Caterva2: On-demand access to Blosc2 data repositories

What is it?

Caterva2 is a distributed system written in Python meant for sharing Blosc2 datasets (either native or converted on-the-fly from HDF5) among different hosts by using a publish–subscribe messaging pattern. Here, publishers categorize datasets into root groups that are announced to the broker and propagated to subscribers. Also, every subscriber exposes a REST interface that allows clients to access the datasets.

Caterva2 subscribers perform on demand data access with local caching (fit for re-publishing), which can be particularly useful for the efficient sharing of remote datasets locally, thus optimizing communication and storage resources within work groups.

Components of Caterva2

A Caterva2 deployment includes:

• One broker service to enable the communication between publishers and subscribers.
• Several publishers, each one providing subscribers with access to one root and the datasets that it contains. The root may be a native Caterva2 directory with Blosc2 and plain files, or an HDF5 file (support for other formats may be added).
• Several subscribers, each one tracking changes in multiple roots and datasets from publishers, and caching them locally for efficient reuse.
• Several clients, each one asking a subscriber to track roots and datasets, and provide access to their data and metadata.

Publishers and subscribers may be apart, in different networks with limited or expensive connectivity between them, while subscribers and clients will usually be close enough to have fast and cheap connectivity (e.g. a local network).

The Caterva2 package includes all the aforementioned components, although its main role is to provide a very simple and lightweight library to build your own Caterva2 clients.

Use with caution

Although this project is in advanced beta stage, it is not meant for production use yet. In case you are interested in Caterva2, please contact us at contact@blosc.org.

Installation

You may install Caterva2 in several ways:.

• Pre-built wheel from PyPI:

  python -m pip install caterva2
  

• Wheel built from source code:

  git clone https://github.com/ironArray/Caterva2
  cd Caterva2
  python -m build
  python -m pip install dist/caterva2-*.whl
  

• Developer setup:

  git clone https://github.com/ironArray/Caterva2
  cd Caterva2
  python -m pip install -e .
  

In any case, if you intend to run Caterva2 services, client programs, or the test suite, you need to enable the proper extra features by appending [feature1,feature2...] to the last argument of pip commands above. The following extras are supported:

• services for running all Caterva2 services (broker, publisher, subscriber)
• base-services for running the Caterva2 broker or publisher services (lighter, less dependencies)
• subscriber for running the Caterva2 subscriber service specifically (heavier, more dependencies)
• clients to use Caterva2 client programs (command-line or terminal)
• hdf5 to enable serving HDF5 files as Caterva2 roots at the publisher
• blosc2-plugins to enable extra Blosc2 features like Btune or JPEG 2000 support
• plugins to enable Web client features like the tomography display
• tools for additional utilities like cat2import and cat2export (see below)
• tests if you want to run the Caterva2 test suite

Testing

After installing with the [tests] extra, you can quickly check that the package is sane by running the test suite (that comes with the package):

python -m caterva2.tests -v

You may also run tests from source code:

cd Caterva2
python -m pytest -v

Tests will use a copy of Caterva2's root-example directory. After they finish, state files will be left under the _caterva2_tests directory for inspection (it will be re-created when tests are run again).

In case you want to run the tests with your own running daemons, you can do:

env CATERVA2_USE_EXTERNAL=1 python -m caterva2.tests -v

Neither root-example nor _caterva2_tests will be used in this case.

Quick start

(Find more detailed step-by-step tutorials in Caterva2 documentation.)

For the purpose of this quick start, let's use the datasets within the root-example folder:

cd Caterva2
ls -F root-example/

README.md  dir1/  dir2/  ds-1d-b.b2nd  ds-1d.b2nd  ds-hello.b2frame

First, create a virtual environment and install Caterva2 with the [services,clients] extras (see above). Then fire up the broker, start publishing a root named foo with root-example datasets, and create a subscriber:

cat2bro &  # broker
cat2pub foo root-example &  # publisher
cat2sub &  # subscriber

(To stop them later on, bring each one to the foreground with fg and press Ctrl+C.)

HDF5 roots

If you want to try and publish your own HDF5 file as a root, you need to include the hdf5 extra in your Caterva2 installation. Then you may just run:

cat2pub foo /path/to/your-file.h5 &

You can also get an example HDF5 file with some datasets by running:

python -m caterva2.services.hdf5root root-example.h5

You may want to test compatibility with silx' HDF5 examples (epics.h5 and grove.h5 are quite illustrative).

The command line client

Now that the services are running, we can use the cat2cli client to talk to the subscriber. In another shell, let's list all the available roots in the system:

cat2cli roots

foo

We only have the foo root that we started publishing. If other publishers were running, we would see them listed here too.

Let's ask our local subscriber to subscribe to the foo root:

cat2cli subscribe foo  # -> Ok

Now, one can list the datasets in the foo root:

cat2cli list foo

foo/README.md
...
foo/ds-hello.b2frame
...
foo/dir2/ds-4d.b2nd

Let's ask the subscriber for more info about the foo/dir2/ds-4d.b2nd dataset:

cat2cli info foo/dir2/ds-4d.b2nd

{
    'shape': [2, 3, 4, 5],
    'chunks': [1, 2, 3, 4],
    'blocks': [1, 2, 2, 2],
    'dtype': 'complex128',
    'schunk': {
        # ...
    }
}

Let's print data from a specified dataset:

cat2cli show foo/ds-hello.b2frame[:12]  # -> Hello world!

It allows printing slices instead of the whole dataset too:

cat2cli show foo/dir2/ds-4d.b2nd[1,2,3]

[115.+115.j 116.+116.j 117.+117.j 118.+118.j 119.+119.j]

Finally, we can tell the subscriber to download the dataset:

cat2cli download foo/dir2/ds-4d.b2nd

Dataset saved to foo/dir2/ds-4d.b2nd

Using a configuration file

All the services mentioned above (and clients, to some limited extent) may get their configuration from a caterva2.toml file at the current directory (or an alternative file given with the --conf option). Caterva2 source code includes a fully documented caterva2.sample.toml file (see also caterva2.toml in Caterva2 tutorials).

Experimental user authentication

The Caterva2 subscriber includes some initial and incomplete support for authenticating users. To enable it, run the subscriber with the environment variable CATERVA2_AUTH_SECRET set to some non-empty, secure string that will be used for various user management operations. After that, accessing the subscriber's Web client will only be possible after logging in with an email address and a password. New accounts may be registered, but their addresses are not verified. Password recovery does not work either.

To tell the command line client to authenticate against a subscriber, add the --username and --password options:

cat2cli --user "user@example.com" --pass "foobar" info foo/README.md

Tools

Although Caterva2 allows publishing an HDF5 file directly as a root (with datasets converted to Blosc2 arrays on-the-fly), it also includes a simple script that can import its full hierarchy to a new Caterva2 root directory. You may use it like:

cat2import existing-hdf5-file.h5 new-caterva2-root

The tool is still pretty limited in its supported input and generated output, please invoke it with --help for more information (see also cat2import in Caterva2 utilities documentation).

Caterva2 also ships a complementary tool to export a Caterva root directory to an HDF5 file; see cat2export in Caterva2 utilities documentation. You may use it like:

cat2export existing-caterva2-root new-hdf5-file.h5

That's all folks!