Thank you for taking the time to write this — this kind of structured feedback is much more useful than an emotional reaction.
I have verified the three points you raised. Here is a brief, direct response:
Image Acceleration — "The version is very old" is actually a polite way of putting it. Currently, our image acceleration for docker.io on mainland servers only has rewrite rules for certain paths; other third-party images (including the New API you mentioned) are rejected with a 401 at the image layer, which manifests as "unable to pull." To be honest, this is harder than it looks: unconditional proxying of Docker Hub in the PRC is a compliance challenge. Services like DaoCloud aren't unconditionally available either, and self-hosting SWR runs into network routing issues. A more reasonable short-term direction is for us to clearly prompt users when launching mainland servers: "Docker Hub pulls may be unstable; please provide your own mirror (push to ACR/SWR/your own registry)," returning that choice to you rather than pretending our acceleration covers 100% of cases.
Zeabur ↔ Mainland Server Connection Stability — This is real and more specific than expected. We already have api.zeabur.cn for the HTTP control plane (which you measured at 4.7ms RTT and 0% packet loss on your mainland machine), and the registry already uses .cn. However, the NATS layer (the message bus for agent event reporting) currently only has nats.zeabur.com and no .cn version. Consequently, the agent on your mainland machine is still connecting across borders to NATS on GCP, resulting in zeabur-kube-watch failing 7 times in 10 days (compared to 0 times for the same agent on your Singapore machine). What's missing here is a nats.zeabur.cn endpoint.
node-exporter Reporting Errors — The pod itself is healthy (1 restart in 10 days). The "error" you see is a downstream effect of the unstable NATS pipeline mentioned above—the agent fails to report, and the Dashboard marks the node as unhealthy. Attributing this to node-exporter is a UX bug.
I will bring these findings back to the relevant internal teams. You are welcome to check back later.
Have fun 👋
