Closed noahlevenson closed 1 year ago
I'm pretty sure I see this -- or something like it -- happening regularly, particularly now that the services are deployed on Heroku.
Are middleboxes closing our extensions' idle websocket connections?
I haven't seen this since we added a keepalive, so I'm closing it for now...
Tonight, I ran a little experiment just to assure myself that our N:M multiplexing still works:
I spun up a single uncensored peer using
./widget
I spun up 3 censored peers using
./derek.sh 3
The 3 censored peers all established connections through the single widget
I made sure that each censored peer could round trip requests by hitting
curl --proxy http://127.0.0.1:108n https://www.nytimes.com
for n = [1, 3]The test passed and everything worked great. I left the widget and the censored peers running while I attended to other tasks. When I came back, I saw that the widget had mysteriously experienced WebSocket errors:
The widget did what it's supposed to do -- it detected the dead WebSocket and dialed up a new WebSocket connection. But what's going on and why?