Add monitoring to devcontainer startup

startupscript/butane/004-git-clone-devcontainer.sh

@@ -17,19 +17,7 @@ if [[ $# -lt 1 ]]; then
     usage
 fi
 
-
-# Map the server to appropriate service path
-function get_service_url() {
-  case "$1" in
-    "dev-stable") echo "https://workbench-dev.verily.com/api/$2" ;;
-    "dev-unstable") echo "https://workbench-dev-unstable.verily.com/api/$2" ;;
-    "test") echo "https://workbench-test.verily.com/api/$2" ;;
-    "prod") echo "https://workbench.verily.com/api/$2" ;;
-    *) return 1 ;;
-  esac
-}
-readonly -f get_service_url
-
+source /home/core/service-utils.sh
 source /home/core/metadata-utils.sh
 
 # To accommodate the use of SSH URLs for public Git repositories, set the following Git configuration:
@@ -58,12 +46,7 @@ api_url="${api_url%.git}"
 private_status=$(curl --retry 5 -s "${api_url}" | jq -r ".status")
 if [[ "${PRIVATE_DEVCONTAINER_ENABLED}" == "TRUE" && "${private_status}" == 404 ]]; then
   # Get ECM service URL
-  SERVER="$(get_metadata_value "terra-cli-server" "")"
-  readonly SERVER
-  if [[ -z "${SERVER}" ]]; then
-    SERVER="prod"
-  fi
-  if ! ECM_SERVICE_URL="$(get_service_url "${SERVER}" "ecm")"; then
+  if ! ECM_SERVICE_URL="$(get_service_url "ecm")"; then
     exit 1
   fi
 

startupscript/butane/monitoring-utils.sh

@@ -0,0 +1,63 @@
+#!/bin/bash
+# monitoring-utils.sh defines helper functions for notifying WSM of VM startup states.
+
+source /home/core/service-utils.sh
+
+WORKSPACE_ID="$(get_metadata_value "terra-workspace-id" "")"
+RESOURCE_ID="$(get_metadata_value "terra-resource-id" "")"
+readonly  WORKSPACE_ID RESOURCE_ID
+
+# Get WSM endpoint URL
+if ! WSM_SERVICE_URL="$(get_service_url "wsm")"; then
+    exit 1
+fi
+LOG_URL="${WSM_SERVICE_URL}/api/workspaces/${WORKSPACE_ID}/resource/${RESOURCE_ID}/instance-state"
+
+function record_devcontainer_end() {
+    if [[ $# -lt 2 || ("$2" != "0" && "$2" != "1") ]]; then
+        echo "usage: record_devcontainer_end <success/fail - 1/0>"
+        exit 1
+    fi
+    SUCCESS="$1"
+    payload=$(cat <<EOF
+{
+  "state": "DEVCONTAINER_END",
+  "success": ${SUCCESS}
+}
+EOF
+)
+    response=$(curl -s -X POST "${LOG_URL}" \
+        -H "Authorization: Bearer $(/home/core/wb.sh auth print-access-token)" \
+        -H "Content-Type: application/json" \
+        -d "${payload}" \
+        -w "\n%{http_code}")
+    http_code=$(echo "$response" | tail -n1)
+    response_body=$(echo "$response" | head -n -1)
+    if [[ $http_code -ne 200 ]]; then
+        echo "Failed to record VM state. HTTP ${http_code}: ${response_body}" >&2
+        return 1
+    fi
+    echo "VM state recorded successfully: ${response_body}"
+}
+
+function record_devcontainer_start() {
+    payload=$(cat <<EOF
+{
+  "state": "DEVCONTAINER_START",
+  "success": 1
+}
+EOF
+)
+    response=$(curl -s -X POST "${LOG_URL}" \
+        -H "Authorization: Bearer $(/home/core/wb.sh auth print-access-token)" \
+        -H "Content-Type: application/json" \
+        -d "${payload}" \
+        -w "\n%{http_code}")
+    http_code=$(echo "$response" | tail -n1)
+    response_body=$(echo "$response" | head -n -1)
+    if [[ $http_code -ne 200 ]]; then
+        echo "Failed to record VM state. HTTP ${http_code}: ${response_body}" >&2
+        return 1
+    fi
+    echo "VM state recorded successfully: ${response_body}"
+}

startupscript/butane/pre-devcontainer.sh

@@ -1,6 +1,16 @@
 #!/bin/bash
 
-# pre-devcontainer.sh creates a file used by the devcontainer service to keep
-# track of the number of service failures.
+# pre-devcontainer.sh creates a file used by the devcontainer service for monitoring
+# and to keep track of the number of service failures.
 
-touch /tmp/devcontainer-failure-count
+touch /tmp/devcontainer-failure-count
+
+MONITORING_UTILS_FILE="/home/core/monitoring-utils.sh"
+FIRST_BOOT_START_FILE="/home/core/first-boot-start"
+if [[ -f "${MONITORING_UTILS_FILE}" && ! -f "${FIRST_BOOT_START_FILE}" ]]; then
+    # First boot file does not exist
+    # Record startup begin for monitoring
+    source "${MONITORING_UTILS_FILE}"
+    record_devcontainer_start "${CLOUD_PLATFORM}"
+fi
+touch "${FIRST_BOOT_START_FILE}"

startupscript/butane/probe-proxy-readiness.sh

@@ -15,9 +15,21 @@ if docker ps -q --filter "name=proxy-agent" | grep -q . \
     && docker ps -q --filter "name=application-server" | grep -q .; then
     echo "Proxy is ready."
     status="$(get_guest_attribute "startup_script/status" "")"
+    success=0
     if [[ "${status}" != "ERROR" ]]; then
         set_metadata "startup_script/status" "COMPLETE"
+        success=1
     fi
+
+    FIRST_BOOT_END_FILE="/home/core/first-boot-end"
+    MONITORING_UTILS_FILE="/home/core/monitoring-utils.sh"
+    if [[ -f "${MONITORING_UTILS_FILE}" && ! -f "${FIRST_BOOT_END_FILE}" ]]; then
+        # first boot file does not exist
+        # record startup end for monitoring
+        source "${MONITORING_UTILS_FILE}"
+        record_devcontainer_end "${success}"
+    fi
+    touch "${FIRST_BOOT_END_FILE}"
 else
     echo "proxy-agent or application-server is not started"
     exit 1

startupscript/butane/service-utils.sh

@@ -0,0 +1,21 @@
+#!/bin/bash
+# service-utils defines helper functions for determining service path.
+
+# Map the server to appropriate service path
+function get_service_url() {
+  SERVER="$(get_metadata_value "terra-cli-server" "")"
+  if [[ -z "${SERVER}" ]]; then
+    SERVER="dev-stable"
+  fi
+  readonly SERVER
+
+
+  case "${SERVER}" in
+    "dev-stable") echo "https://workbench-dev.verily.com/api/${SERVER}" ;;
+    "dev-unstable") echo "https://workbench-dev-unstable.verily.com/api/${SERVER}" ;;
+    "test") echo "https://workbench-test.verily.com/api/${SERVER}" ;;
+    "prod") echo "https://workbench.verily.com/api/${SERVER}" ;;
+    *) return 1 ;;
+  esac
+}
+readonly -f get_service_url