Dual-Stack: Access IPv6 addresses from pod

[slaecker]

Hi all - I'm having a hard time connecting to IPv6 addresses from pods. In some pod logs I noticed errors regarding connection timeouts to certain hostnames. I tracked this down to hosts resolving to IPv4 addresses work fine but IPv6 fails.

My Setup:

• K3s 1.28.6 on 1 server / 2 agents installed via K3s install script
• Server and agents have IPv4 (static) and IPv6 (dhcp) addresses assigned
• Server is installed with

--cluster-cidr=10.42.0.0/16,2001:cafe:42::/56 --service-cidr=10.43.0.0/16,2001:cafe:43::/112
--kubelet-arg node-ip=0.0.0.0

• Agents are installed with

--kubelet-arg node-ip=0.0.0.0

• Nodes have IPv4 and IPv6 addresses assigned
• IP forwarding is enabled on server and agents

net.ipv4.ip_forward = 1
net.ipv4.conf.all.forwarding = 1
net.ipv4.conf.default.forwarding = 1

net.ipv6.conf.all.forwarding = 1
net.ipv6.conf.default.forwarding = 1

When I'm starting an alpine test deployment and access the console I can ping IPv4 addresses normally but any IPv6 address fails.

Examples:

/ # ping amazon.de
PING amazon.de (176.32.108.185): 56 data bytes
64 bytes from 176.32.108.185: seq=0 ttl=239 time=35.165 ms
64 bytes from 176.32.108.185: seq=1 ttl=239 time=33.502 ms
64 bytes from 176.32.108.185: seq=2 ttl=239 time=33.723 ms
^C
--- amazon.de ping statistics ---
3 packets transmitted, 3 packets received, 0% packet loss
round-trip min/avg/max = 33.502/34.130/35.165 ms

/ # ping cloudflare.com
PING cloudflare.com (2606:4700::6810:85e5): 56 data bytes

Server logs don't show any issues regarding IPv6. Server and agents can perform both ping commands successfully, only from within the cluster it doesn't work.

Any hints are highly appreciated.

[brandond]

Have you seen: https://docs.k3s.io/installation/network-options?_highlight=flannel-ipv6-masq#flannel-options

[slaecker]

No, indeed I haven't seen this. Thanks a lot, it solved the issue.