[Services] Restart Sflow service upon unexpected critical process exit. (#3751)

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

dockers/docker-sflow/Dockerfile.j2

@@ -29,5 +29,7 @@ RUN sed -ri '/^DAEMON_ARGS=""/c DAEMON_ARGS="-c /var/log/hsflowd.crash"' /etc/in
 
 COPY ["start.sh", "/usr/bin/"]
 COPY ["supervisord.conf", "/etc/supervisor/conf.d/"]
+COPY ["files/supervisor-proc-exit-listener", "/usr/bin"]
+COPY ["critical_processes", "/etc/supervisor"]
 
 ENTRYPOINT ["/usr/bin/supervisord"]

dockers/docker-sflow/critical_processes

@@ -0,0 +1 @@
+sflowmgrd

dockers/docker-sflow/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/sflow.service.j2

@@ -3,12 +3,16 @@ Description=sFlow container
 Requires=swss.service
 After=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-sflow.mk

@@ -32,4 +32,4 @@ $(DOCKER_SFLOW)_RUN_OPT += -v /host/warmboot:/var/warmboot
 
 $(DOCKER_SFLOW)_BASE_IMAGE_FILES += psample:/usr/bin/psample
 $(DOCKER_SFLOW)_BASE_IMAGE_FILES += sflowtool:/usr/bin/sflowtool
-
+$(DOCKER_SFLOW)_FILES += $(SUPERVISOR_PROC_EXIT_LISTENER_SCRIPT)