Add LandSAR cluster worker deployment

This commit is contained in:
2026-06-24 14:10:58 +08:00
parent 71c524967c
commit da09ba05cb
88 changed files with 9300 additions and 3011 deletions
+11
View File
@@ -126,6 +126,17 @@ _STATS_CACHE_DATA: Optional[Dict[str, Any]] = None
_STATS_CACHE_EXPIRES_AT = 0.0
_STATS_CACHE_GENERATED_AT_UTC: Optional[str] = None
DASHBOARD_STATS_CACHE_TTL_SECONDS = read_int_env(
"DASHBOARD_STATS_CACHE_TTL_SECONDS",
STATS_CACHE_TTL_SECONDS,
minimum=0,
maximum=3600,
)
_DASHBOARD_STATS_CACHE_LOCK = asyncio.Lock()
_DASHBOARD_STATS_CACHE_DATA: Optional[Dict[str, Any]] = None
_DASHBOARD_STATS_CACHE_EXPIRES_AT = 0.0
_DASHBOARD_STATS_CACHE_GENERATED_AT_UTC: Optional[str] = None
# ---------------------------------------------------------------------------
# AOI token store
# ---------------------------------------------------------------------------
+114
View File
@@ -42,6 +42,12 @@ LANDSAR_PRODUCTION_JOB_MAX_ATTEMPTS = read_int_env(
minimum=1,
maximum=10,
)
LANDSAR_CLUSTER_ITEM_JOB_MAX_ATTEMPTS = read_int_env(
"LANDSAR_CLUSTER_ITEM_JOB_MAX_ATTEMPTS",
1,
minimum=1,
maximum=10,
)
class RunJobRequest(BaseModel):
@@ -441,6 +447,114 @@ async def submit_run(
}
@router.post("/landsar-cluster/run")
async def submit_landsar_cluster_run(
req: RunJobRequest,
current_user: AuthUserORM = Depends(_get_current_user),
):
if str(req.engine_code or "").strip().lower() != "landsar":
raise HTTPException(status_code=400, detail="LandSAR cluster only accepts engine_code='landsar'.")
registry = _get_registry()
engine = registry.get_engine("landsar")
if not engine:
raise HTTPException(status_code=400, detail="Engine 'landsar' not found.")
valid_profiles = {profile.code for profile in engine.get_profiles()}
if req.profile not in valid_profiles:
raise HTTPException(
status_code=400,
detail=(
f"Engine 'landsar' does not support profile '{req.profile}'. "
f"Available profiles: {sorted(valid_profiles)}"
),
)
validation_summary = None
if hasattr(engine, "validate_root_dir"):
try:
validation_summary = await asyncio.to_thread(
engine.validate_root_dir,
req.root_dir,
req.num_to_process,
req.rerun_mode,
)
except ValueError as exc:
raise HTTPException(status_code=400, detail=str(exc)) from exc
if int(validation_summary.get("task_count", 0) or 0) <= 0:
if (
req.rerun_mode == "unfinished_only"
and int(validation_summary.get("skipped_completed_count", 0) or 0) > 0
):
raise HTTPException(
status_code=400,
detail=(
f"All discovered Task_* directories already have completed "
f"landsar/{req.profile} results under: {req.root_dir}"
),
)
raise HTTPException(status_code=400, detail="No valid LandSAR task directories selected.")
normalized_extra = dict(req.extra or {})
if hasattr(engine, "normalize_extra"):
try:
normalized_extra = engine.normalize_extra(normalized_extra)
except ValueError as exc:
raise HTTPException(status_code=400, detail=str(exc)) from exc
if validation_summary is not None:
validated_task_count = validation_summary.get("task_count", 0)
normalized_extra.update(
{
"__validated_task_count": validated_task_count,
"__validated_mode": validation_summary.get("mode", ""),
"__rerun_mode": req.rerun_mode,
"__discovered_task_count": int(validation_summary.get("discovered_task_count", validated_task_count) or 0),
"__skipped_completed_count": int(validation_summary.get("skipped_completed_count", 0) or 0),
"__cluster": True,
}
)
effective_timeout_seconds = req.timeout_seconds
if effective_timeout_seconds is None:
engine_default_timeout = getattr(engine, "default_timeout_seconds", None)
if engine_default_timeout:
effective_timeout_seconds = int(engine_default_timeout)
try:
async with _new_session() as db:
result = await dinsar_production_service.create_landsar_cluster_run(
profile_code=req.profile,
root_dir=req.root_dir,
num_to_process=req.num_to_process,
rerun_mode=req.rerun_mode,
timeout_seconds=effective_timeout_seconds,
extra=normalized_extra,
created_by=getattr(current_user, "username", None),
max_attempts=LANDSAR_CLUSTER_ITEM_JOB_MAX_ATTEMPTS,
db=db,
)
except ValueError as exc:
message = str(exc)
status_code = 409 if "浠诲姟鍐茬獊" in message else 400
raise HTTPException(status_code=status_code, detail=message) from exc
return {
"task_id": result["task_id"],
"job_id": None,
"run_id": result["run_id"],
"workflow_run_id": None,
"job_type": "LANDSAR_CLUSTER_ITEM",
"engine_code": "landsar",
"profile": req.profile,
"selected_task_count": result.get("selected_task_count", 0),
"discovered_task_count": result.get("discovered_task_count", result.get("selected_task_count", 0)),
"skipped_completed_count": result.get("skipped_completed_count", 0),
"rerun_mode": result.get("rerun_mode", req.rerun_mode),
"message": "LandSAR cluster items queued.",
}
@router.get("/runs")
async def list_runs(limit: int = 20, offset: int = 0):
async with _new_session() as db:
+3 -8
View File
@@ -318,10 +318,7 @@ async def _build_radar_preview_cache(
geo_error = "invalid_bbox"
has_geo_cache = False
else:
source_corner_mapping = await asyncio.to_thread(
data_service.get_radar_source_corner_mapping,
record.file_path,
)
source_corner_mapping = data_service.get_radar_record_corner_mapping(record)
ok_geo, geo_error = await asyncio.to_thread(
image_service.create_geocorrected_radar_cached_image,
preview_source,
@@ -408,6 +405,7 @@ async def _get_cached_radar_preview(data_id: int, db: AsyncSession):
raw_cache_path, geo_cache_path = _radar_preview_paths(record)
if (
(record.preview_cache_status or "NONE") == "READY"
and (record.preview_cache_version or "") == settings.RADAR_GEO_CACHE_VERSION
and record.preview_cache_path
and str(record.preview_cache_path).lower().endswith(".webp")
and os.path.exists(record.preview_cache_path)
@@ -418,7 +416,7 @@ async def _get_cached_radar_preview(data_id: int, db: AsyncSession):
headers={"Cache-Control": "public, max-age=31536000"},
)
if os.path.exists(geo_cache_path):
if os.path.exists(geo_cache_path) and (record.preview_cache_version or "") == settings.RADAR_GEO_CACHE_VERSION:
return FileResponse(
geo_cache_path,
media_type="image/webp",
@@ -523,7 +521,6 @@ async def search_radar_data_endpoint(
product_unique_id: Optional[str] = Form(None),
orbit_direction: Optional[str] = Form(None),
has_orbit_data: Optional[bool] = Form(None),
is_envi_processed: Optional[bool] = Form(None),
imaging_date_from: Optional[str] = Form(None),
imaging_date_to: Optional[str] = Form(None),
region_tree_id: Optional[str] = Form(None),
@@ -609,8 +606,6 @@ async def search_radar_data_endpoint(
filters.append(RadarDataORM.orbit_direction.ilike(f"%{n_orbit_direction}%"))
if has_orbit_data is not None:
filters.append(RadarDataORM.has_orbit_data == has_orbit_data)
if is_envi_processed is not None:
filters.append(RadarDataORM.is_envi_processed == is_envi_processed)
if n_date_from:
filters.append(RadarDataORM.imaging_date >= n_date_from)
if n_date_to:
File diff suppressed because it is too large Load Diff