Fix-uniqueness-of-hdf5-based-dask-arrays

satpy/readers/hdf5_utils.py

@@ -17,7 +17,9 @@
 # satpy.  If not, see <http://www.gnu.org/licenses/>.
 """Helpers for reading hdf5-based files."""
 
+import hashlib
 import logging
+import os
 
 import dask.array as da
 import h5py
@@ -102,7 +104,7 @@ def __getitem__(self, key):
             # these datasets are closed and inaccessible when the file is closed, need to reopen
             f_obj = open_file_or_filename(self.filename)
             dset = h5py.File(f_obj, "r")[key]
-            dset_data = da.from_array(dset, chunks=CHUNK_SIZE)
+            dset_data = from_h5_array(dset)
             attrs = self._attrs_cache.get(key, dset.attrs)
             if dset.ndim == 2:
                 return xr.DataArray(dset_data, dims=["y", "x"], attrs=attrs)
@@ -120,3 +122,11 @@ def get(self, item, default=None):
             return self[item]
         else:
             return default
+
+
+def from_h5_array(h5dset):
+    """Create a dask array from an h5py dataset, ensuring uniqueness of the dask array name."""
+    name_str = os.fspath(h5dset.file.filename) + "-" + h5dset.name
+    name = hashlib.md5(name_str.encode(), usedforsecurity=False).hexdigest()
+    dset_data = da.from_array(h5dset, chunks=CHUNK_SIZE, name=name)
+    return dset_data

satpy/tests/reader_tests/test_hdf5_utils.py

@@ -150,3 +150,12 @@ def test_all_basic(self):
         assert "fake_ds" not in file_handler
 
         assert isinstance(file_handler["ds2_f/attr/test_ref"], np.ndarray)
+
+    def test_array_name_uniqueness(self):
+        """Test the dask array generated from an hdf5 dataset stay constant and unique."""
+        from satpy.readers.hdf5_utils import HDF5FileHandler
+        file_handler = HDF5FileHandler("test.h5", {}, {})
+
+        dsname = "test_group/ds1_f"
+
+        assert file_handler[dsname].data.name == file_handler[dsname].data.name