[Services] Restart Router-advertiser service upon unexpected critical process exit (#3681)

Signed-off-by: Yong Zhao <[email protected]>

Author: yozhao101

dockers/docker-router-advertiser/Dockerfile.j2

@@ -27,5 +27,7 @@ RUN apt-get clean -y        && \
 COPY ["start.sh", "/usr/bin/"]
 COPY ["docker-router-advertiser.supervisord.conf", "/etc/supervisor/conf.d/"]
 COPY ["radvd.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/supervisord"]

dockers/docker-router-advertiser/critical_processes

@@ -0,0 +1 @@
+radvd

dockers/docker-router-advertiser/docker-router-advertiser.supervisord.conf

@@ -3,6 +3,12 @@ logfile_maxbytes=1MB
 logfile_backups=2
 nodaemon=true
 
+[eventlistener:supervisor-proc-exit-script]
+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/radv.service.j2

@@ -3,12 +3,16 @@ Description=Router advertiser container
 Requires=updategraph.service
 After=updategraph.service swss.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

rules/docker-router-advertiser.mk

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