260 lines
8.3 KiB
Python
260 lines
8.3 KiB
Python
"""
|
|
Source browser for detection pipeline.
|
|
|
|
Lists available media sources from blob storage (MinIO).
|
|
All file-based sources go through MinIO — no host filesystem access.
|
|
The pipeline downloads chunks to a temp path before processing.
|
|
|
|
Source types (current and future):
|
|
- chunk_job: pre-chunked segments in MinIO (current)
|
|
- upload: user-uploaded file, lands in MinIO via upload endpoint (future)
|
|
- device: local camera/capture card via ffmpeg, no MinIO (future)
|
|
- stream: RTMP/HLS URL via ffmpeg, no MinIO (future)
|
|
|
|
GET /detect/sources — list chunk jobs from blob store
|
|
GET /detect/sources/{job_id}/chunks — list chunks for a specific job
|
|
POST /detect/run — launch pipeline on selected source
|
|
"""
|
|
|
|
from __future__ import annotations
|
|
|
|
import logging
|
|
import os
|
|
import threading
|
|
import uuid
|
|
|
|
from fastapi import APIRouter, HTTPException
|
|
from pydantic import BaseModel
|
|
|
|
logger = logging.getLogger(__name__)
|
|
|
|
router = APIRouter(prefix="/detect", tags=["detect"])
|
|
|
|
# In-process pipeline tracking
|
|
_running_jobs: dict[str, "threading.Thread"] = {}
|
|
_cancelled_jobs: set[str] = set()
|
|
|
|
|
|
class ChunkInfo(BaseModel):
|
|
filename: str
|
|
key: str
|
|
size_bytes: int
|
|
|
|
|
|
class SourceInfo(BaseModel):
|
|
job_id: str
|
|
source_type: str = "chunk_job"
|
|
chunk_count: int
|
|
total_bytes: int = 0
|
|
|
|
|
|
class RunRequest(BaseModel):
|
|
video_path: str # storage key
|
|
profile_name: str = "soccer_broadcast"
|
|
source_asset_id: str = ""
|
|
checkpoint: bool = True
|
|
skip_vlm: bool = False
|
|
skip_cloud: bool = False
|
|
log_level: str = "INFO" # INFO | DEBUG
|
|
|
|
|
|
class RunResponse(BaseModel):
|
|
status: str
|
|
job_id: str
|
|
video_path: str
|
|
|
|
|
|
# ---------------------------------------------------------------------------
|
|
# Source listing
|
|
# ---------------------------------------------------------------------------
|
|
|
|
def _list_sources() -> list[SourceInfo]:
|
|
"""List chunk jobs from blob storage."""
|
|
from core.storage.blob import get_store
|
|
|
|
store = get_store("out")
|
|
try:
|
|
objects = store.list(prefix="chunks/")
|
|
except Exception as e:
|
|
logger.warning("Failed to list blob sources: %s", e)
|
|
return []
|
|
|
|
jobs: dict[str, int] = {}
|
|
job_bytes: dict[str, int] = {}
|
|
for obj in objects:
|
|
# Keys include store prefix: out/chunks/{job_id}/file.mp4
|
|
# Strip prefix to get: chunks/{job_id}/file.mp4
|
|
rel_key = obj.key.removeprefix(store.prefix)
|
|
parts = rel_key.split("/")
|
|
if len(parts) >= 3 and parts[0] == "chunks":
|
|
job_id = parts[1]
|
|
jobs[job_id] = jobs.get(job_id, 0) + 1
|
|
job_bytes[job_id] = job_bytes.get(job_id, 0) + obj.size_bytes
|
|
|
|
sources = []
|
|
for job_id, count in sorted(jobs.items()):
|
|
source = SourceInfo(
|
|
job_id=job_id,
|
|
source_type="chunk_job",
|
|
chunk_count=count,
|
|
total_bytes=job_bytes.get(job_id, 0),
|
|
)
|
|
sources.append(source)
|
|
return sources
|
|
|
|
|
|
@router.get("/sources", response_model=list[SourceInfo])
|
|
def list_sources():
|
|
"""List available chunk jobs from blob storage."""
|
|
return _list_sources()
|
|
|
|
|
|
@router.get("/sources/{source_job_id}/chunks", response_model=list[ChunkInfo])
|
|
def list_chunks(source_job_id: str):
|
|
"""List chunks for a specific source job."""
|
|
from core.storage.blob import get_store
|
|
|
|
store = get_store("out")
|
|
try:
|
|
objects = store.list(prefix=f"chunks/{source_job_id}/", extensions={".mp4"})
|
|
except Exception as e:
|
|
logger.warning("Failed to list chunks for %s: %s", source_job_id, e)
|
|
raise HTTPException(status_code=503, detail=f"Blob storage unavailable: {e}")
|
|
|
|
if not objects:
|
|
raise HTTPException(status_code=404, detail=f"Source not found: {source_job_id}")
|
|
|
|
chunks = []
|
|
for obj in objects:
|
|
info = ChunkInfo(filename=obj.filename, key=obj.key, size_bytes=obj.size_bytes)
|
|
chunks.append(info)
|
|
return sorted(chunks, key=lambda c: c.filename)
|
|
|
|
|
|
@router.get("/sources/{source_job_id}/chunks/{filename}/url")
|
|
def get_chunk_url(source_job_id: str, filename: str):
|
|
"""Return a presigned URL for previewing a chunk in the browser."""
|
|
from core.storage.blob import get_store
|
|
|
|
store = get_store("out")
|
|
key = f"chunks/{source_job_id}/{filename}"
|
|
try:
|
|
url = store.get_url(key, expires=3600)
|
|
except Exception as e:
|
|
raise HTTPException(status_code=503, detail=f"Could not generate URL: {e}")
|
|
return {"url": url}
|
|
|
|
|
|
# ---------------------------------------------------------------------------
|
|
# Run pipeline
|
|
# ---------------------------------------------------------------------------
|
|
|
|
def _resolve_video_path(video_path: str) -> str:
|
|
"""Download a chunk from blob storage to a temp file."""
|
|
from core.storage.blob import get_store
|
|
|
|
store = get_store("out")
|
|
try:
|
|
return store.download_to_temp(video_path)
|
|
except Exception as e:
|
|
raise HTTPException(status_code=400, detail=f"Failed to download chunk: {e}")
|
|
|
|
|
|
@router.post("/run", response_model=RunResponse)
|
|
def run_pipeline(req: RunRequest):
|
|
"""Launch a detection pipeline run on a source chunk."""
|
|
from detect import emit
|
|
from detect.graph import get_pipeline
|
|
from detect.state import DetectState
|
|
|
|
local_path = _resolve_video_path(req.video_path)
|
|
job_id = str(uuid.uuid4())[:8]
|
|
|
|
if req.skip_vlm:
|
|
os.environ["SKIP_VLM"] = "1"
|
|
elif "SKIP_VLM" in os.environ:
|
|
del os.environ["SKIP_VLM"]
|
|
|
|
if req.skip_cloud:
|
|
os.environ["SKIP_CLOUD"] = "1"
|
|
elif "SKIP_CLOUD" in os.environ:
|
|
del os.environ["SKIP_CLOUD"]
|
|
|
|
# Clear any stale events from a previous run with same job_id
|
|
from core.events import _get_redis
|
|
from detect.events import DETECT_EVENTS_PREFIX
|
|
r = _get_redis()
|
|
r.delete(f"{DETECT_EVENTS_PREFIX}:{job_id}")
|
|
|
|
emit.set_run_context(
|
|
run_id=job_id, parent_job_id=job_id, run_type="initial",
|
|
log_level=req.log_level,
|
|
)
|
|
|
|
pipeline = get_pipeline(checkpoint=req.checkpoint)
|
|
|
|
initial_state = DetectState(
|
|
video_path=local_path,
|
|
job_id=job_id,
|
|
profile_name=req.profile_name,
|
|
source_asset_id=req.source_asset_id,
|
|
)
|
|
|
|
import traceback
|
|
|
|
from detect.graph import PipelineCancelled, set_cancel_check, clear_cancel_check
|
|
|
|
set_cancel_check(job_id, lambda: job_id in _cancelled_jobs)
|
|
|
|
def _run():
|
|
try:
|
|
emit.log(job_id, "Pipeline", "INFO",
|
|
f"Starting pipeline: {req.video_path} (profile={req.profile_name})")
|
|
pipeline.invoke(initial_state)
|
|
emit.log(job_id, "Pipeline", "INFO", "Pipeline completed successfully")
|
|
emit.job_complete(job_id, {"status": "completed"})
|
|
except PipelineCancelled:
|
|
emit.log(job_id, "Pipeline", "INFO", "Pipeline cancelled")
|
|
emit.job_complete(job_id, {"status": "cancelled"})
|
|
except Exception as e:
|
|
logger.exception("Pipeline run %s failed: %s", job_id, e)
|
|
tb = traceback.format_exc()
|
|
emit.log(job_id, "Pipeline", "ERROR", str(e))
|
|
emit.log(job_id, "Pipeline", "DEBUG", tb)
|
|
emit.job_complete(job_id, {"status": "failed", "error": str(e)})
|
|
finally:
|
|
_running_jobs.pop(job_id, None)
|
|
_cancelled_jobs.discard(job_id)
|
|
clear_cancel_check(job_id)
|
|
emit.clear_run_context()
|
|
|
|
thread = threading.Thread(target=_run, daemon=True, name=f"pipeline-{job_id}")
|
|
_running_jobs[job_id] = thread
|
|
thread.start()
|
|
|
|
return RunResponse(status="started", job_id=job_id, video_path=req.video_path)
|
|
|
|
|
|
@router.post("/stop/{job_id}")
|
|
def stop_pipeline(job_id: str):
|
|
"""Stop a running pipeline. Signals cancellation; the thread checks on next stage."""
|
|
from detect import emit
|
|
|
|
if job_id not in _running_jobs:
|
|
raise HTTPException(status_code=404, detail=f"No running pipeline: {job_id}")
|
|
|
|
_cancelled_jobs.add(job_id)
|
|
emit.log(job_id, "Pipeline", "INFO", "Stop requested — cancelling after current stage")
|
|
return {"status": "stopping", "job_id": job_id}
|
|
|
|
|
|
@router.post("/clear/{job_id}")
|
|
def clear_pipeline(job_id: str):
|
|
"""Clear events for a job from Redis."""
|
|
from core.events import _get_redis
|
|
from detect.events import DETECT_EVENTS_PREFIX
|
|
|
|
r = _get_redis()
|
|
r.delete(f"{DETECT_EVENTS_PREFIX}:{job_id}")
|
|
return {"status": "cleared", "job_id": job_id}
|