Fix hangup on reading of closed socket #151
Conversation
We cannot get more date on closed socket, return what we have, if anythink
|
This makes sense as far as I can tell but I lack the specific knowledge about the lower level details so I may not be the best to review. I did test this branch successfully with the simpletest script in this repo on a Feather RP2040 + Ethernet Featherwing. I attempted to test the specific path of code that contains the change but couldn't figure out how. I tried adding some sleep inside of recv_into (to give myself more time to react) and then unplugging the ethernet cable while a request was being received. I'm not really sure if that is a correct way to test this, but in my case it was raising this exception from inside the |
|
I think to test, we would need a way to kill the socket from the other side. I wonder if you setup one device as a webserver, and then during that wait, power that device off, if you would get the right thing (because then the socket would be closed). You could also spin up something local if that's easier. |
There was a problem hiding this comment.
Thank you @justmobilize! That did the trick.
I tested this successfully with this flask handler:
def hello():
print("inc request to hello")
time.sleep(10)
return "Hot Peppers"
During the sleep() I stopped the server with keyboard interrupt.
Using this branch the microcontroller notices the severed connection immediately raising RuntimeError: Error connecting socket: Failed to establish connection.
Using current main there is a very long delay and then eventually the request times out.
The version from this branch responds much faster, pretty much instantly when the flask server is stopped.
Updating https://github.com/adafruit/Adafruit_CircuitPython_floppy to 2.0.0 from 1.0.11: > Merge pull request adafruit/Adafruit_CircuitPython_floppy#11 from adafruit/updates-incompatible-floppyio-floppsy Updating https://github.com/adafruit/Adafruit_CircuitPython_SharpMemoryDisplay to 1.4.12 from 1.4.11: > Merge pull request adafruit/Adafruit_CircuitPython_SharpMemoryDisplay#24 from makermelissa/master Updating https://github.com/adafruit/Adafruit_CircuitPython_Thermistor to 3.4.11 from 3.4.10: > Merge pull request adafruit/Adafruit_CircuitPython_Thermistor#22 from adafruit/dhalbert-patch-1 Updating https://github.com/adafruit/Adafruit_CircuitPython_Wiznet5k to 5.2.1 from 5.0.9: > Merge pull request adafruit/Adafruit_CircuitPython_Wiznet5k#151 from pinkavaj/pi-fix-recv > Merge pull request adafruit/Adafruit_CircuitPython_Wiznet5k#157 from adafruit/core-compatible-socket-type-numbers > Merge pull request adafruit/Adafruit_CircuitPython_Wiznet5k#153 from us3r64/fix/socket-httpserver-poll > Merge pull request adafruit/Adafruit_CircuitPython_Wiznet5k#149 from pinkavaj/pi-fix-close-wait > Merge pull request adafruit/Adafruit_CircuitPython_Wiznet5k#156 from kevin-tritz/main Updating https://github.com/adafruit/Adafruit_CircuitPython_BLE to 10.0.7 from 10.0.6: > Merge pull request adafruit/Adafruit_CircuitPython_BLE#195 from adafruit/dhalbert-patch-1 Updating https://github.com/adafruit/Adafruit_CircuitPython_BusDevice to 5.2.9 from 5.2.8: > Merge pull request adafruit/Adafruit_CircuitPython_BusDevice#99 from RetiredWizard/spicsinit Updating https://github.com/adafruit/Adafruit_CircuitPython_Bundle/circuitpython_library_list.md to NA from NA: > Updated download stats for the libraries
failures. The issue is caused by the code checking socket status *after* checking if bytes are available. Bytes may have become available between the last byte available check and the socket status check. This causes us to incorrectly ignore the newly available bytes and return 0 to the caller, which in any blocking/timeout scenario tells the caller no more bytes will ever be available. This was introduced in commit 89b9f10 See also adafruit#151 as a fix for adafruit#135
We cannot get more date on closed socket, return what we have, if anythink.
Closes #135