Based on the behavior you described, I’d try to isolate the issue by destination IP first.

Since the exact same endpoint sometimes returns HTTP 401 and other times fails with IPv4 ETIMEDOUT, the first step is to check whether the failures are tied to a specific IPv4 address returned by DNS.

You can start with:

getent ahostsv4 ws-douf5b4ymvzf09du.cn-beijing.maas.aliyuncs.com

Then run multiple requests while logging the remote IP used for each connection:

for i in $(seq 1 30); do
  curl -4 \
    --connect-timeout 5 \
    --max-time 10 \
    -sS \
    -o /dev/null \
    -w 'IP=%{remote_ip} HTTP=%{http_code} CONNECT=%{time_connect}s TOTAL=%{time_total}s\n' \
    https://ws-douf5b4ymvzf09du.cn-beijing.maas.aliyuncs.com/ \
    || echo "FAILED"
  sleep 1
done

If you notice that one destination IP consistently fails while another works, test each IP directly using --resolve:

curl -4 -v \
  --connect-timeout 5 \
  --resolve "ws-douf5b4ymvzf09du.cn-beijing.maas.aliyuncs.com:443:DESTINATION_IP" \
  https://ws-douf5b4ymvzf09du.cn-beijing.maas.aliyuncs.com/

That should help confirm whether the problem is tied to a specific Alibaba IP or route.

I wouldn’t focus too much on the IPv6 ENETUNREACH for now. The main issue to isolate is the intermittent IPv4 ETIMEDOUT.

If all IPv4 addresses fail randomly, I’d also run mtr or traceroute toward the destination IPs and record the exact UTC timestamps of the failures.

At that point, you should have enough information to open a Railway support ticket with useful evidence:

• exact UTC timestamps
• destination IP for each attempt
• which IPs succeed and which fail
• curl results
• MTR/traceroute results
• service region: US West

If the failures consistently follow one destination IP, it’s likely a route/backend-specific issue. If they happen randomly across all IPs, I’d suspect the Railway egress/NAT/upstream path instead.

One more useful test would be deploying the same service temporarily in another Railway region. If the Alibaba endpoint becomes stable there, that would strongly suggest the issue is specific to the US West outbound path.