[Services] Restart NAT service upon unexpected critical process exit. (…

…sonic-net#4208)
SONIC-DEV · Apr 24, 2020 · 8689613 · 8689613
1 parent dd499ee
commit 8689613
Show file tree

Hide file tree

Showing 5 changed files with 17 additions and 2 deletions.
diff --git a/dockers/docker-nat/Dockerfile.j2 b/dockers/docker-nat/Dockerfile.j2
@@ -38,6 +38,8 @@ RUN apt-get update        \
 COPY ["start.sh", "/usr/bin/"]
 COPY ["supervisord.conf", "/etc/supervisor/conf.d/"]
 COPY ["restore_nat_entries.py", "/usr/bin/"]
+COPY ["files/supervisor-proc-exit-listener", "/usr/bin"]
+COPY ["critical_processes", "/etc/supervisor"]
 
 RUN apt-get clean -y; apt-get autoclean -y; apt-get autoremove -y
 RUN rm -rf /debs

diff --git a/dockers/docker-nat/critical_processes b/dockers/docker-nat/critical_processes
@@ -0,0 +1,2 @@
+natmgrd
+natsyncd
diff --git a/dockers/docker-nat/supervisord.conf b/dockers/docker-nat/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 --container-name nat
+events=PROCESS_STATE_EXITED
+autostart=true
+autorestart=unexpected
+
 [program:start.sh]
 command=/usr/bin/start.sh
 priority=1
@@ -15,7 +21,7 @@ stderr_logfile=syslog
 command=/usr/sbin/rsyslogd -n
 priority=2
 autostart=false
-autorestart=false
+autorestart=unexpected
 stdout_logfile=syslog
 stderr_logfile=syslog
 

diff --git a/files/build_templates/nat.service.j2 b/files/build_templates/nat.service.j2
@@ -3,12 +3,16 @@ Description=NAT container
 Requires=updategraph.service swss.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 swss.service

diff --git a/rules/docker-nat.mk b/rules/docker-nat.mk
@@ -30,5 +30,6 @@ $(DOCKER_NAT)_RUN_OPT += --privileged -t
 $(DOCKER_NAT)_RUN_OPT += -v /etc/sonic:/etc/sonic:ro
 $(DOCKER_NAT)_RUN_OPT += -v /host/warmboot:/var/warmboot
 
-$(DOCKER_NAT)_BASE_IMAGE_FILES += natctl:/usr/bin/natctl
+$(DOCKER_NAT)_FILES += $(SUPERVISOR_PROC_EXIT_LISTENER_SCRIPT)
 
+$(DOCKER_NAT)_BASE_IMAGE_FILES += natctl:/usr/bin/natctl