hugedict 2.13.1

A dictionary-like object that is friendly with multiprocessing and uses key-value databases (e.g., RocksDB) as the underlying storage.

hugedict

hugedict provides a drop-in replacement for dictionary objects that are too big to fit in memory. hugedict's dictionary-like objects implement typing.Mapping and typing.MutableMapping interfaces using key-value databases (e.g., RocksDB) as the underlying storage. Moreover, they are friendly with Python's multiprocessing.

• Documentation

Installation

From PyPI (using pre-built binaries):

pip install hugedict

To compile the source, run: maturin build -r inside the project directory. You need Rust, Maturin, CMake and CLang (to build Rust-RocksDB).

Features

1. Create a mutable mapping backed by RocksDB

from functools import partial
from hugedict.prelude import RocksDBDict, RocksDBOptions

# replace [str, str] for the types of keys and values you want
# as well as deser_key, deser_value, ser_value
mapping: MutableMapping[str, str] = RocksDBDict(
    path=dbpath,  # path (str) to db file
    options=RocksDBOptions(create_if_missing=create_if_missing),  # whether to create database if missing, check other options
    deser_key=partial(str, encoding="utf-8"),  # decode the key from memoryview
    deser_value=partial(str, encoding="utf-8"),  # decode the value from memoryview
    ser_value=str.encode,  # encode the value to bytes
    readonly=False,  # open database in read only mode
    secondary_mode=False,  # open database in secondary mode
    secondary_path=None,  # when secondary_mode is True, it's a string pointing to a directory for storing data required to operate in secondary mode
)

2. Load huge data from files into RocksDB in parallel: from hugedict.prelude import rocksdb_load. This function creates SST files in parallel, ingests into the db and (optionally) compacts them.

3. Cache a function when doing parallel processing

from hugedict.prelude import Parallel

pp = Parallel()

@pp.cache_func("/tmp/test.db")
def heavy_computing(seconds: float):
    time.sleep(seconds)
    return seconds * 2


output = pp.map(heavy_computing, [0.5, 1, 0.7, 0.3, 0.6], n_processes=3)

4. Create dictionary backed by Sqlite: hugedict.sqlite.SqliteDict
5. Chain multiple dictionaries: hugedict.chained_mapping.ChainedMapping
6. Cache a dictionary so previously accessed keys are stored in memory: hugedict.cachedict.CacheDict or called hugedict.types.HugeMapping.cache