I've been investigating a special issue on our docker host server (17.09.0-ce) and found that every 3-5 minutes, many connections disappear from /proc/net/nf_conntrack causing the client side of the connection to time out.
I'm on CentOS 6.
I didn't see any special packets that may cause this behavior (I used tshark).
I don't see any errors in /var/log/messages or in dmesg.
Also, it looks like the sockets are not closed though as they seem to be remain in the output of netstat -atp.
I'm kind of stuck here at the moment and I have no other idea how to find the root cause of this. How should I debug this issue further?