The same output a minute later:
tim@ferrari# sudo conntrack -S
cpu=0 found=11150 invalid=48766 insert=0 insert_failed=1691
drop=1691 early_drop=0 error=1792 search_restart=0 clash_resolve=901897
chaintoolong=0
cpu=1 found=10497 invalid=48631 insert=0 insert_failed=1748
drop=1748 early_drop=0 error=1422 search_restart=0 clash_resolve=907859
chaintoolong=0
So ~300 clash_resolve's a minute.
Replying to myself, but documenting it for future reference in case
anyone else stumbles on this.
My router has a DNS server on it. Every query from a LAN client to the
DNS server, even though it's the same subnet, created a conntrack entry.
I have a Zabbix server that generates A LOT of DNS queries, it was this
Zabbix server and its huge amount of DNS queries causing the
clash_resolve to kick in.
The fix has been to put in a conntrack ignore rule, so that queries to
the DNS server from the local LAN aren't conntracked (They don't need to
be).
This has
a) Removed a bunch of DNS conntrack entries I didn't need taking up
conntrack space
b) Stopped clash_resolve from incrementing at all.
Hope this helps someone in the future.
Tim
