Websocket Protocol handler and HTTP Endpoints for Connection Node

Private HTTP Endpoints

These HTTP endpoints are only for communication from endpoint nodes and must not be publicly exposed.

PUT /push/(uuid: uaid)

Send a notification to a connected client with the given uaid.

Status Codes:
PUT /notif/(uuid: uaid)

Trigger a stored notification check for a connected client.

Status Codes:
  • 200 OK – Client is connected, and has started checking.
  • 202 Accepted – Client is connected but busy, will check notifications when not busy.
  • 404 Not Found – Client is not connected to this node.
DELETE /notif/(uuid: uaid)/(int: connected_at)

Immediately drop a client of this uaid if its connection time matches the connected_at provided.

Websocket Protocol

class autopush.websocket.PushServerProtocol[source]

Main Websocket Connection Protocol


alias of autobahn.twisted.websocket.WebSocketServerProtocol

classmethod randrange(start, stop=None, step=1, _int=<type 'int'>, _maxwidth=9007199254740992L)

Choose a random item from range(start, stop[, step]).

This fixes the problem with randint() which includes the endpoint; in Python this is usually not what you want.

deferToThread(func, *args, **kwargs)[source]

deferToThread helper that tracks defers outstanding

deferToLater(when, func, *args, **kwargs)[source]

deferToLater helper that tracks defers outstanding

force_retry(func, *args, **kwargs)[source]

Forcefully retry a function in a thread until it doesn’t error

Note that this does not use self.deferToThread, so this will continue to retry even if the client drops.


Property that uses None if there’s no tags due to a DataDog library bug

log_failure(failure, **kwargs)[source]

Log a twisted failure out through twisted’s log.failure


Indicates if we are paused for output production or not

_sendAutoPing(*args, **kwargs)[source]

Override for sanity checking during auto-ping interval

sendClose(*args, **kwargs)[source]

Override to add tracker that ensures the connection is truly torn down

nukeConnection(*args, **kwargs)[source]

Aggressive connection shutdown using abortConnection if onClose still hadn’t run by this point

onConnect(*args, **kwargs)[source]

autobahn onConnect handler for when a connection has started

processHandshake(*args, **kwargs)[source]

Disable host port checking on nonstandard ports since some clients are buggy and don’t provide it

onMessage(*args, **kwargs)[source]

autobahn onMessage processor for incoming messages


Idle timer fired.


Override to track that this shut-down is from a ping timeout

onClose(*args, **kwargs)[source]

autobahn onClose handler for shutting down the connection and any outstanding deferreds related to this connection

cleanUp(wasClean, code, reason)[source]

Thorough clean-up method to cancel all remaining deferreds, and send connection metrics in


Save a direct_update webpush style notification


Looks up the node to send a notify for it to check storage if connected


Traps UAID not found error


Checks the result of lookup node to send the notify if the client is connected elsewhere now

returnError(messageType, reason, statusCode, close=True, url='')[source]

Return an error to a client, and optionally shut down the connection safely

error_overload(failure, message_type, disconnect=True)[source]

Handle database overloads and errors

If disconnect is False, the an overload error is returned and the client is not disconnected.

Otherwise, pause producing to cease incoming notifications while we wait a random interval up to 8 seconds before closing down the connection. Most clients wait up to 10 seconds for a command, but this is not a guarantee, so rather than never reply, we still shut the connection down.

Parameters:disconnect – Whether the client should be disconnected or not.

Close the connection down and resume consuming input after the random interval from a db overload


Send a Python dict as a JSON string in a websocket message


Process a hello message


Register a returning or new user


Verify a user record is valid

Returns a record that is ready for registering in the database if the user record was found.

Return type:Item or None

errBack for hello failures

_check_other_nodes(result, url='')[source]

callback to check other nodes for clients and send them a delete as needed


callback for successful hello message, that sends hello reply


Run a notification check against storage


Helper to return an appropriate function to fetch messages


errBack for notification check failing


errBack for provisioned errors during notification check


errBack for handling excessive messages per UAID


callback for processing notifications from storage


WebPush notification processor


Function to fire off a message table copy of channels + update the router current_month entry


Transition the client to use a new message month

Utilized to migrate a users channels to a new message month and update the router record reflecting the proper month.

This is a blocking function that does not run on the event loop.


Mark the client as successfully transitioned and resume


Capture overload on monthly table rotation attempt

If a provision exceeded error hits while attempting monthly table rotation, schedule it all over and re-scan the messages. Normal websocket client flow is returned in the meantime.


Helper for ping sending that tracks when the ping was sent


Ping Handling

Clients in the wild have a bug that lowers their ping interval to 0. It will never increase for them, as there is no way to remedy this without causing the client to use drastically more battery/data-usage we send them a code 4774 close to signify that they should stop until network change.

No other client should ping more than once per minute, or we tell them to go away.


Process a register message


errBack handler for registering to fail

finish_register(endpoint, chid)[source]

callback for successful endpoint creation, sends register reply


Process an unregister message


Helper function for tracking ack’d updates

Returns either None, if no delete_notification call is needed, or a deferred for the delete_notification call if it was needed.

_handle_webpush_ack(chid, version, code)[source]

Handle clearing out a webpush ack

_handle_webpush_update_remove(result, chid, notif)[source]

Handle clearing out the updates_sent

It’s possible the client may leave before this runs, so this is wrapped in a try/except in case the tear-down of self has started.


Process an ack message, delete notifications from storage if needed


Process a nack message and log its contents

check_missed_notifications(results, resume=False)[source]

Check to see if notifications were missed

bad_message(typ, message=None, url='')[source]

Error helper for sending a 401 status back


Utility function for external use

This function is called by the HTTP handler to deliver an incoming update notification from an endpoint.

HTTP Handlers

class autopush.websocket.RouterHandler(application, request, **kwargs)[source]

Router Handler

Handles routing a notification to a connected client from an endpoint.



Attempt delivery of a notification to a connected client.

class autopush.websocket.NotificationHandler(application, request, **kwargs)[source]
put(uaid, *args)[source]


Notify a connected client that it should check storage for new notifications.

delete(uaid, connected_at)[source]

HTTP Delete

Drop a connected client as the client has connected to a new node.

Utility Functions


Return current time.time call as ms and a Python int


Exception Logger Decorator for protocol methods