From 67fc68513ee784e12e4588f7eb6c8f51135bd16b Mon Sep 17 00:00:00 2001
From: yozhao101 <56170650+yozhao101@users.noreply.github.com>
Date: Mon, 25 Nov 2019 13:02:00 -0800
Subject: [PATCH] [Services] Restart Sflow service upon unexpected critical
 process exit. (#3751)

Signed-off-by: Yong Zhao <yozhao@microsoft.com>
---
 dockers/docker-sflow/Dockerfile.j2      | 2 ++
 dockers/docker-sflow/critical_processes | 1 +
 dockers/docker-sflow/supervisord.conf   | 6 ++++++
 files/build_templates/sflow.service.j2  | 4 ++++
 rules/docker-sflow.mk                   | 2 +-
 5 files changed, 14 insertions(+), 1 deletion(-)
 create mode 100644 dockers/docker-sflow/critical_processes

diff --git a/dockers/docker-sflow/Dockerfile.j2 b/dockers/docker-sflow/Dockerfile.j2
index e88789fb69..75da64e02e 100644
--- a/dockers/docker-sflow/Dockerfile.j2
+++ b/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"]
diff --git a/dockers/docker-sflow/critical_processes b/dockers/docker-sflow/critical_processes
new file mode 100644
index 0000000000..5b24e2d8e1
--- /dev/null
+++ b/dockers/docker-sflow/critical_processes
@@ -0,0 +1 @@
+sflowmgrd
diff --git a/dockers/docker-sflow/supervisord.conf b/dockers/docker-sflow/supervisord.conf
index e4f9259712..50986f197d 100644
--- a/dockers/docker-sflow/supervisord.conf
+++ b/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
diff --git a/files/build_templates/sflow.service.j2 b/files/build_templates/sflow.service.j2
index 3a5752412b..de08f027ad 100644
--- a/files/build_templates/sflow.service.j2
+++ b/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
diff --git a/rules/docker-sflow.mk b/rules/docker-sflow.mk
index 19b0f290bf..94b5684818 100644
--- a/rules/docker-sflow.mk
+++ b/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)