After upgrading to the latest prometheus-node-exporter build from today's master branch (commit 6523fdfc4b20f42269c849cde95a413c232d38a0) with procfs 0.7.3, we see some clusters have failing lnstat collectors:
prometheus-node-exporter[59966]: level=error ts=2021-10-28T15:16:33.821Z caller=collector.go:169 msg="collector failed" name=lnstat duration_seconds=0.000423308 err="Lnstat error: strconv.ParseUint: parsing \"125ee882a\": value out of range"
The hex values comes from /proc/net/stat/arp_cache:
$ head -n2 /proc/net/stat/arp_cache
entries allocs destroys hash_grows lookups hits res_failed rcv_probes_mcast rcv_probes_ucast periodic_gc_runs forced_gc_runs unresolved_discards table_fulls
00000431 00223378 00425a10 00000007 126152242 122ba4ce5 000352a9 00000000 00000000 00028463 00000000 00000000 00000000
These hosts use a custom linux 5.4 kernel.