Initial commit: hardened DeerFlow factory

Vendored deer-flow upstream (bytedance/deer-flow) plus prompt-injection
hardening:

- New deerflow.security package: content_delimiter, html_cleaner,
  sanitizer (8 layers — invisible chars, control chars, symbols, NFC,
  PUA, tag chars, horizontal whitespace collapse with newline/tab
  preservation, length cap)
- New deerflow.community.searx package: web_search, web_fetch,
  image_search backed by a private SearX instance, every external
  string sanitized and wrapped in <<<EXTERNAL_UNTRUSTED_CONTENT>>>
  delimiters
- All native community web providers (ddg_search, tavily, exa,
  firecrawl, jina_ai, infoquest, image_search) replaced with hard-fail
  stubs that raise NativeWebToolDisabledError at import time, so a
  misconfigured tool.use path fails loud rather than silently falling
  back to unsanitized output
- Native client back-doors (jina_client.py, infoquest_client.py)
  stubbed too
- Native-tool tests quarantined under tests/_disabled_native/
  (collect_ignore_glob via local conftest.py)
- Sanitizer Layer 7 fix: only collapse horizontal whitespace, preserve
  newlines and tabs so list/table structure survives
- Hardened runtime config.yaml references only the searx-backed tools
- Factory overlay (backend/) kept in sync with deer-flow tree as a
  reference / source

See HARDENING.md for the full audit trail and verification steps.
This commit is contained in:
2026-04-12 14:23:57 +02:00
commit 6de0bf9f5b
889 changed files with 173052 additions and 0 deletions

View File

@@ -0,0 +1,253 @@
# DeerFlow Development Environment
# Usage: docker-compose -f docker-compose-dev.yaml up --build
#
# Services:
# - nginx: Reverse proxy (port 2026)
# - frontend: Frontend Next.js dev server (port 3000)
# - gateway: Backend Gateway API (port 8001)
# - langgraph: LangGraph server (port 2024)
# - provisioner (optional): Sandbox provisioner (creates Pods in host Kubernetes)
#
# Prerequisites:
# - Kubernetes cluster + kubeconfig are only required when using provisioner mode.
#
# Access: http://localhost:2026
services:
# ── Sandbox Provisioner ────────────────────────────────────────────────
# Manages per-sandbox Pod + Service lifecycle in the host Kubernetes
# cluster via the K8s API.
# Backend accesses sandboxes directly via host.docker.internal:{NodePort}.
provisioner:
build:
context: ./provisioner
dockerfile: Dockerfile
args:
APT_MIRROR: ${APT_MIRROR:-}
container_name: deer-flow-provisioner
volumes:
- ~/.kube/config:/root/.kube/config:ro
environment:
- K8S_NAMESPACE=deer-flow
- SANDBOX_IMAGE=enterprise-public-cn-beijing.cr.volces.com/vefaas-public/all-in-one-sandbox:latest
# Host paths for K8s HostPath volumes (must be absolute paths accessible by K8s node)
# On Docker Desktop/OrbStack, use your actual host paths like /Users/username/...
# Set these in your shell before running docker-compose:
# export DEER_FLOW_ROOT=/absolute/path/to/deer-flow
- SKILLS_HOST_PATH=${DEER_FLOW_ROOT}/skills
- THREADS_HOST_PATH=${DEER_FLOW_ROOT}/backend/.deer-flow/threads
# Production: use PVC instead of hostPath to avoid data loss on node failure.
# When set, hostPath vars above are ignored for the corresponding volume.
# USERDATA_PVC_NAME uses subPath (threads/{thread_id}/user-data) automatically.
# - SKILLS_PVC_NAME=deer-flow-skills-pvc
# - USERDATA_PVC_NAME=deer-flow-userdata-pvc
- KUBECONFIG_PATH=/root/.kube/config
- NODE_HOST=host.docker.internal
# Override K8S API server URL since kubeconfig uses 127.0.0.1
# which is unreachable from inside the container
- K8S_API_SERVER=https://host.docker.internal:26443
env_file:
- ../.env
extra_hosts:
- "host.docker.internal:host-gateway"
networks:
- deer-flow-dev
restart: unless-stopped
healthcheck:
test: ["CMD", "curl", "-f", "http://localhost:8002/health"]
interval: 10s
timeout: 5s
retries: 6
start_period: 15s
# ── Reverse Proxy ──────────────────────────────────────────────────────
# Routes API traffic to gateway/langgraph and (optionally) provisioner.
# LANGGRAPH_UPSTREAM and LANGGRAPH_REWRITE control gateway vs standard
# routing (processed by envsubst at container start).
nginx:
image: nginx:alpine
container_name: deer-flow-nginx
ports:
- "2026:2026"
volumes:
- ./nginx/nginx.conf:/etc/nginx/nginx.conf.template:ro
environment:
- LANGGRAPH_UPSTREAM=${LANGGRAPH_UPSTREAM:-langgraph:2024}
- LANGGRAPH_REWRITE=${LANGGRAPH_REWRITE:-/}
command:
- sh
- -c
- |
set -e
envsubst '$$LANGGRAPH_UPSTREAM $$LANGGRAPH_REWRITE' \
< /etc/nginx/nginx.conf.template > /etc/nginx/nginx.conf
test -e /proc/net/if_inet6 || sed -i '/^[[:space:]]*listen[[:space:]]\+\[::\]:2026;/d' /etc/nginx/nginx.conf
exec nginx -g 'daemon off;'
depends_on:
- frontend
- gateway
networks:
- deer-flow-dev
restart: unless-stopped
# Frontend - Next.js Development Server
frontend:
build:
context: ../
dockerfile: frontend/Dockerfile
target: dev
args:
PNPM_STORE_PATH: ${PNPM_STORE_PATH:-/root/.local/share/pnpm/store}
NPM_REGISTRY: ${NPM_REGISTRY:-}
container_name: deer-flow-frontend
command: sh -c "cd frontend && pnpm run dev > /app/logs/frontend.log 2>&1"
volumes:
- ../frontend/src:/app/frontend/src
- ../frontend/public:/app/frontend/public
- ../frontend/next.config.js:/app/frontend/next.config.js:ro
- ../logs:/app/logs
# Mount pnpm store for caching
- ${PNPM_STORE_PATH:-~/.local/share/pnpm/store}:/root/.local/share/pnpm/store
working_dir: /app
environment:
- NODE_ENV=development
- WATCHPACK_POLLING=true
- CI=true
- DEER_FLOW_INTERNAL_GATEWAY_BASE_URL=http://gateway:8001
- DEER_FLOW_INTERNAL_LANGGRAPH_BASE_URL=http://langgraph:2024
env_file:
- ../frontend/.env
networks:
- deer-flow-dev
restart: unless-stopped
# Backend - Gateway API
gateway:
build:
context: ../
dockerfile: backend/Dockerfile
target: dev
# cache_from disabled - requires manual setup: mkdir -p /tmp/docker-cache-gateway
args:
APT_MIRROR: ${APT_MIRROR:-}
UV_IMAGE: ${UV_IMAGE:-ghcr.io/astral-sh/uv:0.7.20}
UV_INDEX_URL: ${UV_INDEX_URL:-https://pypi.org/simple}
container_name: deer-flow-gateway
command: sh -c "{ cd backend && (uv sync || (echo '[startup] uv sync failed; recreating .venv and retrying once' && uv venv --allow-existing .venv && uv sync)) && PYTHONPATH=. uv run uvicorn app.gateway.app:app --host 0.0.0.0 --port 8001 --reload --reload-include='*.yaml .env'; } > /app/logs/gateway.log 2>&1"
volumes:
- ../backend/:/app/backend/
# Preserve the .venv built during Docker image build — mounting the full backend/
# directory above would otherwise shadow it with the (empty) host directory.
- gateway-venv:/app/backend/.venv
- ../config.yaml:/app/config.yaml
- ../extensions_config.json:/app/extensions_config.json
- ../skills:/app/skills
- ../logs:/app/logs
# Use a Docker-managed uv cache volume instead of a host bind mount.
# On macOS/Docker Desktop, uv may fail to create symlinks inside shared
# host directories, which causes startup-time `uv sync` to crash.
- gateway-uv-cache:/root/.cache/uv
# DooD: same as gateway — AioSandboxProvider runs inside LangGraph process.
- /var/run/docker.sock:/var/run/docker.sock
# CLI auth directories for auto-auth (Claude Code + Codex CLI)
- type: bind
source: ${HOME:?HOME must be set}/.claude
target: /root/.claude
read_only: true
bind:
create_host_path: true
- type: bind
source: ${HOME:?HOME must be set}/.codex
target: /root/.codex
read_only: true
bind:
create_host_path: true
working_dir: /app
environment:
- CI=true
- DEER_FLOW_HOME=/app/backend/.deer-flow
- DEER_FLOW_CHANNELS_LANGGRAPH_URL=${DEER_FLOW_CHANNELS_LANGGRAPH_URL:-http://langgraph:2024}
- DEER_FLOW_CHANNELS_GATEWAY_URL=${DEER_FLOW_CHANNELS_GATEWAY_URL:-http://gateway:8001}
- DEER_FLOW_HOST_BASE_DIR=${DEER_FLOW_ROOT}/backend/.deer-flow
- DEER_FLOW_HOST_SKILLS_PATH=${DEER_FLOW_ROOT}/skills
- DEER_FLOW_SANDBOX_HOST=host.docker.internal
env_file:
- ../.env
extra_hosts:
# For Linux: map host.docker.internal to host gateway
- "host.docker.internal:host-gateway"
networks:
- deer-flow-dev
restart: unless-stopped
# Backend - LangGraph Server
langgraph:
build:
context: ../
dockerfile: backend/Dockerfile
target: dev
# cache_from disabled - requires manual setup: mkdir -p /tmp/docker-cache-langgraph
args:
APT_MIRROR: ${APT_MIRROR:-}
UV_IMAGE: ${UV_IMAGE:-ghcr.io/astral-sh/uv:0.7.20}
UV_INDEX_URL: ${UV_INDEX_URL:-https://pypi.org/simple}
container_name: deer-flow-langgraph
command: sh -c "cd backend && { (uv sync || (echo '[startup] uv sync failed; recreating .venv and retrying once' && uv venv --allow-existing .venv && uv sync)) && allow_blocking='' && if [ \"\${LANGGRAPH_ALLOW_BLOCKING:-0}\" = '1' ]; then allow_blocking='--allow-blocking'; fi && uv run langgraph dev --no-browser \${allow_blocking} --host 0.0.0.0 --port 2024 --n-jobs-per-worker \${LANGGRAPH_JOBS_PER_WORKER:-10}; } > /app/logs/langgraph.log 2>&1"
volumes:
- ../backend/:/app/backend/
# Preserve the .venv built during Docker image build — mounting the full backend/
# directory above would otherwise shadow it with the (empty) host directory.
- langgraph-venv:/app/backend/.venv
- ../config.yaml:/app/config.yaml
- ../extensions_config.json:/app/extensions_config.json
- ../skills:/app/skills
- ../logs:/app/logs
# Use a Docker-managed uv cache volume instead of a host bind mount.
# On macOS/Docker Desktop, uv may fail to create symlinks inside shared
# host directories, which causes startup-time `uv sync` to crash.
- langgraph-uv-cache:/root/.cache/uv
# DooD: same as gateway — AioSandboxProvider runs inside LangGraph process.
- /var/run/docker.sock:/var/run/docker.sock
# CLI auth directories for auto-auth (Claude Code + Codex CLI)
- type: bind
source: ${HOME:?HOME must be set}/.claude
target: /root/.claude
read_only: true
bind:
create_host_path: true
- type: bind
source: ${HOME:?HOME must be set}/.codex
target: /root/.codex
read_only: true
bind:
create_host_path: true
working_dir: /app
environment:
- CI=true
- DEER_FLOW_HOME=/app/backend/.deer-flow
- DEER_FLOW_HOST_BASE_DIR=${DEER_FLOW_ROOT}/backend/.deer-flow
- DEER_FLOW_HOST_SKILLS_PATH=${DEER_FLOW_ROOT}/skills
- DEER_FLOW_SANDBOX_HOST=host.docker.internal
env_file:
- ../.env
extra_hosts:
# For Linux: map host.docker.internal to host gateway
- "host.docker.internal:host-gateway"
networks:
- deer-flow-dev
restart: unless-stopped
volumes:
# Persist .venv across container restarts so dependencies installed during
# image build are not shadowed by the host backend/ directory mount.
gateway-venv:
langgraph-venv:
gateway-uv-cache:
langgraph-uv-cache:
networks:
deer-flow-dev:
driver: bridge
ipam:
config:
- subnet: 192.168.200.0/24

View File

@@ -0,0 +1,202 @@
# DeerFlow Production Environment
# Usage: make up
#
# Services:
# - nginx: Reverse proxy (port 2026, configurable via PORT env var)
# - frontend: Next.js production server
# - gateway: FastAPI Gateway API
# - langgraph: LangGraph production server (Dockerfile generated by langgraph dockerfile)
# - provisioner: (optional) Sandbox provisioner for Kubernetes mode
#
# Key environment variables (set via environment/.env or scripts/deploy.sh):
# DEER_FLOW_HOME — runtime data dir, default $REPO_ROOT/backend/.deer-flow
# DEER_FLOW_CONFIG_PATH — path to config.yaml
# DEER_FLOW_EXTENSIONS_CONFIG_PATH — path to extensions_config.json
# DEER_FLOW_DOCKER_SOCKET — Docker socket path, default /var/run/docker.sock
# DEER_FLOW_REPO_ROOT — repo root (used for skills host path in DooD)
# BETTER_AUTH_SECRET — required for frontend auth/session security
#
# LangSmith tracing is disabled by default (LANGSMITH_TRACING=false).
# Set LANGSMITH_TRACING=true and LANGSMITH_API_KEY in .env to enable it.
#
# Access: http://localhost:${PORT:-2026}
services:
# ── Reverse Proxy ──────────────────────────────────────────────────────────
nginx:
image: nginx:alpine
container_name: deer-flow-nginx
ports:
- "${PORT:-2026}:2026"
volumes:
- ./nginx/nginx.conf:/etc/nginx/nginx.conf.template:ro
environment:
- LANGGRAPH_UPSTREAM=${LANGGRAPH_UPSTREAM:-langgraph:2024}
- LANGGRAPH_REWRITE=${LANGGRAPH_REWRITE:-/}
command: >
sh -c "envsubst '$$LANGGRAPH_UPSTREAM $$LANGGRAPH_REWRITE'
< /etc/nginx/nginx.conf.template > /etc/nginx/nginx.conf
&& nginx -g 'daemon off;'"
depends_on:
- frontend
- gateway
networks:
- deer-flow
restart: unless-stopped
# ── Frontend: Next.js Production ───────────────────────────────────────────
frontend:
build:
context: ../
dockerfile: frontend/Dockerfile
target: prod
args:
PNPM_STORE_PATH: ${PNPM_STORE_PATH:-/root/.local/share/pnpm/store}
NPM_REGISTRY: ${NPM_REGISTRY:-}
container_name: deer-flow-frontend
environment:
- BETTER_AUTH_SECRET=${BETTER_AUTH_SECRET}
- DEER_FLOW_INTERNAL_GATEWAY_BASE_URL=http://gateway:8001
- DEER_FLOW_INTERNAL_LANGGRAPH_BASE_URL=http://langgraph:2024
env_file:
- ../frontend/.env
networks:
- deer-flow
restart: unless-stopped
# ── Gateway API ────────────────────────────────────────────────────────────
gateway:
build:
context: ../
dockerfile: backend/Dockerfile
args:
APT_MIRROR: ${APT_MIRROR:-}
UV_IMAGE: ${UV_IMAGE:-ghcr.io/astral-sh/uv:0.7.20}
UV_INDEX_URL: ${UV_INDEX_URL:-https://pypi.org/simple}
container_name: deer-flow-gateway
command: sh -c "cd backend && PYTHONPATH=. uv run uvicorn app.gateway.app:app --host 0.0.0.0 --port 8001 --workers ${GATEWAY_WORKERS:-4}"
volumes:
- ${DEER_FLOW_CONFIG_PATH}:/app/backend/config.yaml:ro
- ${DEER_FLOW_EXTENSIONS_CONFIG_PATH}:/app/backend/extensions_config.json:ro
- ../skills:/app/skills:ro
- ${DEER_FLOW_HOME}:/app/backend/.deer-flow
# DooD: AioSandboxProvider starts sandbox containers via host Docker daemon
- ${DEER_FLOW_DOCKER_SOCKET}:/var/run/docker.sock
# CLI auth directories for auto-auth (Claude Code + Codex CLI)
- type: bind
source: ${HOME:?HOME must be set}/.claude
target: /root/.claude
read_only: true
bind:
create_host_path: true
- type: bind
source: ${HOME:?HOME must be set}/.codex
target: /root/.codex
read_only: true
bind:
create_host_path: true
working_dir: /app
environment:
- CI=true
- DEER_FLOW_HOME=/app/backend/.deer-flow
- DEER_FLOW_CONFIG_PATH=/app/backend/config.yaml
- DEER_FLOW_EXTENSIONS_CONFIG_PATH=/app/backend/extensions_config.json
- DEER_FLOW_CHANNELS_LANGGRAPH_URL=${DEER_FLOW_CHANNELS_LANGGRAPH_URL:-http://langgraph:2024}
- DEER_FLOW_CHANNELS_GATEWAY_URL=${DEER_FLOW_CHANNELS_GATEWAY_URL:-http://gateway:8001}
# DooD path/network translation
- DEER_FLOW_HOST_BASE_DIR=${DEER_FLOW_HOME}
- DEER_FLOW_HOST_SKILLS_PATH=${DEER_FLOW_REPO_ROOT}/skills
- DEER_FLOW_SANDBOX_HOST=host.docker.internal
env_file:
- ../.env
extra_hosts:
- "host.docker.internal:host-gateway"
networks:
- deer-flow
restart: unless-stopped
# ── LangGraph Server ───────────────────────────────────────────────────────
# TODO: switch to langchain/langgraph-api (licensed) once a license key is available.
# For now, use `langgraph dev` (no license required) with the standard backend image.
langgraph:
build:
context: ../
dockerfile: backend/Dockerfile
args:
APT_MIRROR: ${APT_MIRROR:-}
UV_IMAGE: ${UV_IMAGE:-ghcr.io/astral-sh/uv:0.7.20}
UV_INDEX_URL: ${UV_INDEX_URL:-https://pypi.org/simple}
container_name: deer-flow-langgraph
command: sh -c 'cd /app/backend && args="--no-browser --no-reload --host 0.0.0.0 --port 2024 --n-jobs-per-worker $${LANGGRAPH_JOBS_PER_WORKER:-10}" && if [ "$${LANGGRAPH_ALLOW_BLOCKING:-0}" = "1" ]; then args="$$args --allow-blocking"; fi && uv run langgraph dev $$args'
volumes:
- ${DEER_FLOW_CONFIG_PATH}:/app/backend/config.yaml:ro
- ${DEER_FLOW_EXTENSIONS_CONFIG_PATH}:/app/backend/extensions_config.json:ro
- ${DEER_FLOW_HOME}:/app/backend/.deer-flow
- ../skills:/app/skills:ro
- ../backend/.langgraph_api:/app/backend/.langgraph_api
# DooD: same as gateway
- ${DEER_FLOW_DOCKER_SOCKET}:/var/run/docker.sock
# CLI auth directories for auto-auth (Claude Code + Codex CLI)
- type: bind
source: ${HOME:?HOME must be set}/.claude
target: /root/.claude
read_only: true
bind:
create_host_path: true
- type: bind
source: ${HOME:?HOME must be set}/.codex
target: /root/.codex
read_only: true
bind:
create_host_path: true
environment:
- CI=true
- DEER_FLOW_HOME=/app/backend/.deer-flow
- DEER_FLOW_CONFIG_PATH=/app/backend/config.yaml
- DEER_FLOW_EXTENSIONS_CONFIG_PATH=/app/backend/extensions_config.json
- DEER_FLOW_HOST_BASE_DIR=${DEER_FLOW_HOME}
- DEER_FLOW_HOST_SKILLS_PATH=${DEER_FLOW_REPO_ROOT}/skills
- DEER_FLOW_SANDBOX_HOST=host.docker.internal
# LangSmith tracing: set LANGSMITH_TRACING=true and LANGSMITH_API_KEY in .env to enable.
env_file:
- ../.env
extra_hosts:
- "host.docker.internal:host-gateway"
networks:
- deer-flow
restart: unless-stopped
# ── Sandbox Provisioner (optional, Kubernetes mode) ────────────────────────
provisioner:
build:
context: ./provisioner
dockerfile: Dockerfile
args:
APT_MIRROR: ${APT_MIRROR:-}
PIP_INDEX_URL: ${PIP_INDEX_URL:-}
container_name: deer-flow-provisioner
volumes:
- ~/.kube/config:/root/.kube/config:ro
environment:
- K8S_NAMESPACE=deer-flow
- SANDBOX_IMAGE=enterprise-public-cn-beijing.cr.volces.com/vefaas-public/all-in-one-sandbox:latest
- SKILLS_HOST_PATH=${DEER_FLOW_REPO_ROOT}/skills
- THREADS_HOST_PATH=${DEER_FLOW_HOME}/threads
- KUBECONFIG_PATH=/root/.kube/config
- NODE_HOST=host.docker.internal
- K8S_API_SERVER=https://host.docker.internal:26443
env_file:
- ../.env
extra_hosts:
- "host.docker.internal:host-gateway"
networks:
- deer-flow
restart: unless-stopped
healthcheck:
test: ["CMD", "curl", "-f", "http://localhost:8002/health"]
interval: 10s
timeout: 5s
retries: 6
networks:
deer-flow:
driver: bridge

View File

@@ -0,0 +1,236 @@
events {
worker_connections 1024;
}
pid /tmp/nginx.pid;
http {
# Basic settings
sendfile on;
tcp_nopush on;
tcp_nodelay on;
keepalive_timeout 65;
types_hash_max_size 2048;
# Logging
access_log /dev/stdout;
error_log /dev/stderr;
# Docker internal DNS (for resolving k3s hostname)
resolver 127.0.0.11 valid=10s ipv6=off;
# Upstream servers (using Docker service names)
# NOTE: `zone` and `resolve` are nginx Plus-only features and are not
# available in the standard nginx:alpine image. Docker's internal DNS
# (127.0.0.11) handles service discovery; upstreams are resolved at
# nginx startup and remain valid for the lifetime of the deployment.
upstream gateway {
server gateway:8001;
}
upstream langgraph {
server ${LANGGRAPH_UPSTREAM};
}
upstream frontend {
server frontend:3000;
}
# ── Main server (path-based routing) ─────────────────────────────────
server {
listen 2026 default_server;
listen [::]:2026 default_server;
server_name _;
# Hide CORS headers from upstream to prevent duplicates
proxy_hide_header 'Access-Control-Allow-Origin';
proxy_hide_header 'Access-Control-Allow-Methods';
proxy_hide_header 'Access-Control-Allow-Headers';
proxy_hide_header 'Access-Control-Allow-Credentials';
# CORS headers for all responses (nginx handles CORS centrally)
add_header 'Access-Control-Allow-Origin' '*' always;
add_header 'Access-Control-Allow-Methods' 'GET, POST, PUT, DELETE, PATCH, OPTIONS' always;
add_header 'Access-Control-Allow-Headers' '*' always;
# Handle OPTIONS requests (CORS preflight)
if ($request_method = 'OPTIONS') {
return 204;
}
# LangGraph API routes
# In standard mode: /api/langgraph/* → langgraph:2024 (rewrite to /*)
# In gateway mode: /api/langgraph/* → gateway:8001 (rewrite to /api/*)
# Controlled by LANGGRAPH_UPSTREAM and LANGGRAPH_REWRITE env vars.
location /api/langgraph/ {
rewrite ^/api/langgraph/(.*) ${LANGGRAPH_REWRITE}$1 break;
proxy_pass http://langgraph;
proxy_http_version 1.1;
# Headers
proxy_set_header Host $host;
proxy_set_header X-Real-IP $remote_addr;
proxy_set_header X-Forwarded-For $proxy_add_x_forwarded_for;
proxy_set_header X-Forwarded-Proto $scheme;
proxy_set_header Connection '';
# SSE/Streaming support
proxy_buffering off;
proxy_cache off;
proxy_set_header X-Accel-Buffering no;
# Timeouts for long-running requests
proxy_connect_timeout 600s;
proxy_send_timeout 600s;
proxy_read_timeout 600s;
# Chunked transfer encoding
chunked_transfer_encoding on;
}
# Custom API: Models endpoint
location /api/models {
proxy_pass http://gateway;
proxy_http_version 1.1;
proxy_set_header Host $host;
proxy_set_header X-Real-IP $remote_addr;
proxy_set_header X-Forwarded-For $proxy_add_x_forwarded_for;
proxy_set_header X-Forwarded-Proto $scheme;
}
# Custom API: Memory endpoint
location /api/memory {
proxy_pass http://gateway;
proxy_http_version 1.1;
proxy_set_header Host $host;
proxy_set_header X-Real-IP $remote_addr;
proxy_set_header X-Forwarded-For $proxy_add_x_forwarded_for;
proxy_set_header X-Forwarded-Proto $scheme;
}
# Custom API: MCP configuration endpoint
location /api/mcp {
proxy_pass http://gateway;
proxy_http_version 1.1;
proxy_set_header Host $host;
proxy_set_header X-Real-IP $remote_addr;
proxy_set_header X-Forwarded-For $proxy_add_x_forwarded_for;
proxy_set_header X-Forwarded-Proto $scheme;
}
# Custom API: Skills configuration endpoint
location /api/skills {
proxy_pass http://gateway;
proxy_http_version 1.1;
proxy_set_header Host $host;
proxy_set_header X-Real-IP $remote_addr;
proxy_set_header X-Forwarded-For $proxy_add_x_forwarded_for;
proxy_set_header X-Forwarded-Proto $scheme;
}
# Custom API: Agents endpoint
location /api/agents {
proxy_pass http://gateway;
proxy_http_version 1.1;
proxy_set_header Host $host;
proxy_set_header X-Real-IP $remote_addr;
proxy_set_header X-Forwarded-For $proxy_add_x_forwarded_for;
proxy_set_header X-Forwarded-Proto $scheme;
}
# Custom API: Uploads endpoint
location ~ ^/api/threads/[^/]+/uploads {
proxy_pass http://gateway;
proxy_http_version 1.1;
proxy_set_header Host $host;
proxy_set_header X-Real-IP $remote_addr;
proxy_set_header X-Forwarded-For $proxy_add_x_forwarded_for;
proxy_set_header X-Forwarded-Proto $scheme;
# Large file upload support
client_max_body_size 100M;
proxy_request_buffering off;
}
# Custom API: Other endpoints under /api/threads
location ~ ^/api/threads {
proxy_pass http://gateway;
proxy_http_version 1.1;
proxy_set_header Host $host;
proxy_set_header X-Real-IP $remote_addr;
proxy_set_header X-Forwarded-For $proxy_add_x_forwarded_for;
proxy_set_header X-Forwarded-Proto $scheme;
}
# API Documentation: Swagger UI
location /docs {
proxy_pass http://gateway;
proxy_http_version 1.1;
proxy_set_header Host $host;
proxy_set_header X-Real-IP $remote_addr;
proxy_set_header X-Forwarded-For $proxy_add_x_forwarded_for;
proxy_set_header X-Forwarded-Proto $scheme;
}
# API Documentation: ReDoc
location /redoc {
proxy_pass http://gateway;
proxy_http_version 1.1;
proxy_set_header Host $host;
proxy_set_header X-Real-IP $remote_addr;
proxy_set_header X-Forwarded-For $proxy_add_x_forwarded_for;
proxy_set_header X-Forwarded-Proto $scheme;
}
# API Documentation: OpenAPI Schema
location /openapi.json {
proxy_pass http://gateway;
proxy_http_version 1.1;
proxy_set_header Host $host;
proxy_set_header X-Real-IP $remote_addr;
proxy_set_header X-Forwarded-For $proxy_add_x_forwarded_for;
proxy_set_header X-Forwarded-Proto $scheme;
}
# Health check endpoint (gateway)
location /health {
proxy_pass http://gateway;
proxy_http_version 1.1;
proxy_set_header Host $host;
proxy_set_header X-Real-IP $remote_addr;
proxy_set_header X-Forwarded-For $proxy_add_x_forwarded_for;
proxy_set_header X-Forwarded-Proto $scheme;
}
# ── Provisioner API (sandbox management) ────────────────────────
# Use a variable so nginx resolves provisioner at request time (not startup).
# This allows nginx to start even when provisioner container is not running.
location /api/sandboxes {
set $provisioner_upstream provisioner:8002;
proxy_pass http://$provisioner_upstream;
proxy_http_version 1.1;
proxy_set_header Host $host;
proxy_set_header X-Real-IP $remote_addr;
proxy_set_header X-Forwarded-For $proxy_add_x_forwarded_for;
proxy_set_header X-Forwarded-Proto $scheme;
}
# All other requests go to frontend
location / {
proxy_pass http://frontend;
proxy_http_version 1.1;
# Headers
proxy_set_header Host $host;
proxy_set_header X-Real-IP $remote_addr;
proxy_set_header X-Forwarded-For $proxy_add_x_forwarded_for;
proxy_set_header X-Forwarded-Proto $scheme;
proxy_set_header Upgrade $http_upgrade;
proxy_set_header Connection 'upgrade';
proxy_cache_bypass $http_upgrade;
# Timeouts
proxy_connect_timeout 600s;
proxy_send_timeout 600s;
proxy_read_timeout 600s;
}
}
}

View File

@@ -0,0 +1,241 @@
events {
worker_connections 1024;
}
pid logs/nginx.pid;
http {
# Basic settings
sendfile on;
tcp_nopush on;
tcp_nodelay on;
keepalive_timeout 65;
types_hash_max_size 2048;
# Logging
access_log logs/nginx-access.log;
error_log logs/nginx-error.log;
# Upstream servers (using 127.0.0.1 for local development)
upstream gateway {
server 127.0.0.1:8001;
}
upstream langgraph {
server 127.0.0.1:2024;
}
upstream frontend {
server 127.0.0.1:3000;
}
server {
listen 2026;
listen [::]:2026;
server_name _;
# Hide CORS headers from upstream to prevent duplicates
proxy_hide_header 'Access-Control-Allow-Origin';
proxy_hide_header 'Access-Control-Allow-Methods';
proxy_hide_header 'Access-Control-Allow-Headers';
proxy_hide_header 'Access-Control-Allow-Credentials';
# CORS headers for all responses (nginx handles CORS centrally)
add_header 'Access-Control-Allow-Origin' '*' always;
add_header 'Access-Control-Allow-Methods' 'GET, POST, PUT, DELETE, PATCH, OPTIONS' always;
add_header 'Access-Control-Allow-Headers' '*' always;
# Handle OPTIONS requests (CORS preflight)
if ($request_method = 'OPTIONS') {
return 204;
}
# LangGraph API routes (served by langgraph dev)
# Rewrites /api/langgraph/* to /* before proxying to LangGraph server
location /api/langgraph/ {
rewrite ^/api/langgraph/(.*) /$1 break;
proxy_pass http://langgraph;
proxy_http_version 1.1;
# Headers
proxy_set_header Host $host;
proxy_set_header X-Real-IP $remote_addr;
proxy_set_header X-Forwarded-For $proxy_add_x_forwarded_for;
proxy_set_header X-Forwarded-Proto $scheme;
proxy_set_header Connection '';
# SSE/Streaming support
proxy_buffering off;
proxy_cache off;
proxy_set_header X-Accel-Buffering no;
# Timeouts for long-running requests
proxy_connect_timeout 600s;
proxy_send_timeout 600s;
proxy_read_timeout 600s;
# Chunked transfer encoding
chunked_transfer_encoding on;
}
# Experimental: Gateway-backed LangGraph-compatible API
# Frontend can opt-in via NEXT_PUBLIC_LANGGRAPH_BASE_URL=/api/langgraph-compat
location /api/langgraph-compat/ {
rewrite ^/api/langgraph-compat/(.*) /api/$1 break;
proxy_pass http://gateway;
proxy_http_version 1.1;
# Headers
proxy_set_header Host $host;
proxy_set_header X-Real-IP $remote_addr;
proxy_set_header X-Forwarded-For $proxy_add_x_forwarded_for;
proxy_set_header X-Forwarded-Proto $scheme;
proxy_set_header Connection '';
# SSE/Streaming support
proxy_buffering off;
proxy_cache off;
proxy_set_header X-Accel-Buffering no;
# Timeouts for long-running requests
proxy_connect_timeout 600s;
proxy_send_timeout 600s;
proxy_read_timeout 600s;
# Chunked transfer encoding
chunked_transfer_encoding on;
}
# Custom API: Models endpoint
location /api/models {
proxy_pass http://gateway;
proxy_http_version 1.1;
proxy_set_header Host $host;
proxy_set_header X-Real-IP $remote_addr;
proxy_set_header X-Forwarded-For $proxy_add_x_forwarded_for;
proxy_set_header X-Forwarded-Proto $scheme;
}
# Custom API: Memory endpoint
location /api/memory {
proxy_pass http://gateway;
proxy_http_version 1.1;
proxy_set_header Host $host;
proxy_set_header X-Real-IP $remote_addr;
proxy_set_header X-Forwarded-For $proxy_add_x_forwarded_for;
proxy_set_header X-Forwarded-Proto $scheme;
}
# Custom API: MCP configuration endpoint
location /api/mcp {
proxy_pass http://gateway;
proxy_http_version 1.1;
proxy_set_header Host $host;
proxy_set_header X-Real-IP $remote_addr;
proxy_set_header X-Forwarded-For $proxy_add_x_forwarded_for;
proxy_set_header X-Forwarded-Proto $scheme;
}
# Custom API: Skills configuration endpoint
location /api/skills {
proxy_pass http://gateway;
proxy_http_version 1.1;
proxy_set_header Host $host;
proxy_set_header X-Real-IP $remote_addr;
proxy_set_header X-Forwarded-For $proxy_add_x_forwarded_for;
proxy_set_header X-Forwarded-Proto $scheme;
}
# Custom API: Agents endpoint
location /api/agents {
proxy_pass http://gateway;
proxy_http_version 1.1;
proxy_set_header Host $host;
proxy_set_header X-Real-IP $remote_addr;
proxy_set_header X-Forwarded-For $proxy_add_x_forwarded_for;
proxy_set_header X-Forwarded-Proto $scheme;
}
# Custom API: Uploads endpoint
location ~ ^/api/threads/[^/]+/uploads {
proxy_pass http://gateway;
proxy_http_version 1.1;
proxy_set_header Host $host;
proxy_set_header X-Real-IP $remote_addr;
proxy_set_header X-Forwarded-For $proxy_add_x_forwarded_for;
proxy_set_header X-Forwarded-Proto $scheme;
# Large file upload support
client_max_body_size 100M;
proxy_request_buffering off;
}
# Custom API: Other endpoints under /api/threads
location ~ ^/api/threads {
proxy_pass http://gateway;
proxy_http_version 1.1;
proxy_set_header Host $host;
proxy_set_header X-Real-IP $remote_addr;
proxy_set_header X-Forwarded-For $proxy_add_x_forwarded_for;
proxy_set_header X-Forwarded-Proto $scheme;
}
# API Documentation: Swagger UI
location /api/docs {
proxy_pass http://gateway/docs ;
proxy_http_version 1.1;
proxy_set_header Host $host;
proxy_set_header X-Real-IP $remote_addr;
proxy_set_header X-Forwarded-For $proxy_add_x_forwarded_for;
proxy_set_header X-Forwarded-Proto $scheme;
}
# API Documentation: ReDoc
location /api/redoc {
proxy_pass http://gateway/redoc;
proxy_http_version 1.1;
proxy_set_header Host $host;
proxy_set_header X-Real-IP $remote_addr;
proxy_set_header X-Forwarded-For $proxy_add_x_forwarded_for;
proxy_set_header X-Forwarded-Proto $scheme;
}
# API Documentation: OpenAPI Schema
location /openapi.json {
proxy_pass http://gateway;
proxy_http_version 1.1;
proxy_set_header Host $host;
proxy_set_header X-Real-IP $remote_addr;
proxy_set_header X-Forwarded-For $proxy_add_x_forwarded_for;
proxy_set_header X-Forwarded-Proto $scheme;
}
# Health check endpoint (gateway)
location /health {
proxy_pass http://gateway;
proxy_http_version 1.1;
proxy_set_header Host $host;
proxy_set_header X-Real-IP $remote_addr;
proxy_set_header X-Forwarded-For $proxy_add_x_forwarded_for;
proxy_set_header X-Forwarded-Proto $scheme;
}
# All other requests go to frontend
location / {
proxy_pass http://frontend;
proxy_http_version 1.1;
# Headers
proxy_set_header Host $host;
proxy_set_header X-Real-IP $remote_addr;
proxy_set_header X-Forwarded-For $proxy_add_x_forwarded_for;
proxy_set_header X-Forwarded-Proto $scheme;
proxy_set_header Upgrade $http_upgrade;
proxy_set_header Connection 'upgrade';
proxy_cache_bypass $http_upgrade;
# Timeouts
proxy_connect_timeout 600s;
proxy_send_timeout 600s;
proxy_read_timeout 600s;
}
}
}

View File

@@ -0,0 +1,29 @@
FROM python:3.12-slim-bookworm
ARG APT_MIRROR
ARG PIP_INDEX_URL
# Optionally override apt mirror for restricted networks (e.g. APT_MIRROR=mirrors.aliyun.com)
RUN if [ -n "${APT_MIRROR}" ]; then \
sed -i "s|deb.debian.org|${APT_MIRROR}|g" /etc/apt/sources.list.d/debian.sources 2>/dev/null || true; \
sed -i "s|deb.debian.org|${APT_MIRROR}|g" /etc/apt/sources.list 2>/dev/null || true; \
fi
# Install system dependencies
RUN apt-get update && apt-get install -y --no-install-recommends \
curl \
&& rm -rf /var/lib/apt/lists/*
# Install Python dependencies
RUN pip install --no-cache-dir \
${PIP_INDEX_URL:+--index-url "$PIP_INDEX_URL"} \
fastapi \
"uvicorn[standard]" \
kubernetes
WORKDIR /app
COPY app.py .
EXPOSE 8002
CMD ["uvicorn", "app:app", "--host", "0.0.0.0", "--port", "8002"]

View File

@@ -0,0 +1,332 @@
# DeerFlow Sandbox Provisioner
The **Sandbox Provisioner** is a FastAPI service that dynamically manages sandbox Pods in Kubernetes. It provides a REST API for the DeerFlow backend to create, monitor, and destroy isolated sandbox environments for code execution.
## Architecture
```
┌────────────┐ HTTP ┌─────────────┐ K8s API ┌──────────────┐
│ Backend │ ─────▸ │ Provisioner │ ────────▸ │ Host K8s │
│ (gateway/ │ │ :8002 │ │ API Server │
│ langgraph) │ └─────────────┘ └──────┬───────┘
└────────────┘ │ creates
┌─────────────┐ ┌────▼─────┐
│ Backend │ ──────▸ │ Sandbox │
│ (via Docker │ NodePort│ Pod(s) │
│ network) │ └──────────┘
└─────────────┘
```
### How It Works
1. **Backend Request**: When the backend needs to execute code, it sends a `POST /api/sandboxes` request with a `sandbox_id` and `thread_id`.
2. **Pod Creation**: The provisioner creates a dedicated Pod in the `deer-flow` namespace with:
- The sandbox container image (all-in-one-sandbox)
- HostPath volumes mounted for:
- `/mnt/skills` → Read-only access to public skills
- `/mnt/user-data` → Read-write access to thread-specific data
- Resource limits (CPU, memory, ephemeral storage)
- Readiness/liveness probes
3. **Service Creation**: A NodePort Service is created to expose the Pod, with Kubernetes auto-allocating a port from the NodePort range (typically 30000-32767).
4. **Access URL**: The provisioner returns `http://host.docker.internal:{NodePort}` to the backend, which the backend containers can reach directly.
5. **Cleanup**: When the session ends, `DELETE /api/sandboxes/{sandbox_id}` removes both the Pod and Service.
## Requirements
Host machine with a running Kubernetes cluster (Docker Desktop K8s, OrbStack, minikube, kind, etc.)
### Enable Kubernetes in Docker Desktop
1. Open Docker Desktop settings
2. Go to "Kubernetes" tab
3. Check "Enable Kubernetes"
4. Click "Apply & Restart"
### Enable Kubernetes in OrbStack
1. Open OrbStack settings
2. Go to "Kubernetes" tab
3. Check "Enable Kubernetes"
## API Endpoints
### `GET /health`
Health check endpoint.
**Response**:
```json
{
"status": "ok"
}
```
### `POST /api/sandboxes`
Create a new sandbox Pod + Service.
**Request**:
```json
{
"sandbox_id": "abc-123",
"thread_id": "thread-456"
}
```
**Response**:
```json
{
"sandbox_id": "abc-123",
"sandbox_url": "http://host.docker.internal:32123",
"status": "Pending"
}
```
**Idempotent**: Calling with the same `sandbox_id` returns the existing sandbox info.
### `GET /api/sandboxes/{sandbox_id}`
Get status and URL of a specific sandbox.
**Response**:
```json
{
"sandbox_id": "abc-123",
"sandbox_url": "http://host.docker.internal:32123",
"status": "Running"
}
```
**Status Values**: `Pending`, `Running`, `Succeeded`, `Failed`, `Unknown`, `NotFound`
### `DELETE /api/sandboxes/{sandbox_id}`
Destroy a sandbox Pod + Service.
**Response**:
```json
{
"ok": true,
"sandbox_id": "abc-123"
}
```
### `GET /api/sandboxes`
List all sandboxes currently managed.
**Response**:
```json
{
"sandboxes": [
{
"sandbox_id": "abc-123",
"sandbox_url": "http://host.docker.internal:32123",
"status": "Running"
}
],
"count": 1
}
```
## Configuration
The provisioner is configured via environment variables (set in [docker-compose-dev.yaml](../docker-compose-dev.yaml)):
| Variable | Default | Description |
|----------|---------|-------------|
| `K8S_NAMESPACE` | `deer-flow` | Kubernetes namespace for sandbox resources |
| `SANDBOX_IMAGE` | `enterprise-public-cn-beijing.cr.volces.com/vefaas-public/all-in-one-sandbox:latest` | Container image for sandbox Pods |
| `SKILLS_HOST_PATH` | - | **Host machine** path to skills directory (must be absolute) |
| `THREADS_HOST_PATH` | - | **Host machine** path to threads data directory (must be absolute) |
| `SKILLS_PVC_NAME` | empty (use hostPath) | PVC name for skills volume; when set, sandbox Pods use PVC instead of hostPath |
| `USERDATA_PVC_NAME` | empty (use hostPath) | PVC name for user-data volume; when set, uses PVC with `subPath: threads/{thread_id}/user-data` |
| `KUBECONFIG_PATH` | `/root/.kube/config` | Path to kubeconfig **inside** the provisioner container |
| `NODE_HOST` | `host.docker.internal` | Hostname that backend containers use to reach host NodePorts |
| `K8S_API_SERVER` | (from kubeconfig) | Override K8s API server URL (e.g., `https://host.docker.internal:26443`) |
### Important: K8S_API_SERVER Override
If your kubeconfig uses `localhost`, `127.0.0.1`, or `0.0.0.0` as the API server address (common with OrbStack, minikube, kind), the provisioner **cannot** reach it from inside the Docker container.
**Solution**: Set `K8S_API_SERVER` to use `host.docker.internal`:
```yaml
# docker-compose-dev.yaml
provisioner:
environment:
- K8S_API_SERVER=https://host.docker.internal:26443 # Replace 26443 with your API port
```
Check your kubeconfig API server:
```bash
kubectl config view --minify -o jsonpath='{.clusters[0].cluster.server}'
```
## Prerequisites
### Host Machine Requirements
1. **Kubernetes Cluster**:
- Docker Desktop with Kubernetes enabled, or
- OrbStack (built-in K8s), or
- minikube, kind, k3s, etc.
2. **kubectl Configured**:
- `~/.kube/config` must exist and be valid
- Current context should point to your local cluster
3. **Kubernetes Access**:
- The provisioner needs permissions to:
- Create/read/delete Pods in the `deer-flow` namespace
- Create/read/delete Services in the `deer-flow` namespace
- Read Namespaces (to create `deer-flow` if missing)
4. **Host Paths**:
- The `SKILLS_HOST_PATH` and `THREADS_HOST_PATH` must be **absolute paths on the host machine**
- These paths are mounted into sandbox Pods via K8s HostPath volumes
- The paths must exist and be readable by the K8s node
### Docker Compose Setup
The provisioner runs as part of the docker-compose-dev stack:
```bash
# Start Docker services (provisioner starts only when config.yaml enables provisioner mode)
make docker-start
# Or start just the provisioner
docker compose -p deer-flow-dev -f docker/docker-compose-dev.yaml up -d provisioner
```
The compose file:
- Mounts your host's `~/.kube/config` into the container
- Adds `extra_hosts` entry for `host.docker.internal` (required on Linux)
- Configures environment variables for K8s access
## Testing
### Manual API Testing
```bash
# Health check
curl http://localhost:8002/health
# Create a sandbox (via provisioner container for internal DNS)
docker exec deer-flow-provisioner curl -X POST http://localhost:8002/api/sandboxes \
-H "Content-Type: application/json" \
-d '{"sandbox_id":"test-001","thread_id":"thread-001"}'
# Check sandbox status
docker exec deer-flow-provisioner curl http://localhost:8002/api/sandboxes/test-001
# List all sandboxes
docker exec deer-flow-provisioner curl http://localhost:8002/api/sandboxes
# Verify Pod and Service in K8s
kubectl get pod,svc -n deer-flow -l sandbox-id=test-001
# Delete sandbox
docker exec deer-flow-provisioner curl -X DELETE http://localhost:8002/api/sandboxes/test-001
```
### Verify from Backend Containers
Once a sandbox is created, the backend containers (gateway, langgraph) can access it:
```bash
# Get sandbox URL from provisioner
SANDBOX_URL=$(docker exec deer-flow-provisioner curl -s http://localhost:8002/api/sandboxes/test-001 | jq -r .sandbox_url)
# Test from gateway container
docker exec deer-flow-gateway curl -s $SANDBOX_URL/v1/sandbox
```
## Troubleshooting
### Issue: "Kubeconfig not found"
**Cause**: The kubeconfig file doesn't exist at the mounted path.
**Solution**:
- Ensure `~/.kube/config` exists on your host machine
- Run `kubectl config view` to verify
- Check the volume mount in docker-compose-dev.yaml
### Issue: "Kubeconfig path is a directory"
**Cause**: The mounted `KUBECONFIG_PATH` points to a directory instead of a file.
**Solution**:
- Ensure the compose mount source is a file (e.g., `~/.kube/config`) not a directory
- Verify inside container:
```bash
docker exec deer-flow-provisioner ls -ld /root/.kube/config
```
- Expected output should indicate a regular file (`-`), not a directory (`d`)
### Issue: "Connection refused" to K8s API
**Cause**: The provisioner can't reach the K8s API server.
**Solution**:
1. Check your kubeconfig server address:
```bash
kubectl config view --minify -o jsonpath='{.clusters[0].cluster.server}'
```
2. If it's `localhost` or `127.0.0.1`, set `K8S_API_SERVER`:
```yaml
environment:
- K8S_API_SERVER=https://host.docker.internal:PORT
```
### Issue: "Unprocessable Entity" when creating Pod
**Cause**: HostPath volumes contain invalid paths (e.g., relative paths with `..`).
**Solution**:
- Use absolute paths for `SKILLS_HOST_PATH` and `THREADS_HOST_PATH`
- Verify the paths exist on your host machine:
```bash
ls -la /path/to/skills
ls -la /path/to/backend/.deer-flow/threads
```
### Issue: Pod stuck in "ContainerCreating"
**Cause**: Usually pulling the sandbox image from the registry.
**Solution**:
- Pre-pull the image: `make docker-init`
- Check Pod events: `kubectl describe pod sandbox-XXX -n deer-flow`
- Check node: `kubectl get nodes`
### Issue: Cannot access sandbox URL from backend
**Cause**: NodePort not reachable or `NODE_HOST` misconfigured.
**Solution**:
- Verify the Service exists: `kubectl get svc -n deer-flow`
- Test from host: `curl http://localhost:NODE_PORT/v1/sandbox`
- Ensure `extra_hosts` is set in docker-compose (Linux)
- Check `NODE_HOST` env var matches how backend reaches host
## Security Considerations
1. **HostPath Volumes**: The provisioner mounts host directories into sandbox Pods by default. Ensure these paths contain only trusted data. For production, prefer PVC-based volumes (set `SKILLS_PVC_NAME` and `USERDATA_PVC_NAME`) to avoid node-specific data loss risks.
2. **Resource Limits**: Each sandbox Pod has CPU, memory, and storage limits to prevent resource exhaustion.
3. **Network Isolation**: Sandbox Pods run in the `deer-flow` namespace but share the host's network namespace via NodePort. Consider NetworkPolicies for stricter isolation.
4. **kubeconfig Access**: The provisioner has full access to your Kubernetes cluster via the mounted kubeconfig. Run it only in trusted environments.
5. **Image Trust**: The sandbox image should come from a trusted registry. Review and audit the image contents.
## Future Enhancements
- [ ] Support for custom resource requests/limits per sandbox
- [x] PersistentVolume support for larger data requirements
- [ ] Automatic cleanup of stale sandboxes (timeout-based)
- [ ] Metrics and monitoring (Prometheus integration)
- [ ] Multi-cluster support (route to different K8s clusters)
- [ ] Pod affinity/anti-affinity rules for better placement
- [ ] NetworkPolicy templates for sandbox isolation

View File

@@ -0,0 +1,582 @@
"""DeerFlow Sandbox Provisioner Service.
Dynamically creates and manages per-sandbox Pods in Kubernetes.
Each ``sandbox_id`` gets its own Pod + NodePort Service. The backend
accesses sandboxes directly via ``{NODE_HOST}:{NodePort}``.
The provisioner connects to the host machine's Kubernetes cluster via a
mounted kubeconfig (``~/.kube/config``). Sandbox Pods run on the host
K8s and are accessed by the backend via ``{NODE_HOST}:{NodePort}``.
Endpoints:
POST /api/sandboxes — Create a sandbox Pod + Service
DELETE /api/sandboxes/{sandbox_id} — Destroy a sandbox Pod + Service
GET /api/sandboxes/{sandbox_id} — Get sandbox status & URL
GET /api/sandboxes — List all sandboxes
GET /health — Provisioner health check
Architecture (docker-compose-dev):
┌────────────┐ HTTP ┌─────────────┐ K8s API ┌──────────────┐
│ remote │ ─────▸ │ provisioner │ ────────▸ │ host K8s │
│ _backend │ │ :8002 │ │ API server │
└────────────┘ └─────────────┘ └──────┬───────┘
│ creates
┌─────────────┐ ┌──────▼───────┐
│ backend │ ────────▸ │ sandbox │
│ │ direct │ Pod(s) │
└─────────────┘ NodePort └──────────────┘
"""
from __future__ import annotations
import logging
import os
import re
import time
from contextlib import asynccontextmanager
import urllib3
from fastapi import FastAPI, HTTPException
from kubernetes import client as k8s_client
from kubernetes import config as k8s_config
from kubernetes.client.rest import ApiException
from pydantic import BaseModel, Field
# Suppress only the InsecureRequestWarning from urllib3
urllib3.disable_warnings(urllib3.exceptions.InsecureRequestWarning)
logger = logging.getLogger(__name__)
logging.basicConfig(
level=logging.INFO,
format="%(asctime)s [%(levelname)s] %(name)s: %(message)s",
)
# ── Configuration (all tuneable via environment variables) ───────────────
K8S_NAMESPACE = os.environ.get("K8S_NAMESPACE", "deer-flow")
SANDBOX_IMAGE = os.environ.get(
"SANDBOX_IMAGE",
"enterprise-public-cn-beijing.cr.volces.com/vefaas-public/all-in-one-sandbox:latest",
)
SKILLS_HOST_PATH = os.environ.get("SKILLS_HOST_PATH", "/skills")
THREADS_HOST_PATH = os.environ.get("THREADS_HOST_PATH", "/.deer-flow/threads")
SKILLS_PVC_NAME = os.environ.get("SKILLS_PVC_NAME", "")
USERDATA_PVC_NAME = os.environ.get("USERDATA_PVC_NAME", "")
SAFE_THREAD_ID_PATTERN = r"^[A-Za-z0-9_\-]+$"
# Path to the kubeconfig *inside* the provisioner container.
# Typically the host's ~/.kube/config is mounted here.
KUBECONFIG_PATH = os.environ.get("KUBECONFIG_PATH", "/root/.kube/config")
# The hostname / IP that the *backend container* uses to reach NodePort
# services on the host Kubernetes node. On Docker Desktop for macOS this
# is ``host.docker.internal``; on Linux it may be the host's LAN IP.
NODE_HOST = os.environ.get("NODE_HOST", "host.docker.internal")
def join_host_path(base: str, *parts: str) -> str:
"""Join host filesystem path segments while preserving native style."""
if not parts:
return base
if re.match(r"^[A-Za-z]:[\\/]", base) or base.startswith("\\\\") or "\\" in base:
from pathlib import PureWindowsPath
result = PureWindowsPath(base)
for part in parts:
result /= part
return str(result)
from pathlib import Path
result = Path(base)
for part in parts:
result /= part
return str(result)
def _validate_thread_id(thread_id: str) -> str:
if not re.match(SAFE_THREAD_ID_PATTERN, thread_id):
raise ValueError(
"Invalid thread_id: only alphanumeric characters, hyphens, and underscores are allowed."
)
return thread_id
# ── K8s client setup ────────────────────────────────────────────────────
core_v1: k8s_client.CoreV1Api | None = None
def _init_k8s_client() -> k8s_client.CoreV1Api:
"""Load kubeconfig from the mounted host config and return a CoreV1Api.
Tries the mounted kubeconfig first, then falls back to in-cluster
config (useful if the provisioner itself runs inside K8s).
"""
if os.path.exists(KUBECONFIG_PATH):
if os.path.isdir(KUBECONFIG_PATH):
raise RuntimeError(
f"KUBECONFIG_PATH points to a directory, expected a file: {KUBECONFIG_PATH}"
)
try:
k8s_config.load_kube_config(config_file=KUBECONFIG_PATH)
logger.info(f"Loaded kubeconfig from {KUBECONFIG_PATH}")
except Exception as exc:
raise RuntimeError(
f"Failed to load kubeconfig from {KUBECONFIG_PATH}: {exc}"
) from exc
else:
logger.warning(
f"Kubeconfig not found at {KUBECONFIG_PATH}; trying in-cluster config"
)
try:
k8s_config.load_incluster_config()
except Exception as exc:
raise RuntimeError(
"Failed to initialize Kubernetes client. "
f"No kubeconfig at {KUBECONFIG_PATH}, and in-cluster config is unavailable: {exc}"
) from exc
# When connecting from inside Docker to the host's K8s API, the
# kubeconfig may reference ``localhost`` or ``127.0.0.1``. We
# optionally rewrite the server address so it reaches the host.
k8s_api_server = os.environ.get("K8S_API_SERVER")
if k8s_api_server:
configuration = k8s_client.Configuration.get_default_copy()
configuration.host = k8s_api_server
# Self-signed certs are common for local clusters
configuration.verify_ssl = False
api_client = k8s_client.ApiClient(configuration)
return k8s_client.CoreV1Api(api_client)
return k8s_client.CoreV1Api()
def _wait_for_kubeconfig(timeout: int = 30) -> None:
"""Wait for kubeconfig file if configured, then continue with fallback support."""
deadline = time.time() + timeout
while time.time() < deadline:
if os.path.exists(KUBECONFIG_PATH):
if os.path.isfile(KUBECONFIG_PATH):
logger.info(f"Found kubeconfig file at {KUBECONFIG_PATH}")
return
if os.path.isdir(KUBECONFIG_PATH):
raise RuntimeError(
"Kubeconfig path is a directory. "
f"Please mount a kubeconfig file at {KUBECONFIG_PATH}."
)
raise RuntimeError(
f"Kubeconfig path exists but is not a regular file: {KUBECONFIG_PATH}"
)
logger.info(f"Waiting for kubeconfig at {KUBECONFIG_PATH}")
time.sleep(2)
logger.warning(
f"Kubeconfig not found at {KUBECONFIG_PATH} after {timeout}s; "
"will attempt in-cluster Kubernetes config"
)
def _ensure_namespace() -> None:
"""Create the K8s namespace if it does not yet exist."""
try:
core_v1.read_namespace(K8S_NAMESPACE)
logger.info(f"Namespace '{K8S_NAMESPACE}' already exists")
except ApiException as exc:
if exc.status == 404:
ns = k8s_client.V1Namespace(
metadata=k8s_client.V1ObjectMeta(
name=K8S_NAMESPACE,
labels={
"app.kubernetes.io/name": "deer-flow",
"app.kubernetes.io/component": "sandbox",
},
)
)
core_v1.create_namespace(ns)
logger.info(f"Created namespace '{K8S_NAMESPACE}'")
else:
raise
# ── FastAPI lifespan ─────────────────────────────────────────────────────
@asynccontextmanager
async def lifespan(_app: FastAPI):
global core_v1
_wait_for_kubeconfig()
core_v1 = _init_k8s_client()
_ensure_namespace()
logger.info("Provisioner is ready (using host Kubernetes)")
yield
app = FastAPI(title="DeerFlow Sandbox Provisioner", lifespan=lifespan)
# ── Request / Response models ───────────────────────────────────────────
class CreateSandboxRequest(BaseModel):
sandbox_id: str
thread_id: str = Field(pattern=SAFE_THREAD_ID_PATTERN)
class SandboxResponse(BaseModel):
sandbox_id: str
sandbox_url: str # Direct access URL, e.g. http://host.docker.internal:{NodePort}
status: str
# ── K8s resource helpers ─────────────────────────────────────────────────
def _pod_name(sandbox_id: str) -> str:
return f"sandbox-{sandbox_id}"
def _svc_name(sandbox_id: str) -> str:
return f"sandbox-{sandbox_id}-svc"
def _sandbox_url(node_port: int) -> str:
"""Build the sandbox URL using the configured NODE_HOST."""
return f"http://{NODE_HOST}:{node_port}"
def _build_volumes(thread_id: str) -> list[k8s_client.V1Volume]:
"""Build volume list: PVC when configured, otherwise hostPath."""
if SKILLS_PVC_NAME:
skills_vol = k8s_client.V1Volume(
name="skills",
persistent_volume_claim=k8s_client.V1PersistentVolumeClaimVolumeSource(
claim_name=SKILLS_PVC_NAME,
read_only=True,
),
)
else:
skills_vol = k8s_client.V1Volume(
name="skills",
host_path=k8s_client.V1HostPathVolumeSource(
path=SKILLS_HOST_PATH,
type="Directory",
),
)
if USERDATA_PVC_NAME:
userdata_vol = k8s_client.V1Volume(
name="user-data",
persistent_volume_claim=k8s_client.V1PersistentVolumeClaimVolumeSource(
claim_name=USERDATA_PVC_NAME,
),
)
else:
userdata_vol = k8s_client.V1Volume(
name="user-data",
host_path=k8s_client.V1HostPathVolumeSource(
path=join_host_path(THREADS_HOST_PATH, thread_id, "user-data"),
type="DirectoryOrCreate",
),
)
return [skills_vol, userdata_vol]
def _build_volume_mounts(thread_id: str) -> list[k8s_client.V1VolumeMount]:
"""Build volume mount list, using subPath for PVC user-data."""
userdata_mount = k8s_client.V1VolumeMount(
name="user-data",
mount_path="/mnt/user-data",
read_only=False,
)
if USERDATA_PVC_NAME:
userdata_mount.sub_path = f"threads/{thread_id}/user-data"
return [
k8s_client.V1VolumeMount(
name="skills",
mount_path="/mnt/skills",
read_only=True,
),
userdata_mount,
]
def _build_pod(sandbox_id: str, thread_id: str) -> k8s_client.V1Pod:
"""Construct a Pod manifest for a single sandbox."""
thread_id = _validate_thread_id(thread_id)
return k8s_client.V1Pod(
metadata=k8s_client.V1ObjectMeta(
name=_pod_name(sandbox_id),
namespace=K8S_NAMESPACE,
labels={
"app": "deer-flow-sandbox",
"sandbox-id": sandbox_id,
"app.kubernetes.io/name": "deer-flow",
"app.kubernetes.io/component": "sandbox",
},
),
spec=k8s_client.V1PodSpec(
containers=[
k8s_client.V1Container(
name="sandbox",
image=SANDBOX_IMAGE,
image_pull_policy="IfNotPresent",
ports=[
k8s_client.V1ContainerPort(
name="http",
container_port=8080,
protocol="TCP",
)
],
readiness_probe=k8s_client.V1Probe(
http_get=k8s_client.V1HTTPGetAction(
path="/v1/sandbox",
port=8080,
),
initial_delay_seconds=5,
period_seconds=5,
timeout_seconds=3,
failure_threshold=3,
),
liveness_probe=k8s_client.V1Probe(
http_get=k8s_client.V1HTTPGetAction(
path="/v1/sandbox",
port=8080,
),
initial_delay_seconds=10,
period_seconds=10,
timeout_seconds=3,
failure_threshold=3,
),
resources=k8s_client.V1ResourceRequirements(
requests={
"cpu": "100m",
"memory": "256Mi",
"ephemeral-storage": "500Mi",
},
limits={
"cpu": "1000m",
"memory": "1Gi",
"ephemeral-storage": "500Mi",
},
),
volume_mounts=_build_volume_mounts(thread_id),
security_context=k8s_client.V1SecurityContext(
privileged=False,
allow_privilege_escalation=True,
),
)
],
volumes=_build_volumes(thread_id),
restart_policy="Always",
),
)
def _build_service(sandbox_id: str) -> k8s_client.V1Service:
"""Construct a NodePort Service manifest (port auto-allocated by K8s)."""
return k8s_client.V1Service(
metadata=k8s_client.V1ObjectMeta(
name=_svc_name(sandbox_id),
namespace=K8S_NAMESPACE,
labels={
"app": "deer-flow-sandbox",
"sandbox-id": sandbox_id,
"app.kubernetes.io/name": "deer-flow",
"app.kubernetes.io/component": "sandbox",
},
),
spec=k8s_client.V1ServiceSpec(
type="NodePort",
ports=[
k8s_client.V1ServicePort(
name="http",
port=8080,
target_port=8080,
protocol="TCP",
# nodePort omitted → K8s auto-allocates from the range
)
],
selector={
"sandbox-id": sandbox_id,
},
),
)
def _get_node_port(sandbox_id: str) -> int | None:
"""Read the K8s-allocated NodePort from the Service."""
try:
svc = core_v1.read_namespaced_service(_svc_name(sandbox_id), K8S_NAMESPACE)
for port in svc.spec.ports or []:
if port.name == "http":
return port.node_port
except ApiException:
pass
return None
def _get_pod_phase(sandbox_id: str) -> str:
"""Return the Pod phase (Pending / Running / Succeeded / Failed / Unknown)."""
try:
pod = core_v1.read_namespaced_pod(_pod_name(sandbox_id), K8S_NAMESPACE)
return pod.status.phase or "Unknown"
except ApiException:
return "NotFound"
# ── API endpoints ────────────────────────────────────────────────────────
@app.get("/health")
async def health():
"""Provisioner health check."""
return {"status": "ok"}
@app.post("/api/sandboxes", response_model=SandboxResponse)
async def create_sandbox(req: CreateSandboxRequest):
"""Create a sandbox Pod + NodePort Service for *sandbox_id*.
If the sandbox already exists, returns the existing information
(idempotent).
"""
sandbox_id = req.sandbox_id
thread_id = req.thread_id
logger.info(
f"Received request to create sandbox '{sandbox_id}' for thread '{thread_id}'"
)
# ── Fast path: sandbox already exists ────────────────────────────
existing_port = _get_node_port(sandbox_id)
if existing_port:
return SandboxResponse(
sandbox_id=sandbox_id,
sandbox_url=_sandbox_url(existing_port),
status=_get_pod_phase(sandbox_id),
)
# ── Create Pod ───────────────────────────────────────────────────
try:
core_v1.create_namespaced_pod(K8S_NAMESPACE, _build_pod(sandbox_id, thread_id))
logger.info(f"Created Pod {_pod_name(sandbox_id)}")
except ApiException as exc:
if exc.status != 409: # 409 = AlreadyExists
raise HTTPException(
status_code=500, detail=f"Pod creation failed: {exc.reason}"
)
# ── Create Service ───────────────────────────────────────────────
try:
core_v1.create_namespaced_service(K8S_NAMESPACE, _build_service(sandbox_id))
logger.info(f"Created Service {_svc_name(sandbox_id)}")
except ApiException as exc:
if exc.status != 409:
# Roll back the Pod on failure
try:
core_v1.delete_namespaced_pod(_pod_name(sandbox_id), K8S_NAMESPACE)
except ApiException:
pass
raise HTTPException(
status_code=500, detail=f"Service creation failed: {exc.reason}"
)
# ── Read the auto-allocated NodePort ─────────────────────────────
node_port: int | None = None
for _ in range(20):
node_port = _get_node_port(sandbox_id)
if node_port:
break
time.sleep(0.5)
if not node_port:
raise HTTPException(
status_code=500, detail="NodePort was not allocated in time"
)
return SandboxResponse(
sandbox_id=sandbox_id,
sandbox_url=_sandbox_url(node_port),
status=_get_pod_phase(sandbox_id),
)
@app.delete("/api/sandboxes/{sandbox_id}")
async def destroy_sandbox(sandbox_id: str):
"""Destroy a sandbox Pod + Service."""
errors: list[str] = []
# Delete Service
try:
core_v1.delete_namespaced_service(_svc_name(sandbox_id), K8S_NAMESPACE)
logger.info(f"Deleted Service {_svc_name(sandbox_id)}")
except ApiException as exc:
if exc.status != 404:
errors.append(f"service: {exc.reason}")
# Delete Pod
try:
core_v1.delete_namespaced_pod(_pod_name(sandbox_id), K8S_NAMESPACE)
logger.info(f"Deleted Pod {_pod_name(sandbox_id)}")
except ApiException as exc:
if exc.status != 404:
errors.append(f"pod: {exc.reason}")
if errors:
raise HTTPException(
status_code=500, detail=f"Partial cleanup: {', '.join(errors)}"
)
return {"ok": True, "sandbox_id": sandbox_id}
@app.get("/api/sandboxes/{sandbox_id}", response_model=SandboxResponse)
async def get_sandbox(sandbox_id: str):
"""Return current status and URL for a sandbox."""
node_port = _get_node_port(sandbox_id)
if not node_port:
raise HTTPException(status_code=404, detail=f"Sandbox '{sandbox_id}' not found")
return SandboxResponse(
sandbox_id=sandbox_id,
sandbox_url=_sandbox_url(node_port),
status=_get_pod_phase(sandbox_id),
)
@app.get("/api/sandboxes")
async def list_sandboxes():
"""List every sandbox currently managed in the namespace."""
try:
services = core_v1.list_namespaced_service(
K8S_NAMESPACE,
label_selector="app=deer-flow-sandbox",
)
except ApiException as exc:
raise HTTPException(
status_code=500, detail=f"Failed to list services: {exc.reason}"
)
sandboxes: list[SandboxResponse] = []
for svc in services.items:
sid = (svc.metadata.labels or {}).get("sandbox-id")
if not sid:
continue
node_port = None
for port in svc.spec.ports or []:
if port.name == "http":
node_port = port.node_port
break
if node_port:
sandboxes.append(
SandboxResponse(
sandbox_id=sid,
sandbox_url=_sandbox_url(node_port),
status=_get_pod_phase(sid),
)
)
return {"sandboxes": sandboxes, "count": len(sandboxes)}