partd 0.1.0

Concurrent appendable key-value storage

Minimal key-value byte storage with appendable values

Partd stores data. Partd arranges data in key-value pairs where the values are just raw bytes. We can append data onto the end of existing values. You can get the entire value associated to a list of keys.

API

Partd supports four operations

1. create: Create a partd dataset:

   >>> create('/path/to/partd-file/')

2. put: Append key-byte pairs to dataset:

   >>> put('/path/to/partd-file/', {'x': b'Hello ', 'y': b'123'})
   >>> put('/path/to/partd-file/', {'x': b'world!', 'y': b'456'})

3. get: Get all bytes associated to a set of keys:

   >>> get('/path/to/partd-file/', ['y', 'x'])
   [b'123456', b'Hello world!']

4. destroy: Destroy partd dataset:

   >>> destroy('/path/to/partd-file')

That’s it.

There is no in-memory state.

Implementations

The reference implementation uses file-based locks. This works surprisingly well as long as you don’t do many small writes

If you do many small writes then you probably want to cache in memory; this is hard to do in parallel while also maintaining consistency. For this we have a centralized server (see partd.zmq) that caches data in memory and writes only large chunks to disk when necessary.

Encodings

partd.pickle and partd.numpy handle encoding/decoding pickled lists and numpy arrays with the same partd get/put interface. They compose well with the zmq solution.

Future plans

Things that will happen if our plan is perfect (it isn’t!)

1. These functions can be implemented in other, non-file-based ways, perhaps with a distributed system that peer-to-peer communication and stuff.

2. We’ll build interfaces around partd that handle dataframes, numpy arrays, pickled lists, etc..

3. Compression