[x-post from (URL Removed by Staff)]
I have been using Cloud Batch and discovered an issue that seems like should be more widely known or handled by Google already. I have been working the GCP support team for many weeks on this, but I also want to check with the community on whether people have missed the issue entirely or it's actually a minor issue not warranting attention.
On a VM with a COS or Debian disk image, no external IP, and a network configured with Cloud Router and NAT, if I try to connect to a remote server (can be any server outside of the Google network) in the startup script, the first connection takes 20-40s. If in my startup script, I just sleep for 40s first and then try to connect, the connection is immediate. The command I used to test was `curl -v --trace-time https://microsoft.com`.
This would normally not be an issue other than the increased startup latency, but several applications (like docker daemon) have lower connection timeouts and steps invoking these apps in the startup script would always fail unless they are put in a retry loop or if the connection timeout is somehow increased.
I would imagine that an extra 40s of network bootstrap delay would be a serious issue for Google to look into, or acknowledge more publicly as an issue.
LinkedIn
Twitter
