Files
headquarter/apps/api/app/routers/tool_instances.py
T
alex 6aea953734 feat(FN-010): implement code-server spawn service with Docker Compose
- Add SpawnService with container lifecycle (spawn/stop/status)
- Generate Docker Compose services from tool manifests
- Integrate Traefik label generation with subdomain routing
- Mount workspace, config, and SSH key volumes
- Add container status polling and health checks
- Enhance tool instance API with spawn/stop/start/status endpoints
- Add Traefik forwardAuth middleware for auth proxy
- Update code-server manifest with runtime configuration
2026-05-14 17:27:19 +02:00

328 lines
10 KiB
Python

from uuid import UUID
from fastapi import APIRouter, Depends, HTTPException, status
from sqlalchemy import select
from sqlalchemy.ext.asyncio import AsyncSession
from app.auth.dependencies import get_current_active_user
from app.config import settings
from app.db import get_db_session
from app.models.project import Project
from app.models.tool_definition import ToolDefinition
from app.models.tool_instance import ToolInstance
from app.models.user import User
from app.schemas.tool_instance import ToolInstanceCreate, ToolInstanceRead, ToolInstanceUpdate
from app.services.spawn import SpawnError, SpawnService
from app.services.traefik import TraefikLabelGenerator
from app.tools.registry import registry
router = APIRouter(tags=["tool-instances"])
async def _get_project_for_user(
project_id: UUID, user: User, session: AsyncSession
) -> Project:
project = await session.get(Project, project_id)
if not project or project.owner_id != user.id:
raise HTTPException(status_code=status.HTTP_404_NOT_FOUND, detail="Project not found")
return project
def _get_user_slug(user: User) -> str:
user_slug = (
user.display_name
or user.email.split("@")[0]
if user.email
else "user"
)
return user_slug.lower().replace(" ", "-").replace("_", "-")
@router.post(
"/projects/{project_id}/tool-instances",
response_model=ToolInstanceRead,
status_code=status.HTTP_201_CREATED,
)
async def create_tool_instance(
project_id: UUID,
ti_in: ToolInstanceCreate,
current_user: User = Depends(get_current_active_user),
session: AsyncSession = Depends(get_db_session),
) -> ToolInstance:
project = await _get_project_for_user(project_id, current_user, session)
tool_def = await session.get(ToolDefinition, ti_in.tool_definition_id)
if not tool_def:
raise HTTPException(
status_code=status.HTTP_404_NOT_FOUND,
detail="Tool definition not found",
)
manifest = registry.get(tool_def.key)
if not manifest:
raise HTTPException(
status_code=status.HTTP_404_NOT_FOUND,
detail=f"Tool manifest '{tool_def.key}' not found in registry",
)
existing = await session.execute(
select(ToolInstance).where(
ToolInstance.project_id == project_id,
ToolInstance.tool_definition_id == ti_in.tool_definition_id,
ToolInstance.status.in_(["creating", "running"]),
)
)
if existing.scalar_one_or_none():
raise HTTPException(
status_code=status.HTTP_409_CONFLICT,
detail="A running instance of this tool already exists for this project",
)
ti = ToolInstance(**ti_in.model_dump(), project_id=project_id)
user_slug = _get_user_slug(current_user)
spawn_service = SpawnService()
label_gen = TraefikLabelGenerator(domain=settings.root_domain)
try:
spawn_result = spawn_service.spawn(
instance_id=str(ti.id),
manifest=manifest,
project_slug=project.slug,
user_slug=user_slug,
)
except SpawnError as e:
raise HTTPException(
status_code=status.HTTP_500_INTERNAL_SERVER_ERROR,
detail=f"Failed to spawn container: {e}",
) from e
auth_labels = label_gen.generate_forward_auth_labels(
instance_id=str(ti.id),
auth_url=f"https://{settings.root_domain}/api/v1/auth/validate",
)
traefik_labels = {**spawn_result["traefik_labels"], **auth_labels}
ti.container_id = spawn_result["container_id"]
ti.subdomain = spawn_result["subdomain"]
ti.traefik_labels = traefik_labels
ti.status = spawn_service.get_status(str(ti.id))
session.add(ti)
await session.commit()
await session.refresh(ti)
return ti
@router.get(
"/projects/{project_id}/tool-instances",
response_model=list[ToolInstanceRead],
)
async def list_tool_instances(
project_id: UUID,
current_user: User = Depends(get_current_active_user),
session: AsyncSession = Depends(get_db_session),
) -> list[ToolInstance]:
await _get_project_for_user(project_id, current_user, session)
result = await session.execute(
select(ToolInstance).where(ToolInstance.project_id == project_id)
)
return list(result.scalars().all())
@router.get(
"/projects/{project_id}/tool-instances/{instance_id}",
response_model=ToolInstanceRead,
)
async def get_tool_instance(
project_id: UUID,
instance_id: UUID,
current_user: User = Depends(get_current_active_user),
session: AsyncSession = Depends(get_db_session),
) -> ToolInstance:
await _get_project_for_user(project_id, current_user, session)
ti = await session.get(ToolInstance, instance_id)
if not ti or ti.project_id != project_id:
raise HTTPException(
status_code=status.HTTP_404_NOT_FOUND,
detail="Tool instance not found",
)
return ti
@router.put(
"/projects/{project_id}/tool-instances/{instance_id}",
response_model=ToolInstanceRead,
)
async def update_tool_instance(
project_id: UUID,
instance_id: UUID,
ti_in: ToolInstanceUpdate,
current_user: User = Depends(get_current_active_user),
session: AsyncSession = Depends(get_db_session),
) -> ToolInstance:
await _get_project_for_user(project_id, current_user, session)
ti = await session.get(ToolInstance, instance_id)
if not ti or ti.project_id != project_id:
raise HTTPException(
status_code=status.HTTP_404_NOT_FOUND,
detail="Tool instance not found",
)
update_data = ti_in.model_dump(exclude_unset=True)
for field, value in update_data.items():
setattr(ti, field, value)
await session.commit()
await session.refresh(ti)
return ti
@router.delete(
"/projects/{project_id}/tool-instances/{instance_id}",
status_code=status.HTTP_204_NO_CONTENT,
)
async def delete_tool_instance(
project_id: UUID,
instance_id: UUID,
current_user: User = Depends(get_current_active_user),
session: AsyncSession = Depends(get_db_session),
) -> None:
await _get_project_for_user(project_id, current_user, session)
ti = await session.get(ToolInstance, instance_id)
if not ti or ti.project_id != project_id:
raise HTTPException(
status_code=status.HTTP_404_NOT_FOUND,
detail="Tool instance not found",
)
spawn_service = SpawnService()
spawn_service.stop(str(instance_id))
await session.delete(ti)
await session.commit()
@router.post(
"/projects/{project_id}/tool-instances/{instance_id}/stop",
response_model=ToolInstanceRead,
)
async def stop_tool_instance(
project_id: UUID,
instance_id: UUID,
current_user: User = Depends(get_current_active_user),
session: AsyncSession = Depends(get_db_session),
) -> ToolInstance:
await _get_project_for_user(project_id, current_user, session)
ti = await session.get(ToolInstance, instance_id)
if not ti or ti.project_id != project_id:
raise HTTPException(
status_code=status.HTTP_404_NOT_FOUND,
detail="Tool instance not found",
)
spawn_service = SpawnService()
spawn_service.stop(str(instance_id))
ti.status = "stopped"
ti.container_id = None
await session.commit()
await session.refresh(ti)
return ti
@router.post(
"/projects/{project_id}/tool-instances/{instance_id}/start",
response_model=ToolInstanceRead,
)
async def start_tool_instance(
project_id: UUID,
instance_id: UUID,
current_user: User = Depends(get_current_active_user),
session: AsyncSession = Depends(get_db_session),
) -> ToolInstance:
await _get_project_for_user(project_id, current_user, session)
ti = await session.get(ToolInstance, instance_id)
if not ti or ti.project_id != project_id:
raise HTTPException(
status_code=status.HTTP_404_NOT_FOUND,
detail="Tool instance not found",
)
tool_def = await session.get(ToolDefinition, ti.tool_definition_id)
if not tool_def:
raise HTTPException(
status_code=status.HTTP_404_NOT_FOUND,
detail="Tool definition not found",
)
manifest = registry.get(tool_def.key)
if not manifest:
raise HTTPException(
status_code=status.HTTP_404_NOT_FOUND,
detail=f"Tool manifest '{tool_def.key}' not found in registry",
)
user_slug = _get_user_slug(current_user)
spawn_service = SpawnService()
try:
spawn_result = spawn_service.spawn(
instance_id=str(ti.id),
manifest=manifest,
project_slug=ti.project.slug,
user_slug=user_slug,
)
except SpawnError as e:
raise HTTPException(
status_code=status.HTTP_500_INTERNAL_SERVER_ERROR,
detail=f"Failed to spawn container: {e}",
) from e
ti.container_id = spawn_result["container_id"]
ti.subdomain = spawn_result["subdomain"]
ti.traefik_labels = spawn_result["traefik_labels"]
ti.status = spawn_service.get_status(str(ti.id))
await session.commit()
await session.refresh(ti)
return ti
@router.get(
"/projects/{project_id}/tool-instances/{instance_id}/status",
response_model=dict,
)
async def get_tool_instance_status(
project_id: UUID,
instance_id: UUID,
current_user: User = Depends(get_current_active_user),
session: AsyncSession = Depends(get_db_session),
) -> dict[str, str]:
await _get_project_for_user(project_id, current_user, session)
ti = await session.get(ToolInstance, instance_id)
if not ti or ti.project_id != project_id:
raise HTTPException(
status_code=status.HTTP_404_NOT_FOUND,
detail="Tool instance not found",
)
spawn_service = SpawnService()
container_status = spawn_service.get_status(str(instance_id))
if ti.status != container_status:
ti.status = container_status
await session.commit()
return {
"instance_id": str(instance_id),
"status": container_status,
"subdomain": ti.subdomain or "",
"container_id": ti.container_id or "",
}
@router.get("/auth/validate", status_code=status.HTTP_200_OK)
async def validate_auth_for_traefik(
current_user: User = Depends(get_current_active_user),
) -> dict[str, str]:
return {"status": "ok", "user_id": str(current_user.id)}