Routing problem. I worked around this problem by adding the following line to the startup-script:

ip -6 route add default via $(curl -fsS http://metadata.google.internal/computeMetadata/v1/instance/network-interfaces/0/gateway-ipv6 -H "Metadata-Flavor: Google") dev eth0

But I still think it should be handled automatically by some system component, so I leave this issue open for now.

Hi @aofei.

I'm not familiar with custom VPCs - could you share either gcloud commands or step by step instruction for the cloud console so I can replicate your setup for testing?

$ gcloud compute networks create archimg-issue-45 --subnet-mode custom
$ gcloud compute networks subnets create archimg-issue-45 --network archimg-issue-45 --region asia-east2 --range 10.10.0.0/24 --stack-type IPV4_IPV6 --ipv6-access-type EXTERNAL
$ gcloud compute firewall-rules create archimg-issue-45-allow-ssh --network archimg-issue-45 --allow tcp:22
$ gcloud compute instances create archimg-issue-45 --zone asia-east2-a --image-project arch-linux-gce --image-family arch --network-interface subnet=archimg-issue-45,stack-type=IPV4_IPV6 --metadata block-project-ssh-keys=true --no-shielded-secure-boot --no-shielded-vtpm --no-shielded-integrity-monitoring

If all goes well, these four lines of commands should create a problematic instance for you.

Then you can execute:

$ gcloud compute ssh archimg-issue-45 --zone asia-east2-a --command "ip a s eth0; ping -6 -c 3 dns.google"

and everything should be fine at this point. But if you then execute:

$ gcloud compute ssh archimg-issue-45 --zone asia-east2-a --command "sudo reboot"

the problem arises at this point. Either the instance is not reachable at all (no internet), or ip a s eth0 shows no external IPv6 address. The problem I'm having at this moment is the first one (no internet). The problem I had three days ago was the second.

the problem arises at this point. Either the instance is not reachable at all (no internet), or ip a s eth0 shows no external IPv6 address.

For the second case (ip a s eth0 shows no external IPv6 address), a workaround is to add the following line to the startup-script:

ip -6 addr add $(curl -fsS http://metadata.google.internal/computeMetadata/v1/instance/network-interfaces/0/ipv6 -H "Metadata-Flavor: Google" | sed "s/\/96/\/128/") dev eth0

I happened to have the same problem today (ip a s eth0 shows no external IPv6 address after sudo reboot) with an instance I created with ubuntu-2204-jammy-v20220528 image.

So I don't know, according to the IPv6 address assignment, this problem seems more likely to be caused by some metadata server bug in asia-east2 region than the VM OS now. Because I just tested the commands in #45 (comment) with us-central1 region, and everything works fine.

If it's a metadata server bug, then this issue can be closed. @toastwaffle