Configurable liveness probe initial delays for etcd and kube-apiserver in GCE

cluster/common.sh

@@ -838,6 +838,16 @@ EOF
     if [ -n "${ETCD_HOSTNAME:-}" ]; then
       cat >>$file <<EOF
 ETCD_HOSTNAME: $(yaml-quote ${ETCD_HOSTNAME})
+EOF
+    fi
+    if [ -n "${ETCD_LIVENESS_PROBE_INITIAL_DELAY_SEC:-}" ]; then
+      cat >>$file <<EOF
+ETCD_LIVENESS_PROBE_INITIAL_DELAY_SEC: $(yaml-quote ${ETCD_LIVENESS_PROBE_INITIAL_DELAY_SEC})
+EOF
+    fi
+    if [ -n "${KUBE_APISERVER_LIVENESS_PROBE_INITIAL_DELAY_SEC:-}" ]; then
+      cat >>$file <<EOF
+KUBE_APISERVER_LIVENESS_PROBE_INITIAL_DELAY_SEC: $(yaml-quote ${KUBE_APISERVER_LIVENESS_PROBE_INITIAL_DELAY_SEC})
 EOF
     fi
     if [ -n "${APISERVER_TEST_ARGS:-}" ]; then

cluster/gce/configure-vm.sh

@@ -494,6 +494,16 @@ EOF
     if [ -n "${KUBE_APISERVER_REQUEST_TIMEOUT_SEC:-}" ]; then
       cat <<EOF >>/srv/salt-overlay/pillar/cluster-params.sls
 kube_apiserver_request_timeout_sec: '$(echo "$KUBE_APISERVER_REQUEST_TIMEOUT_SEC" | sed -e "s/'/''/g")'
+EOF
+    fi
+    if [ -n "${ETCD_LIVENESS_PROBE_INITIAL_DELAY_SEC:-}" ]; then
+      cat <<EOF >>/srv/salt-overlay/pillar/cluster-params.sls
+etcd_liveness_probe_initial_delay: '$(echo "$ETCD_LIVENESS_PROBE_INITIAL_DELAY_SEC" | sed -e "s/'/''/g")'
+EOF
+    fi
+    if [ -n "${KUBE_APISERVER_LIVENESS_PROBE_INITIAL_DELAY_SEC:-}" ]; then
+      cat <<EOF >>/srv/salt-overlay/pillar/cluster-params.sls
+kube_apiserver_liveness_probe_initial_delay: '$(echo "$KUBE_APISERVER_LIVENESS_PROBE_INITIAL_DELAY_SEC" | sed -e "s/'/''/g")'
 EOF
     fi
     if [ -n "${ADMISSION_CONTROL:-}" ] && [ ${ADMISSION_CONTROL} == *"ImagePolicyWebhook"* ]; then

cluster/gce/container-linux/configure-helper.sh

@@ -778,6 +778,7 @@ function prepare-etcd-manifest {
   sed -i -e "s@{{ *hostname *}}@$host_name@g" "${temp_file}"
   sed -i -e "s@{{ *srv_kube_path *}}@/etc/srv/kubernetes@g" "${temp_file}"
   sed -i -e "s@{{ *etcd_cluster *}}@$etcd_cluster@g" "${temp_file}"
+  sed -i -e "s@{{ *liveness_probe_initial_delay *}}@${ETCD_LIVENESS_PROBE_INITIAL_DELAY_SEC:-15}@g" "${temp_file}"
   # Get default storage backend from manifest file.
   local -r default_storage_backend=$(cat "${temp_file}" | \
     grep -o "{{ *pillar\.get('storage_backend', '\(.*\)') *}}" | \
@@ -1060,6 +1061,7 @@ function start-kube-apiserver {
   sed -i -e "s@{{pillar\['kube_docker_registry'\]}}@${DOCKER_REGISTRY}@g" "${src_file}"
   sed -i -e "s@{{pillar\['kube-apiserver_docker_tag'\]}}@${kube_apiserver_docker_tag}@g" "${src_file}"
   sed -i -e "s@{{pillar\['allow_privileged'\]}}@true@g" "${src_file}"
+  sed -i -e "s@{{liveness_probe_initial_delay}}@${KUBE_APISERVER_LIVENESS_PROBE_INITIAL_DELAY_SEC:-15}@g" "${src_file}"
   sed -i -e "s@{{secure_port}}@443@g" "${src_file}"
   sed -i -e "s@{{secure_port}}@8080@g" "${src_file}"
   sed -i -e "s@{{additional_cloud_config_mount}}@@g" "${src_file}"

cluster/gce/gci/configure-helper.sh

@@ -1372,6 +1372,7 @@ function prepare-etcd-manifest {
   sed -i -e "s@{{ *hostname *}}@$host_name@g" "${temp_file}"
   sed -i -e "s@{{ *srv_kube_path *}}@/etc/srv/kubernetes@g" "${temp_file}"
   sed -i -e "s@{{ *etcd_cluster *}}@$etcd_cluster@g" "${temp_file}"
+  sed -i -e "s@{{ *liveness_probe_initial_delay *}}@${ETCD_LIVENESS_PROBE_INITIAL_DELAY_SEC:-15}@g" "${temp_file}"
   # Get default storage backend from manifest file.
   local -r default_storage_backend=$(cat "${temp_file}" | \
     grep -o "{{ *pillar\.get('storage_backend', '\(.*\)') *}}" | \
@@ -1747,6 +1748,7 @@ function start-kube-apiserver {
   sed -i -e "s@{{pillar\['kube_docker_registry'\]}}@${DOCKER_REGISTRY}@g" "${src_file}"
   sed -i -e "s@{{pillar\['kube-apiserver_docker_tag'\]}}@${kube_apiserver_docker_tag}@g" "${src_file}"
   sed -i -e "s@{{pillar\['allow_privileged'\]}}@true@g" "${src_file}"
+  sed -i -e "s@{{liveness_probe_initial_delay}}@${KUBE_APISERVER_LIVENESS_PROBE_INITIAL_DELAY_SEC:-15}@g" "${src_file}"
   sed -i -e "s@{{secure_port}}@443@g" "${src_file}"
   sed -i -e "s@{{secure_port}}@8080@g" "${src_file}"
   sed -i -e "s@{{additional_cloud_config_mount}}@@g" "${src_file}"

cluster/saltbase/salt/etcd/etcd.manifest

@@ -22,6 +22,7 @@
 {% if pillar.get('storage_backend', 'etcd3') == 'etcd3' -%}
   {% set quota_bytes = '--quota-backend-bytes=4294967296' -%}
 {% endif -%}
+{% set liveness_probe_initial_delay = pillar.get('etcd_liveness_probe_initial_delay', 15) -%}
 {% set srv_kube_path = "/srv/kubernetes" -%}
 
 {
@@ -67,7 +68,7 @@
         "port": {{ port }},
         "path": "/health"
       },
-      "initialDelaySeconds": 15,
+      "initialDelaySeconds": {{ liveness_probe_initial_delay }},
       "timeoutSeconds": 15
     },
     "ports": [

cluster/saltbase/salt/kube-apiserver/kube-apiserver.manifest

@@ -66,6 +66,8 @@
   {% set storage_media_type = "--storage-media-type=" + pillar['storage_media_type'] -%}
 {% endif -%}
 
+{% set liveness_probe_initial_delay = pillar.get('kube_apiserver_liveness_probe_initial_delay', 15) -%}
+
 {% set request_timeout = "" -%}
 {% if pillar['kube_apiserver_request_timeout_sec'] is defined -%}
   {% set request_timeout = "--request-timeout=" + pillar['kube_apiserver_request_timeout_sec'] + "s" -%}
@@ -248,7 +250,7 @@
         "port": 8080,
         "path": "/healthz"
       },
-      "initialDelaySeconds": 15,
+      "initialDelaySeconds": {{liveness_probe_initial_delay}},
       "timeoutSeconds": 15
     },
     "ports":[