Description of problem:
One of our customers is running performance tests for his Web portal built on top of Shift on Stack environment. One of the problems we have found which has perfect correlation with client errors is spam of "nf_conntrack: table full, dropping packet" messages in amphora's console.
From the tests we can see that Octavia starts spamming this errors when "/proc/sys/net/netfilter/nf_conntrack_count" shows around 32000.
I have found two related bugs fixed in newer versions:
Bug/fix 1:
nf_conntrack: table full, dropping packet
https://bugzilla.redhat.com/show_bug.cgi?id=1869771 (fixed in RHOSP 16.2)
https://review.opendev.org/c/openstack/octavia/+/748749/ (fix)
Bug/fix 2:
https://storyboard.openstack.org/#!/story/2008979 (is not backported to RHOSP 16)
https://review.opendev.org/c/openstack/octavia/+/796608
It doesn't look like these fixes will be released for RHOSP 13, so I am wondering if there is some supported way to apply some workaround for this problem and prevent DoS situation for Amphora?
Version-Release number of selected component (if applicable):
Red Hat OpenStack Platform release 13.0.13 (Queens)