Files
headquarter/apps/api/src/services/instance_lifecycle.py
T
Developer 5ed5e1c84b fix: resolve four frontend/backend issues
- Fix ProjectsPage tests by wrapping renders in MemoryRouter (9 passing)
- Improve session auto-naming to 'project / repo / tool' format
- Add missing /users/me/sessions endpoint for sidebar session loading
- Handle git history 500s: catch RuntimeError in endpoints, graceful empty repo handling
- Add git status badge and discard-changes button to FileEditor toolbar

Quality gates: tsc pass, build pass, Python syntax pass
2026-06-03 08:30:28 +00:00

421 lines
15 KiB
Python

"""High-level tool instance lifecycle orchestration.
Coordinates Docker compose, container, tunnel, and config staging services
to create, start, stop, restart, and delete tool instances.
"""
import logging
import os
import shutil
from datetime import datetime
from typing import Any
from fastapi import HTTPException, status
from sqlalchemy import select
from sqlalchemy.ext.asyncio import AsyncSession
from src.models.config_folder import ConfigFolder
from src.models.config_profile import ConfigProfile
from src.models.git_repository import GitRepository
from src.models.project import Project
from src.models.tool_config import ToolConfig
from src.models.tool_instance import ToolInstance
from src.models.tool_type import ToolType
from src.models.user import User
from src.services.docker import compose as compose_svc
from src.services.docker import config_staging
from src.services.docker import container as container_svc
from src.services.docker import tunnel as tunnel_svc
from src.services.docker_build import build_image
from src.services.readiness_probe import execute_probe
logger = logging.getLogger(__name__)
async def create_new_instance(
session: AsyncSession,
project: Project,
repo: GitRepository,
tool_type: ToolType,
user: User,
display_name: str | None,
selected_profile: ConfigProfile | None,
) -> ToolInstance:
"""Create a new tool instance record and its compose file."""
instance_name = await compose_svc._generate_instance_name(
session, project.name, tool_type.name
)
instance_dir = compose_svc.ensure_instance_directory(instance_name)
tool_port = container_svc.find_free_port()
compose_path = await _build_or_render_compose(
tool_type, instance_name, instance_dir, repo, user, project.id, tool_port
)
instance = ToolInstance(
name=instance_name,
display_name=display_name or f"{project.name} / {repo.name} / {tool_type.display_name}",
tool_type_id=tool_type.id,
repository_id=repo.id,
project_id=project.id,
owner_id=user.id,
status="pending",
compose_path=compose_path,
port=tool_port,
selected_profile_id=selected_profile.id if selected_profile else None,
)
session.add(instance)
await session.commit()
await session.refresh(instance)
return instance
async def start_existing_instance(
session: AsyncSession,
instance: ToolInstance,
user: User,
project_id: Any,
) -> dict:
"""Start an existing instance: stage configs, compose up, probe, tunnel."""
if not instance.compose_path or not os.path.exists(instance.compose_path):
raise HTTPException(
status_code=status.HTTP_400_BAD_REQUEST, detail="compose file not found"
)
instance.status = "building"
await session.commit()
env_vars, config_files, port_override, start_command, working_directory, _extra_env, extra_volumes = await _fetch_tool_configs(
session, user.id, instance.tool_type_id, project_id
)
selected_profile = None
if instance.selected_profile_id:
selected_profile = await session.get(ConfigProfile, instance.selected_profile_id)
if selected_profile and selected_profile.user_id == user.id:
instance_dir = os.path.dirname(instance.compose_path)
env_vars, port_override, start_command, working_directory, extra_volumes = await compose_svc._apply_resolved_profile(
selected_profile,
instance_dir,
env_vars,
port_override,
start_command,
working_directory,
extra_volumes,
)
env_file_path, extra_volumes = await _stage_configs_and_folders(
session, user.id, project_id, os.path.dirname(instance.compose_path),
env_vars, config_files, extra_volumes
)
if port_override or start_command or working_directory or extra_volumes:
compose_svc._modify_compose_file(
instance.compose_path, port_override, start_command, working_directory, extra_volumes
)
returncode, _stdout, stderr = compose_svc.execute_compose_command(
instance.compose_path, "up", env_file=env_file_path
)
if returncode != 0:
instance.status = "error"
await session.commit()
raise HTTPException(
status_code=status.HTTP_500_INTERNAL_SERVER_ERROR,
detail=f"failed to start instance: {stderr}",
)
container_id = container_svc.get_container_id(instance.name)
if container_id:
instance.container_id = container_id
container_name = container_svc.get_container_name(instance.name)
if container_name:
instance.container_name = container_name
container_svc.connect_container_to_network(container_name, "backend")
instance.status = "starting"
instance.last_started_at = datetime.now()
await session.commit()
tool_type = await session.get(ToolType, instance.tool_type_id)
success, probe_logs = await _run_readiness_probe(instance, tool_type)
if not success:
instance.status = "failed"
instance.url = None
instance.public_url = None
await session.commit()
return {
"status": "failed",
"error": f"Readiness probe failed: {' '.join(probe_logs)}",
}
instance.status = "running"
await session.commit()
await _start_tunnel_if_web(instance, tool_type)
await session.commit()
return {"status": instance.status, "url": instance.url}
async def restart_existing_instance(
session: AsyncSession,
instance: ToolInstance,
user: User,
project_id: Any,
) -> dict:
"""Restart an instance: re-stage configs, compose restart, tunnel."""
if instance.tunnel_id:
try:
tunnel_svc.stop_cloudflared_tunnel(instance.tunnel_id)
except Exception as exc:
logger.warning("Failed to stop old tunnel: %s", exc)
if not instance.compose_path or not os.path.exists(instance.compose_path):
instance.status = "error"
await session.commit()
return {"status": instance.status}
env_vars, config_files, port_override, start_command, working_directory, _extra_env, extra_volumes = await _fetch_tool_configs(
session, user.id, instance.tool_type_id, project_id
)
stored_profile = None
if instance.selected_profile_id:
stored_profile = await session.get(ConfigProfile, instance.selected_profile_id)
if stored_profile and stored_profile.user_id == user.id:
instance_dir = os.path.dirname(instance.compose_path)
env_vars, port_override, start_command, working_directory, extra_volumes = await compose_svc._apply_resolved_profile(
stored_profile, instance_dir, env_vars, port_override, start_command, working_directory, extra_volumes
)
env_file_path, extra_volumes = await _stage_configs_and_folders(
session, user.id, project_id, os.path.dirname(instance.compose_path),
env_vars, config_files, extra_volumes
)
if port_override or start_command or working_directory or extra_volumes:
compose_svc._modify_compose_file(
instance.compose_path, port_override, start_command, working_directory, extra_volumes
)
returncode, _stdout, _stderr = compose_svc.execute_compose_command(
instance.compose_path, "restart", env_file=env_file_path
)
if returncode != 0:
instance.status = "error"
await session.commit()
return {"status": instance.status}
instance.status = "running"
instance.last_started_at = datetime.now()
tool_type = await session.get(ToolType, instance.tool_type_id)
await _start_tunnel_if_web(instance, tool_type)
await session.commit()
return {"status": instance.status, "url": instance.url}
async def stop_existing_instance(session: AsyncSession, instance: ToolInstance) -> None:
"""Stop an instance and its tunnel."""
if instance.tunnel_id:
try:
tunnel_svc.stop_cloudflared_tunnel(instance.tunnel_id)
except Exception as exc:
logger.warning("Failed to stop tunnel: %s", exc)
if instance.compose_path and os.path.exists(instance.compose_path):
compose_svc.execute_compose_command(instance.compose_path, "stop")
instance.status = "stopped"
instance.last_stopped_at = datetime.now()
instance.url = None
instance.public_url = None
instance.tunnel_id = None
await session.commit()
async def delete_existing_instance(session: AsyncSession, instance: ToolInstance) -> None:
"""Delete an instance, its containers, and its directory."""
if instance.tunnel_id:
try:
tunnel_svc.stop_cloudflared_tunnel(instance.tunnel_id)
except Exception as exc:
logger.warning("Failed to stop tunnel: %s", exc)
if instance.compose_path and os.path.exists(instance.compose_path):
compose_svc.execute_compose_command(instance.compose_path, "down")
instance_dir = os.path.dirname(instance.compose_path)
if os.path.exists(instance_dir):
shutil.rmtree(instance_dir)
await session.delete(instance)
await session.commit()
# ── Internal helpers ───────────────────────────────────────────────────────
async def _build_or_render_compose(
tool_type: ToolType,
instance_name: str,
instance_dir: str,
repo: GitRepository,
user: User,
project_id: Any,
tool_port: int,
) -> str:
"""Build Dockerfile or render compose template."""
if tool_type.definition_type == "dockerfile":
image_tag = f"headquarter/{instance_name}:latest"
if tool_type.dockerfile_template:
returncode, _stdout, stderr = build_image(
instance_dir=instance_dir,
dockerfile=tool_type.dockerfile_template,
tag=image_tag,
build_context=tool_type.build_context,
)
if returncode != 0:
logger.error("Build failed for %s: %s", instance_name, stderr)
raise HTTPException(
status_code=status.HTTP_500_INTERNAL_SERVER_ERROR,
detail=f"Failed to build Docker image: {stderr[:500]}",
)
compose_content = (
f'version: "3.8"\nservices:\n app:\n'
f' image: {image_tag}\n'
f' container_name: {instance_name}\n'
f' ports:\n - "{tool_port}:{tool_type.default_port}"\n'
f' volumes:\n - {repo.path}:/workspace\n'
f' restart: unless-stopped\n'
)
else:
variables = {
"REPO_PATH": repo.path,
"INSTANCE_NAME": instance_name,
"INSTANCE_ID": instance_name,
"TOOL_NAME": instance_name,
"TOOL_PORT": tool_port,
"USER_ID": str(user.id),
"PROJECT_ID": str(project_id),
}
compose_content = compose_svc.render_compose_template(
tool_type.compose_template, variables
)
compose_svc.write_compose_file(instance_dir, compose_content)
return os.path.join(instance_dir, "docker-compose.yml")
async def _fetch_tool_configs(
session: AsyncSession,
user_id: Any,
tool_type_id: Any,
project_id: Any,
) -> tuple[dict, dict, Any, Any, Any, dict, list]:
"""Fetch tool configs and return parsed values."""
env_vars: dict[str, str] = {}
config_files: dict[str, str] = {}
port_override = None
start_command = None
working_directory = None
extra_env_vars: dict[str, str] = {}
extra_volumes: list[dict] = []
query = (
select(ToolConfig)
.where(ToolConfig.user_id == user_id, ToolConfig.tool_type_id == tool_type_id)
.where((ToolConfig.project_id == project_id) | (ToolConfig.project_id.is_(None)))
)
configs = (await session.execute(query)).scalars().all()
for cfg in configs:
if cfg.config_type == "env":
env_vars[cfg.key] = cfg.value
elif cfg.config_type == "file" and cfg.file_path:
config_files[cfg.file_path] = cfg.value
if cfg.port_override:
port_override = cfg.port_override
if cfg.start_command:
start_command = cfg.start_command
if cfg.working_directory:
working_directory = cfg.working_directory
if cfg.environment_variables:
extra_env_vars.update(cfg.environment_variables)
if cfg.volumes:
extra_volumes.extend(cfg.volumes)
env_vars.update(extra_env_vars)
return env_vars, config_files, port_override, start_command, working_directory, extra_env_vars, extra_volumes
async def _stage_configs_and_folders(
session: AsyncSession,
user_id: Any,
project_id: Any,
instance_dir: str,
env_vars: dict[str, str],
config_files: dict[str, str],
extra_volumes: list[dict],
) -> tuple[str | None, list[dict]]:
"""Write env/config files and config folders."""
env_file_path: str | None = None
if env_vars:
env_file_path = compose_svc.write_env_file(instance_dir, env_vars)
if config_files:
config_staging.write_config_files(instance_dir, config_files)
folder_query = select(ConfigFolder).where(
ConfigFolder.user_id == user_id, ConfigFolder.is_active.is_(True)
)
folders = (await session.execute(folder_query)).scalars().all()
if folders:
folder_volumes = config_staging.write_config_folder_files(
instance_dir, folders, str(project_id)
)
extra_volumes.extend(folder_volumes)
return env_file_path, extra_volumes
async def _start_tunnel_if_web(instance: ToolInstance, tool_type: ToolType) -> None:
"""Create Cloudflare tunnel for web-enabled tools."""
if "web" not in tool_type.interfaces or not tool_type.default_port:
instance.url = None
instance.public_url = None
return
try:
tunnel_info = tunnel_svc.start_cloudflared_tunnel(
container_name=instance.container_name or instance.name,
port=tool_type.default_port,
)
instance.tunnel_id = tunnel_info["pid"]
instance.public_url = tunnel_info["url"]
instance.url = tunnel_info["url"]
logger.info("Created tunnel for instance %s: %s", instance.id, tunnel_info["url"])
except Exception as exc:
logger.error("Failed to create tunnel for instance %s: %s", instance.id, exc)
instance.status = "error"
instance.url = None
async def _run_readiness_probe(
instance: ToolInstance, tool_type: ToolType
) -> tuple[bool, list[str]]:
"""Run readiness probe if configured."""
if not tool_type.readiness_probe or not instance.container_id:
return True, []
probe = tool_type.readiness_probe
command = probe.get("command", "")
if not command:
return True, []
return await execute_probe(
container_id=instance.container_id,
command=command,
timeout=probe.get("timeout", 30),
interval=probe.get("interval", 2),
)