Streaming add remfile

CHANGELOG.md

@@ -16,6 +16,9 @@
 - Fix bug where namespaces were loaded in "w-" mode. @h-mayorquin [#1795](https://github.com/NeurodataWithoutBorders/pynwb/pull/1795)
 - Fix bug where pynwb version was reported as "unknown" to readthedocs @stephprince [#1810](https://github.com/NeurodataWithoutBorders/pynwb/pull/1810)
 
+### Documentation and tutorial enhancements
+- Add RemFile to streaming tutorial @bendichter [#1761](https://github.com/NeurodataWithoutBorders/pynwb/pull/1761)
+
 ## PyNWB 2.5.0 (August 18, 2023)
 
 ### Enhancements and minor changes

docs/gallery/advanced_io/streaming.py

@@ -90,6 +90,9 @@
 # `fsspec documentation on known implementations <https://filesystem-spec.readthedocs.io/en/latest/api.html?highlight=S3#other-known-implementations>`_
 # for a full updated list of supported store formats.
 #
+# One downside of this fsspec method is that fsspec is not optimized for reading HDF5 files, and so streaming data
+# using this method can be slow. A faster alternative is ``remfile`` described below.
+#
 # Streaming Method 2: ROS3
 # ------------------------
 # ROS3 stands for "read only S3" and is a driver created by the HDF5 Group that allows HDF5 to read HDF5 files stored
@@ -120,15 +123,43 @@
 #
 #        pip uninstall h5py
 #        conda install -c conda-forge "h5py>=3.2"
+#
+# Besides the extra burden of installing h5py from a non-PyPI source, one downside of this ROS3 method is that
+# this method does not support automatic retries in case the connection fails.
+
+
+##################################################
+# Method 3: remfile
+# -----------------
+# ``remfile`` is another library that enables indexing and streaming of files in s3. remfile is simple, fast, and
+# allows for caching of data in the local filesystem. The caveats of ``remfile`` are that it is a very new project
+# that has not been tested in a variety of use-cases and caching options are limited compared to ``fsspec``.
+# You can install ``remfile`` with pip:
+#
+# .. code-block:: bash
+#
+#   pip install remfile
+#
+
+import h5py
+from pynwb import NWBHDF5IO
+import remfile
+
+rem_file = remfile.File(s3_url)
 
+with h5py.File(rem_file, "r") as h5py_file:
+    with NWBHDF5IO(file=h5py_file, load_namespaces=True) as io:
+        nwbfile = io.read()
+        print(nwbfile.acquisition["lick_times"].time_series["lick_left_times"].data[:])
 
 ##################################################
 # Which streaming method to choose?
 # ---------------------------------
 #
 # From a user perspective, once opened, the :py:class:`~pynwb.file.NWBFile` works the same with
-# both fsspec and ros3.  However, in general, we currently recommend using fsspec for streaming
-# NWB files because it is more performant and reliable than ros3. In particular fsspec:
+# fsspec, ros3, or remfile.  However, in general, we currently recommend using fsspec for streaming
+# NWB files because it is more performant and reliable than ros3 and more widely tested than remfile.
+# In particular, fsspec:
 #
 # 1. supports caching, which will dramatically speed up repeated requests for the
 #    same region of data,

environment-ros3.yml

@@ -16,3 +16,6 @@ dependencies:
   - fsspec==2023.6.0
   - requests==2.28.1
   - aiohttp==3.8.3
+  - pip
+  - pip:
+    - remfile==0.1.9