Revert "[201803] [services] Restart SwSS service upon unexpected critical process exit (#2546)"

This reverts commit 2a8af27.

Author: lguohan

dockers/docker-dhcp-relay/wait_for_intf.sh.j2

@@ -1,40 +1,42 @@
 #!/usr/bin/env bash
 
-STATE_DB_IDX="6"
-
-PORT_TABLE_PREFIX="PORT_TABLE"
-VLAN_TABLE_PREFIX="VLAN_TABLE"
-LAG_TABLE_PREFIX="LAG_TABLE"
-
 function wait_until_iface_ready
 {
-    TABLE_PREFIX=$1
-    IFACE=$2
+    IFACE=$1
+
+    echo "Waiting until interface $IFACE is up..."
+
+    # Wait for the interface to come up (i.e., 'ip link show' returns 0)
+    until ip link show dev $IFACE up > /dev/null 2>&1; do
+        sleep 1
+    done
 
-    echo "Waiting until interface $IFACE is ready..."
+    echo "Interface $IFACE is up"
 
-    # Wait for the interface to come up
-    # (i.e., interface is present in STATE_DB and state is "ok")
+    echo "Waiting until interface $IFACE has an IPv4 address..."
+
+    # Wait until the interface gets assigned an IPv4 address
     while true; do
-        RESULT=$(redis-cli -n ${STATE_DB_IDX} HGET "${TABLE_PREFIX}|${IFACE}" "state" 2> /dev/null)
-        if [ x"$RESULT" == x"ok" ]; then
+        IP=$(ip -4 addr show dev $IFACE | grep "inet " | awk '{ print $2 }' | cut -d '/' -f1)
+
+        if [ -n "$IP" ]; then
             break
         fi
 
         sleep 1
     done
 
-    echo "Interface ${IFACE} is ready!"
+    echo "Interface $IFACE is configured with IP $IP"
 }
 
 
-# Wait for all interfaces to be up and ready
+# Wait for all interfaces to come up and have IPv4 addresses assigned
 {% for (name, prefix) in INTERFACE %}
-wait_until_iface_ready ${PORT_TABLE_PREFIX} {{ name }}
+wait_until_iface_ready {{ name }}
 {% endfor %}
 {% for (name, prefix) in VLAN_INTERFACE %}
-wait_until_iface_ready ${VLAN_TABLE_PREFIX} {{ name }}
+wait_until_iface_ready {{ name }}
 {% endfor %}
 {% for (name, prefix) in PORTCHANNEL_INTERFACE %}
-wait_until_iface_ready ${LAG_TABLE_PREFIX} {{ name }}
+wait_until_iface_ready {{ name }}
 {% endfor %}

dockers/docker-orchagent/Dockerfile.j2

@@ -30,8 +30,6 @@ COPY ["files/arp_update", "/usr/bin"]
 COPY ["enable_counters.py", "/usr/bin"]
 COPY ["start.sh", "orchagent.sh", "swssconfig.sh", "/usr/bin/"]
 COPY ["supervisord.conf", "/etc/supervisor/conf.d/"]
-COPY ["files/supervisor-proc-exit-listener", "/usr/bin"]
-COPY ["critical_processes", "/etc/supervisor/"]
 
 ## Copy all Jinja2 template files into the templates folder
 COPY ["*.j2", "/usr/share/sonic/templates/"]

dockers/docker-orchagent/critical_processes

@@ -3,12 +3,6 @@ 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
@@ -21,7 +15,7 @@ stderr_logfile=syslog
 command=/usr/sbin/rsyslogd -n
 priority=2
 autostart=false
-autorestart=unexpected
+autorestart=false
 stdout_logfile=syslog
 stderr_logfile=syslog
 

dockers/docker-orchagent/supervisord.conf

@@ -11,4 +11,4 @@ ExecStart=/usr/bin/{{ docker_container_name }}.sh attach
 ExecStop=/usr/bin/{{ docker_container_name }}.sh stop
 
 [Install]
-WantedBy=multi-user.target swss.service teamd.service
+WantedBy=multi-user.target teamd.service

files/build_templates/dhcp_relay.service.j2

@@ -11,4 +11,4 @@ ExecStart=/usr/bin/{{ docker_container_name }}.sh attach
 ExecStop=/usr/bin/{{ docker_container_name }}.sh stop
 
 [Install]
-WantedBy=multi-user.target swss.service
+WantedBy=multi-user.target

files/build_templates/radv.service.j2

@@ -8,6 +8,3 @@ Before=ntp-config.service
 ExecStartPre=/usr/bin/{{docker_container_name}}.sh start
 ExecStart=/usr/bin/{{docker_container_name}}.sh attach
 ExecStop=/usr/bin/{{docker_container_name}}.sh stop
-
-[Install]
-WantedBy=multi-user.target swss.service

files/build_templates/snmp.service.j2

@@ -14,8 +14,6 @@ After=opennsl-modules-3.16.0-6-amd64.service
 After=nps-modules-3.16.0-6-amd64.service
 {% endif %}
 Before=ntp-config.service
-StartLimitInterval=1200
-StartLimitBurst=3
 
 [Service]
 User=root
@@ -54,8 +52,6 @@ ExecStopPost=/usr/bin/mst stop
 ExecStopPost=/etc/init.d/xpnet.sh stop
 ExecStopPost=/etc/init.d/xpnet.sh start
 {% endif %}
-Restart=always
-RestartSec=30
 
 [Install]
 WantedBy=multi-user.target

files/build_templates/swss.service.j2

@@ -1,7 +1,7 @@
 [Unit]
 Description=TEAMD container
-Requires=updategraph.service swss.service
-After=updategraph.service swss.service
+Requires=updategraph.service
+After=updategraph.service
 Before=ntp-config.service
 
 [Service]
@@ -11,4 +11,4 @@ ExecStart=/usr/bin/{{docker_container_name}}.sh attach
 ExecStop=/usr/bin/{{docker_container_name}}.sh stop
 
 [Install]
-WantedBy=multi-user.target swss.service
+WantedBy=multi-user.target

files/build_templates/teamd.service.j2

@@ -16,4 +16,4 @@ $(DOCKER_ORCHAGENT_BRCM)_RUN_OPT += -v /etc/sonic:/etc/sonic:ro
 $(DOCKER_ORCHAGENT_BRCM)_RUN_OPT += -v /var/log/swss:/var/log/swss:rw
 
 $(DOCKER_ORCHAGENT_BRCM)_BASE_IMAGE_FILES += swssloglevel:/usr/bin/swssloglevel
-$(DOCKER_ORCHAGENT_BRCM)_FILES += $(ARP_UPDATE_SCRIPT) $(SUPERVISOR_PROC_EXIT_LISTENER_SCRIPT)
+$(DOCKER_ORCHAGENT_BRCM)_FILES += $(ARP_UPDATE_SCRIPT)

files/scripts/supervisor-proc-exit-listener

@@ -16,4 +16,4 @@ $(DOCKER_ORCHAGENT_CAVM)_RUN_OPT += -v /etc/sonic:/etc/sonic:ro
 $(DOCKER_ORCHAGENT_CAVM)_RUN_OPT += -v /var/log/swss:/var/log/swss:rw
 
 $(DOCKER_ORCHAGENT_CAVM)_BASE_IMAGE_FILES += swssloglevel:/usr/bin/swssloglevel
-$(DOCKER_ORCHAGENT_CAVM)_FILES += $(ARP_UPDATE_SCRIPT) $(SUPERVISOR_PROC_EXIT_LISTENER_SCRIPT)
+$(DOCKER_ORCHAGENT_CAVM)_FILES += $(ARP_UPDATE_SCRIPT)

platform/broadcom/docker-orchagent-brcm.mk

@@ -16,4 +16,4 @@ $(DOCKER_ORCHAGENT_CENTEC)_RUN_OPT += -v /etc/sonic:/etc/sonic:ro
 $(DOCKER_ORCHAGENT_CENTEC)_RUN_OPT += -v /var/log/swss:/var/log/swss:rw
 
 $(DOCKER_ORCHAGENT_CENTEC)_BASE_IMAGE_FILES += swssloglevel:/usr/bin/swssloglevel
-$(DOCKER_ORCHAGENT_CENTEC)_FILES += $(ARP_UPDATE_SCRIPT) $(SUPERVISOR_PROC_EXIT_LISTENER_SCRIPT)
+$(DOCKER_ORCHAGENT_CENTEC)_FILES += $(ARP_UPDATE_SCRIPT)

platform/cavium/docker-orchagent-cavm.mk

@@ -15,4 +15,4 @@ $(DOCKER_ORCHAGENT_MRVL)_RUN_OPT += -v /host/machine.conf:/host/machine.conf
 $(DOCKER_ORCHAGENT_MRVL)_RUN_OPT += -v /etc/sonic:/etc/sonic:ro
 
 $(DOCKER_ORCHAGENT_MRVL)_BASE_IMAGE_FILES += swssloglevel:/usr/bin/swssloglevel
-$(DOCKER_ORCHAGENT_MRVL)_FILES += $(ARP_UPDATE_SCRIPT) $(SUPERVISOR_PROC_EXIT_LISTENER_SCRIPT)
+$(DOCKER_ORCHAGENT_MRVL)_FILES += $(ARP_UPDATE_SCRIPT)

platform/centec/docker-orchagent-centec.mk

@@ -16,4 +16,4 @@ $(DOCKER_ORCHAGENT_MLNX)_RUN_OPT += -v /etc/sonic:/etc/sonic:ro
 $(DOCKER_ORCHAGENT_MLNX)_RUN_OPT += -v /var/log/swss:/var/log/swss:rw
 
 $(DOCKER_ORCHAGENT_MLNX)_BASE_IMAGE_FILES += swssloglevel:/usr/bin/swssloglevel
-$(DOCKER_ORCHAGENT_MLNX)_FILES += $(ARP_UPDATE_SCRIPT) $(SUPERVISOR_PROC_EXIT_LISTENER_SCRIPT)
+$(DOCKER_ORCHAGENT_MLNX)_FILES += $(ARP_UPDATE_SCRIPT)

platform/marvell/docker-orchagent-mrvl.mk

@@ -16,4 +16,4 @@ $(DOCKER_ORCHAGENT_NEPHOS)_RUN_OPT += -v /etc/sonic:/etc/sonic:ro
 $(DOCKER_ORCHAGENT_NEPHOS)_RUN_OPT += -v /var/log/swss:/var/log/swss:rw
 
 $(DOCKER_ORCHAGENT_NEPHOS)_BASE_IMAGE_FILES += swssloglevel:/usr/bin/swssloglevel
-$(DOCKER_ORCHAGENT_NEPHOS)_FILES += $(ARP_UPDATE_SCRIPT) $(SUPERVISOR_PROC_EXIT_LISTENER_SCRIPT)
+$(DOCKER_ORCHAGENT_NEPHOS)_FILES += $(ARP_UPDATE_SCRIPT)

platform/mellanox/docker-orchagent-mlnx.mk

@@ -2,7 +2,7 @@
 
 DOCKER_DHCP_RELAY = docker-dhcp-relay.gz
 $(DOCKER_DHCP_RELAY)_PATH = $(DOCKERS_PATH)/docker-dhcp-relay
-$(DOCKER_DHCP_RELAY)_DEPENDS += $(ISC_DHCP_COMMON) $(ISC_DHCP_RELAY) $(ISC_DHCP_CLIENT) $(REDIS_TOOLS)
+$(DOCKER_DHCP_RELAY)_DEPENDS += $(ISC_DHCP_COMMON) $(ISC_DHCP_RELAY) $(ISC_DHCP_CLIENT)
 $(DOCKER_DHCP_RELAY)_LOAD_DOCKERS = $(DOCKER_CONFIG_ENGINE)
 SONIC_DOCKER_IMAGES += $(DOCKER_DHCP_RELAY)
 SONIC_INSTALL_DOCKER_IMAGES += $(DOCKER_DHCP_RELAY)

platform/nephos/docker-orchagent-nephos.mk

@@ -5,11 +5,7 @@ $(ARP_UPDATE_SCRIPT)_PATH = files/scripts
 CONFIGDB_LOAD_SCRIPT = configdb-load.sh
 $(CONFIGDB_LOAD_SCRIPT)_PATH = files/scripts
 
-SUPERVISOR_PROC_EXIT_LISTENER_SCRIPT = supervisor-proc-exit-listener
-$(SUPERVISOR_PROC_EXIT_LISTENER_SCRIPT)_PATH = files/scripts
-
 SONIC_COPY_FILES += $(CONFIGDB_LOAD_SCRIPT) \
-                    $(ARP_UPDATE_SCRIPT) \
-                    $(SUPERVISOR_PROC_EXIT_LISTENER_SCRIPT)
+                    $(ARP_UPDATE_SCRIPT)
 
 

rules/docker-dhcp-relay.mk

@@ -1,41 +1,43 @@
 #!/usr/bin/env bash
 
-STATE_DB_IDX="6"
-
-PORT_TABLE_PREFIX="PORT_TABLE"
-VLAN_TABLE_PREFIX="VLAN_TABLE"
-LAG_TABLE_PREFIX="LAG_TABLE"
-
 function wait_until_iface_ready
 {
-    TABLE_PREFIX=$1
-    IFACE=$2
+    IFACE=$1
+
+    echo "Waiting until interface $IFACE is up..."
+
+    # Wait for the interface to come up (i.e., 'ip link show' returns 0)
+    until ip link show dev $IFACE up > /dev/null 2>&1; do
+        sleep 1
+    done
 
-    echo "Waiting until interface $IFACE is ready..."
+    echo "Interface $IFACE is up"
 
-    # Wait for the interface to come up
-    # (i.e., interface is present in STATE_DB and state is "ok")
+    echo "Waiting until interface $IFACE has an IPv4 address..."
+
+    # Wait until the interface gets assigned an IPv4 address
     while true; do
-        RESULT=$(redis-cli -n ${STATE_DB_IDX} HGET "${TABLE_PREFIX}|${IFACE}" "state" 2> /dev/null)
-        if [ x"$RESULT" == x"ok" ]; then
+        IP=$(ip -4 addr show dev $IFACE | grep "inet " | awk '{ print $2 }' | cut -d '/' -f1)
+
+        if [ -n "$IP" ]; then
             break
         fi
 
         sleep 1
     done
 
-    echo "Interface ${IFACE} is ready!"
+    echo "Interface $IFACE is configured with IP $IP"
 }
 
 
-# Wait for all interfaces to be up and ready
-wait_until_iface_ready ${VLAN_TABLE_PREFIX} Vlan1000
-wait_until_iface_ready ${LAG_TABLE_PREFIX} PortChannel04
-wait_until_iface_ready ${LAG_TABLE_PREFIX} PortChannel02
-wait_until_iface_ready ${LAG_TABLE_PREFIX} PortChannel03
-wait_until_iface_ready ${LAG_TABLE_PREFIX} PortChannel03
-wait_until_iface_ready ${LAG_TABLE_PREFIX} PortChannel01
-wait_until_iface_ready ${LAG_TABLE_PREFIX} PortChannel02
-wait_until_iface_ready ${LAG_TABLE_PREFIX} PortChannel04
-wait_until_iface_ready ${LAG_TABLE_PREFIX} PortChannel01
+# Wait for all interfaces to come up and have IPv4 addresses assigned
+wait_until_iface_ready Vlan1000
+wait_until_iface_ready PortChannel04
+wait_until_iface_ready PortChannel02
+wait_until_iface_ready PortChannel03
+wait_until_iface_ready PortChannel03
+wait_until_iface_ready PortChannel01
+wait_until_iface_ready PortChannel02
+wait_until_iface_ready PortChannel04
+wait_until_iface_ready PortChannel01