Extendable zarr arrays (#802)

* add parameter to from_uri

* growable zarr array

* pass max_depth to server

* revert max_shaape

* checkpoint

* checkpoint

* Add endpoint link for append

* fix errant code in write_block

* Add append link to arrays, client sends shape and axis

* zarr adapter is called for write_block

* fix growth of zarr array

* update database for size in array append_block

* precommit cleanup

* fix test

* fix database issue

* add client call docstring

* WIP: Move type_aliases to root and fix misspelled name

* WIP: Rework to use PATCH /array/full

* Appending works

* Rename 'grow' to 'extend'.

* Raise if assumptions are not met

* Improve usability. Test.

* Update imports after rebase.

* Use Python 3.9 compatibile typing.

* Fix regression in refresh

* Reference documentation

* Add example to docstring.

* Test overwrite and out-of-order updates.

* Change from 'slice' to 'offset'.

* update array test

* add extend array to writing tutorial

* whitespace

* Finesse docs

* Data type of patch must match.

* Update CHANGELOG

---------

Co-authored-by: Dan Allan <[email protected]>

Author: dylanmcreynolds

CHANGELOG.md

@@ -3,6 +3,14 @@ Write the date in place of the "Unreleased" in the case a new version is release
 
 # Changelog
 
+## Unreleased
+
+### Added
+
+- Add HTTP endpoint `PATCH /array/full/{path}` to enable updating and
+  optionally _extending_ an existing array.
+- Add associated Python client method `ArrayClient.patch`.
+
 ## v0.1.0b11 (2024-11-14)
 
 ### Added

docs/source/reference/python-client.md

@@ -165,6 +165,7 @@ Tiled currently includes two clients for each structure family:
    tiled.client.array.DaskArrayClient.export
    tiled.client.array.DaskArrayClient.write
    tiled.client.array.DaskArrayClient.write_block
+   tiled.client.array.DaskArrayClient.patch
 ```
 
 ```{eval-rst}

docs/source/tutorials/writing.md

@@ -1,12 +1,5 @@
 # Writing Data
 
-```{warning}
-
-This is a highly experimental feature, recently introduced and included for
-evaluation by early users. At this time we do not recommend using it for
-anything important.
-```
-
 In this tutorial we will start Tiled in a mode where the client can
 write (upload) data for later retrieving, search, or sharing.
 
@@ -48,21 +41,38 @@ where `...` will be whatever secret was printed at server startup above.
 
 ## Write data
 
-Write array and tabular data.
+Write array.
 
 ```python
-# Write simple Python list (which gets converted to numpy.array).
->>> client.write_array([1, 2, 3], metadata={"color": "red", "barcode": 10})
-<ArrayClient shape=(3,) chunks=((3,),) dtype=int64>
-
 # Write an array.
 >>> import numpy
 >>> client.write_array(numpy.array([4, 5, 6]), metadata={"color": "blue", "barcode": 11})
 <ArrayClient shape=(3,) chunks=((3,),) dtype=int64>
 
-# Write a table (DataFrame).
+# Write a Python list (which gets converted to numpy array).
+>>> client.write_array([1, 2, 3], metadata={"color": "red", "barcode": 10})
+<ArrayClient shape=(3,) chunks=((3,),) dtype=int64>
+
+# Create an array and grow it by one.
+>>> new_array = client.write_array([1, 2, 3])
+>>> new_array
+<ArrayClient shape=(3,) chunks=((3,),) dtype=int64>
+
+# Extend the array. This array has only one dimension, here we extend by one
+# along that dimension.
+>>> new_array.patch([4], offset=(3,), extend=True)
+>>> new_array
+<ArrayClient shape=(4,) chunks=((3, 1),) dtype=int64>
+>>> new_array.read()
+array([1, 2, 3, 4])
+```
+
+Write tabular data in a pandas DataFrame.
+
+```python
 >>> import pandas
->>> client.write_dataframe(pandas.DataFrame({"x": [1, 2, 3], "y": [4, 5, 6]}), metadata={"color": "green", "barcode": 12})
+>>> df = pandas.DataFrame({"x": [1, 2, 3], "y": [4, 5, 6]})
+>>> client.write_dataframe(df, metadata={"color": "green", "barcode": 12})
 <DataFrameClient ['x', 'y']>
 ```
 

tiled/_tests/test_protocols.py

@@ -18,13 +18,13 @@
     SparseAdapter,
     TableAdapter,
 )
-from tiled.adapters.type_alliases import JSON, Filters, NDSlice, Scopes
 from tiled.server.schemas import Principal, PrincipalType
 from tiled.structures.array import ArrayStructure, BuiltinDtype
 from tiled.structures.awkward import AwkwardStructure
 from tiled.structures.core import Spec, StructureFamily
 from tiled.structures.sparse import COOStructure
 from tiled.structures.table import TableStructure
+from tiled.type_aliases import JSON, Filters, NDSlice, Scopes
 
 
 class CustomArrayAdapter:

tiled/_tests/test_writing.py

@@ -121,6 +121,59 @@ def test_write_array_chunked(tree):
         assert result.specs == specs
 
 
+def test_extend_array(tree):
+    "Extend an array with additional data, expanding its shape."
+    with Context.from_app(
+        build_app(tree, validation_registry=validation_registry)
+    ) as context:
+        client = from_context(context)
+
+        a = numpy.ones((3, 2, 2))
+        new_data = numpy.ones((1, 2, 2)) * 2
+        full_array = numpy.concatenate((a, new_data), axis=0)
+
+        # Upload a (3, 2, 2) array.
+        ac = client.write_array(a)
+        assert ac.shape == a.shape
+
+        # Patching data into a region beyond the current extent of the array
+        # raises a ValueError (catching a 409 from the server).
+        with pytest.raises(ValueError):
+            ac.patch(new_data, offset=(3,))
+        # With extend=True, the array is expanded.
+        ac.patch(new_data, offset=(3,), extend=True)
+        # The local cache of the structure is updated.
+        assert ac.shape == full_array.shape
+        actual = ac.read()
+        # The array has the expected shape and data.
+        assert actual.shape == full_array.shape
+        numpy.testing.assert_equal(actual, full_array)
+
+        # Overwrite data (do not extend).
+        revised_data = numpy.ones((1, 2, 2)) * 3
+        revised_array = full_array.copy()
+        revised_array[3, :, :] = 3
+        ac.patch(revised_data, offset=(3,))
+        numpy.testing.assert_equal(ac.read(), revised_array)
+
+        # Extend out of order.
+        ones = numpy.ones((1, 2, 2))
+        ac.patch(ones * 7, offset=(7,), extend=True)
+        ac.patch(ones * 5, offset=(5,), extend=True)
+        ac.patch(ones * 6, offset=(6,), extend=True)
+        numpy.testing.assert_equal(ac[5:6], ones * 5)
+        numpy.testing.assert_equal(ac[6:7], ones * 6)
+        numpy.testing.assert_equal(ac[7:8], ones * 7)
+
+        # Offset given as an int is acceptable.
+        ac.patch(ones * 8, offset=8, extend=True)
+        numpy.testing.assert_equal(ac[8:9], ones * 8)
+
+        # Data type must match.
+        with pytest.raises(ValueError):
+            ac.patch(ones.astype("uint8"), offset=9, extend=True)
+
+
 def test_write_dataframe_full(tree):
     with Context.from_app(
         build_app(tree, validation_registry=validation_registry)

tiled/adapters/array.py

@@ -5,8 +5,8 @@
 
 from ..structures.array import ArrayStructure
 from ..structures.core import Spec, StructureFamily
+from ..type_aliases import JSON, NDSlice
 from .protocols import AccessPolicy
-from .type_alliases import JSON, NDSlice
 
 
 class ArrayAdapter:

tiled/adapters/arrow.py

@@ -9,10 +9,10 @@
 from ..structures.core import Spec, StructureFamily
 from ..structures.data_source import Asset, DataSource, Management
 from ..structures.table import TableStructure
+from ..type_aliases import JSON
 from ..utils import ensure_uri, path_from_uri
 from .array import ArrayAdapter
 from .protocols import AccessPolicy
-from .type_alliases import JSON
 
 
 class ArrowAdapter:

tiled/adapters/awkward.py

@@ -6,9 +6,9 @@
 
 from ..structures.awkward import AwkwardStructure
 from ..structures.core import Spec, StructureFamily
+from ..type_aliases import JSON
 from .awkward_directory_container import DirectoryContainer
 from .protocols import AccessPolicy
-from .type_alliases import JSON
 
 
 class AwkwardAdapter:

tiled/adapters/awkward_buffers.py

@@ -9,11 +9,11 @@
 from ..server.schemas import Asset
 from ..structures.awkward import AwkwardStructure
 from ..structures.core import Spec, StructureFamily
+from ..type_aliases import JSON
 from ..utils import path_from_uri
 from .awkward import AwkwardAdapter
 from .awkward_directory_container import DirectoryContainer
 from .protocols import AccessPolicy
-from .type_alliases import JSON
 
 
 class AwkwardBuffersAdapter(AwkwardAdapter):

tiled/adapters/csv.py

@@ -7,12 +7,12 @@
 from ..structures.core import Spec, StructureFamily
 from ..structures.data_source import Asset, DataSource, Management
 from ..structures.table import TableStructure
+from ..type_aliases import JSON
 from ..utils import ensure_uri, path_from_uri
 from .array import ArrayAdapter
 from .dataframe import DataFrameAdapter
 from .protocols import AccessPolicy
 from .table import TableAdapter
-from .type_alliases import JSON
 
 
 def read_csv(