Watch node object for podCIDR instead of crash + wait for restart

This can save up to 10 seconds before Kubelet decides to restart the pod after the crash.
GoogleCloudPlatform · May 24, 2024 · c7585b0 · c7585b0
1 parent f19530f
commit c7585b0
Show file tree

Hide file tree

Showing 32 changed files with 383 additions and 108 deletions.
diff --git a/scripts/install-cni.sh b/scripts/install-cni.sh
@@ -18,6 +18,11 @@ log() {
   echo "$@"
 }
 
+fatal() {
+  echo FATAL: "$@" >&2
+  exit 1
+}
+
 # shellcheck disable=SC2317,SC2329 # when called with $1=calico_ready
 calico_ready() {
   log "Listing items matching /host/etc/cni/net.d/*calico*.conflist"
@@ -110,21 +115,41 @@ else
   cni_spec=${cni_spec//@cniBandwidthPlugin/}
 fi
 
-token=$(</var/run/secrets/kubernetes.io/serviceaccount/token)
-host=${KUBERNETES_SERVICE_HOST}
-# If host contains a colon (:), it is an IPv6 address, hence needs wrapping
-# with [..].
-if [[ "${host}" =~ : ]]; then
-  host="[$host]"
-fi
-node_url="https://$host:${KUBERNETES_SERVICE_PORT}/api/v1/nodes/${HOSTNAME}"
-response=$(curl -k -s -H "Authorization: Bearer $token" "$node_url")
+fetch_node_object() {
+  local attempts=$1
+  local timeout=$2
+
+  local host=${KUBERNETES_SERVICE_HOST}
+  # If host contains a colon (:), it is an IPv6 address, hence needs wrapping
+  # with [..].
+  if [[ "${host}" =~ : ]]; then
+    host="[${host}]"
+  fi
+
+  local token
+  local node_url="https://${host}:${KUBERNETES_SERVICE_PORT}/api/v1/nodes?watch=1&timeoutSeconds=${timeout}&fieldSelector=metadata.name=${HOSTNAME}"
+
+  for ((i=1; i<=attempts; i++)); do
+    log "Watching attempt #${i} at ${node_url}"
+    token=$(</var/run/secrets/kubernetes.io/serviceaccount/token)
+    node_object=$(grep --line-buffered -m1 . <(curl -k -s -N -H "Authorization: Bearer ${token}" "${node_url}" | jq --unbuffered -c '.object | select(.spec.podCIDR != null)')) || node_object=
+    [[ -n "${node_object}" ]] && return
+  done
+
+  fatal "Could not successfully watch node and wait for podCIDR."
+}
+
+# Watch for up to 1 minute, we don't expect podCIDR to be not populated for too
+# long, but this can also be three continuous retries and failures, then we wait
+# for kubelet to retry the whole container if node_object is still not fetched.
+fetch_node_object 3 20
+log "Node object fetched:"
+log "${node_object}"
 
-if [ "${MIGRATE_TO_DPV2:-}" == "true" ]; then
-  DPV2_MIGRATION_READY=$(jq '.metadata.labels."cloud.google.com/gke-dpv2-migration-ready"' <<<"$response")
+if [[ "${MIGRATE_TO_DPV2:-}" == "true" ]]; then
+  DPV2_MIGRATION_READY=$(jq -r '.metadata.labels."cloud.google.com/gke-dpv2-migration-ready"' <<<"${node_object}")
   log "Migration to DPv2 in progress; node ready: '${DPV2_MIGRATION_READY}'"
-  if [ "${DPV2_MIGRATION_READY}" != '"true"' ] # DPV2_MIGRATION_READY is a JSON string thus double quotes
-  then
+  if [[ "${DPV2_MIGRATION_READY}" != "true" ]]; then
     ENABLE_CILIUM_PLUGIN=false
   fi
 fi
@@ -134,7 +159,8 @@ if [[ "${ENABLE_CILIUM_PLUGIN}" == "true" ]]; then
   if [[ -n "${CILIUM_FAST_START_NAMESPACES:-}" ]]; then
     cilium_cni_config=$(jq --arg namespaces "${CILIUM_FAST_START_NAMESPACES:-}" '.["dpv2-fast-start-namespaces"] = $namespaces' <<<"${cilium_cni_config}")
   fi
-  log "Adding Cilium plug-in to the CNI config: ${cilium_cni_config//$'\n'/ }"
+  log "Adding Cilium plug-in to the CNI config:"
+  log "${cilium_cni_config//$'\n'/ }"
   cni_spec=${cni_spec//@cniCiliumPlugin/, ${cilium_cni_config}}
 else
   log "Not using Cilium plug-in."
@@ -218,8 +244,7 @@ function fillSubnetsInCniSpecV2Template {
       SUBNETS_REPLACEMENT+=("$(jq -nc --arg subnet "${subnet}" '[{"subnet": $subnet}]')")
       ROUTES_REPLACEMENT+=('{"dst": "::/0"}')
     else
-      log "[ERROR] Subnet detected in .spec.podCIDRs '${subnet}' is not a valid IP range"
-      exit 1
+      fatal "Subnet detected in .spec.podCIDRs '${subnet}' is not a valid IP range"
     fi
   done
 
@@ -261,13 +286,9 @@ function fillSubnetsInCniSpecLegacyTemplate {
     log "PodCIDR IPv4 detected: '${primary_subnet:-}'"
     cni_spec=${cni_spec//@ipv4Subnet/[{\"subnet\": \"${primary_subnet:-}\"\}]}
   elif is_ipv6_range "${primary_subnet:-}" ; then
-    log "Primary IPv6 pod range detected '${primary_subnet:-}'. It will only work with new spec template."
-    exit 1
+    fatal "Primary IPv6 pod range detected '${primary_subnet:-}'. It will only work with new spec template."
   else
-    log "Response from $node_url"
-    log "$node"
-    log "Failed to fetch PodCIDR from K8s API server, primary_subnet=${primary_subnet:-}. Exiting (1)..."
-    exit 1
+    fatal "Failed to fetch PodCIDR from K8s API server, primary_subnet=${primary_subnet:-}."
   fi
 
   if [ -n "${node_ipv6_addr:-}" ] && [ "${node_ipv6_addr}" != "null" ]; then
@@ -296,15 +317,15 @@ function fillSubnetsInCniSpec {
 }
 
 
-CLUSTER_STACK_TYPE=$(jq -r '.metadata.labels."cloud.google.com/gke-stack-type"' <<<"$response")
+CLUSTER_STACK_TYPE=$(jq -r '.metadata.labels."cloud.google.com/gke-stack-type"' <<<"${node_object}")
 log "Node's cluster stack type label: '${CLUSTER_STACK_TYPE:-}'"
 
 node_ipv6_addr=''
 if [ "$ENABLE_IPV6" == "true" ] || [ "${CLUSTER_STACK_TYPE:-}" == "IPV4_IPV6" ]; then
   node_ipv6_addr=$(curl -s -k --fail "http://metadata.google.internal/computeMetadata/v1/instance/network-interfaces/0/?recursive=true" -H "Metadata-Flavor: Google" | jq -r '.ipv6s[0]' ) ||:
 fi
 
-fillSubnetsInCniSpec "$response" "$node_ipv6_addr"
+fillSubnetsInCniSpec "${node_object}" "${node_ipv6_addr}"
 
 if [ "$POPULATE_IP6TABLES" == "true" ] ; then
   # Ensure the IPv6 firewall rules are as expected.
@@ -433,7 +454,8 @@ cilium_wait_or_ignore() {
 }
 
 write_and_success() {
-  log "Creating CNI spec at '${output_file}' with content: ${cni_spec//$'\n'/ }"
+  log "Creating CNI spec at '${output_file}' with content:"
+  log "${cni_spec//$'\n'/ }"
   write_file "${output_file}" "${cni_spec}"
   success
 }
@@ -449,7 +471,8 @@ if [[ "${RUN_CNI_WATCHDOG:-}" != "true" ]]; then
   write_and_success
 fi
 
-log "Running CNI watchdog to watch Cilium and manage CNI config at '${output_file}' with content: ${cni_spec//$'\n'/ }"
+log "Running CNI watchdog to watch Cilium and manage CNI config at '${output_file}' with content:"
+log "${cni_spec//$'\n'/ }"
 cilium_watchdog_success_wait=${CILIUM_WATCHDOG_SUCCESS_WAIT:-300}
 cilium_watchdog_failure_retry=${CILIUM_WATCHDOG_FAILURE_RETRY:-60}
 cilium_watchdog_fast_start_wait=${CILIUM_WATCHDOG_FAST_START_WAIT:-60}

diff --git a/scripts/testcase/testcase-basic-v2-ipv4.sh b/scripts/testcase/testcase-basic-v2-ipv4.sh
@@ -21,8 +21,8 @@ function before_test() {
       *http://metadata.google.internal/computeMetadata/v1/instance/network-interfaces/0*)
         echo '{"ipv6s": ["2600:1900:4000:318:0:7:0:0"]}'
         ;;
-      *https://10.0.0.1:443/api/v1/nodes/*)
-        echo '{
+      *https://10.0.0.1:443/api/v1/nodes*)
+        echo '{"object":{
                 "metadata": {
                   "labels": {
                   },
@@ -38,7 +38,7 @@ function before_test() {
                   ],
                   "providerID": "gce://my-gke-project/us-central1-c/gke-my-cluster-default-pool-128bc25d-9c94"
                 }
-              }'
+              }}'
         ;;
       *)
         #unsupported

diff --git a/scripts/testcase/testcase-basic-v2-ipv6.sh b/scripts/testcase/testcase-basic-v2-ipv6.sh
@@ -21,8 +21,8 @@ function before_test() {
       *http://metadata.google.internal/computeMetadata/v1/instance/network-interfaces/0*)
         echo '{"ipv6s": ["2600:1900:4000:318:0:7:0:0"]}'
         ;;
-      *"https://[fd20::1]:443/api/v1/nodes/"*)
-        echo '{
+      *"https://[fd20::1]:443/api/v1/nodes"*)
+        echo '{"object":{
                 "metadata": {
                   "labels": {
                   },
@@ -38,7 +38,7 @@ function before_test() {
                   ],
                   "providerID": "gce://my-gke-project/us-central1-c/gke-my-cluster-default-pool-128bc25d-9c94"
                 }
-              }'
+              }}'
         ;;
       *)
         #unsupported

diff --git a/scripts/testcase/testcase-basic-v2.sh b/scripts/testcase/testcase-basic-v2.sh
@@ -21,8 +21,8 @@ function before_test() {
       *http://metadata.google.internal/computeMetadata/v1/instance/network-interfaces/0*)
         echo '{"ipv6s": ["2600:1900:4000:318:0:7:0:0"]}'
         ;;
-      *https://kubernetes.default.svc:443/api/v1/nodes/*)
-        echo '{
+      *https://kubernetes.default.svc:443/api/v1/nodes*)
+        echo '{"object":{
                 "metadata": {
                   "labels": {
                   },
@@ -38,7 +38,7 @@ function before_test() {
                   ],
                   "providerID": "gce://my-gke-project/us-central1-c/gke-my-cluster-default-pool-128bc25d-9c94"
                 }
-              }'
+              }}'
         ;;
       *)
         #unsupported

diff --git a/scripts/testcase/testcase-basic.sh b/scripts/testcase/testcase-basic.sh
@@ -19,8 +19,8 @@ function before_test() {
       *http://metadata.google.internal/computeMetadata/v1/instance/network-interfaces/0*)
         echo '{"ipv6s": ["2600:1900:4000:318:0:7:0:0"]}'
         ;;
-      *https://kubernetes.default.svc:443/api/v1/nodes/*)
-        echo '{
+      *https://kubernetes.default.svc:443/api/v1/nodes*)
+        echo '{"object":{
                 "metadata": {
                   "labels": {
                   },
@@ -36,7 +36,7 @@ function before_test() {
                   ],
                   "providerID": "gce://my-gke-project/us-central1-c/gke-my-cluster-default-pool-128bc25d-9c94"
                 }
-              }'
+              }}'
         ;;
       *)
         #unsupported

diff --git a/scripts/testcase/testcase-calico-v2.sh b/scripts/testcase/testcase-calico-v2.sh
@@ -26,8 +26,8 @@ function before_test() {
       *http://metadata.google.internal/computeMetadata/v1/instance/network-interfaces/0*)
         echo '{"ipv6s": ["2600:1900:4000:318:0:7:0:0"]}'
         ;;
-      *https://kubernetes.default.svc:443/api/v1/nodes/*)
-        echo '{
+      *https://kubernetes.default.svc:443/api/v1/nodes*)
+        echo '{"object":{
                 "metadata": {
                   "labels": {
                   },
@@ -43,7 +43,7 @@ function before_test() {
                   ],
                   "providerID": "gce://my-gke-project/us-central1-c/gke-my-cluster-default-pool-128bc25d-9c94"
                 }
-              }'
+              }}'
         ;;
       *)
         #unsupported

diff --git a/scripts/testcase/testcase-calico.sh b/scripts/testcase/testcase-calico.sh
@@ -24,8 +24,8 @@ function before_test() {
       *http://metadata.google.internal/computeMetadata/v1/instance/network-interfaces/0*)
         echo '{"ipv6s": ["2600:1900:4000:318:0:7:0:0"]}'
         ;;
-      *https://kubernetes.default.svc:443/api/v1/nodes/*)
-        echo '{
+      *https://kubernetes.default.svc:443/api/v1/nodes*)
+        echo '{"object":{
                 "metadata": {
                   "labels": {
                   },
@@ -41,7 +41,7 @@ function before_test() {
                   ],
                   "providerID": "gce://my-gke-project/us-central1-c/gke-my-cluster-default-pool-128bc25d-9c94"
                 }
-              }'
+              }}'
         ;;
       *)
         #unsupported

diff --git a/scripts/testcase/testcase-cilium-faststart-v2.sh b/scripts/testcase/testcase-cilium-faststart-v2.sh
@@ -23,8 +23,8 @@ function before_test() {
       *http://metadata.google.internal/computeMetadata/v1/instance/network-interfaces/0*)
         echo '{"ipv6s": ["2600:1900:4000:318:0:7:0:0"]}'
         ;;
-      *https://kubernetes.default.svc:443/api/v1/nodes/*)
-        echo '{
+      *https://kubernetes.default.svc:443/api/v1/nodes*)
+        echo '{"object":{
                 "metadata": {
                   "labels": {
                   },
@@ -40,7 +40,7 @@ function before_test() {
                   ],
                   "providerID": "gce://my-gke-project/us-central1-c/gke-my-cluster-default-pool-128bc25d-9c94"
                 }
-              }'
+              }}'
         ;;
       *http://localhost:63197/*)
         echo 'healthz'

diff --git a/scripts/testcase/testcase-cilium-faststart.sh b/scripts/testcase/testcase-cilium-faststart.sh
@@ -21,8 +21,8 @@ function before_test() {
       *http://metadata.google.internal/computeMetadata/v1/instance/network-interfaces/0*)
         echo '{"ipv6s": ["2600:1900:4000:318:0:7:0:0"]}'
         ;;
-      *https://kubernetes.default.svc:443/api/v1/nodes/*)
-        echo '{
+      *https://kubernetes.default.svc:443/api/v1/nodes*)
+        echo '{"object":{
                 "metadata": {
                   "labels": {
                   },
@@ -38,7 +38,7 @@ function before_test() {
                   ],
                   "providerID": "gce://my-gke-project/us-central1-c/gke-my-cluster-default-pool-128bc25d-9c94"
                 }
-              }'
+              }}'
         ;;
       *http://localhost:63197/*)
         echo 'healthz'

diff --git a/scripts/testcase/testcase-cilium-v2.sh b/scripts/testcase/testcase-cilium-v2.sh
@@ -23,8 +23,8 @@ function before_test() {
       *http://metadata.google.internal/computeMetadata/v1/instance/network-interfaces/0*)
         echo '{"ipv6s": ["2600:1900:4000:318:0:7:0:0"]}'
         ;;
-      *https://kubernetes.default.svc:443/api/v1/nodes/*)
-        echo '{
+      *https://kubernetes.default.svc:443/api/v1/nodes*)
+        echo '{"object":{
                 "metadata": {
                   "labels": {
                   },
@@ -40,7 +40,7 @@ function before_test() {
                   ],
                   "providerID": "gce://my-gke-project/us-central1-c/gke-my-cluster-default-pool-128bc25d-9c94"
                 }
-              }'
+              }}'
         ;;
       *http://localhost:63197/*)
         echo 'healthz'

diff --git a/scripts/testcase/testcase-cilium.sh b/scripts/testcase/testcase-cilium.sh
@@ -21,8 +21,8 @@ function before_test() {
       *http://metadata.google.internal/computeMetadata/v1/instance/network-interfaces/0*)
         echo '{"ipv6s": ["2600:1900:4000:318:0:7:0:0"]}'
         ;;
-      *https://kubernetes.default.svc:443/api/v1/nodes/*)
-        echo '{
+      *https://kubernetes.default.svc:443/api/v1/nodes*)
+        echo '{"object":{
                 "metadata": {
                   "labels": {
                   },
@@ -38,7 +38,7 @@ function before_test() {
                   ],
                   "providerID": "gce://my-gke-project/us-central1-c/gke-my-cluster-default-pool-128bc25d-9c94"
                 }
-              }'
+              }}'
         ;;
       *http://localhost:63197/*)
         echo 'healthz'

diff --git a/scripts/testcase/testcase-directpath-v2.sh b/scripts/testcase/testcase-directpath-v2.sh
@@ -24,9 +24,9 @@ function before_test() {
         # call to GCE metadata server
         echo '{"ipv6s": ["2600:1900:4000:318:0:7:0:0"]}'
         ;;
-      *https://kubernetes.default.svc:443/api/v1/nodes/*)
+      *https://kubernetes.default.svc:443/api/v1/nodes*)
         # call to kube-apiserver
-        echo '{
+        echo '{"object":{
                 "metadata": {
                   "labels": {
                   },
@@ -42,7 +42,7 @@ function before_test() {
                   ],
                   "providerID": "gce://my-gke-project/us-central1-c/gke-my-cluster-default-pool-128bc25d-9c94"
                 }
-              }'
+              }}'
         ;;
       *)
         # unmatched call

diff --git a/scripts/testcase/testcase-directpath.sh b/scripts/testcase/testcase-directpath.sh
@@ -22,9 +22,9 @@ function before_test() {
         # call to GCE metadata server
         echo '{"ipv6s": ["2600:1900:4000:318:0:7:0:0"]}'
         ;;
-      *https://kubernetes.default.svc:443/api/v1/nodes/*)
+      *https://kubernetes.default.svc:443/api/v1/nodes*)
         # call to kube-apiserver
-        echo '{
+        echo '{"object":{
                 "metadata": {
                   "labels": {
                   },
@@ -40,7 +40,7 @@ function before_test() {
                   ],
                   "providerID": "gce://my-gke-project/us-central1-c/gke-my-cluster-default-pool-128bc25d-9c94"
                 }
-              }'
+              }}'
         ;;
       *)
         # unmatched call