Outbound WebSocket disconnections every 7-8 minutes on K3s (Hetzner Nürnberg)

Resolved

TthetimecabinetsDEVELOPER4/28/2026

Hi,

I'm running a Python service on Zeabur using a Hetzner Nürnberg server (K3s). My service maintains a persistent outbound WebSocket connection to Binance (wss://stream.binance.com:9443).

I'm experiencing disconnections every 7-8 minutes consistently with this error: sent 1011 (internal error) keepalive ping timeout; no close frame received

What I've already tried:

Set ping_interval=5 on the WebSocket client (websockets library)
Set ping_timeout=10
Added open_timeout=30
Exponential backoff on reconnect

The ping fix reduced drops from every 3-5 minutes to every 7-8 minutes but did not eliminate them. This suggests the issue is at the K3s network layer, not the application layer.

My question: Is there a TCP idle timeout or keepalive policy configured at the K3s level for outbound connections on Hetzner servers? If so, can it be disabled or extended for persistent WebSocket connections?

My service requires a 24/7 stable outbound WebSocket connection. The reconnects are causing in-memory state to reset every 7-8 minutes which affects my application logic.

Thank you

Sonnet 4.6

2 Replies

BCBohan ChenEMPLOYEE4/29/2026

Hi, thanks for the detailed write-up. Quick answer:

Zeabur's K3s on Hetzner Nürnberg has no 7-8 min idle/keepalive timeout for outbound traffic. Checked your host: nf_conntrack_tcp_timeout_established=86400 (24h), standard flannel CNI, no egress rate-limiting. Pod is healthy, data flowing right now. So nothing on our side is introducing this.

Likely cause: stream.binance.com resolves to AWS Tokyo (ap-northeast-1). Hetzner Nürnberg → Tokyo is ~250ms RTT across long-haul backbones — persistent WebSockets over that distance are sensitive to transit routing hiccups and AWS edge LB connection recycling. 1011 keepalive ping timeout = pongs stopped coming back, which is a network/upstream symptom.

Recommended fixes, in order:

Move closer to Binance. A Zeabur dedicated server in Tokyo / Singapore / Hong Kong puts you same-region with Binance's AWS backend (RTT ~5-30ms). This is the standard fix for crypto-streaming services and almost always eliminates flap-style disconnects.
Drop ping_interval=5 to ~20-30s. 5s is ~36 pings/min — Binance's own server pings every 3 min and there are reports of overly-pingy clients getting silently throttled.
A/B test if you want to confirm the layer: run openssl s_client -connect stream.binance.com:9443 from inside the same pod and leave it idle. If it also drops at 7-8 min, the issue is upstream (network/Binance). If raw TLS holds but WS drops, it's WS-protocol-level. Either way it isolates Zeabur out of the picture.

Most reliable fix is #1 — moving to an Asian region.

AAnonymous5/3/2026

This post has been inactive for a while. We will be closing it in 2 days if there is no new activity.

This issue has been marked as resolved

New replies are disabled for resolved issues.

Forum

2 Replies