Now the upsert is also going to call the store if it is necessary

setup.py

@@ -5,7 +5,7 @@
 
 setup(
     name='TensorDB',
-    version='0.25',
+    version='0.26',
     description='Database based in a file system storage combined with Xarray and Zarr',
     author='Joseph Nowak',
     author_email='josephgonowak97@gmail.com',

tensordb/storages/zarr_storage.py

@@ -297,7 +297,7 @@ def update(
             new_data: Union[xr.DataArray, xr.Dataset],
             compute: bool = True,
             complete_update_dims: Union[List[str], str] = None,
-    ) -> xr.backends.ZarrStore:
+    ) -> Union[xr.backends.ZarrStore, None]:
         """
         Replace data on an existing Zarr files based on the new_data, internally calls the method
         `to_zarr <https://xr.pydata.org/en/stable/generated/xr.Dataset.to_zarr.html>`_ using the
@@ -339,6 +339,8 @@ def update(
             k: new_data.coords[k].isin(v)
             for k, v in act_coords.items()
         })
+        if any(size == 0 for size in new_data.sizes.values()):
+            return None
 
         if complete_update_dims is not None:
             if isinstance(complete_update_dims, str):
@@ -383,12 +385,18 @@ def upsert(
         A list of xr.backends.ZarrStore produced by the append and update methods
 
         """
+        if not self.exist():
+            return [
+                self.store(new_data, compute=compute)
+            ]
+
         delayed_writes = [
             self.update(new_data, compute=compute, complete_update_dims=complete_update_dims)
         ]
         delayed_writes.extend(
             self.append(new_data, compute=compute)
         )
+        delayed_writes = [write for write in delayed_writes if write is not None]
         return delayed_writes
 
     def drop(

tensordb/tests/test_algorithms.py

@@ -139,34 +139,35 @@ def test_bitmask_topk_tie_breaker(top_size, dim):
         coords={'c': list(range(2)), 'a': list(range(6)), 'b': list(range(3))}
     ).chunk((1, 3, 1))
 
-    df = pd.DataFrame({
-        "data1": arr.isel(c=0).to_series(),
-        "data2": arr.isel(c=1).to_series()
-    }).reset_index()
-
-    group_dim = "b" if dim == "a" else "a"
-
-    ranked = df.groupby(group_dim, group_keys=False).apply(
-        lambda x: x[["data1", "data2"]].fillna(-np.inf).apply(
-            tuple, axis=1
-        ).rank(
-            ascending=False, method="min"
-        ).astype(int)
-    )
-    df["rank"] = ranked
-    df.loc[df["data1"].isna(), "rank"] = np.nan
-
-    expected = df.pivot_table(
-        index="a", columns="b", values="rank", dropna=False
-    ) <= top_size
-
     result = Algorithms.bitmask_topk(
         arr,
         dim=dim,
         tie_breaker_dim="c",
         top_size=top_size
     ).isel(c=0, drop=True)
 
+    total_dfs = [
+        arr.sel(c=index, drop=True).to_pandas()
+        for index in arr.c.values
+    ]
+    df = (
+        pd.concat(total_dfs)
+        .fillna(-np.inf)
+        .stack(dropna=False)
+        .groupby(level=[0, 1])
+        .apply(tuple)
+        .unstack()
+    )
+
+    df = df.where(~total_dfs[0].isna())
+
+    expected = df.rank(
+        # Subtract 1 due that the C dim is at the first position
+        arr.dims.index(dim) - 1,
+        method="min",
+        ascending=False
+    ) <= top_size
+
     assert result.equals(
         xr.DataArray(
             expected.values,

tensordb/tests/test_tensor_client.py

@@ -117,6 +117,23 @@ def test_update(self, use_local):
         tensor_client.update(new_data=self.arr2, path='data_one')
         assert tensor_client.read(path='data_one').equals(self.arr2)
 
+        assert tensor_client.update(
+            new_data=self.arr2.reindex(index=[100]),
+            path='data_one'
+        ) is None
+
+    @pytest.mark.parametrize('use_local', [True, False])
+    def test_upsert(self, use_local):
+        tensor_client = self.local_tensor_client if use_local else self.tensor_client
+        tensor_client.create_tensor({"path": "data_upsert"})
+        # Test creating from strach the tensor, this should call the store method
+        tensor_client.upsert(new_data=self.arr2, path="data_upsert")
+        assert tensor_client.read(path='data_upsert').equals(self.arr2)
+
+        # This method should call the append and the update method
+        tensor_client.upsert(new_data=self.arr3, path="data_upsert")
+        assert tensor_client.read(path='data_upsert').equals(self.arr3)
+
     @pytest.mark.parametrize('use_local', [True, False])
     def test_drop(self, use_local):
         tensor_client = self.local_tensor_client if use_local else self.tensor_client