Background Tasks¶
Some times it is necessary to run background tasks. For example to poll a data source, or listen to a data stream.
The following example starts a background task when the server starts up, then gracefully terminates it when the server shuts down.
The source code for the following example can be found here (and here here with typing).
The Task¶
The following code provides a dummy task, which just waits for a second then gets the time. This time I've used the typed code.
async def time_ticker(info: Dict[str, Any], shutdown_event: Event) -> None:
print('Starting the time ticker')
while not shutdown_event.is_set():
info['now'] = datetime.now()
print(f"time: {info['now']}")
try:
await asyncio.wait_for(shutdown_event.wait(), timeout=1)
except asyncio.TimeoutError:
print('Timeout - normal behaviour when waiting with a timeout')
except:
print('Failure - we should not see this exception')
print('The time ticker has stopped')
Rather than simply cancelling the task, I've used the asyncio Event to shutdown more gracefully.
The Startup Handler¶
First we create the event and store it in the application's info
.
Then we create the task with
asyncio.create_task
and store the task in the application's info
.
When the task is created it will be scheduled to run. Any time the task awaits, it gives up control and other tasks which are ready to run can proceed.
async def time_ticker_startup_handler(request):
# Create an event that can be set when the background task should shutdown.
shutdown_event = Event()
request.info['shutdown_event'] = shutdown_event
# Create the background task.
request.info['time_ticker_task'] = asyncio.create_task(
time_ticker(info, shutdown_event)
)
Note that the shutdown event was created in the startup handler. This is critical. I'll discuss why in the "Gotcha!" section.
The Shutdown Handler¶
Here is the code for the shutdown handler.
async def time_ticker_shutdown_handler(request):
# Set the shutdown event so the background task can stop gracefully.
shutdown_event: Event = request.info['shutdown_event']
shutdown_event.set()
# Wait for the background task to finish.
time_ticker_task: asyncio.Task = request.info['time_ticker_task']
await time_ticker_task
First we retrieve the event from the application's info and set it. In the background task the loop will exit when it next checks the event.
while not shutdown_event.is_set():
...
Next we fetch the task from the applications info and await
it. When the
next timeout occurs in the background task it will exit it's loop, and the
background task gracefully shuts down.
The Program¶
Here is the full code.
I've added a request handler which returns the time stored by the background task to demonstrate that the web server is still able to handle requests.
import asyncio
from asyncio import Event
from datetime import datetime
import logging
import uvicorn
from bareasgi import Application, text_writer, HttpResponse
logging.basicConfig(level=logging.DEBUG)
LOGGER = logging.getLogger('background_tasks')
async def time_ticker(info, shutdown_event):
LOGGER.debug('Starting the time ticker')
while not shutdown_event.is_set():
# Store the time
info['now'] = datetime.now()
LOGGER.debug('time: %s', info['now'])
try:
await asyncio.wait_for(shutdown_event.wait(), timeout=1)
except asyncio.TimeoutError:
LOGGER.debug(
'Timeout - normal behaviour when waiting with a timeout')
except:
LOGGER.exception('Failure - we should not see this exception')
LOGGER.debug('The time ticker has stopped')
async def time_ticker_startup_handler(request):
# Create an event that can be set when the background task should shutdown.
shutdown_event = Event()
request.info['shutdown_event'] = shutdown_event
# Create the background task.
request.info['time_ticker_task'] = asyncio.create_task(
time_ticker(info, shutdown_event)
)
async def time_ticker_shutdown_handler(request):
# Set the shutdown event so the background task can stop gracefully.
shutdown_event: Event = request.info['shutdown_event']
LOGGER.debug('Stopping the time_ticker')
shutdown_event.set()
# Wait for the background task to finish.
time_ticker_task: asyncio.Task = request.info['time_ticker_task']
LOGGER.debug('Waiting for time_ticker')
await time_ticker_task
LOGGER.debug('time_ticker shutdown')
async def http_request_callback(request):
headers = [
(b'content-type', b'text/plain')
]
return HttpResponse(200, headers, text_writer(f"Last time tick: {request.info.get('now')}"))
app = Application(
startup_handlers=[time_ticker_startup_handler],
shutdown_handlers=[time_ticker_shutdown_handler]
)
app.http_router.add({'GET'}, '/{rest:path}', http_request_callback)
uvicorn.run(app, port=9009)
Canceling¶
Rather than using an event we could have just cancelled the task.
async def time_ticker_shutdown_handler(scope, info, request):
time_ticker_task: asyncio.Task = info['time_ticker_task']
try:
time_ticker_task.cancel()
await time_ticker_task
except asyncio.CancellationError:
pass
The background task might then look as follows.
async def time_ticker(info, shutdown_event):
while True:
# Store the time
info['now'] = datetime.now()
print(f"time: {info['now']}")
try:
await asyncio.sleep(1)
except asyncio.CancellationError:
return # Catch the task.cancel() and exit
except:
print('Failure - we should not see this exception')
print('The time ticker has stopped')
Cancel or Shutdown Event?¶
In the above example cancelling the task would have been appropriate. Nothing bad would happen, and no data would be lost.
The event approach is useful where cancelling the task would result in some kind of corrupt state. For example if a database write was in progress, or a message queue was being serviced.
Gotcha!¶
If we had a large program running many background tasks with multiple startup handlers, it would seem reasonable to create the shutdown event right at the start.
# This won't work!
app = Application(info={'shutdown_event': asyncio.Event()})
Unfortunately this won't work and will lead to hours of frustration.
When an ASGI server starts it creates a new event loop. However the
asyncio.Event()
call attached the event to the existing event loop. At the
point the event is awaited you will get an error telling you the coroutine was
attached to another event loop, which is true, but not helpful!
This is true for anything that can be awaited. They must all be created in the context of the ASGI server's event loop.
What next?¶
Either go back to the table of contents or go to the https tutorial.