Martynas Pumputis
Martynas Pumputis
> I try it with the latest release. It does not happen again. Thanks for confirming!
Is it still WIP (as the tittle suggests) or is it ready for a review?
I wouldn't bother with checking the kernel version, and I would enable the safe-delete for all. Also, as we already bookkeeping all ipset elements in NPC, wouldn't it better to...
E.g. https://github.com/weaveworks/weave/blob/master/npc/selector.go#L170. I'm suggesting to always sync with the tracked by NPC ipsets as these are the source of truth.
Yep, we can close it - ci-e2e-upgrade, ci-ipsec-ugprade, ci-clustermesh do test the upgrade/downgrade with non-defaults.
The [ci-e2e-upgrade](https://github.com/cilium/cilium/pull/27322) PR is hitting this too (1 out of ~3 times). For example, https://github.com/cilium/cilium/actions/runs/6147703736/job/16679772837.
@giorio94 Mentioned that the drop happens almost at the same time when the related client endpoint gets regenerated. The drop was observed only when both IPv{4,6} and the bpf_lxc per-packet...
Small update. Added bpftrace to the rescue: ``` cat > trace.bt id == $1) { printf("%s %s:%d->%s:%d : %x\n", strftime("%H:%M:%S:%f", nsecs), ntop(((struct key *)((struct bpf_map*)args->key))->saddr), bswap(((struct key *)((struct bpf_map*)args->key))->dport), ntop(((struct...
> (If pwru has ability to trace helper functions and parse kv automatically.... Yeah... /me is looking at you :sweat_smile: