implement Zarr v3 spec support

.github/workflows/upstream-dev-ci.yaml

@@ -80,6 +80,7 @@ jobs:
         if: success()
         id: status
         run: |
+          export ZARR_V3_EXPERIMENTAL_API=1
           python -m pytest --timeout=60 -rf \
             --report-log output-${{ matrix.python-version }}-log.jsonl
       - name: Generate and publish the report

doc/whats-new.rst

@@ -14,7 +14,6 @@ What's New
 
     np.random.seed(123456)
 
-
 .. _whats-new.2022.11.1:
 
 v2022.11.1 (unreleased)
@@ -23,6 +22,8 @@ v2022.11.1 (unreleased)
 New Features
 ~~~~~~~~~~~~
 
+- Add experimental support for Zarr's in-progress V3 specification. (:pull:`6475`).
+  By `Gregory Lee  <https://github.com/grlee77>`_ and `Joe Hamman <https://github.com/jhamman>`_.
 
 Breaking changes
 ~~~~~~~~~~~~~~~~

xarray/backends/api.py

@@ -1503,6 +1503,7 @@ def to_zarr(
     region: Mapping[str, slice] | None = None,
     safe_chunks: bool = True,
     storage_options: dict[str, str] | None = None,
+    zarr_version: int | None = None,
 ) -> backends.ZarrStore:
     ...
 
@@ -1524,6 +1525,7 @@ def to_zarr(
     region: Mapping[str, slice] | None = None,
     safe_chunks: bool = True,
     storage_options: dict[str, str] | None = None,
+    zarr_version: int | None = None,
 ) -> Delayed:
     ...
 
@@ -1542,6 +1544,7 @@ def to_zarr(
     region: Mapping[str, slice] | None = None,
     safe_chunks: bool = True,
     storage_options: dict[str, str] | None = None,
+    zarr_version: int | None = None,
 ) -> backends.ZarrStore | Delayed:
     """This function creates an appropriate datastore for writing a dataset to
     a zarr ztore
@@ -1608,6 +1611,13 @@ def to_zarr(
                 f"``region`` with to_zarr(), got {append_dim} in both"
             )
 
+    if zarr_version is None:
+        # default to 2 if store doesn't specify it's version (e.g. a path)
+        zarr_version = int(getattr(store, "_store_version", 2))
+
+    if consolidated is None and zarr_version > 2:
+        consolidated = False
+
     if mode == "r+":
         already_consolidated = consolidated
         consolidate_on_close = False
@@ -1626,6 +1636,7 @@ def to_zarr(
         write_region=region,
         safe_chunks=safe_chunks,
         stacklevel=4,  # for Dataset.to_zarr()
+        zarr_version=zarr_version,
     )
 
     if mode in ["a", "r+"]:

xarray/backends/zarr.py

@@ -353,19 +353,37 @@ def open_group(
         write_region=None,
         safe_chunks=True,
         stacklevel=2,
+        zarr_version=None,
     ):
         import zarr
 
         # zarr doesn't support pathlib.Path objects yet. zarr-python#601
         if isinstance(store, os.PathLike):
             store = os.fspath(store)
 
+        if zarr_version is None:
+            # default to 2 if store doesn't specify it's version (e.g. a path)
+            zarr_version = getattr(store, "_store_version", 2)
+
         open_kwargs = dict(
             mode=mode,
             synchronizer=synchronizer,
             path=group,
         )
         open_kwargs["storage_options"] = storage_options
+        if zarr_version > 2:
+            open_kwargs["zarr_version"] = zarr_version
+
+            if consolidated or consolidate_on_close:
+                raise ValueError(
+                    "consolidated metadata has not been implemented for zarr "
+                    f"version {zarr_version} yet. Set consolidated=False for "
+                    f"zarr version {zarr_version}. See also "
+                    "https://github.com/zarr-developers/zarr-specs/issues/136"
+                )
+
+            if consolidated is None:
+                consolidated = False
 
         if chunk_store:
             open_kwargs["chunk_store"] = chunk_store
@@ -440,6 +458,11 @@ def open_store_variable(self, name, zarr_array):
             zarr_array, DIMENSION_KEY, try_nczarr
         )
         attributes = dict(attributes)
+
+        # TODO: this should not be needed once
+        # https://github.com/zarr-developers/zarr-python/issues/1269 is resolved.
+        attributes.pop("filters", None)
+
         encoding = {
             "chunks": zarr_array.chunks,
             "preferred_chunks": dict(zip(dimensions, zarr_array.chunks)),
@@ -668,6 +691,7 @@ def open_zarr(
     storage_options=None,
     decode_timedelta=None,
     use_cftime=None,
+    zarr_version=None,
     **kwargs,
 ):
     """Load and decode a dataset from a Zarr store.
@@ -725,6 +749,9 @@ def open_zarr(
         capability. Only works for stores that have already been consolidated.
         By default (`consolidate=None`), attempts to read consolidated metadata,
         falling back to read non-consolidated metadata if that fails.
+
+        When the experimental ``zarr_version=3``, ``consolidated`` must be
+        either be ``None`` or ``False``.
     chunk_store : MutableMapping, optional
         A separate Zarr store only for chunk data.
     storage_options : dict, optional
@@ -745,6 +772,10 @@ def open_zarr(
         represented using ``np.datetime64[ns]`` objects.  If False, always
         decode times to ``np.datetime64[ns]`` objects; if this is not possible
         raise an error.
+    zarr_version : int or None, optional
+        The desired zarr spec version to target (currently 2 or 3). The default
+        of None will attempt to determine the zarr version from ``store`` when
+        possible, otherwise defaulting to 2.
 
     Returns
     -------
@@ -782,6 +813,7 @@ def open_zarr(
         "chunk_store": chunk_store,
         "storage_options": storage_options,
         "stacklevel": 4,
+        "zarr_version": zarr_version,
     }
 
     ds = open_dataset(
@@ -798,6 +830,7 @@ def open_zarr(
         backend_kwargs=backend_kwargs,
         decode_timedelta=decode_timedelta,
         use_cftime=use_cftime,
+        zarr_version=zarr_version,
     )
     return ds
 
@@ -842,6 +875,7 @@ def open_dataset(
         chunk_store=None,
         storage_options=None,
         stacklevel=3,
+        zarr_version=None,
     ):
 
         filename_or_obj = _normalize_path(filename_or_obj)
@@ -855,6 +889,7 @@ def open_dataset(
             chunk_store=chunk_store,
             storage_options=storage_options,
             stacklevel=stacklevel + 1,
+            zarr_version=zarr_version,
         )
 
         store_entrypoint = StoreBackendEntrypoint()

xarray/core/dataset.py

@@ -1930,6 +1930,7 @@ def to_zarr(
         region: Mapping[str, slice] | None = None,
         safe_chunks: bool = True,
         storage_options: dict[str, str] | None = None,
+        zarr_version: int | None = None,
     ) -> ZarrStore:
         ...
 
@@ -1967,6 +1968,7 @@ def to_zarr(
         region: Mapping[str, slice] | None = None,
         safe_chunks: bool = True,
         storage_options: dict[str, str] | None = None,
+        zarr_version: int | None = None,
     ) -> ZarrStore | Delayed:
         """Write dataset contents to a zarr group.
 
@@ -2017,6 +2019,9 @@ def to_zarr(
             metadata; if False, do not. The default (`consolidated=None`) means
             write consolidated metadata and attempt to read consolidated
             metadata for existing stores (falling back to non-consolidated).
+
+            When the experimental ``zarr_version=3``, ``consolidated`` must be
+            either be ``None`` or ``False``.
         append_dim : hashable, optional
             If set, the dimension along which the data will be appended. All
             other dimensions on overridden variables must remain the same size.
@@ -2048,6 +2053,10 @@ def to_zarr(
         storage_options : dict, optional
             Any additional parameters for the storage backend (ignored for local
             paths).
+        zarr_version : int or None, optional
+            The desired zarr spec version to target (currently 2 or 3). The
+            default of None will attempt to determine the zarr version from
+            ``store`` when possible, otherwise defaulting to 2.
 
         Returns
         -------
@@ -2092,6 +2101,7 @@ def to_zarr(
             append_dim=append_dim,
             region=region,
             safe_chunks=safe_chunks,
+            zarr_version=zarr_version,
         )
 
     def __repr__(self) -> str: