I’ve been running a standalone Snowflake proxy on EndeavourOS (Arch Linux). It was working fine, usually around 10-30 conn/hour.
On 24 July, it suddenly dropped to 0 successful connections (Jul 24 18:02:31 in logs). Strange part is no package updates, system updates, configuration changes, network changes, etc.
Environment — Snowflake Proxy 2.13.1 (75251a4c*), snowflake-pt-proxy downloaded from AUR and ran as Systemd service.
Logs from starting proxy in verbose mode (snowflake-proxy -unsafe-logging -verbose
)
2026/07/26 02:28:44 Received Offer From Broker:
...
2026/07/26 02:28:44 Generating answer...
2026/07/26 02:28:47 ICE gathering is not yet complete, but let's send the answer before the client times out
2026/07/26 02:28:47 Answer:
...
2026/07/26 02:28:48 error sending answer to client through broker: error sending answer to broker: remote returned status code 400
What could be the problem and how to resolve it?
I publish connection number logs on github repo (https://github.com/LakshitSinghBishtTM/snowflake) in case need connection drop logs. (check phase 2 daily logs)
I checked your logs and see nothing which stands out. but this looks weird: Jul 25 17:56:36 mac snowflake-proxy[853]: 2026/07/25 12:26:36 NAT type: unknown
It should be able to figure that out.
Error code 400 suggests a problem on your side.
I would not put much faith in the message: In the last 1h0m0s, there were 22 completed successful connections.
My analysis show that many connections die within 30 seconds but are reflected in that message.
The real count is how many people are connected NOW.
This is the count of how many connections you have 01-snowflake and 02-snowflake (193.187.88.42:443 and 141.212.118.18:443 respectively)
An example from mine is: 2026/07/25 04:26:25 In the last 1h0m0s, there were 23 completed successful connections. Traffic Relayed ↓ 34290 KB (9.53 KB/s), ↑ 6649 KB (1.85 KB/s).
which is impossible since I have use -capacity 14 on the startup command.
To get a real idea try: netstat -anp | grep -i 'proxy' and count them
OR
netstat -anp 2> /dev/null | grep -i ‘proxy’ | grep -i -E -c '141.212.118.18|193.187.88.42
OR
ss -tpn | grep -i ‘proxy’ | grep -i -E -c ‘141.212.118.18|193.187.88.42’
It’s almost as if you lost contact with the broker as impossible as that seems.
Thanks for replying. Yesterday evening, I was trying to fix my internet speed. I was having an unrelated issue of too slow internet like wifi is around 400mbps but I was getting only 2kbps) and my dns was 61.1.1.1 (bsnl dns auto selected, I didn’t wrote it as it is bsnl wifi. I first thought 6 is typo but on internet search found out that its bsnl dns).
GPT said it won’t do anything useful, and problem may be in my drivers and kernel as we had done checks and tested a lot of things but nothing had worked. I thought to change to 1.1.1.1 as I had nothing to lose. After changing it, the internet speed increased and snowflake also automatically started working.
However, it’s not formally proved it was dns but it happened at same time, so I guess it was dns issue to be honest. I would say it’s more like hypothesis than proof. But at the end, snowflake started working so I am glad. Also, thanks again for the help and reply.