[ricardoasmarques]

The kind of change this PR does introduce

• a bug fix
• a new feature
• an update to the documentation
• a code change that improves performance
• other

Current behavior

With @socket.io/postgres-emitter@0.1.1 and @socket.io/postgres-adapter@0.4.0 or later, the servers never receive anything the emitter publishes.

Since 0.4.0, the Postgres adapter extends ClusterAdapter from socket.io-adapter, which drops messages whose top-level nsp doesn't match:

if (message.nsp !== this.nsp.name) {
  return debug("[%s] ignore message from another namespace (%s)", this.uid, message.nsp);
}

0.5.0 also routes incoming messages by msg.nsp (namespaces.get(msg.nsp)) before reaching that check.

The emitter sets nsp only on the inner packet (data.packet.nsp), never on the message itself. So every broadcast, socketsJoin / socketsLeave, disconnectSockets and serverSideEmit is dropped, with this debug log on each server:

socket.io-adapter [<uid>] ignore message from another namespace (undefined)

The emitter's tests didn't catch it because the root package.json pins @socket.io/postgres-adapter to ^0.1.0, which never checked the namespace.

New behavior

publish() sets document.nsp = this.emitter.nsp next to document.uid. Every message goes through publish(), so this covers both the direct pg_notify path and the attachment path (payload over payloadThreshold, or with binary data).

The dev dependency moves to @socket.io/postgres-adapter@^0.5.0, so the existing tests now run against a current adapter. Locally with Postgres 14 (as in CI):

• adapter 0.5.0 without the fix: the broadcast tests time out
• adapter 0.5.0 with the fix: 12 passing

Other information (e.g. related issues)

I couldn't find an existing issue. We hit this in production after moving an emit-only worker from the full adapter to the emitter: the worker's events stopped reaching clients without any error.