404, Emitter should consider the associated subscription terminated by Router.
  • If the error code is 400 in response to a heartbeat message, Emitter should terminate all subscriptions listed in the response body's invalid_ids field.
  • In other error cases, Emitter should consider the subscription terminated due to an unexpected error.
• Emitter fails to send Router a heartbeat or check message for an active subscription every five seconds, causing Router to terminate that subscription.