Description of problem: During a performance test over an Octavia loadbalancer customer encountered nf_conntrack table full errors. which resulted in errors in the requests. [root@amphora-a0ce9751-247d-48ab-9b13-53c6162beba1 ~]# dmesg -T [Mon Aug 17 06:16:28 2020] nf_conntrack: table full, dropping packet [Mon Aug 17 06:16:28 2020] nf_conntrack: table full, dropping packet Version-Release number of selected component (if applicable): - RHOSP15 - Customer using octavia-amphora-image provided by RedHat CDN and not any customized image - [cloud-user@amphora-d1351773-8d87-46ef-aeca-146c2460c159 ~]$ rpm -qa | grep octavia openstack-octavia-amphora-agent-3.1.0-3.el7ost.noarch openstack-octavia-common-3.1.0-3.el7ost.noarch python-octavia-3.1.0-3.el7ost.noarch - [root@srv-zmb15-12 (PRD-ZM) ~]# rpm -qa | grep octavia openstack-octavia-common-4.1.2-0.20200114080449.5a71643.el8ost.noarch python3-octavia-lib-1.1.1-0.20190312101220.6b55d62.el8ost.noarch python3-octavia-4.1.2-0.20200114080449.5a71643.el8ost.noarch openstack-octavia-health-manager-4.1.2-0.20200114080449.5a71643.el8ost.noarch openstack-octavia-worker-4.1.2-0.20200114080449.5a71643.el8ost.noarch openstack-octavia-housekeeping-4.1.2-0.20200114080449.5a71643.el8ost.noarch How reproducible: Steps to Reproduce: 1. 2. 3. Actual results: getting nf_conntrack: table full, dropping packet error. Expected results: should not get 'dropping packet error ' as nf_contrack_bucket in amphora namespace should have the increased value loaded from the sysctl.d conf file. Additional info: -- Customer has opened an storyboard with more details: https://storyboard.openstack.org/#!/story/2008028 The sysctl file to increase nf_conntrack_buckets is set: [root@amphora-a0ce9751-247d-48ab-9b13-53c6162beba1 ~]# cat /etc/sysctl.d/net.netfilter.nf_conntrack_buckets.conf # This file was created by diskimage-builder. net.netfilter.nf_conntrack_buckets = 125000 But when checked the apmhora-netns we can see that setting is ignored: [root@amphora-a0ce9751-247d-48ab-9b13-53c6162beba1 ~]# ip net exec amphora-haproxy cat /proc/sys/net/netfilter/nf_conntrack_buckets 8192 [root@amphora-a0ce9751-247d-48ab-9b13-53c6162beba1 ~]# cat /proc/sys/net/netfilter/nf_conntrack_buckets 8192 Note: The current error disappeared when the RAM of amphora instances were increased then customer could see the increased bucket size in amphora namespace. but it has to be cal
Lowered priority because workaround identified in customer case.
I have the same issue when the file '/etc/sysctl.d/net.netfilter.nf_conntrack_buckets.conf' exists with 125000 value inside, but after amphora VM boot this setting is not applied. In journalctl I can see: Dec 02 11:11:08 ubuntu systemd-sysctl[253]: Couldn't write '125000' to 'net/netfilter/nf_conntrack_buckets', ignoring: No such file or directory It seems the issue in how sysctl settings are applied and depends on if nf_conntrack module is loaded or not.
(In reply to Anton from comment #5) > I have the same issue when the file > '/etc/sysctl.d/net.netfilter.nf_conntrack_buckets.conf' exists with 125000 > value inside, but after amphora VM boot this setting is not applied. > > In journalctl I can see: > Dec 02 11:11:08 ubuntu systemd-sysctl[253]: Couldn't write '125000' to > 'net/netfilter/nf_conntrack_buckets', ignoring: No such file or directory > > It seems the issue in how sysctl settings are applied and depends on if > nf_conntrack module is loaded or not. I proposed https://review.opendev.org/c/openstack/octavia/+/748749/ it should fix the issue but it's still in review
Since the problem described in this bug report should be resolved in a recent advisory, it has been closed with a resolution of ERRATA. For information on the advisory (Red Hat OpenStack Platform (RHOSP) 16.2 enhancement advisory), and where to find the updated files, follow the link below. If the solution does not work for you, open a new bug report. https://access.redhat.com/errata/RHEA-2021:3483