Skip to content

Latest commit

 

History

History
191 lines (141 loc) · 6.32 KB

Using_Stream_Management.md

File metadata and controls

191 lines (141 loc) · 6.32 KB

Using Stream Management

AKA: Handling 'Message failed to send' situations

XEP-0198: Stream Management can be used with StanzaJS, and it allows for detecting and recovering some situations where connection loss would prevent messages you thought were sent actually failed.

Configuring Stream Management

Stream management is enabled automatically if your server offers it during stream negotiation.

You can turn it off, if necessary, when configuring a client:

import * as XMPP from 'stanza';

const client = XMPP.createClient({
    ...opts,
    useStreamManagement: false
});

Stream resumption is also enabled by default, if available. It can be disabled in the client configuration:

import * as XMPP from 'stanza';

const client = XMPP.createClient({
    ...opts,
    allowResumption: false,
    useStreamManagement: true
});

Caching State

In order for stream management to recover from a connection loss, the state of unacked stanzas must be cached. StanzaJS does this automatically in memory, but that information can be lost on a page navigation, or if an app gets suspended and unloaded.

The client.sm object provides two methods for persisting the stream management state:

  • load()
  • cache()

Before calling client.connect(), you can use client.sm.load() to restore the stream management state from a previous session:

import { Utils } from 'stanza';

const cachedSM = sessionStorage.cachedSM;
if (cachedSM) {
    client.sm.load(JSON.parse(cachedSM, Utils.reviveData));
}

The Utils.reviveData() helper function is available to restore Date and Buffer objects after they've been serialized with JSON.stringify().

If you pass a function to client.sm.cache(), it will be called on every state change:

client.sm.cache(state => {
    sessionStorage.cachedSM = JSON.stringify(state);
});

You can also return a promise if your caching process needs to be async:

client.sm.cache(async state => {
    await someExternalDataStorage.save(state);
});

Detecting Message Failures

Detecting message loss requires having some sort of storage for tracking the messages you know you have sent locally. How you do that, is up to you.

For now, we'll assume a basic Map:

const messageStorage = new Map();

Then we need to track every message that we send:

client.on('message:sent', (msg, viaCarbon) => {
    messageStorage.set(msg.id, {
        serverReceived: viaCarbon,
        mucReceived: false,
        hibernated: false,
        failedToSend: false,
        retrying: false
    });
});

Now, there are several things that should be noted at this point.

First, we use the message:sent event to capture the id of outgoing messages instead of using the return value of client.sendMessage() or manually generating and passing id values to sendMessage(). Doing it this way ensures we capture all outgoing messages, without worrying about where sendMessage() was called.

You might notice that the message:sent event gets fired immediately when calling sendMessage(), even before the message actually gets written to a connection. This is by design: the moment you call sendMessage() you must treat the message as if it is in flight over the network.

Second, we are looking at the viaCarbon flag to distinguish if this was a message sent locally, or it came from a XEP-0280 message carbon. If it came from a carbon, then the server has clearly received the original message.

Third, we are storing four flags to represent multiple situations:

  • The server has received our message.
  • If we are sending a message to a MUC, the MUC service has received our message.
  • The network connection was down when we tried to send the message.
  • The message failed to send entirely.
  • We are attempting to (automatically) resend the message.

Since MUCs are a very common use case, tracking that your message made it to the room, and not just to your server, is a good practice. MUCs reflect the messages you send back to you, so that state can be tracked regardless of if stream management is in use.

Let's start with the easier MUC tracking:

client.on('groupchat', msg => {
    if (msg.from === yourInRoomNick && messageStorage.has(stanza.id)) {
        messageStorage.set(msg.id, {
            serverReceived: true,
            mucReceived: true
        });
    }
});

We can set multiple states to true here, since if the MUC has received your message, then your server must have as well.

To detect that your server has received your message, we use the message:acked event:

client.on('message:acked', msg => {
    if (messageStorage.has(msg.id)) {
        messageStorage.set(msg.id, {
            serverReceived: true,
            mucReceived: false
        });
    }
});

We are using the message:acked event, which is a special case of the more general stanza:acked event. The stanza:acked event fires for all stanza types, not just messages.

In the event that you lose connection, there are two possibilities:

  1. Stream resumption works, and messages get resent as needed, automatically.
  2. Stream resumption fails.

If stream resumption was enabled, the event message:hibernated is emitted instead if the client is disconnected when it attempts to send the message:

client.on('message:hibernated', msg => {
    if (messageStorage.has(msg.id)) {
        messageStorage.set(msg.id, {
            hibernated: true
        });
    }
});

Hibernated messages will be resent automatically if the session is resumed, and a message:retry event is emitted when that happens:

client.on('message:retry', msg => {
    if (messageStorage.has(msg.id)) {
        messageStorage.set(msg.id, {
            retrying: true
        });
    }
});

If stream resumption fails (or was never enabled), then any still-pending messages never made it to the server:

client.on('message:failed', msg => {
    if (messageStorage.has(msg.id)) {
        messageStorage.set(msg.id, {
            failedToSend: true
        });
    }
});

These messages will need to be resent. Probably via a direct user interaction to confirm resending, as the messages could be old and no longer needed.