RFC: Async client disconnect #2372
Open
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
added 3 commits
November 19, 2021 10:48
Enabling the EPOLLRDHUP event when watching sockets makes epoll signal closed connections when they happen.
After processing protocol data, the client socket is marked as "idle" in the event queue, which enables connection-closed events in the event loop. Keeping "idle" connections in the event loop requires special casing when enabling read and/or write events too, as the event queue complains if a filedescriptor is added twice to the event loop. Avoid such errors by modifying request sockets (wsgi_req->fd) instead of adding/deleting them when handling response reads and writes. Once a request connection is closed, the async event loop marks the request as closed (wsgi_req->async_closed = 1) and continues request processing to allow the request handler to clean up gracefully.
Check the wsgi_req->async_closed flag when processing response iterations, and raise an OSError is the flag is set. This solves the case, where a long running (but idle) client connection is closed by the client. Since no data is written to the socket, the close event is not detected, and the backend handler keeps running. One case of an idle long running connection is log file streaming: The client reads the log contents, and keeps hanging around for new log messages. If no new messages occur, nothing is written to the connection. When the client disconnects, the request handler should be stopped.
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
I'm scratching my own itch here.
I'm implementing log-file streaming over HTTP using the usual
Client - NGINX - uWSGI - Flaskstack. When clients requests the stream, they receive the full log, and then new lines added to logs are streamed as they come. Those new lines might occur with several minutes in between. If the client disconnects in such an interval, the handler is blocked waiting for new log messages, and doesn't detect the closed connection. This effectively exhausts the "worker" pool if many such disconnects happen. Writing "dummy" data to the socket regularly isn't an option, as the dummy data would appear in the streamed log contents.To fix this, the async core can detect the disconnected clients by keeping the protocol sockets in the event queue, and enable
EPOLLRDHUPevents. (I haven't checked similar flags for other event queue implementations.) The python plugin then checks a flag to detect closed connections in the response handler.