1. Implement the WebPush specification for Mozilla services/products.
2. Provide a bridge between a mobile or desktop client and server app without the server needing to know about ecosystem-specific push providers (FCM/APN/ADM).
3. End-to-end encryption between client app and server app.
### Key actors & processes
Below is a sequence diagram for the four main processes that take place for push messaging:
* **Client App**: this is your Android device that includes the `AutoPushFeature` that contains the AutoPush rust component to create/delete subscriptions and encrypt/decrypt messages. As well as the`lib-push-firebase` which is the push service bridge.
* **Server App** - the application server that has a web push server implementation.
* **AutoPush** - the server bridge between app servers and their clients.
* **Push Provider** - the platform push service that does the "last-mile" message delivered.
Device->>Server: Unsubscribe (notify server that the subscription is dead)
end
```
</details>
### Miscellaneous
Q. Why do we need to verify connections, and what happens when we do?
- Various services may need to communicate with us via push messages. Examples: FxA events (send tab, etc), WebPush (a web app receives a push message from its server).
- To send these push messages, services (FxA, random internet servers talking to their web apps) post an HTTP request to a "push endpoint" maintained by [Mozilla's Autopush service][0]. This push endpoint is specific to its recipient - so one instance of an app may have many endpoints associated with it: one for the current FxA device, a few for web apps, etc.
- Important point here: servers (FxA, services behind web apps, etc.) need to be told about subscription info we get from Autopush.
- Here is where things start to get complicated: client (us) and server (Autopush) may disagree on which channels are associated with the current UAID (remember: our subscriptions are per-channel). Channels may expire (TTL'd) or may be deleted by some server's Cron job if they're unused. For example, if this happens, services that use this subscription info (e.g. FxA servers) to communication with their clients (FxA devices) will fail to deliver push messages.
- So the client needs to be able to find out that this is the case, re-create channel subscriptions on Autopush, and update any dependent services with new subscription info (e.g. update the FxA device record for `PushType.Services`, or notify the JS code with a `pushsubscriptionchanged` event for WebPush).
- The Autopush side of this is `verify_connection` API - we're expected to call this periodically, and that library will compare channel registrations that the server knows about vs those that the client knows about.
- If those are misaligned, we need to re-register affected (or, all?) channels, and notify related services so that they may update their own server-side records.
- For FxA, this means that we need to have an instance of the rust FirefoxAccount object around in order to call `setDevicePushSubscriptionAsync` once we re-generate our push subscription.
- For consumers such as Fenix, easiest way to access that method is via an `account manager`.
- However, neither account object itself, nor the account manager, aren't available from within a Worker. It's possible to "re-hydrate" (instantiate rust object from the locally persisted state) a FirefoxAccount instance, but that's a separate can of worms, and needs to be carefully considered.
- Similarly for WebPush (in the future), we will need to have Gecko around in order to fire `pushsubscriptionchanged` javascript events.
Q. Where do we find more details about AutoPush?
- The Cloud Services team have [an architecture doc][0] for in-depth details on how the AutoPush server works with clients.