Files
NODEDC_MISSION_CORE/src/k1link/web/app.py
T

501 lines
17 KiB
Python

from __future__ import annotations
import asyncio
import os
import shutil
from collections.abc import AsyncIterator, Iterable
from contextlib import asynccontextmanager, suppress
from pathlib import Path
from typing import Any
from fastapi import FastAPI, HTTPException, Request, WebSocket, WebSocketDisconnect
from fastapi.exceptions import RequestValidationError
from fastapi.responses import JSONResponse
from fastapi.staticfiles import StaticFiles
from pydantic import ValidationError
from k1link import __version__
from k1link.compute import (
IntegratedPerceptionOverlayStore,
RecordedCalibratedFusionStore,
RecordedPerceptionEpochStore,
RecordedPerceptionOverlayMux,
RecordedPerceptionOverlayStore,
)
from k1link.sessions import (
MaterializedRecording,
RecordedMediaInspector,
RecordedMediaManifest,
RecordingPreparationQueueFull,
ReplayCommand,
SessionRecordingMaterializer,
SessionRecordingPreparationManager,
SessionStore,
)
from k1link.web.device_plugin_composition import load_installed_device_plugins
from k1link.web.environment_api import build_environment_router
from k1link.web.laboratory_api import build_laboratory_router
from k1link.web.lidar_api import build_lidar_router
from k1link.web.plugin_catalog import DevicePluginCatalog, PluginCatalogError
from k1link.web.plugin_runtime import (
STATE_READ_ACTION_ID,
DevicePluginActionRequest,
DevicePluginDispatcher,
PluginActionNotFoundError,
PluginExecutionError,
PluginNotFoundError,
PluginRuntimeUnavailableError,
)
from k1link.web.polygon_api import build_polygon_router, configured_polygon_runs_root
from k1link.web.session_api import build_session_router
REPOSITORY_ROOT = Path(__file__).resolve().parents[3]
INVALID_REQUEST_DETAIL = "Некорректные параметры запроса."
def _resolve_media_tool(name: str) -> Path | None:
"""Resolve media tools under interactive shells and minimal launchd PATHs."""
discovered = shutil.which(name)
candidates = (
Path(discovered) if discovered is not None else None,
Path("/opt/homebrew/bin") / name,
Path("/usr/local/bin") / name,
Path("/usr/bin") / name,
)
for candidate in candidates:
if candidate is not None and candidate.is_file() and os.access(candidate, os.X_OK):
return candidate
return None
plugin_environment = load_installed_device_plugins(REPOSITORY_ROOT)
plugin_catalog: DevicePluginCatalog = plugin_environment.catalog
plugin_dispatcher: DevicePluginDispatcher = plugin_environment.dispatcher
session_store = SessionStore(REPOSITORY_ROOT)
session_recording_materializer = SessionRecordingMaterializer(
session_store.data_dir,
exporters=plugin_environment.recording_exporters,
)
session_recorded_media_inspector = RecordedMediaInspector(
session_store.data_dir / "recorded-media-preparations"
)
_ffmpeg = _resolve_media_tool("ffmpeg")
_ffprobe = _resolve_media_tool("ffprobe")
session_legacy_perception_overlay_store = (
RecordedPerceptionOverlayStore(
jobs_root=REPOSITORY_ROOT / ".runtime" / "compute-jobs",
results_root=REPOSITORY_ROOT / ".runtime" / "compute-results",
cache_root=session_store.data_dir / "perception-overlays",
ffmpeg_path=_ffmpeg,
ffprobe_path=_ffprobe,
)
if _ffmpeg is not None and _ffprobe is not None
else None
)
session_calibrated_fusion_store = RecordedCalibratedFusionStore(
jobs_root=REPOSITORY_ROOT / ".runtime" / "compute-jobs",
perception_results_root=REPOSITORY_ROOT / ".runtime" / "compute-results",
fusion_results_root=REPOSITORY_ROOT / ".runtime" / "compute-fusions",
cache_root=session_store.data_dir / "calibrated-fusion-overlays",
)
session_previous_perception_overlay_store = RecordedPerceptionOverlayMux(
session_calibrated_fusion_store, session_legacy_perception_overlay_store
)
session_integrated_perception_store = (
IntegratedPerceptionOverlayStore(
jobs_root=REPOSITORY_ROOT / ".runtime" / "compute-jobs",
results_root=(
REPOSITORY_ROOT
/ ".runtime"
/ "compute-experiments"
/ "e10"
/ "worker-results"
),
lidar_packs_root=(
REPOSITORY_ROOT
/ ".runtime"
/ "compute-experiments"
/ "e10"
/ "lidar-packs"
),
cache_root=session_store.data_dir / "integrated-perception-overlays",
ffmpeg_path=_ffmpeg,
)
if _ffmpeg is not None
else None
)
session_perception_overlay_store = RecordedPerceptionOverlayMux(
session_integrated_perception_store or session_previous_perception_overlay_store,
(
session_previous_perception_overlay_store
if session_integrated_perception_store is not None
else None
),
)
session_perception_epoch_store = (
RecordedPerceptionEpochStore(
jobs_root=REPOSITORY_ROOT / ".runtime" / "compute-jobs",
results_root=REPOSITORY_ROOT / ".runtime" / "compute-results",
ffprobe_path=_ffprobe,
)
if _ffprobe is not None
else None
)
def _prepare_recorded_media_for_launch(
command: ReplayCommand,
_: MaterializedRecording,
) -> tuple[RecordedMediaManifest, ...]:
"""Prepare all camera descriptors inside the background job."""
return tuple(
session_recorded_media_inspector.inspect(artifact, command)
for artifact in session_store.list_recorded_media(command.session_id)
)
def _restore_recorded_media_for_launch(
command: ReplayCommand,
_: MaterializedRecording,
) -> tuple[RecordedMediaManifest, ...] | None:
"""Restore only previously published media descriptors."""
restored: list[RecordedMediaManifest] = []
for artifact in session_store.list_recorded_media(command.session_id):
manifest = session_recorded_media_inspector.restore_prepared(artifact, command)
if manifest is None:
return None
restored.append(manifest)
return tuple(restored)
session_recording_preparation_manager = SessionRecordingPreparationManager(
session_recording_materializer,
ready_preparer=_prepare_recorded_media_for_launch,
ready_restorer=_restore_recorded_media_for_launch,
)
def refresh_observation_catalog() -> tuple[str, ...]:
"""Discover completed or recoverable local evidence without copying payloads."""
imported = [
session_id
for archive in plugin_environment.observation_archives
for session_id in session_store.reconcile_archive(archive)
]
return tuple(dict.fromkeys(imported))
def finalized_replayable_recording_ids() -> tuple[str, ...]:
"""List finalized replayable catalog identities without scheduling work."""
finalized: list[str] = []
cursor: str | None = None
while True:
page = session_store.list_recent(limit=100, cursor=cursor)
finalized.extend(
summary.session_id
for summary in page.items
if summary.replayable and summary.status in {"ready", "interrupted", "failed"}
)
cursor = page.next_cursor
if cursor is None:
return tuple(finalized)
def enqueue_replayable_recordings(session_ids: Iterable[str]) -> tuple[str, ...]:
"""Schedule only explicitly selected newly finalized sessions."""
enqueued: list[str] = []
for session_id in dict.fromkeys(session_ids):
try:
command = session_store.prepare_replay(session_id)
session_recording_preparation_manager.enqueue(
command,
retry_interrupted=True,
)
except RecordingPreparationQueueFull:
return tuple(enqueued)
except Exception:
# One stale/corrupt row must not starve a later newly completed
# session. Historical cold caches remain operator-triggered.
continue
enqueued.append(session_id)
return tuple(enqueued)
def newly_finalized_recording_ids(
known_finalized: set[str] | None,
current_finalized: Iterable[str],
) -> tuple[str, ...]:
"""Return only post-startup completions; the first scan is a baseline."""
if known_finalized is None:
return ()
return tuple(sorted(set(current_finalized) - known_finalized))
async def _recording_preparation_reconciler() -> None:
"""Prepare sessions finalized during this process, never historical rows."""
known_finalized: set[str] | None = None
while True:
try:
await asyncio.to_thread(refresh_observation_catalog)
finalized = set(await asyncio.to_thread(finalized_replayable_recording_ids))
newly_finalized = newly_finalized_recording_ids(
known_finalized,
finalized,
)
if newly_finalized:
await asyncio.to_thread(
enqueue_replayable_recordings,
newly_finalized,
)
known_finalized = finalized
except Exception:
# A transient filesystem/catalog failure must not permanently
# disable preparation of sessions completed later in the run.
pass
await asyncio.sleep(2.0)
@asynccontextmanager
async def app_lifespan(_: FastAPI) -> AsyncIterator[None]:
reconciler: asyncio.Task[None] | None = None
try:
session_recording_preparation_manager.start()
# Recovery is intentionally a one-shot startup phase. The archive
# helper owns a cross-process lease, while ordinary catalog requests
# only perform discovery and therefore never touch a live writer.
for archive in plugin_environment.observation_archives:
await asyncio.to_thread(archive.recover, archive.root)
# Full evidence discovery, hashing and RRD queue reconciliation can be
# expensive on field captures. Start it immediately in the background
# instead of holding the ASGI startup gate.
reconciler = asyncio.create_task(_recording_preparation_reconciler())
yield
finally:
if reconciler is not None:
reconciler.cancel()
with suppress(asyncio.CancelledError):
await reconciler
await asyncio.to_thread(session_recording_preparation_manager.close)
plugin_environment.close()
app = FastAPI(
title="NODEDC MISSION CORE API",
version=__version__,
docs_url="/api/docs",
redoc_url=None,
openapi_url="/api/openapi.json",
lifespan=app_lifespan,
)
@app.exception_handler(RequestValidationError)
async def request_validation_error_handler(
_: Request,
__: RequestValidationError,
) -> JSONResponse:
"""Return validation failures without reflecting request values or credentials."""
return JSONResponse(status_code=422, content={"detail": INVALID_REQUEST_DETAIL})
@app.get("/api/health")
def health() -> dict[str, Any]:
runtime_health = plugin_environment.runtime_health
runtimes_ready = all(item["status"] == "ready" for item in runtime_health)
return {
"ok": runtimes_ready,
"status": "ok" if runtimes_ready else "degraded",
"service": "mission-core-control-plane",
"version": __version__,
"plugin_runtimes": {
"ready": sum(item["status"] == "ready" for item in runtime_health),
"total": len(runtime_health),
},
}
@app.get("/api/v1/device-plugins")
def get_device_plugins() -> dict[str, Any]:
try:
return {"items": plugin_catalog.plugin_documents()}
except PluginCatalogError as exc:
raise HTTPException(status_code=500, detail=str(exc)) from exc
@app.get("/api/v1/device-models")
def get_device_models() -> dict[str, Any]:
try:
return {"items": plugin_catalog.model_documents()}
except PluginCatalogError as exc:
raise HTTPException(status_code=500, detail=str(exc)) from exc
@app.get("/api/v1/device-plugin-runtimes")
def get_device_plugin_runtimes() -> dict[str, Any]:
return {"items": list(plugin_environment.runtime_health)}
@app.post("/api/v1/device-plugins/{plugin_id}/actions/{action_id}")
async def invoke_device_plugin_action(
plugin_id: str,
action_id: str,
request: DevicePluginActionRequest,
) -> dict[str, Any]:
try:
state = await plugin_dispatcher.invoke(plugin_id, action_id, request.input)
return {"state": state}
except (PluginNotFoundError, PluginActionNotFoundError) as exc:
raise HTTPException(status_code=404, detail=str(exc)) from exc
except ValidationError as exc:
raise HTTPException(status_code=422, detail=INVALID_REQUEST_DETAIL) from exc
except ValueError as exc:
raise HTTPException(status_code=400, detail=str(exc)) from exc
except PluginExecutionError as exc:
raise HTTPException(status_code=502, detail=str(exc)) from exc
except PluginRuntimeUnavailableError as exc:
raise HTTPException(status_code=503, detail=str(exc)) from exc
@app.websocket("/api/v1/device-plugins/{plugin_id}/events")
async def device_plugin_events(websocket: WebSocket, plugin_id: str) -> None:
await websocket.accept()
sequence = 0
try:
while True:
state = await plugin_dispatcher.invoke(plugin_id, STATE_READ_ACTION_ID, {})
sequence += 1
await websocket.send_json({"pluginId": plugin_id, "sequence": sequence, "state": state})
await asyncio.sleep(0.5)
except WebSocketDisconnect:
return
except (PluginNotFoundError, PluginActionNotFoundError):
await websocket.close(code=1008, reason="Device plugin is not available")
except (PluginExecutionError, PluginRuntimeUnavailableError):
await websocket.close(code=1011, reason="Device plugin state stream failed")
for legacy_router in plugin_environment.legacy_routers:
app.include_router(legacy_router)
app.include_router(
build_session_router(
session_store,
# Production discovery belongs to the startup/background reconciler.
# HTTP list/replay paths must never rescan evidence roots inline.
catalog_refresher=None,
recording_materializer=session_recording_materializer,
recording_preparation_manager=session_recording_preparation_manager,
media_inspector=session_recorded_media_inspector,
perception_overlay_provider=session_perception_overlay_store,
perception_media_provider=session_perception_epoch_store,
point_color_renderers=plugin_environment.point_color_renderers,
)
)
app.include_router(
build_environment_router(
root_provider=lambda: session_store.data_dir / "ui-environment"
)
)
app.include_router(
build_polygon_router(
root_provider=lambda: configured_polygon_runs_root()
or REPOSITORY_ROOT / ".runtime" / "polygon-runs"
)
)
app.include_router(
build_lidar_router(
root_provider=lambda: (
REPOSITORY_ROOT
/ ".runtime"
/ "compute-experiments"
/ "lidar-replay-v2"
/ "packs"
),
ground_root_provider=lambda: (
REPOSITORY_ROOT
/ ".runtime"
/ "compute-experiments"
/ "lidar-ground-v1"
/ "benchmarks"
),
field_review_root_provider=lambda: (
REPOSITORY_ROOT
/ ".runtime"
/ "compute-experiments"
/ "lidar-field-review-v1"
/ "reviews"
),
local_surface_root_provider=lambda: (
REPOSITORY_ROOT
/ ".runtime"
/ "compute-experiments"
/ "k1-local-surface-v1"
/ "models"
),
e10_source_root_provider=lambda: (
REPOSITORY_ROOT
/ ".runtime"
/ "compute-experiments"
/ "e10"
/ "lidar-packs"
),
dataset_admission_provider=lambda: (
REPOSITORY_ROOT / ".runtime" / "dataset-gateway" / "admission.json"
),
dataset_preview_provider=lambda: (
REPOSITORY_ROOT / ".runtime" / "dataset-gateway" / "preview.json"
),
dataset_rellis_preview_provider=lambda: (
REPOSITORY_ROOT / ".runtime" / "dataset-gateway" / "rellis-preview.json"
),
dataset_rellis_admission_provider=lambda: (
REPOSITORY_ROOT / ".runtime" / "dataset-gateway" / "rellis-admission.json"
),
dataset_ground_preview_provider=lambda: (
REPOSITORY_ROOT / ".runtime" / "dataset-gateway" / "ground-comparison.json"
),
)
)
app.include_router(
build_laboratory_router(
e29_root_provider=lambda: (
REPOSITORY_ROOT / ".runtime" / "compute-experiments" / "e29" / "results"
),
local_surface_root_provider=lambda: (
REPOSITORY_ROOT
/ ".runtime"
/ "compute-experiments"
/ "k1-local-surface-v1"
/ "models"
),
source_pack_root_provider=lambda: (
REPOSITORY_ROOT
/ ".runtime"
/ "compute-experiments"
/ "e10"
/ "lidar-packs"
),
source_result_root_provider=lambda: (
REPOSITORY_ROOT
/ ".runtime"
/ "compute-experiments"
/ "e10"
/ "worker-results"
),
)
)
frontend_dist = REPOSITORY_ROOT / "apps" / "control-station" / "dist"
if frontend_dist.is_dir():
app.mount("/", StaticFiles(directory=frontend_dist, html=True), name="frontend")