I keep getting “Failed to establish connection to TeamSpeak server! Reason: failed connection initialization” on Windows when I connect right after Tailscale comes up. The server is on my tailnet too (100.x.x.x:9987), so all the traffic goes over Tailscale. I captured packets on a failing Windows run and a working Mac run.
In the failed run the first server reply took 823 ms. The client had already resent its first two packets at about 500 ms, so the server answered twice and the client got duplicate 35- and 201-byte replies. After that it kept sending the 38-byte packet and never sent the 482-byte one. Pressing Retry about 35 s later (same client session, same local UDP port 49961) failed too, even though the reply came back in 189 ms that time.
My Mac connects to the same server with the same client build in 0.6 s, but Tailscale on the Mac had been up for a while (first reply in about 190 ms). I tried the Mac once right after restarting Tailscale too: the first attempt got no reply at all and the retry about 11 s later worked. I didn’t capture packets for that one.
On Windows, I also tried pinging the server a few times before connecting, and it connected on the first try that time. I’ve only tested that once.
| Windows | Mac | |
|---|---|---|
| Client | 6.0.0-beta5 (2026-10-13 14:20:51) | same build |
| OS | Windows 11 26H2 (10.0.26300) | macOS, Darwin 27.0.0 (arm64) |
| Tailscale | 1.102.4 | 1.104.4 |
Server: 6.0.0-beta13.1
Steps:
- Connect Tailscale.
- Right away, connect to the server from a bookmark.
- When the error shows up, hit Retry.
I’d expect it to connect, or at least the retry to work. Both attempts fail.
Captures below are trimmed (pktmon on Windows, tcpdump on the Mac). Sizes are UDP payload bytes, times are local.
Working, Mac:
21:29:12.857 client -> server 34
21:29:12.858 client -> server 184
21:29:13.050 server -> client 35
21:29:13.050 client -> server 38
21:29:13.243 server -> client 201
21:29:13.247 client -> server 482
21:29:13.441 server -> client 13, 500, 138 ... connected
Failed attempt 1, Windows:
21:20:18.233 client -> server 34
21:20:18.233 client -> server 183
21:20:18.740 client -> server 183 <- resent, no reply yet
21:20:18.740 client -> server 34 <- resent
21:20:19.056 server -> client 35
21:20:19.056 server -> client 35 <- duplicate
21:20:19.056 client -> server 38
21:20:19.056 client -> server 38
21:20:19.249 server -> client 201
21:20:19.249 server -> client 201
21:20:19.562 client -> server 38 <- never sends the 482
21:20:19.751 server -> client 201
21:20:20.230 client -> server 38
21:20:20.424 server -> client 201
... 38 -> 201 keeps going until 21:20:24, then the error
Client log for that attempt. There are three “Selected cipher” lines at the times of the first three 201 replies, then no more, although those replies keep arriving:
[2026-11-04 21:20:18.230] [info] Connecting to resolved address 100.x.x.x:9987
[2026-11-04 21:20:19.056] [info] server sent ciphers:35
[2026-11-04 21:20:19.056] [info] server sent ciphers:35
[2026-11-04 21:20:19.249] [info] Selected cipher:1
[2026-11-04 21:20:19.249] [info] Selected cipher:1
[2026-11-04 21:20:19.751] [info] Selected cipher:1
Failed attempt 2, Windows (the Retry):
21:20:53.458 client -> server 34
21:20:53.458 client -> server 183
21:20:53.647 server -> client 35 <- 189 ms, but no "server sent ciphers" in the log
21:20:53.962 client -> server 183 <- resent after ~500 ms
21:20:53.962 client -> server 34
21:20:54.155 server -> client 35
... same until 21:20:58, then the error
The client never sends the 38 here. I don’t know why it skips that reply. Maybe something is left over from attempt 1 on the same socket, but I can’t tell from my captures.
DNS resolved fine both times, and the client didn’t crash or restart in between.
My guess for the first attempt is that the 500 ms resend duplicates the handshake when the first round trip is slow, and the client doesn’t cope with the doubled replies. Happy to retest on a newer build, and I can send the full client logs and the raw captures to the team privately.