add retries for systemctl commands

parts/k8s/kubernetesagentcustomdata.yml

@@ -191,8 +191,8 @@ runcmd:
 - echo `date`,`hostname`, aptinstall>>/opt/m
 - systemctl enable rpcbind
 - systemctl enable rpc-statd
-- systemctl start rpcbind
-- systemctl start rpc-statd
+- retrycmd_if_failure 100 1 10 systemctl daemon-reload && systemctl restart rpcbind
+- retrycmd_if_failure 100 1 10 systemctl daemon-reload && systemctl restart rpc-statd
 - echo `date`,`hostname`, predockerinstall>>/opt/m
 - retrycmd_if_failure_no_stats 180 1 5 curl -fsSL https://aptdocker.azureedge.net/gpg > /tmp/aptdocker.gpg
 - cat /tmp/aptdocker.gpg | apt-key add -
@@ -204,7 +204,7 @@ runcmd:
 - echo "ExecStartPost=/sbin/iptables -P FORWARD ACCEPT" >> /etc/systemd/system/docker.service.d/exec_start.conf
 - systemctl daemon-reload
 - echo `date`,`hostname`, postdockerinstall>>/opt/m
-- systemctl restart docker
+- retrycmd_if_failure 100 1 10 systemctl daemon-reload && systemctl restart docker
 - mkdir -p /etc/kubernetes/manifests
 - usermod -aG docker {{WrapAsVariable "username"}}
 {{if IsNSeriesSKU .}}

parts/k8s/kubernetescustomscript.sh

@@ -249,8 +249,8 @@ function installClearContainersRuntime() {
 
 	# Enable and start Clear Containers proxy service
 	echo "Enabling and starting Clear Containers proxy service..."
-	systemctl enable cc-proxy
-	systemctl start cc-proxy
+	systemctlEnableAndCheck cc-proxy
+	retrycmd_if_failure 100 1 10 systemctl daemon-reload && systemctl restart cc-proxy
 
 	setKubeletOpts " --container-runtime=remote --runtime-request-timeout=15m --container-runtime-endpoint=unix:///run/containerd/containerd.sock"
 }
@@ -290,7 +290,7 @@ function ensureContainerd() {
 			# Make sure this is done after networking plugins are installed
 			echo "Enabling and starting cri-containerd service..."
 			systemctl enable containerd
-			systemctl start containerd
+			retrycmd_if_failure 100 1 10 systemctl daemon-reload && systemctl restart containerd
 		fi
 	fi
 }
@@ -321,31 +321,15 @@ function ensureDocker() {
     systemctlEnableAndCheck docker
     # only start if a reboot is not required
     if ! $REBOOTREQUIRED; then
-        dockerStarted=1
-        for i in {1..900}; do
-            if ! timeout 10s $DOCKER info; then
-                echo "status $?"
-                timeout 60s /bin/systemctl restart docker
-            else
-                echo "docker started, took $i seconds"
-                dockerStarted=0
-                break
-            fi
-            sleep 1
-        done
-        if [ $dockerStarted -ne 0 ]
-        then
-            echo "docker did not start"
-            exit 2
-        fi
+        retrycmd_if_failure 900 1 60 systemctl daemon-reload && systemctl restart docker
     fi
 }
 
 function ensureKubelet() {
     systemctlEnableAndCheck kubelet
     # only start if a reboot is not required
     if ! $REBOOTREQUIRED; then
-        systemctl restart kubelet
+        retrycmd_if_failure 100 1 10 systemctl daemon-reload && systemctl restart kubelet
     fi
 }
 
@@ -354,7 +338,7 @@ function extractHyperkube(){
     systemctlEnableAndCheck hyperkube-extract
     # only start if a reboot is not required
     if ! $REBOOTREQUIRED; then
-        systemctl restart hyperkube-extract
+        retrycmd_if_failure 100 1 10 systemctl daemon-reload && systemctl restart hyperkube-extract
     fi
 }
 
@@ -367,7 +351,7 @@ function ensureJournal(){
     echo "ForwardToSyslog=no" >> /etc/systemd/journald.conf
     # only start if a reboot is not required
     if ! $REBOOTREQUIRED; then
-        systemctl restart systemd-journald.service
+        retrycmd_if_failure 100 1 10 systemctl daemon-reload && systemctl restart systemd-journald.service
     fi
 }
 

parts/k8s/kubernetesmastercustomdata.yml

@@ -382,7 +382,7 @@ runcmd:
 - /bin/chown -R etcd:etcd /var/lib/etcddisk
 - systemctl stop etcd
 - systemctl daemon-reload
-- systemctl restart etcd
+- retrycmd_if_failure 100 1 10 systemctl daemon-reload && systemctl restart etcd
 - MEMBER="$(sudo etcdctl member list | grep -E {{WrapAsVerbatim "variables('masterVMNames')[copyIndex(variables('masterOffset'))]"}} | cut -d{{WrapAsVariable "singleQuote"}}:{{WrapAsVariable "singleQuote"}} -f 1)"
 - sudo etcdctl member update ${MEMBER} {{WrapAsVerbatim "variables('masterEtcdPeerURLs')[copyIndex(variables('masterOffset'))]"}}
 - retrycmd_if_failure 5 5 10 curl --cacert /etc/kubernetes/certs/ca.crt --cert /etc/kubernetes/certs/etcdclient.crt --key /etc/kubernetes/certs/etcdclient.key --retry 5 --retry-delay 10 --retry-max-time 10 --max-time 60 "{{WrapAsVerbatim "variables('masterEtcdClientURLs')[copyIndex(variables('masterOffset'))]"}}"/v2/machines
@@ -397,7 +397,7 @@ runcmd:
 - touch /opt/azure/containers/dockerinstall.complete
 - echo "ExecStartPost=/sbin/iptables -P FORWARD ACCEPT" >> /etc/systemd/system/docker.service.d/exec_start.conf
 - systemctl daemon-reload
-- systemctl restart docker
+- retrycmd_if_failure 100 1 10 systemctl daemon-reload && systemctl restart docker
 - mkdir -p /etc/kubernetes/manifests
 - usermod -aG docker {{WrapAsVariable "username"}}
 - /usr/lib/apt/apt.systemd.daily