Listeners

Sanic provides you with eight (8) opportunities to inject an operation into the life cycle of your application server. This does not include the signals, which allow further injection customization.

There are two (2) that run only on your main Sanic process (ie, once per call to sanic server.app.)

  • main_process_start
  • main_process_stop

There are also two (2) that run only in a reloader process if auto-reload has been turned on.

  • reload_process_start
  • reload_process_stop

Added reload_process_start and reload_process_stop in v22.3

There are four (4) that enable you to execute startup/teardown code as your server starts or closes.

  • before_server_start
  • after_server_start
  • before_server_stop
  • after_server_stop

The life cycle of a worker process looks like this:

sequenceDiagram autonumber participant Process participant Worker participant Listener participant Handler Note over Process: sanic server.app loop Process->>Listener: @app.main_process_start Listener->>Handler: Invoke event handler end Process->>Worker: Run workers loop Start each worker loop Worker->>Listener: @app.before_server_start Listener->>Handler: Invoke event handler end Note over Worker: Server status: started loop Worker->>Listener: @app.after_server_start Listener->>Handler: Invoke event handler end Note over Worker: Server status: ready end Process->>Worker: Graceful shutdown loop Stop each worker loop Worker->>Listener: @app.before_server_stop Listener->>Handler: Invoke event handler end Note over Worker: Server status: stopped loop Worker->>Listener: @app.after_server_stop Listener->>Handler: Invoke event handler end Note over Worker: Server status: closed end loop Process->>Listener: @app.main_process_stop Listener->>Handler: Invoke event handler end Note over Process: exit

The reloader process live outside of this worker process inside of a process that is responsible for starting and stopping the Sanic processes. Consider the following example:

@app.reload_process_start
async def reload_start(*_):
    print(">>>>>> reload_start <<<<<<")

@app.main_process_start
async def main_start(*_):
    print(">>>>>> main_start <<<<<<")
	
@app.before_server_start
async def before_start(*_):
	print(">>>>>> before_start <<<<<<")

If this application were run with auto-reload turned on, the reload_start function would be called once when the reloader process starts. The main_start function would also be called once when the main process starts. HOWEVER, the before_start function would be called once for each worker process that is started, and subsequently every time that a file is saved and the worker is restarted.

Attaching a listener#

The process to setup a function as a listener is similar to declaring a route.

The currently running Sanic() instance is injected into the listener.

async def setup_db(app):
    app.ctx.db = await db_setup()

app.register_listener(setup_db, "before_server_start")

The Sanic app instance also has a convenience decorator.

@app.listener("before_server_start")
async def setup_db(app):
    app.ctx.db = await db_setup()

Prior to v22.3, both the application instance and the current event loop were injected into the function. However, only the application instance is injected by default. If your function signature will accept both, then both the application and the loop will be injected as shown here.

@app.listener("before_server_start")
async def setup_db(app, loop):
    app.ctx.db = await db_setup()

You can shorten the decorator even further. This is helpful if you have an IDE with autocomplete.

@app.before_server_start
async def setup_db(app):
    app.ctx.db = await db_setup()

Order of execution#

Listeners are executed in the order they are declared during startup, and reverse order of declaration during teardown

Phase Order
main_process_start main startup regular 🙂 ⬇️
before_server_start worker startup regular 🙂 ⬇️
after_server_start worker startup regular 🙂 ⬇️
before_server_stop worker shutdown 🙃 ⬆️ reverse
after_server_stop worker shutdown 🙃 ⬆️ reverse
main_process_stop main shutdown 🙃 ⬆️ reverse

Given the following setup, we should expect to see this in the console if we run two workers.

@app.listener("before_server_start")
async def listener_1(app, loop):
    print("listener_1")

@app.before_server_start
async def listener_2(app, loop):
    print("listener_2")

@app.listener("after_server_start")
async def listener_3(app, loop):
    print("listener_3")

@app.after_server_start
async def listener_4(app, loop):
    print("listener_4")

@app.listener("before_server_stop")
async def listener_5(app, loop):
    print("listener_5")

@app.before_server_stop
async def listener_6(app, loop):
    print("listener_6")

@app.listener("after_server_stop")
async def listener_7(app, loop):
    print("listener_7")

@app.after_server_stop
async def listener_8(app, loop):
    print("listener_8")

[pid: 1000000] [INFO] Goin' Fast @ http://127.0.0.1:9999
[pid: 1000000] [INFO] listener_0
[pid: 1111111] [INFO] listener_1
[pid: 1111111] [INFO] listener_2
[pid: 1111111] [INFO] listener_3
[pid: 1111111] [INFO] listener_4
[pid: 1111111] [INFO] Starting worker [1111111]
[pid: 1222222] [INFO] listener_1
[pid: 1222222] [INFO] listener_2
[pid: 1222222] [INFO] listener_3
[pid: 1222222] [INFO] listener_4
[pid: 1222222] [INFO] Starting worker [1222222]
[pid: 1111111] [INFO] Stopping worker [1111111]
[pid: 1222222] [INFO] Stopping worker [1222222]
[pid: 1222222] [INFO] listener_6
[pid: 1222222] [INFO] listener_5
[pid: 1222222] [INFO] listener_8
[pid: 1222222] [INFO] listener_7
[pid: 1111111] [INFO] listener_6
[pid: 1111111] [INFO] listener_5
[pid: 1111111] [INFO] listener_8
[pid: 1111111] [INFO] listener_7
[pid: 1000000] [INFO] listener_9
[pid: 1000000] [INFO] Server Stopped

In the above example, notice how there are three processes running:

  • pid: 1000000 - The main process
  • pid: 1111111 - Worker 1
  • pid: 1222222 - Worker 2

Just because our example groups all of one worker and then all of another, in reality since these are running on separate processes, the ordering between processes is not guaranteed. But, you can be sure that a single worker will always maintain its order.

FYI

The practical result of this is that if the first listener in before_server_start handler setups a database connection, listeners that are registered after it can rely upon that connection being alive both when they are started and stopped.

Priority#

In v23.12, the priority keyword argument was added to listeners. This allows for fine-tuning the order of execution of listeners. The default priority is 0. Listeners with a higher priority will be executed first. Listeners with the same priority will be executed in the order they were registered. Furthermore, listeners attached to the app instance will be executed before listeners attached to a Blueprint instance.

Overall the rules for deciding the order of execution are as follows:

  1. Priority in descending order
  2. Application listeners before Blueprint listeners
  3. Registration order

As an example, consider the following, which will print:

third
bp_third
second
bp_second
first
fourth
bp_first

@app.before_server_start
async def first(app):
    print("first")

@app.listener("before_server_start", priority=2)
async def second(app):
    print("second")

@app.before_server_start(priority=3)
async def third(app):
    print("third")

@bp.before_server_start
async def bp_first(app):
    print("bp_first")

@bp.listener("before_server_start", priority=2)
async def bp_second(app):
    print("bp_second")

@bp.before_server_start(priority=3)
async def bp_third(app):
    print("bp_third")

@app.before_server_start
async def fourth(app):
    print("fourth")

app.blueprint(bp)

ASGI Mode#

If you are running your application with an ASGI server, then make note of the following changes:

  • reload_process_start and reload_process_stop will be ignored
  • main_process_start and main_process_stop will be ignored
  • before_server_start will run as early as it can, and will be before after_server_start, but technically, the server is already running at that point
  • after_server_stop will run as late as it can, and will be after before_server_stop, but technically, the server is still running at that point