Task was destroyed but it is pending: WebsocketImplProtocol.async_data_received (Sentry-wrapped)

[cnicodeme]

How do you use Sentry?

Sentry Saas (sentry.io)

Version

2.43.0

Steps to Reproduce

Hi team!

I'm having quite a few errors like these popping up in my Sentry account. At first, I thought it was an issue with my Websocket implementation, but looking at the stacktrace, some of the errors are from a standard GET request, such as this one:
https://cnicodeme.sentry.io/share/issue/7f8aa28baa194938bd13f44953014819/
(For the Sentry team, feel free to access this ticket, and all the other that have the "Task was destroyed but it is pending" currently open in my organization).

I suspect that the Sentry system doesn't have the time to send the data to Sentry.io and is killed before, somehow.
Unfortunately the error doesn't contain the original one, making it harder to trace where it originated and/or fix it.

Expected Result

It shouldn't throw a "Task was destroyed but it is pending" but instead the actual underlying error. I suspect the task of sending that error to Sentry is killed before it has the chance to deliver the payload.

Actual Result

I'm using Sanic 24.6.

Happy to share anything more that is needed

[linear]

PY-1966

[alexander-alderman-webb]

Hi @cnicodeme ,

Thanks for the link!

In general, the errors you are seeing have the form

Task was destroyed but it is pending!
task: <Task pending name='...' coro=<...> wait_for=<Future pending cb=[...]>>

The Sentry AsyncioIntegration

• modifies the task name by appending a (Sentry-wrapped); and
• wraps the coroutine, so some of the metadata is lost. Ideally this information should not be erased, we'll look into this.

I would be reluctant to think that the exception originates from the sentry_sdk event flushing mechanism, as events are sent in non-async code paths up to and including version 2.43.0 of sentry_sdk.

Could you provide a reproduction where the exception is raised? It would be useful to know

• how you are running the Sanic server; in particular whether you pass loop or protocol arguments to Sanic.run(); and
• if you are creating your own tasks, how you schedule them, and what you do with coroutines on application shutdown, or generally before the event loop is closed. For example, do you cancel coroutines on shutdown?

[cnicodeme]

Hi @alexander-alderman-webb , thank you for your response (and sorry for my late reply, I was out for a long weekend and couldn't reply properly).

It's hard for me to provide a reproducible issue since I don't have the stack of when this is happening from the "Task was destroyed but it is pending!" issue. I know that these have been increasing a lot on my account.

All of the issues I have logged at Sentry have the same 4 last (recent) stack items:

__init__.py in handle at line 948
__init__.py in handle at line 1595
__init__.py in _log at line 1585
__init__.py in error at line 1471

Then the (probably) actual file that threw (or logged) the error.

Not all, but many seems to be coming from issues related to the database, even though that might be issues from my end I need to resolve.

If I can modify the Sentry code to have a more debugging output of the situation, even by writing on /tmp while we debug this and find the original issue, that would be very helpful to me.

Let me know what I can do else.

Thanks!