Graceful shutdown can (I think) truncate connections that it shouldn't #1369
Labels
Comments
|
Fixed by #1478 |
This was referenced May 29, 2019
Merged
komuw
added a commit
to komuw/naz
that referenced
this issue
May 31, 2019
What: - Introduce a configurable timeout when trying to connect to SMSC - Try and detect when the connection to SMSC is disconnected and attempt to re-connect & re-bind - bugfix; `asyncio.streams.StreamWriter.drain` should not be called concurrently by multiple coroutines[1] - when shutting down, `naz` now tries to make sure that write buffers are properly flushed[2][3] Why: - Make `naz` more failure tolerant - Fixes: #67 - Fixes: #114 - Fixes: #116 - Fixes: #117 - Fixes: #120 References: 1. https://bugs.python.org/issue29930 2. https://vorpus.org/blog/some-thoughts-on-asynchronous-api-design-in-a-post-asyncawait-world/ 3. aio-libs/aiohttp#1369
This was referenced Jun 24, 2019
Sign up for free
to subscribe to this conversation on GitHub.
Already have an account?
Sign in.
The idea of graceful shutdown is that we finish sending out all existing responses before shutting down. But from code inspection, I think it can happen that in certain cases aiohttp will truncate a response, even though it could easily have finished within the time limit.
Specifically, the problem is, how do we know that we've actually finished sending the complete response body to the client? AFAICT there are two ways to do this in asyncio: (a) wait for a
connection_lostcallback, (b) callawait stream_writer.drain(), after callingstream_writer._transport.set_write_buffer_limits(0).As far as I can tell, aiohttp calls
await stream_writer.drain(), but without the crucial second step. Without this,draincan and will return while there is still unsent data buffered inside the event loop, and if we then proceed with shutting down this data can silently lost.The text was updated successfully, but these errors were encountered: