[Services] Restart database service upon unexpected critical process …

…exit. (sonic-net#4138)

* [database] Implement the auto-restart feature for database container.

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

* [database] Remove the duplicate dependency in service files. Since we
already have updategraph ---> config_setup ---> database, we do not need
explicitly add database.service in all other container service files.

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

* [event listener] Reorganize the line 73 in event listener script.

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

* [database] update the file sflow.service.j2 to remove the duplicate
dependency.

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

* [event listener] Add comments in event listener.

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

* [event listener] Update the comments in line 56.

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

* [event listener] Add parentheses for if statement in line 76 in event listener.

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

dockers/docker-database/Dockerfile.j2

@@ -36,5 +36,7 @@ COPY ["supervisord.conf.j2", "/usr/share/sonic/templates/"]
 COPY ["docker-database-init.sh", "/usr/local/bin/"]
 COPY ["ping_pong_db_insts", "/usr/local/bin/"]
 COPY ["database_config.json", "/etc/default/sonic-db/"]
+COPY ["files/supervisor-proc-exit-listener", "/usr/bin"]
+COPY ["critical_processes", "/etc/supervisor"]
 
 ENTRYPOINT ["/usr/local/bin/docker-database-init.sh"]

dockers/docker-database/critical_processes

@@ -0,0 +1 @@
+redis

dockers/docker-database/supervisord.conf.j2

@@ -3,6 +3,13 @@ logfile_maxbytes=1MB
 logfile_backups=2
 nodaemon=true
 
+[eventlistener:supervisor-proc-exit-listener]
+command=/usr/bin/supervisor-proc-exit-listener --container-name database
+events=PROCESS_STATE_EXITED
+autostart=true
+autorestart=unexpected
+
+
 [program:rsyslogd]
 command=/bin/bash -c "rm -f /var/run/rsyslogd.pid && /usr/sbin/rsyslogd -n"
 priority=1

files/build_templates/single_instance/database.service.j2

@@ -3,12 +3,16 @@ Description=Database container
 Requires=docker.service
 After=docker.service
 After=rc-local.service
+StartLimitIntervalSec=1200
+StartLimitBurst=3
 
 [Service]
 User=root
 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

files/scripts/supervisor-proc-exit-listener

@@ -52,24 +52,28 @@ def main(argv):
             processname = payload_headers['processname']
             groupname = payload_headers['groupname']
 
-            config_db = swsssdk.ConfigDBConnector()
-            config_db.connect()
-            container_features_table = config_db.get_table(CONTAINER_FEATURE_TABLE_NAME)
-            if not container_features_table:
-                syslog.syslog(syslog.LOG_ERR, "Unable to retrieve container features table from Config DB. Exiting...")
-                sys.exit(2)
-
-            if not container_features_table.has_key(container_name):
-                syslog.syslog(syslog.LOG_ERR, "Unable to retrieve features for container '{}'. Exiting...".format(container_name))
-                sys.exit(3)
-
-            restart_feature = container_features_table[container_name].get('auto_restart')
-            if not restart_feature:
-                syslog.syslog(syslog.LOG_ERR, "Unable to determine auto-restart feature status for container '{}'. Exiting...".format(container_name))
-                sys.exit(4)
-
-            # If auto-restart feature is enabled and a critical process exited unexpectedly, terminate supervisor
-            if restart_feature == 'enabled' and expected == 0 and (processname in critical_processes or groupname in critical_processes):
+            # Read the status of auto-restart feature from Config_DB.
+            if container_name != 'database':
+                config_db = swsssdk.ConfigDBConnector()
+                config_db.connect()
+                container_features_table = config_db.get_table(CONTAINER_FEATURE_TABLE_NAME)
+                if not container_features_table:
+                    syslog.syslog(syslog.LOG_ERR, "Unable to retrieve container features table from Config DB. Exiting...")
+                    sys.exit(2)
+
+                if not container_features_table.has_key(container_name):
+                    syslog.syslog(syslog.LOG_ERR, "Unable to retrieve features for container '{}'. Exiting...".format(container_name))
+                    sys.exit(3)
+
+                restart_feature = container_features_table[container_name].get('auto_restart')
+                if not restart_feature:
+                    syslog.syslog(syslog.LOG_ERR, "Unable to determine auto-restart feature status for container '{}'. Exiting...".format(container_name))
+                    sys.exit(4)
+
+            # If container is database or auto-restart feature is enabled and at the same time 
+            # a critical process exited unexpectedly, terminate supervisor
+            if ((container_name == 'database' or restart_feature == 'enabled') and 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-database.mk

@@ -28,3 +28,4 @@ $(DOCKER_DATABASE)_RUN_OPT += -v /etc/sonic:/etc/sonic:ro
 
 $(DOCKER_DATABASE)_BASE_IMAGE_FILES += redis-cli:/usr/bin/redis-cli
 $(DOCKER_DATABASE)_BASE_IMAGE_FILES += monit_database:/etc/monit/conf.d
+$(DOCKER_DATABASE)_FILES += $(SUPERVISOR_PROC_EXIT_LISTENER_SCRIPT)