[Services] Restart DHCP-Relay service upon unexpected critical process exit. (#3667)

Signed-off-by: Yong Zhao <yozhao@microsoft.com>

dockers/docker-dhcp-relay/Dockerfile.j2

@@ -26,5 +26,7 @@ RUN apt-get clean -y         && \
 
 COPY ["docker_init.sh", "start.sh", "/usr/bin/"]
 COPY ["docker-dhcp-relay.supervisord.conf.j2", "wait_for_intf.sh.j2", "/usr/share/sonic/templates/"]
+COPY ["files/supervisor-proc-exit-listener", "/usr/bin"]
+COPY ["critical_processes", "/etc/supervisor"]
 
 ENTRYPOINT ["/usr/bin/docker_init.sh"]

dockers/docker-dhcp-relay/critical_processes

@@ -0,0 +1 @@
+isc-dhcp-relay

dockers/docker-dhcp-relay/docker-dhcp-relay.supervisord.conf.j2

@@ -3,6 +3,12 @@ logfile_maxbytes=1MB
 logfile_backups=2
 nodaemon=true
 
+[eventlistener:supervisor-proc-exit-listener]
+command=/usr/bin/supervisor-proc-exit-listener
+events=PROCESS_STATE_EXITED
+autostart=true
+autorestart=unexpected
+
 [program:start.sh]
 command=/usr/bin/start.sh
 priority=1

files/build_templates/dhcp_relay.service.j2

@@ -3,12 +3,16 @@ Description=DHCP relay container
 Requires=updategraph.service swss.service teamd.service
 After=updategraph.service swss.service syncd.service teamd.service
 Before=ntp-config.service
+StartLimitIntervalSec=1200
+StartLimitBurst=3
 
 [Service]
 User={{ sonicadmin_user }}
 ExecStartPre=/usr/bin/{{ docker_container_name }}.sh start
 ExecStart=/usr/bin/{{ docker_container_name }}.sh wait
 ExecStop=/usr/bin/{{ docker_container_name }}.sh stop
+Restart=always
+RestartSec=30
 
 [Install]
 WantedBy=multi-user.target swss.service teamd.service

files/scripts/supervisor-proc-exit-listener

@@ -33,9 +33,10 @@ def main():
 
             expected = int(payload_headers['expected'])
             processname = payload_headers['processname']
+            groupname = payload_headers['groupname']
 
             # If a critical process exited unexpectedly, terminate supervisor
-            if expected == 0 and processname in critical_processes:
+            if expected == 0 and processname in critical_processes or groupname in critical_processes:
                 MSG_FORMAT_STR = "Process {} exited unxepectedly. Terminating supervisor..."
                 msg = MSG_FORMAT_STR.format(payload_headers['processname'])
                 syslog.syslog(syslog.LOG_INFO, msg)

rules/docker-dhcp-relay.mk

@@ -25,3 +25,4 @@ SONIC_STRETCH_DBG_DOCKERS += $(DOCKER_DHCP_RELAY_DBG)
 $(DOCKER_DHCP_RELAY)_CONTAINER_NAME = dhcp_relay
 $(DOCKER_DHCP_RELAY)_RUN_OPT += --net=host --privileged -t
 $(DOCKER_DHCP_RELAY)_RUN_OPT += -v /etc/sonic:/etc/sonic:ro
+$(DOCKER_DHCP_RELAY)_FILES += $(SUPERVISOR_PROC_EXIT_LISTENER_SCRIPT)

src/sonic-config-engine/tests/sample_output/docker-dhcp-relay.supervisord.conf

@@ -3,6 +3,12 @@ logfile_maxbytes=1MB
 logfile_backups=2
 nodaemon=true
 
+[eventlistener:supervisor-proc-exit-listener]
+command=/usr/bin/supervisor-proc-exit-listener
+events=PROCESS_STATE_EXITED
+autostart=true
+autorestart=unexpected
+
 [program:start.sh]
 command=/usr/bin/start.sh
 priority=1