<XioNoX> unrelated I think, but seems like machines catergorized as "cluster:labs" in eqiad had a steady rate of IP hardware IN errors: https://grafana.wikimedia.org/d/000000366/network-performances-global?panelId=21&fullscreen&orgId=1&from=now-1h&to=now
https://tools.ietf.org/html/rfc1213#page-27
8:35 PM this is a bit more clear
8:35 PM but tldr, something wrong in the ip header
8:35 PM "The number of input datagrams discarded due to
8:35 PM errors in their IP headers, including bad
8:35 PM checksums, version number mismatch, other format
8:35 PM errors, time-to-live exceeded, errors discovered
8:35 PM in processing their IP options, etc."