re sync 2

Author: azep-ninja

Dockerfile

@@ -1,4 +1,4 @@
-# Use a specific Node.js version for better reproducibility
+# Use Node.js 23.3.0 as specified in package.json
 FROM node:23.3.0-slim AS builder
 
 # Install pnpm globally and install necessary build tools
@@ -14,19 +14,19 @@ RUN ln -s /usr/bin/python3 /usr/bin/python
 # Set the working directory
 WORKDIR /app
 
-# Copy package.json and other configuration files
+# Copy package files first for better caching
 COPY package.json pnpm-lock.yaml pnpm-workspace.yaml .npmrc turbo.json ./
 
+# Install dependencies and build the project
+RUN pnpm install
+
 # Copy the rest of the application code
 COPY agent ./agent
 COPY packages ./packages
 COPY scripts ./scripts
-# Don't copy characters folder as we'll get it from GCS
 
-# Install dependencies and build the project
-RUN pnpm install --frozen-lockfile \
-    && pnpm build-docker \
-    && pnpm prune --prod
+# Build the project
+RUN pnpm build-docker && pnpm prune --prod
 
 # Create a new stage for the final image
 FROM node:23.3.0-slim
@@ -57,12 +57,19 @@ COPY --from=builder /app/scripts ./scripts
 # Create characters directory
 RUN mkdir -p characters
 
-# Set default environment variables
-ENV PORT=8080
-ENV SERVER_PORT=8080
-
-# Expose the port
-EXPOSE 8080
-
-# Modified command to fetch character file from GCS before starting
-CMD sh -c "gsutil cp gs://${AGENTS_BUCKET_NAME}/${CHARACTER_FILE} characters/${CHARACTER_FILE} && pnpm start --non-interactive --characters=characters/$CHARACTER_FI${CHARACTER_FILE}"
+# Add debugging and error handling to startup command
+CMD sh -c '\
+    echo "Debug: Starting container" && \
+    echo "Debug: AGENTS_BUCKET_NAME=${AGENTS_BUCKET_NAME}" && \
+    echo "Debug: CHARACTER_FILE=${CHARACTER_FILE}" && \
+    echo "Debug: Full GCS path=gs://${AGENTS_BUCKET_NAME}/${CHARACTER_FILE}" && \
+    if [ -z "${AGENTS_BUCKET_NAME}" ]; then \
+        echo "Error: AGENTS_BUCKET_NAME is empty" && exit 1; \
+    fi && \
+    if [ -z "${CHARACTER_FILE}" ]; then \
+        echo "Error: CHARACTER_FILE is empty" && exit 1; \
+    fi && \
+    echo "Debug: Attempting to copy character file..." && \
+    gsutil cp gs://${AGENTS_BUCKET_NAME}/${CHARACTER_FILE} characters/${CHARACTER_FILE} && \
+    echo "Debug: Starting application..." && \
+    pnpm start --non-interactive --characters=characters/${CHARACTER_FILE}'

cloudbuild.yaml

@@ -26,19 +26,29 @@ steps:
     ls -la
     pwd
 
-# Verify secrets access
+# Verify secrets and bucket access
 - name: 'gcr.io/cloud-builders/gcloud'
   entrypoint: 'bash'
   args:
   - '-c'
   - |
     echo "Verifying secrets access..."
-    echo "Testing secret access..."
-    if ! gcloud secrets versions access latest --secret="small-google-model"; then
-      echo "Failed to access secrets"
+    echo "Testing secrets..."
+    if ! gcloud secrets versions access latest --secret="small-google-model" > /dev/null; then
+      echo "Failed to access small-google-model secret"
       exit 1
     fi
-    echo "Secret access verified"
+    if ! gcloud secrets versions access latest --secret="agents-bucket-name" > /dev/null; then
+      echo "Failed to access agents-bucket-name secret"
+      exit 1
+    fi
+    echo "Verifying bucket access..."
+    bucket_name=$(gcloud secrets versions access latest --secret="agents-bucket-name")
+    if ! gsutil ls gs://${bucket_name} > /dev/null; then
+      echo "Failed to access bucket"
+      exit 1
+    fi
+    echo "Secret and bucket access verified"
 
 # Build the container image
 - name: 'gcr.io/cloud-builders/docker'
@@ -104,7 +114,7 @@ substitutions:
   _DISK_SIZE: 30GB
   _NETWORK: agents-vpc
   _SUBNET: agents-subnet
-  _CHARACTER_FILE: defalt.character.json
+  _CHARACTER_FILE: default.character.json
   _SUFFIX: agent
 
 options:

startup.sh

@@ -59,49 +59,59 @@ echo "Verifying directory structure..."
 ls -la "$AGENT_DATA_DIR"
 ls -la "$AGENT_DATA_DIR/data"
 
-# Pull latest image with verification
-echo "Pulling latest image..."
-docker pull us-central1-docker.pkg.dev/${PROJECT_ID}/${FULL_NAME}:${version} 2>&1 | tee /tmp/docker-pull.log || {
-    echo "Failed to pull image, logs:"
-    cat /tmp/docker-pull.log
+# Get secrets from Secret Manager with error checking
+echo "Fetching secrets..."
+AGENTS_BUCKET_NAME=$(gcloud secrets versions access latest --secret="agents-bucket-name") || {
+    echo "Failed to fetch agents-bucket-name secret"
+    exit 1
+}
+SMALL_GOOGLE_MODEL=$(gcloud secrets versions access latest --secret="small-google-model") || {
+    echo "Failed to fetch small-google-model secret"
+    exit 1
+}
+MEDIUM_GOOGLE_MODEL=$(gcloud secrets versions access latest --secret="medium-google-model") || {
+    echo "Failed to fetch medium-google-model secret"
+    exit 1
+}
+GOOGLE_GENERATIVE_AI_API_KEY=$(gcloud secrets versions access latest --secret="google-generative-ai-key") || {
+    echo "Failed to fetch google-generative-ai-key secret"
     exit 1
 }
 
-# Get secrets from Secret Manager
-echo "Fetching secrets..."
-AGENTS_BUCKET_NAME=$(gcloud secrets versions access latest --secret="agents-bucket-name")
-SMALL_GOOGLE_MODEL=$(gcloud secrets versions access latest --secret="small-google-model")
-MEDIUM_GOOGLE_MODEL=$(gcloud secrets versions access latest --secret="medium-google-model")
-GOOGLE_GENERATIVE_AI_API_KEY=$(gcloud secrets versions access latest --secret="google-generative-ai-key")
+# Debug environment variables
+echo "Debug: Environment variables:"
+echo "AGENTS_BUCKET_NAME: ${AGENTS_BUCKET_NAME:-not set}"
+echo "CHARACTER_FILE: ${CHARACTER_FILE:-not set}"
+echo "SMALL_GOOGLE_MODEL: ${SMALL_GOOGLE_MODEL:0:5}... (truncated)"
+echo "MEDIUM_GOOGLE_MODEL: ${MEDIUM_GOOGLE_MODEL:0:5}... (truncated)"
+echo "GOOGLE_GENERATIVE_AI_API_KEY: ${GOOGLE_GENERATIVE_AI_API_KEY:0:5}... (truncated)"
 
 # Run container with persistent storage and restart policy
 echo "Starting container..."
 docker run -d \
     --name ${FULL_NAME} \
     --restart=always \
     -v "$AGENT_DATA_DIR/data":/app/agent/data \
-    -e AGENTS_BUCKET_NAME="${AGENTS_BUCKET_NAME}" \
-    -e CHARACTER_FILE="${CHARACTER_FILE}" \
-    -e SMALL_GOOGLE_MODEL="${SMALL_GOOGLE_MODEL}" \
-    -e MEDIUM_GOOGLE_MODEL="${MEDIUM_GOOGLE_MODEL}" \
-    -e GOOGLE_GENERATIVE_AI_API_KEY="${GOOGLE_GENERATIVE_AI_API_KEY}" \
+    -e AGENTS_BUCKET_NAME="${AGENTS_BUCKET_NAME:?'AGENTS_BUCKET_NAME not set'}" \
+    -e CHARACTER_FILE="${CHARACTER_FILE:?'CHARACTER_FILE not set'}" \
+    -e SMALL_GOOGLE_MODEL="${SMALL_GOOGLE_MODEL:?'SMALL_GOOGLE_MODEL not set'}" \
+    -e MEDIUM_GOOGLE_MODEL="${MEDIUM_GOOGLE_MODEL:?'MEDIUM_GOOGLE_MODEL not set'}" \
+    -e GOOGLE_GENERATIVE_AI_API_KEY="${GOOGLE_GENERATIVE_AI_API_KEY:?'GOOGLE_GENERATIVE_AI_API_KEY not set'}" \
     -e PORT="8080" \
     -e SERVER_PORT="8080" \
     us-central1-docker.pkg.dev/${PROJECT_ID}/${FULL_NAME}:${version}
 
-# Verify container is running
-echo "Container started. Checking status..."
+# Verify container is running and check logs
+echo "Checking container status and logs..."
+sleep 5  # Give container a moment to start
 if ! docker ps | grep ${FULL_NAME}; then
     echo "Container failed to start. Checking logs:"
     docker logs ${FULL_NAME}
     exit 1
 fi
 
-# Final verification
-echo "Verifying final state..."
-df -h
-mount
-docker ps
-ls -la "$AGENT_DATA_DIR"
+# Show container logs
+echo "Initial container logs:"
+docker logs ${FULL_NAME}
 
 echo "Startup script completed successfully"