Skip to content

Commit 95cb760

Browse files
azep-ninjaazep-ninja
committed
sync fix
1 parent 8b8de44 commit 95cb760

File tree

2 files changed

+124
-72
lines changed

2 files changed

+124
-72
lines changed

‎Dockerfile

+120-72
Original file line numberDiff line numberDiff line change
@@ -1,72 +1,120 @@
1-
# Use Node.js 23.3.0 as specified in package.json
2-
FROM node:23.3.0-slim AS builder
3-
4-
# Install pnpm globally and install necessary build tools
5-
RUN npm install -g pnpm@9.4.0 && \
6-
apt-get update && \
7-
apt-get install -y git python3 make g++ && \
8-
apt-get clean && \
9-
rm -rf /var/lib/apt/lists/*
10-
11-
# Set Python 3 as the default python
12-
RUN ln -s /usr/bin/python3 /usr/bin/python
13-
14-
# Set the working directory
15-
WORKDIR /app
16-
17-
# Copy all workspace files first
18-
COPY package.json pnpm-lock.yaml pnpm-workspace.yaml .npmrc turbo.json ./
19-
COPY packages ./packages
20-
COPY agent ./agent
21-
COPY scripts ./scripts
22-
23-
# Install dependencies and build the project
24-
RUN pnpm install && \
25-
pnpm build-docker && \
26-
pnpm prune --prod
27-
28-
# Create a new stage for the final image
29-
FROM node:23.3.0-slim
30-
31-
# Install runtime dependencies and Google Cloud SDK
32-
RUN npm install -g pnpm@9.4.0 && \
33-
apt-get update && \
34-
apt-get install -y git python3 curl gnupg && \
35-
echo "deb [signed-by=/usr/share/keyrings/cloud.google.gpg] http://packages.cloud.google.com/apt cloud-sdk main" | tee -a /etc/apt/sources.list.d/google-cloud-sdk.list && \
36-
curl https://packages.cloud.google.com/apt/doc/apt-key.gpg | apt-key --keyring /usr/share/keyrings/cloud.google.gpg add - && \
37-
apt-get update && \
38-
apt-get install -y google-cloud-sdk && \
39-
apt-get clean && \
40-
rm -rf /var/lib/apt/lists/*
41-
42-
WORKDIR /app
43-
44-
# Copy built artifacts and production dependencies from the builder stage
45-
COPY --from=builder /app/package.json ./
46-
COPY --from=builder /app/pnpm-workspace.yaml ./
47-
COPY --from=builder /app/.npmrc ./
48-
COPY --from=builder /app/turbo.json ./
49-
COPY --from=builder /app/node_modules ./node_modules
50-
COPY --from=builder /app/agent ./agent
51-
COPY --from=builder /app/packages ./packages
52-
COPY --from=builder /app/scripts ./scripts
53-
54-
# Create characters directory
55-
RUN mkdir -p characters
56-
57-
# Add debugging and error handling to startup command
58-
CMD sh -c '\
59-
echo "Debug: Starting container" && \
60-
echo "Debug: AGENTS_BUCKET_NAME=${AGENTS_BUCKET_NAME}" && \
61-
echo "Debug: CHARACTER_FILE=${CHARACTER_FILE}" && \
62-
echo "Debug: Full GCS path=gs://${AGENTS_BUCKET_NAME}/${CHARACTER_FILE}" && \
63-
if [ -z "${AGENTS_BUCKET_NAME}" ]; then \
64-
echo "Error: AGENTS_BUCKET_NAME is empty" && exit 1; \
65-
fi && \
66-
if [ -z "${CHARACTER_FILE}" ]; then \
67-
echo "Error: CHARACTER_FILE is empty" && exit 1; \
68-
fi && \
69-
echo "Debug: Attempting to copy character file..." && \
70-
gsutil cp gs://${AGENTS_BUCKET_NAME}/${CHARACTER_FILE} characters/${CHARACTER_FILE} && \
71-
echo "Debug: Starting application..." && \
72-
pnpm start --non-interactive --characters=characters/${CHARACTER_FILE}'
1+
#!/bin/bash
2+
3+
# Enable error checking and debugging
4+
set -e # Exit on any error
5+
set -x # Print each command before executing
6+
7+
# Log startup for debugging
8+
echo "Starting initialization script..."
9+
10+
# Check if directory exists first and show current permissions
11+
echo "Checking current state..."
12+
ls -la /mnt || true
13+
ls -la /mnt/stateful_partition || true
14+
15+
# Try creating base directory first
16+
echo "Creating base directory..."
17+
if ! sudo mkdir -p /mnt/stateful_partition; then
18+
echo "Failed to create /mnt/stateful_partition"
19+
# Try to identify the issue
20+
df -h
21+
mount | grep /mnt
22+
ls -la /mnt
23+
exit 1
24+
fi
25+
26+
# Try alternative locations if stateful_partition isn't working
27+
if [ ! -d "/mnt/stateful_partition" ]; then
28+
echo "Using alternative directory /var/lib/qi-agents"
29+
sudo mkdir -p /var/lib/qi-agents || {
30+
echo "Failed to create alternative directory"
31+
exit 1
32+
}
33+
AGENT_DATA_DIR="/var/lib/qi-agents"
34+
else
35+
AGENT_DATA_DIR="/mnt/stateful_partition/qi-agents"
36+
# Create qi-agents directory
37+
sudo mkdir -p "$AGENT_DATA_DIR" || {
38+
echo "Failed to create qi-agents directory"
39+
exit 1
40+
}
41+
fi
42+
43+
# Create data directory
44+
echo "Creating data directory in $AGENT_DATA_DIR..."
45+
sudo mkdir -p "$AGENT_DATA_DIR/data" || {
46+
echo "Failed to create data directory"
47+
exit 1
48+
}
49+
50+
# Set permissions with verbose output
51+
echo "Setting directory permissions..."
52+
sudo chmod -Rv 777 "$AGENT_DATA_DIR" || {
53+
echo "Failed to set permissions"
54+
exit 1
55+
}
56+
57+
# Verify directory structure and permissions
58+
echo "Verifying directory structure..."
59+
ls -la "$AGENT_DATA_DIR"
60+
ls -la "$AGENT_DATA_DIR/data"
61+
62+
# Pull latest image with verification
63+
echo "Pulling latest image..."
64+
if ! docker pull us-central1-docker.pkg.dev/${PROJECT_ID}/${FULL_NAME}:${version}; then
65+
echo "Failed to pull image"
66+
exit 1
67+
fi
68+
69+
# Get secrets from Secret Manager
70+
echo "Fetching secrets..."
71+
SMALL_GOOGLE_MODEL=$(gcloud secrets versions access latest --secret="small-google-model") || {
72+
echo "Failed to fetch small-google-model secret"
73+
exit 1
74+
}
75+
MEDIUM_GOOGLE_MODEL=$(gcloud secrets versions access latest --secret="medium-google-model") || {
76+
echo "Failed to fetch medium-google-model secret"
77+
exit 1
78+
}
79+
GOOGLE_GENERATIVE_AI_API_KEY=$(gcloud secrets versions access latest --secret="google-generative-ai-key") || {
80+
echo "Failed to fetch google-generative-ai-key secret"
81+
exit 1
82+
}
83+
84+
# Debug existing environment variables
85+
echo "Debug: Checking existing environment variables:"
86+
echo "AGENTS_BUCKET_NAME from environment: ${AGENTS_BUCKET_NAME}"
87+
echo "CHARACTER_FILE from environment: ${CHARACTER_FILE}"
88+
89+
# Run container with persistent storage and restart policy
90+
echo "Starting container with environment variables..."
91+
docker run -d \
92+
--name ${FULL_NAME} \
93+
--restart=always \
94+
-v "$AGENT_DATA_DIR/data":/app/agent/data \
95+
-e AGENTS_BUCKET_NAME="${AGENTS_BUCKET_NAME}" \
96+
-e CHARACTER_FILE="${CHARACTER_FILE}" \
97+
-e SMALL_GOOGLE_MODEL="${SMALL_GOOGLE_MODEL}" \
98+
-e MEDIUM_GOOGLE_MODEL="${MEDIUM_GOOGLE_MODEL}" \
99+
-e GOOGLE_GENERATIVE_AI_API_KEY="${GOOGLE_GENERATIVE_AI_API_KEY}" \
100+
-e PORT="8080" \
101+
-e SERVER_PORT="8080" \
102+
us-central1-docker.pkg.dev/${PROJECT_ID}/${FULL_NAME}:${version}
103+
104+
# Verify container is running and check logs
105+
echo "Checking container status and logs..."
106+
sleep 5 # Give container a moment to start
107+
if ! docker ps | grep ${FULL_NAME}; then
108+
echo "Container failed to start. Checking logs:"
109+
docker logs ${FULL_NAME}
110+
exit 1
111+
fi
112+
113+
# Show container logs and verify environment variables
114+
echo "Container environment variables:"
115+
docker exec ${FULL_NAME} env | grep -E 'AGENTS_BUCKET_NAME|CHARACTER_FILE'
116+
117+
echo "Initial container logs:"
118+
docker logs ${FULL_NAME}
119+
120+
echo "Startup script completed successfully"

‎cloudbuild.yaml

+4
Original file line numberDiff line numberDiff line change
@@ -49,6 +49,7 @@ steps:
4949
exit 1
5050
fi
5151
echo "Secret and bucket access verified"
52+
echo "BUCKET_NAME=${bucket_name}" > /workspace/env_vars
5253
5354
# Build the container image
5455
- name: 'gcr.io/cloud-builders/docker'
@@ -86,12 +87,14 @@ steps:
8687
echo "Starting instance creation/update..."
8788
full_name=$(cat /workspace/full_name)
8889
version=$(cat /workspace/version)
90+
bucket_name=$(gcloud secrets versions access latest --secret="agents-bucket-name")
8991
echo "Working with instance: ${full_name}"
9092
echo "Checking for existing instance..."
9193
if [[ $(gcloud compute instances list --filter="name=${full_name}" --format="get(name)") ]]; then
9294
echo "Updating existing instance ${full_name}"
9395
gcloud compute instances update-container ${full_name} \
9496
--container-image us-central1-docker.pkg.dev/$PROJECT_ID/qi-agents/${full_name}:${version} \
97+
--container-env AGENTS_BUCKET_NAME=${bucket_name},CHARACTER_FILE=${_CHARACTER_FILE} \
9598
--zone $_ZONE || { echo "Instance update failed"; exit 1; }
9699
echo "Instance update completed"
97100
else
@@ -104,6 +107,7 @@ steps:
104107
--subnet $_SUBNET \
105108
--no-address \
106109
--boot-disk-size $_DISK_SIZE \
110+
--container-env AGENTS_BUCKET_NAME=${bucket_name},CHARACTER_FILE=${_CHARACTER_FILE} \
107111
--metadata-from-file startup-script=startup.sh || { echo "Instance creation failed"; exit 1; }
108112
echo "Instance creation completed"
109113
fi

0 commit comments

Comments
 (0)