From dab40c88dd5724311c8f48bfb40eb252f231ecd9 Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?Taus=20M=C3=B8ller?= <teamo@dtu.dk>
Date: Mon, 20 Jul 2026 12:29:58 +0200
Subject: [PATCH 1/2] Add unit test for sharded zarr roundtrip

---
 xarray/tests/test_backends_shards.py | 66 ++++++++++++++++++++++++++++
 1 file changed, 66 insertions(+)
 create mode 100644 xarray/tests/test_backends_shards.py

diff --git a/xarray/tests/test_backends_shards.py b/xarray/tests/test_backends_shards.py
new file mode 100644
index 00000000000..5ea780e8392
--- /dev/null
+++ b/xarray/tests/test_backends_shards.py
@@ -0,0 +1,66 @@
+import numpy as np
+import pytest
+
+import xarray as xr
+from xarray.tests import requires_dask
+
+
+@requires_dask
+@pytest.mark.parametrize(
+    "shape, enc_chunks, enc_shards",
+    [
+        (
+            (10, 10),
+            (2, 2),
+            (4, 4),
+        ),
+        (
+            (100, 10),
+            (2, 2),
+            (4, 4),
+        ),
+        (
+            (100, 10, 1),
+            (2, 2, 1),
+            (4, 4, 1),
+        ),
+    ],
+)
+def test_zarr_shard_roundtrip(shape, enc_chunks, enc_shards, tmp_path):
+
+    dimensions = {f"dim_{i}" for i in range(len(shape))}
+    sizes = dict(zip(dimensions, shape, strict=True))
+
+    data_values = np.arange(np.prod(shape)).reshape(shape)
+    coords = {dim: np.arange(sizes[dim]) for dim in dimensions}
+
+    chunks = dict(zip(dimensions, enc_chunks, strict=True))
+    shards = dict(zip(dimensions, enc_shards, strict=True))
+
+    ds = xr.Dataset(
+        {
+            "data": (dimensions, data_values),
+        },
+        coords=coords,
+    )
+
+    # Make the xarray chunks match the zarr shards
+    ds_sharded = ds.chunk(shards)
+    encoding = {
+        dim: {"chunks": (chunks[dim],), "shards": (shards[dim],)} for dim in dimensions
+    }
+    encoding["data"] = {"chunks": enc_shards, "shards": enc_shards}
+
+    store = tmp_path / "test.zarr"
+    ds_sharded.to_zarr(
+        store,
+        mode="w",
+        encoding=encoding,
+        zarr_format=3,
+        consolidated=False,
+    )
+
+    # reopen dataset and store it back
+    reopend_ds = xr.open_zarr(store, consolidated=False)
+    reopend_ds.to_zarr(store, mode="a")
+    assert reopend_ds.chunks == ds_sharded.chunks

From 97fcc89b701f9d9665d13ae58eecb56a34c63e34 Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?Taus=20M=C3=B8ller?= <teamo@dtu.dk>
Date: Mon, 20 Jul 2026 14:48:27 +0200
Subject: [PATCH 2/2] Use shards as prefered chunks if present

---
 xarray/backends/zarr.py | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/xarray/backends/zarr.py b/xarray/backends/zarr.py
index 6f3e1ad4eb4..d425ef28861 100644
--- a/xarray/backends/zarr.py
+++ b/xarray/backends/zarr.py
@@ -491,7 +491,6 @@ def extract_zarr_variable_encoding(
     """
 
     encoding = variable.encoding.copy()
-
     safe_to_drop = {"source", "original_shape", "preferred_chunks"}
     valid_encodings = {
         "chunks",
@@ -909,10 +908,11 @@ def open_store_variable(self, name):
             zarr_array, DIMENSION_KEY, try_nczarr
         )
         attributes = dict(attributes)
+        preferred_chunks = zarr_array.shards or zarr_array.chunks
 
         encoding = {
             "chunks": zarr_array.chunks,
-            "preferred_chunks": dict(zip(dimensions, zarr_array.chunks, strict=True)),
+            "preferred_chunks": dict(zip(dimensions, preferred_chunks, strict=True)),
         }
 
         if _zarr_v3():