feat(prometheus): Phase 2 - Chrome & Chrome-Go metrics endpoints

docker/Dockerfile.chrome

@@ -211,6 +211,15 @@ WORKDIR /home/runner
 COPY --chown=runner:runner entrypoint-chrome.sh /entrypoint.sh
 RUN chmod +x /entrypoint.sh
 
+# Copy Prometheus metrics scripts (Phase 2: Prometheus Monitoring)
+# TASK-014: Add metrics scripts to Chrome runner
+COPY --chown=runner:runner metrics-server.sh /usr/local/bin/metrics-server.sh
+COPY --chown=runner:runner metrics-collector.sh /usr/local/bin/metrics-collector.sh
+RUN chmod +x /usr/local/bin/metrics-server.sh /usr/local/bin/metrics-collector.sh
+
+# TASK-014: Expose Prometheus metrics port
+EXPOSE 9091
+
 HEALTHCHECK --interval=30s --timeout=10s --start-period=60s --retries=3 \
   CMD pgrep -f "Runner.Listener" > /dev/null || exit 1
 

docker/Dockerfile.chrome-go

@@ -151,7 +151,7 @@
 # --- INSTALL GO ---
 # Use BuildKit cache for Go download
 # Go supports both amd64 and arm64 architectures
 RUN --mount=type=cache,target=/tmp/downloads \
     GO_VERSION="1.25.5" \
     && case ${TARGETARCH} in \
         "amd64")  GO_ARCH="amd64"  ;; \
@@ -243,6 +243,15 @@
 COPY --chown=runner:runner entrypoint-chrome.sh /entrypoint.sh
 RUN chmod +x /entrypoint.sh
 
+# Copy Prometheus metrics scripts (Phase 2: Prometheus Monitoring)
+# TASK-015: Add metrics scripts to Chrome-Go runner
+COPY --chown=runner:runner metrics-server.sh /usr/local/bin/metrics-server.sh
+COPY --chown=runner:runner metrics-collector.sh /usr/local/bin/metrics-collector.sh
+RUN chmod +x /usr/local/bin/metrics-server.sh /usr/local/bin/metrics-collector.sh
+
+# TASK-015: Expose Prometheus metrics port
+EXPOSE 9091
+
 HEALTHCHECK --interval=30s --timeout=10s --start-period=60s --retries=3 \
   CMD pgrep -f "Runner.Listener" > /dev/null || exit 1
 

docker/docker-compose.chrome-go.yml

@@ -3,6 +3,9 @@ services:
     image: ghcr.io/grammatonic/github-runner:chrome-go-latest
     container_name: github-runner-chrome-go
     restart: unless-stopped
+    # TASK-017: Expose metrics port on unique host port to avoid conflicts
+    ports:
+      - "9093:9091"  # Prometheus metrics endpoint (host:container)
     environment:
       - GITHUB_TOKEN=${GITHUB_TOKEN}
       - GITHUB_REPOSITORY=${GITHUB_REPOSITORY}
@@ -14,6 +17,10 @@ services:
       - RUNNER_REPLACE_EXISTING=${RUNNER_REPLACE_EXISTING:-true}
       - DISPLAY=${DISPLAY:-:99}
       - CHROME_FLAGS=${CHROME_FLAGS:---headless --no-sandbox --disable-dev-shm-usage --disable-gpu}
+      # TASK-019: Prometheus metrics configuration
+      - RUNNER_TYPE=chrome-go
+      - METRICS_PORT=9091
+      - METRICS_UPDATE_INTERVAL=${METRICS_UPDATE_INTERVAL:-30}
       # Go-specific environment variables
       - GO_VERSION=${GO_VERSION:-1.25.4}
       - GOPATH=${GOPATH:-/home/runner/go}
@@ -41,6 +48,8 @@ services:
       - chrome-go-cache-pip:/home/runner/.runnercache/pip
       - chrome-go-cache-go:/home/runner/go/pkg
       - /dev/shm:/dev/shm
+      # TASK-017: Persist job log for metrics across restarts
+      - chrome-go-jobs-log:/tmp
     # Drop all capabilities except those needed for Chrome and Docker socket
     cap_drop:
       - ALL
@@ -81,6 +90,9 @@ volumes:
     driver: local
   chrome-go-cache-go:
     driver: local
+  # TASK-017: Volume for persistent job log
+  chrome-go-jobs-log:
+    driver: local
 
 networks:
   runner-network:

docker/docker-compose.chrome.yml

@@ -3,6 +3,9 @@ services:
     image: ghcr.io/grammatonic/github-runner:chrome-latest
     container_name: github-runner-chrome
     restart: unless-stopped
+    # TASK-016: Expose metrics port on unique host port to avoid conflicts
+    ports:
+      - "9092:9091"  # Prometheus metrics endpoint (host:container)
     environment:
       - GITHUB_TOKEN=${GITHUB_TOKEN}
       - GITHUB_REPOSITORY=${GITHUB_REPOSITORY}
@@ -16,13 +19,19 @@ services:
       - RUNNER_REPLACE_EXISTING=${RUNNER_REPLACE_EXISTING:-true}
       - DISPLAY=${DISPLAY:-:99}
       - CHROME_FLAGS=${CHROME_FLAGS:---headless --no-sandbox --disable-dev-shm-usage --disable-gpu}
+      # TASK-018: Prometheus metrics configuration
+      - RUNNER_TYPE=chrome
+      - METRICS_PORT=9091
+      - METRICS_UPDATE_INTERVAL=${METRICS_UPDATE_INTERVAL:-30}
     volumes:
       - /var/run/docker.sock:/var/run/docker.sock
       - chrome-cache:/home/runner/_work
       - chrome-config:/home/runner/.config
       - chrome-cache-npm:/home/runner/.runnercache/npm
       - chrome-cache-pip:/home/runner/.runnercache/pip
       - /dev/shm:/dev/shm
+      # TASK-016: Persist job log for metrics across restarts
+      - chrome-jobs-log:/tmp
     # Drop all capabilities except those needed for Chrome and Docker socket
     cap_drop:
       - ALL
@@ -61,6 +70,9 @@ volumes:
     driver: local
   chrome-cache-pip:
     driver: local
+  # TASK-016: Volume for persistent job log
+  chrome-jobs-log:
+    driver: local
 
 networks:
   runner-network:

docker/entrypoint-chrome.sh

@@ -44,6 +44,47 @@ GITHUB_HOST="${GITHUB_HOST:-github.com}" # For GitHub Enterprise
 # Validate GitHub host
 validate_hostname "$GITHUB_HOST" || exit 1
 
+# --- METRICS SETUP (Phase 2: Prometheus Monitoring) ---
+# Start metrics services BEFORE token validation to enable standalone testing
+# TASK-013: Initialize job log
+JOBS_LOG="${JOBS_LOG:-/tmp/jobs.log}"
+echo "Initializing job log: ${JOBS_LOG}"
+touch "${JOBS_LOG}"
+
+# TASK-013: Start metrics collection services
+METRICS_PORT="${METRICS_PORT:-9091}"
+METRICS_FILE="${METRICS_FILE:-/tmp/runner_metrics.prom}"
+RUNNER_TYPE="${RUNNER_TYPE:-chrome}"
+
+echo "Starting Prometheus metrics services..."
+echo "  - Metrics endpoint: http://localhost:${METRICS_PORT}/metrics"
+echo "  - Runner type: ${RUNNER_TYPE}"
+
+# Start metrics collector in background
+if [ -f "/usr/local/bin/metrics-collector.sh" ]; then
+	RUNNER_NAME="${RUNNER_NAME}" \
+		RUNNER_TYPE="${RUNNER_TYPE}" \
+		METRICS_FILE="${METRICS_FILE}" \
+		JOBS_LOG="${JOBS_LOG}" \
+		UPDATE_INTERVAL="${METRICS_UPDATE_INTERVAL:-30}" \
+		/usr/local/bin/metrics-collector.sh &
+	COLLECTOR_PID=$!
+	echo "Metrics collector started (PID: ${COLLECTOR_PID})"
+else
+	echo "Warning: metrics-collector.sh not found, metrics collection disabled"
+fi
+
+# Start metrics HTTP server in background
+if [ -f "/usr/local/bin/metrics-server.sh" ]; then
+	METRICS_PORT="${METRICS_PORT}" \
+		METRICS_FILE="${METRICS_FILE}" \
+		/usr/local/bin/metrics-server.sh &
+	SERVER_PID=$!
+	echo "Metrics server started (PID: ${SERVER_PID})"
+else
+	echo "Warning: metrics-server.sh not found, metrics endpoint disabled"
+fi
+
 # Change to the runner's directory
 cd /actions-runner
 
@@ -87,7 +128,21 @@ echo "Configuring runner..."
 
 # Function to clean up the runner on exit
 cleanup() {
-	echo "Signal received, removing runner registration..."
+	echo "Signal received, shutting down..."
+
+	# Stop metrics services
+	if [ -n "${COLLECTOR_PID:-}" ]; then
+		echo "Stopping metrics collector (PID: ${COLLECTOR_PID})..."
+		kill -TERM "${COLLECTOR_PID}" 2>/dev/null || true
+	fi
+
+	if [ -n "${SERVER_PID:-}" ]; then
+		echo "Stopping metrics server (PID: ${SERVER_PID})..."
+		kill -TERM "${SERVER_PID}" 2>/dev/null || true
+	fi
+
+	# Remove runner registration
+	echo "Removing runner registration..."
 	./config.sh remove --token "${RUNNER_TOKEN}"
 	echo "Runner registration removed."
 }