Description of problem: When performing neutron scale testing on OSP16.1, we see OVSDB Error: transaction failed messages in neutron/server.log and even the ovn is dead. https://gist.github.com/asyedham/ad441ce7ce7c46c2dd0e6d6aec1bd096 This is the rally plugin used(create network, subnet, boot server with fip and ping)for a total of 2000 times at a concurrency of 16: https://github.com/openstack/browbeat/blob/master/rally/rally-plugins/netcreate-boot-ping/netcreate_nova-boot-fip-ping.py Note: Neutron resources are still persistent where as nova is able to delete the resource after the workload is run. Version-Release number of selected component (if applicable): Red Hat OpenStack Platform release 16.1.4 GA (Train) How reproducible: 100% Steps to Reproduce: 1. Run the scenario netcreate-boot-ping 2. Observe the results and neutron logs Rally test and errors log is at http://perf1.perf.lab.eng.bos.redhat.com/pub/asyedham/neutron-scale-testing/results/20210225-094725/rally/simple-plugins/netcreate-boot-ping/20210225-094725-browbeat-netcreate-boot-ping-16-1-iteration-0.log PFA attached neutron, openvswitch and nova logs captured during rally run http://perf1.perf.lab.eng.bos.redhat.com/pub/asyedham/neutron-scale-testing/neutron-scale-test/ grafana snapshot: https://snapshot.raintank.io/dashboard/snapshot/VaU3D8L7o5HPnCYkim0iXrNblkRYJlZT Additional info: ovn2.13 scratch build based on ovn2.13-20.12.0-19 with logical datapath groups enabled by default: http://brew-task-repos.usersys.redhat.com/repos/scratch/dceara/ovn2.13/20.12.0/20pvt_dp_groups.el8fdp/ Also, we have increased the timeouts from 300s to 3000s nova_server_boot_timeout = 3000.0 vm_ping_timeout = 3000.0