Merge pull request #477 from niaid/use_pytools_compute_args

Improve HedwigZARRImages performance in CZI flow

Author: philipmac

em_workflows/czi/flow.py

@@ -21,7 +21,7 @@
 
 import SimpleITK as sitk
 from prefect import flow, task
-from pytools.HedwigZarrImages import HedwigZarrImage, HedwigZarrImages
+from pytools import HedwigZarrImage, HedwigZarrImages
 
 from em_workflows.file_path import FilePath
 from em_workflows.utils import utils
@@ -78,8 +78,11 @@ def copy_zarr_to_assets_dir(file_path: FilePath) -> None:
 
 
 @task
-def generate_imageset(file_path: FilePath) -> List[Dict]:
+def generate_imageset(file_path: FilePath,
+                      use_default_dask=False) -> List[Dict]:
     """
+    :param: use_default_dask: If True, reuses the Prefect Dask Scheduler for the ZARR and Dask array operations.
+
     | ImageSet consists of all the assets for a particular zarr sub-image and label images
     | Macro image is ignored
     | Label image is added as an thumbnail asset
@@ -91,16 +94,17 @@ def generate_imageset(file_path: FilePath) -> List[Dict]:
     """
     zarr_fp = f"{file_path.assets_dir}/{file_path.base}.zarr"
     image_set = list()
-    zarr_images = HedwigZarrImages(Path(zarr_fp))
+
+    if use_default_dask:
+        compute_args = {}
+    else:
+        # This task is used in a sub-flow where it's the only task running.
+        # use all the cores in a thread pool.
+        compute_args = {"scheduler": "threads"}
+    zarr_images = HedwigZarrImages(Path(zarr_fp), compute_args=compute_args)
     # for image_name, image in zarr_images.series():
     for k_idx, image_name in enumerate(zarr_images.get_series_keys()):
-        # The relative path of the zarr group from the root zarr
-        # this assumes a valid zarr group with OME directory inside
-        ome_index_to_zarr_group = zarr_images.zarr_root["OME"].attrs["series"]
-        zarr_idx = ome_index_to_zarr_group[k_idx]
-        image = HedwigZarrImage(
-            zarr_images.zarr_root[zarr_idx], zarr_images.ome_info, k_idx
-        )
+        image = zarr_images[k_idx]
         # single image element
         image_elt = dict()
         image_elt["imageMetadata"] = None
@@ -119,7 +123,7 @@ def generate_imageset(file_path: FilePath) -> List[Dict]:
 
         else:
             ng_asset = file_path.gen_asset(
-                asset_type="neuroglancerZarr", asset_fp=Path(zarr_fp) / zarr_idx
+                asset_type="neuroglancerZarr", asset_fp=image.path
             )
             # note - dims should be image.dims, but GUI does not want XYC
             # hardcoding in XY for now.
@@ -156,7 +160,9 @@ async def generate_czi_imageset(file_path: FilePath) -> List[Dict]:
     copy_to_assets = copy_zarr_to_assets_dir.submit(
         file_path, wait_for=[rechunk_result]
     )
-    return generate_imageset.submit(file_path, wait_for=[copy_to_assets])
+    return generate_imageset.submit(file_path,
+                                    use_default_dask=True,
+                                    wait_for=[copy_to_assets])
 
 
 @task