Fix rechunk bug after spatial query (#861)

* fix rechunk bug after spatial query

* code cleanup

pyproject.toml

@@ -127,7 +127,7 @@ exclude = [
     "docs/_build",
     "dist",
     "setup.py",
-    
+
 ]
 line-length = 120
 target-version = "py310"

src/spatialdata/_core/query/_utils.py

@@ -140,10 +140,32 @@ def _process_data_tree_query_result(query_result: DataTree) -> DataTree | None:
     d = {k: Dataset({"image": d[k]}) for k in scales_to_keep}
     result = DataTree.from_dict(d)
 
-    # Rechunk the data to avoid irregular chunks
+    # rechunk the data to avoid irregular chunks
+    coords = list(result["scale0"].coords.keys())
+    result = result.chunk({c: "auto" for c in coords})
+
+    from dask.array.core import _check_regular_chunks
+
+    # check that the rechunking into regular chunks worked
+    chunks_still_irregular = False
     for scale in result:
-        result[scale]["image"] = result[scale]["image"].chunk("auto")
+        data = result[scale]["image"].data
+        chunks_still_irregular = chunks_still_irregular or not _check_regular_chunks(data.chunks)
+
+    if chunks_still_irregular:
+        # reported here: https://github.com/scverse/spatialdata/issues/821#issuecomment-2632201695
+        # seemingly due to this bug: https://github.com/dask/dask/issues/11713
+        CHUNK_SIZE = 1024
+        rechunk_strategy = {c: CHUNK_SIZE for c in coords}
+        if "c" in coords:
+            rechunk_strategy["c"] = result["scale0"]["image"].chunks[0][0]
+        result = result.chunk(rechunk_strategy)
 
+    for scale in result:
+        data = result[scale]["image"].data
+        assert _check_regular_chunks(data.chunks), (
+            f"Chunks are not regular for the {scale} of the queried data: {data.chunks}. Please report this bug."
+        )
     return result
 
 

tests/io/test_pyramids_performance.py

@@ -82,5 +82,5 @@ def test_write_image_multiscale_performance(sdata_with_image: SpatialData, tmp_p
 
     actual_num_chunk_writes = zarr_chunk_write_spy.call_count
     actual_num_chunk_reads = zarr_chunk_read_spy.call_count
-    assert actual_num_chunk_writes == num_chunks_all_scales
-    assert actual_num_chunk_reads == num_chunks_scale0
+    assert actual_num_chunk_writes == num_chunks_all_scales.item()
+    assert actual_num_chunk_reads == num_chunks_scale0.item()