Ok it might not be a bug per say and maybe it was designed like that but I have some interrogations on how the presence is being updated on the js side. The file app/assets/javascripts/discourse/app/services/presence.js
What I understood so far is that the JS client:
Will throttle the update to 1 per second when the queue of updates is not empty
Then one request every 30 seconds when there is no changes
What I noticed is that if the server errors for any reason the js will keep making requests every seconds. Also creating errors in the console Uncaught (in promise) line 565.
Shouldnât it stop at some point after N errors to not overload the server ?
How to reproduce ?
Simulate an error on the presence_controller.rbupdate method. (In my case I rendered status 405 just for the sake of the demo)
def update
# JS client is designed to throttle to one request per second
# When no changes are being made, it makes one request every 30 seconds
RateLimiter.new(nil, "update-presence-#{current_user.id}", 20, 10.seconds).performed!
render json: { error: "Not authorized" }, status: 405
return
...
Then on the client go to any screen with a presence management. I chose the topic reply screen and started replying.
Even without the scheduled update every 30 seconds if there is any error from the backend the algorithm will try again every second because the queue of events will not be empty:
In _scheduleNextUpdate
} else if (this._queuedEvents.length > 0) {
this._cancelTimer();
cancel(this._debounceTimer);
this._debounceTimer = debounce(
this,
this._throttledUpdateServer,
this._presenceDebounceMs // This is 1 sec I believe
);
In fact in the _updateServer method â events are put back in the queue on error:
} catch (e) {
// Put the failed events back in the queue for next time
this._queuedEvents.unshift(...queue);
This was just something I noticed afterwards⌠When everything goes well ( meaning no errors ), is it really necessary to keep updating the server every other 30 seconds ?
} else if (
!this._nextUpdateTimer &&
this._presentChannels.size > 0 &&
!isTesting()
) {
this._nextUpdateTimer = discourseLater(
this,
this._throttledUpdateServer,
PRESENCE_INTERVAL_S * 1000 // This is 30 seconds
);
}
Ok I think I understood why it needs to âcheck-inâ again every 30 seconds.
There is a background job that is triggered every minutes PresenceChannelAutoLeave that will make expired members leave the channel.
It is also commented in presence_channel.rb for the present(user:, client_id:) method:
# Mark a user's client as present in this channel. The client_id should be unique per
# browser tab. This method should be called repeatedly (at least once every DEFAULT_TIMEOUT)
# while the user is present in the channel.
def present(user:, client_id:)
Still wondering about the error part that is triggered every second
We were already handling 429 ârate limitedâ errors correctly. But for other, unexpected, errors Iâve added some exponential backoff logic (capped at 30s). Thanks for bringing this to our attention @AhmedLoud
