Race condition

[maxpain]

https://github.com/metallb/metallb/blob/main/internal/layer2/announcer.go#L158
There is a race condition.

We have a big cluster with many pod creation/deletion operations, so the interfaces list on a node changes very often.
In our cluster MetalLB sometimes doesn't announce IPs (L2/ARP) because, at the moment when gratuitous function executes, one of the interfaces in a.arps array becomes dead and this loop crashes with an error, so no more gratuitous announces will be sent.

I think we should ignore these errors and don't crash the loop.

[maxpain]

@danderson

[harii9]

HI @Maxpain177,

We want to reproduce this issue. Can you share the details of the test environment(Subsequent deletion/Creation of pods)