Currently, the sync script searches for running processes and if the target one is not running, it'll start it.
This is the logic:
IFS=$'\n'
for LINE in $(cat {{ tripleo_systemd_wrapper_service_dir }}/{{ tripleo_systemd_wrapper_service_name }}/processes); do
NETNS=$(echo $LINE | awk '{ print $1 }')
IFS=$' ' ARGS=$(echo $LINE | sed -e "s|$NETNS ||" | xargs)
# TODO(emilien) investigate if we should rather run docker/podman ps instead of ps on the host
if ! ps -e -o pid,command | grep "$(echo $NETNS | sed 's|^[^-]*\-||')" | grep -v grep &> /dev/null; then
start_service $NETNS $ARGS
fi
done
However, the command itself invoked by the Neutron agent may still show up in the 'ps' output which makes the sync script to think it's running and hence skip its start. This may delay the execution of the sidecar container until the next iteration (1 minute) which in the case of the metadata container (for both ML2/OVS and ML2/OVN) it may be already too late as the cloud-init of the instance had given up.
Example of the 'ps' output when the issue is being hit:
Mar 19 11:12:31 compute-0 sync[92924]: 92914 /usr/bin/python3 /usr/bin/neutron-rootwrap /etc/neutron/rootwrap.conf ip netns exec ovnmeta-31c07b24-f920-4a24-a499-8a5bc16ed44d haproxy -f /var/lib/neutron/ovn-metadata-proxy/31c07b24-f920-4a24-a499-8a5bc16ed44d.conf
Fix proposed to branch: master
Review: https:/