Prevent neutron-l3-agent killing keepalived on restart

Systemd processes use a default KillMode of 'control-group'
which causes all other processes spawned during execution to be
killed on service stop. Neutron expects the keepalived processes
it starts to remain running in order to prevent data-plane
interruptions for HA routers.

This change switches the systemd KillMode to process in order to
prevent this issue. In doing so we also have to clean up
non-keepalived processes started by neutron so that upon restart
everything is running from the latest virtualenv which may have
changed during an upgrade.

Change-Id: I958fda17e6207553466d8a7512e35c30b122c22c
Closes-Bug: #1846198
Depends-On: https://review.opendev.org/771770
(cherry picked from commit a640ae8cc9)
This commit is contained in:
Andrew Bonney 2021-01-21 10:19:05 +00:00
parent 8928157640
commit d12a8aa6c5
2 changed files with 31 additions and 1 deletions

View File

@ -49,6 +49,27 @@
- "Restart neutron services"
- "venv changed"
# NOTE
# When restarting neutron-l3-agent, a non-default systemd KillMode of 'process' is used
# to prevent Keepalived from exiting and causing a data-plane outage. As a result of this
# some neutron processes remain running. In the case of an upgrade, these remaining
# processes will be running code from the previous version. This step ensures these
# orphaned processes are cleaned up correctly.
- name: Run neutron-l3-agent process cleanup
shell: |
for ns_pid in $(cat /sys/fs/cgroup/pids/neutron.slice/neutron-l3-agent.service/cgroup.procs); do
echo $(readlink -f "/proc/$ns_pid/exe") | grep -qv "keepalived"
if [ $? -eq 0 ]; then
if kill -9 "$ns_pid"; then
logger -s "old neutron-l3-agent pid found and has been cleaned up on: \"$ns_pid\""
fi
fi
done
when: "'neutron-l3-agent' in (filtered_neutron_services | map(attribute='service_key') | list)"
listen:
- "Restart neutron services"
- "venv changed"
- name: Perform a DB contract
command: "{{ neutron_bin }}/neutron-db-manage upgrade --contract"
become: yes

View File

@ -415,7 +415,7 @@ neutron_services:
execstarts: "{{ neutron_bin }}/neutron-l3-agent --config-file {{ neutron_conf_dir }}/neutron.conf --config-file {{ neutron_conf_dir }}/l3_agent.ini"
config_overrides: "{{ neutron_l3_agent_ini_overrides }}"
config_type: "ini"
init_config_overrides: "{{ neutron_l3_agent_init_overrides }}"
init_config_overrides: "{{ _neutron_l3_agent_init_defaults | combine(neutron_l3_agent_init_overrides) }}"
start_order: 3
neutron-bgp-dragent:
group: neutron_bgp_dragent
@ -508,6 +508,15 @@ neutron_services:
init_config_overrides: "{{ neutron_ovn_metadata_agent_init_overrides }}"
start_order: 3
###
### Overrides for services where required
###
# Ensure that keepalived processes are not stopped on exit
_neutron_l3_agent_init_defaults:
Service:
KillMode: process
###
### Internals: Drivers mappings
###