I found an interesting pattern. Even if I tried all the solutions on the forum (administrators start/stop http2), I couldn’t solve this long-standing problem.
Often when this error occurs, no matter how many times you retry, if the first return fails immediately, it will never succeed again.
But as long as you are willing to spend the cost of a new request, it will work normally immediately.
I want to say that this looks like a dirty trick. It takes more requests to complete a request task. In theory, the cursor should be responsible for the failed request, but in reality, the user needs to spend more.
I think for some reason, the cursor defaults to this bug, or maybe it is part of the design.