[Services] Restart LLDP service upon unexpected critical process exit. (sonic-net#3713)

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

Author: yozhao101

dockers/docker-lldp-sv2/Dockerfile.j2

@@ -40,5 +40,7 @@ COPY ["supervisord.conf", "/etc/supervisor/conf.d/"]
 COPY ["lldpd.conf.j2", "/usr/share/sonic/templates/"]
 COPY ["lldpd", "/etc/default/"]
 COPY ["lldpmgrd", "/usr/bin/"]
+COPY ["files/supervisor-proc-exit-listener", "/usr/bin"]
+COPY ["critical_processes", "/etc/supervisor"]
 
 ENTRYPOINT ["/usr/bin/supervisord"]

dockers/docker-lldp-sv2/critical_processes

@@ -0,0 +1,3 @@
+lldpd
+lldp-syncd
+lldpmgrd

dockers/docker-lldp-sv2/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

files/build_templates/lldp.service.j2

@@ -3,12 +3,16 @@ Description=LLDP container
 Requires=updategraph.service
 After=updategraph.service swss.service syncd.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-lldp-sv2.mk

@@ -30,3 +30,4 @@ $(DOCKER_LLDP_SV2)_RUN_OPT += -v /etc/sonic:/etc/sonic:ro
 
 $(DOCKER_LLDP_SV2)_BASE_IMAGE_FILES += lldpctl:/usr/bin/lldpctl
 $(DOCKER_LLDP_SV2)_BASE_IMAGE_FILES += lldpcli:/usr/bin/lldpcli
+$(DOCKER_LLDP_SV2)_FILES += $(SUPERVISOR_PROC_EXIT_LISTENER_SCRIPT)