when connecting to OGS, got the following error message
Unknown speed rapid, cannot check challenge, please try again. (mostly)
TypeError: NetworkError when attempting to fetch resource. (occasionally)
and a sign of no signal showing on the top right corner:
Are you able to test this against LHR ipv6 forward proxies:
The connection reliably stalls via CF only. It has nothing to do with OGS specifically AFAICT, but something either in LHR forward proxies, or I guess it’ts feasible BT could be doing something weird for ipv6 only and for CF peering only, but that sounds pretty tenuous.
Anyway, it’s just not correct that CF is the most stable path for BT users, the CF path is in fact 100% broken for any connection that still needs to send requests more than 20s into the connection.
It is an ordinary HTTPS (HTTP over TLS) request. It is not a websocket.
If you have issues with ipv6 try to disable it. But it is definitely not related to ipv6. I saw network errors on OGS. I use ipv4.
P. S. provider cannot distinguish between websocket or non websocket connection over HTTPS. Transmission is fully encrypted. Provider does not see headers or any data. The only exposed is addresses and domain name in SNI. Persistent TCP connection on 80/443 does not mean usage of websocket. Since http 1.1 there is a keep alive mechanism allows a single TCP connection to remain open for multiple HTTP requests. OGS supports http 2 and http 3. Both allow multiple concurrent requests/responses to be multiplexed over a single connection. There is no way for provider to detect websocket.
Yes, the OP problem is definitely unrelated to the network issue, I’m not sure what’s going on with that. I just responded to you to let you know that the connections through CF are definitely broken for websockets, so you shouldn’t advise people to switch to CF.
But the OGS connections via CF are definitely terminated by CF and then proxied to OGS from CF, the response headers contain a bunch of CF stuff:
And you are right that the problem isn’t specific to websockets, it’s just ipv6 connections to any of online-go.com IPs, from BT (but I suspect just anyone terminating in LHR):
online-go.com. 300 IN AAAA 2606:4700:20::681a:24
online-go.com. 300 IN AAAA 2606:4700:20::681a:124
online-go.com. 300 IN AAAA 2606:4700:20::ac43:4912
The connections just stall after 20s (no more TCP packet payloads get through and there are a bunch of retransmission attempts from both server to client and client to server). It just only really affects websockets, because they expect to be able to send/receive after 20s while most other connections have sent/received all their content by that point.
Yes, all other connections I’ve been able to test are fine. Even connecting to any other CF IPs (eg: 2606:4700:20::681a:23 rather than 2606:4700:20::681a:24) is fine. Connecting to these specific IPs from elsewhere is also fine.
So as far as I can tell it’s just these specific CF IPs, when you connect from BT (or if my suspicion is correct, anywhere that terminates in LHR) that the connections stall.
This looks like very suspicious. There is only last 3 bits difference.
I sent a very silly question to CF about maintaining tcp over different routes to different edge endpoints. I do not believe I am lucky enough to get a response. But there is a chance.