Routing device discovery SIGSEGV about cilium HOT 2 CLOSED

Hey, the direct routing device detection picks the device that has Node IP address assigned on it when there are multiple devices. It seems like in your case it's assigned onto a bridge device. The device detection code however currently ignores bridge devices (this was already the case in v1.14), but the recent change is that now the code to find the device on which the Node IP is assigned to is relying on the detected devices when previously it considered all devices. I would argue this is the correct behavior as a lot of features depend on the correct devices being chosen for datapath use, so it's better to be stricter and require the "--devices" to be specified when the default detection fails.

The other issue here is the crash when the agent is being stopped. This is due to DNSProxy not being properly stopped and it accessing a global variable that was already cleaned up. We'll look into that.
EDIT: Fix for the crash on stop in #32673. I'll look into backport to v1.15 a bit later.

from cilium.

If I remember correctly, 1.14.7 cilium was indeed choosing the bride as direct routing device. You are totally right this was not supposed to be the case, as the node IP was supposed to be taken from another vlan interface.

AFAIK supplying --devices did not work at the time but I'm pretty sure the bridge was also in the list (something like --devices="bridge01,vlan01,vlan02" where vlan01 was holding the right IP). Maybe that's were the issue was coming from.

Thanks for the clarification though ! Again this issue was mostly to report the SIGSEGV in case that wasn't discovered or fixed already, I'm glad everything works as expected then.

from cilium.