A rebuilt host that dropped connections under load, and a kernel parameter set by hand in 2020.
/etc/sysctl.d/ on the old host:
net.core.somaxconn = 4096
net.ipv4.tcp_max_syn_backlog = 8192
on the new host: the distribution defaults, 4096 and
1024 respectively.
the symptom: connection resets at about 1,200 concurrent,
under a load test that the old host passed.
and nothing in git, no comment, no ticket. found by
diffing sysctl -a between the two hosts.
Diffing the full kernel parameter set between the old and new host is the technique, and it produced two genuine settings out of about forty differences that were distribution defaults changing between releases. Both are now in the provisioning script with a comment naming the load test that justifies them, which is the documentation that should have existed in 2020.