Merge pull request #7938 from mbforbes/rollingUpgrade

Rolling node upgrade

cluster/gce/coreos/helper.sh

@@ -131,8 +131,15 @@ function create-master-instance {
 
 # TODO(dawnchen): Check $CONTAINER_RUNTIME to decide which
 # cloud_config yaml file should be passed
+# TODO(mbforbes): Make $1 required.
+# TODO(mbforbes): Document required vars (for this and call chain).
+# $1 version
 function create-node-instance-template {
-   create-node-template "${NODE_INSTANCE_PREFIX}-template" "${scope_flags[*]}" \
+  local suffix=""
+  if [[ -n ${1:-} ]]; then
+    suffix="-${1}"
+  fi
+   create-node-template "${NODE_INSTANCE_PREFIX}-template${suffix}" "${scope_flags[*]}" \
     "kube-env=${KUBE_TEMP}/node-kube-env.yaml" \
     "user-data=${KUBE_ROOT}/cluster/gce/coreos/node.yaml"
 }

cluster/gce/debian/helper.sh

@@ -107,8 +107,15 @@ function create-master-instance {
     --disk name="${MASTER_NAME}-pd" device-name=master-pd mode=rw boot=no auto-delete=no
 }
 
+# TODO(mbforbes): Make $1 required.
+# TODO(mbforbes): Document required vars (for this and call chain).
+# $1 version
 function create-node-instance-template {
-  create-node-template "${NODE_INSTANCE_PREFIX}-template" "${scope_flags[*]}" \
+  local suffix=""
+  if [[ -n ${1:-} ]]; then
+    suffix="-${1}"
+  fi
+  create-node-template "${NODE_INSTANCE_PREFIX}-template${suffix}" "${scope_flags[*]}" \
     "startup-script=${KUBE_ROOT}/cluster/gce/configure-vm.sh" \
     "kube-env=${KUBE_TEMP}/node-kube-env.yaml"
 }

cluster/gce/upgrade.sh

@@ -66,9 +66,10 @@ function usage() {
 function upgrade-master() {
   echo "== Upgrading master to '${SERVER_BINARY_TAR_URL}'. Do not interrupt, deleting master instance. =="
 
-  detect-master
   get-password
-  set-master-htpasswd
+  get-bearer-token
+
+  detect-master
 
   # Delete the master instance. Note that the master-pd is created
   # with auto-delete=no, so it should not be deleted.
@@ -111,13 +112,58 @@ function prepare-upgrade() {
   fi
 }
 
+# Reads kube-env metadata from master and extracts value from provided key.
+#
+# Assumed vars:
+#   MASTER_NAME
+#   ZONE
+#
+# Args:
+# $1 env key to use
+function get-env-val() {
+  # TODO(mbforbes): Make this more reliable with retries.
+  gcloud compute ssh --zone ${ZONE} ${MASTER_NAME} --command \
+    "curl --fail --silent -H 'Metadata-Flavor: Google' \
+      'http://metadata/computeMetadata/v1/instance/attributes/kube-env'" 2>/dev/null \
+    | grep ${1} | cut -d : -f 2 | cut -d \' -f 2
+}
+
+# $1 veresion
 function upgrade-nodes() {
-  echo "== Upgrading nodes to ${SERVER_BINARY_TAR_URL}. =="
+  local version=${1}
+  local sanitized_version=$(echo ${version} | sed s/"\."/-/g)
+  echo "== Upgrading nodes to ${version}. =="
 
   detect-minion-names
-  get-password
+
-  set-master-htpasswd
+  # TODO(mbforbes): Refactor setting scope flags.
-  kube-update-nodes upgrade
+  local -a scope_flags=()
+  if (( "${#MINION_SCOPES[@]}" > 0 )); then
+    scope_flags=("--scopes" "${MINION_SCOPES[@]}")
+  else
+    scope_flags=("--no-scopes")
+  fi
+
+  # Get required node tokens.
+  KUBELET_TOKEN=$(get-env-val "KUBELET_TOKEN")
+  KUBE_PROXY_TOKEN=$(get-env-val "KUBE_PROXY_TOKEN")
+
+  # TODO(mbforbes): How do we ensure kube-env is written in a ${version}-
+  #                 compatible way?
+  write-node-env
+  # TODO(mbforbes): Get configure-vm script from ${version}. (Must plumb this
+  #                 through all create-node-instance-template implementations).
+  create-node-instance-template ${sanitized_version}
+
+  # Do the actual upgrade.
+  gcloud preview rolling-updates start \
+      --group "${NODE_INSTANCE_PREFIX}-group" \
+      --max-num-concurrent-instances 1 \
+      --max-num-failed-instances 0 \
+      --project "${PROJECT}" \
+      --zone "${ZONE}" \
+      --template "${NODE_INSTANCE_PREFIX}-template-${sanitized_version}"
+
   echo "== Done =="
 }
 
@@ -192,7 +238,12 @@ if [[ "${master_upgrade}" == "true" ]]; then
 fi
 
 if [[ "${node_upgrade}" == "true" ]]; then
-  upgrade-nodes
+  if [[ "${local_binaries}" == "true" ]]; then
+    echo "Upgrading nodes to local binaries is not yet supported." >&2
+  else
+    upgrade-nodes ${binary_version}
+  fi
 fi
 
+echo "== Validating cluster post-upgrade =="
 "${KUBE_ROOT}/cluster/validate-cluster.sh"

cluster/gce/util.sh

@@ -383,6 +383,16 @@ function create-firewall-rule {
 # $4: The kube-env metadata.
 function create-node-template {
   detect-project
+
+  # First, ensure the template doesn't exist.
+  # TODO(mbforbes): To make this really robust, we need to parse the output and
+  #                 add retries. Just relying on a non-zero exit code doesn't
+  #                 distinguish an ephemeral failed call from a "not-exists".
+  if gcloud compute instance-templates describe "$1"; then
+    echo "Instance template ${1} already exists; continuing." >&2
+    return
+  fi
+
   local attempt=0
   while true; do
     if ! gcloud compute instance-templates create "$1" \
@@ -398,10 +408,10 @@ function create-node-template {
       --can-ip-forward \
       --metadata-from-file "$3" "$4"; then
         if (( attempt > 5 )); then
-          echo -e "${color_red}Failed to create instance template $1 ${color_norm}"
+          echo -e "${color_red}Failed to create instance template $1 ${color_norm}" >&2
           exit 2
         fi
-        echo -e "${color_yellow}Attempt $(($attempt+1)) failed to create instance template $1. Retrying.${color_norm}"
+        echo -e "${color_yellow}Attempt $(($attempt+1)) failed to create instance template $1. Retrying.${color_norm}" >&2
         attempt=$(($attempt+1))
     else
         break
@@ -624,6 +634,7 @@ function kube-up {
 
   echo "Creating minions."
 
+  # TODO(mbforbes): Refactor setting scope flags.
   local -a scope_flags=()
   if (( "${#MINION_SCOPES[@]}" > 0 )); then
     scope_flags=("--scopes" "${MINION_SCOPES[@]}")