backpressure on LISTEN?

• asyncpg version: 0.18.3
• PostgreSQL version: 9.5
• Do you use a PostgreSQL SaaS? If so, which? Can you reproduce
  the issue with a local PostgreSQL install?:
• Python version: 3.6
• Platform:
• Do you use pgbouncer?:
• Did you install asyncpg with pip?:
• If you built asyncpg locally, which version of Cython did you use?:
• Can the issue be reproduced under both asyncio and
  uvloop?:

My application subscribes to postgres notifications and fans them out to subscribers using streaming http.

I see that notifications emerge in connection._process_notification, where they are dispatched via call_soon(), each time calling _call_listener(), which will synchronously call the callback that I specify in conn.add_listener(). In the callback, I insert a task into the queue for asynchronous processing...

I see a lot of tasks inserted this way, e.g. ~5000 before my application has a chance to process them. Therefore application memory has big unpredictable spikes, like +300k

Question: how do you apply back-pressure in this setup? is there a way to limit the number of postgres notifications that are being converted to tasks, to control overall number of in-flight tasks? Thanks!

Read connection._process_notification, call_soon(), _call_listener(), and conn.add_listener() first, then trace how the application’s queue creates tasks. The issue does not name a target file, test, or acceptance behavior; define whether the outcome should be an API-level limit on in-flight notifications and add coverage for that behavior.