A python key-value file database
Project description
Introduction
Booklet is a pure python key-value file database. It allows for multiple serializers for both the keys and values. Booklet uses the MutableMapping class API which is the same as python’s dictionary in addition to some dbm methods (i.e. sync and prune). It is thread-safe (using thread locks on writes) and multiprocessing-safe (using file locks).
Deletes do not remove data from the file directly. Similarly, reassigning a value to an existing key adds a new key/value set to the file. During normal usage, the user will not notice a difference when requesting a key/value set, but the file size will grow. If size becomes an issue because of lots of deletes or reassignments, then the user should run the “prune” method to remove old values.
When an error occurs and is caught by the module (e.g. trying to access a key that doesn’t exist), booklet will properly close the file. This means it will sync any changes and “unlock” the file. There will be errors that can occur that are not caught and in these circumstances there are no guarantees for what happens to the file.
Installation
Install via pip:
pip install booklet
Or conda:
conda install -c mullenkamp booklet
I’ll probably put it on conda-forge once I feel like it’s up to an appropriate standard…
Serialization
Both the keys and values stored in Booklet must be bytes when written to disk. This is the default when “open” is called. Booklet allows for various serializers to be used for taking input keys and values and converting them to bytes. There are many in-built serializers. Check the booklet.available_serializers list for what’s available. Some serializers require additional packages to be installed (e.g. orjson, zstd, etc). If you want to serialize to json, then it is highly recommended to use orjson or msgpack as they are substantially faster than the standard json python module. If in-built serializers are assigned at initial file creation, then they will be saved on future reading and writing on the same file (i.e. they don’t need to be passed after the first time). Setting a serializer to None will not do any serializing, and the input must be bytes. The user can also pass custom serializers to the key_serializer and value_serializer parameters. These must have “dumps” and “loads” static methods. This allows the user to chain a serializer and a compressor together if desired. Custom serializers must be passed for writing and reading as they are not stored in the booklet file.
import booklet
print(booklet.available_serializers)
Usage
The docstrings have a lot of info about the classes and methods. Files should be opened with the booklet.open function. Read the docstrings of the open function for more details.
Write data using the context manager
import booklet
with booklet.open('test.blt', 'n', value_serializer='pickle', key_serializer='str') as db:
db['test_key'] = ['one', 2, 'three', 4]
Read data using the context manager
with booklet.open('test.blt', 'r') as db:
test_data = db['test_key']
Notice that you don’t need to pass serializer parameters when reading (and additional writing) when in-built serializers are used. Booklet stores this info on the initial file creation.
In most cases, the user should use python’s context manager “with” when reading and writing data. This will ensure data is properly written and locks are released on the file. If the context manager is not used, then the user must be sure to run the db.sync() (or db.close()) at the end of a series of writes to ensure the data has been fully written to disk. Only after the writes have been synced can additional reads occur. Make sure you close your file or you’ll run into file deadlocks!
Write data without using the context manager
import booklet
db = booklet.open('test.blt', 'n', value_serializer='pickle', key_serializer='str')
db['test_key'] = ['one', 2, 'three', 4]
db['2nd_test_key'] = ['five', 6, 'seven', 8]
db.sync() # Normally not necessary if the user closes the file after writing
db.close() # Will also run sync as part of the closing process
Read data without using the context manager
db = booklet.open('test.blt') # 'r' is the default flag
test_data1 = db['test_key']
test_data2 = db['2nd_test_key']
db.close()
Custom serializers
import orjson
class Orjson:
def dumps(obj):
return orjson.dumps(obj, option=orjson.OPT_NON_STR_KEYS | orjson.OPT_OMIT_MICROSECONDS | orjson.OPT_SERIALIZE_NUMPY)
def loads(obj):
return orjson.loads(obj)
with booklet.open('test.blt', 'n', value_serializer=Orjson, key_serializer='str') as db:
db['test_key'] = ['one', 2, 'three', 4]
The Orjson class is actually already built into the package. You can pass the string ‘orjson’ to either serializer parameters to use the above serializer. This is just an example of a serializer.
Here’s another example with compression.
import orjson
import zstandard as zstd
class OrjsonZstd:
def dumps(obj):
return zstd.compress(orjson.dumps(obj, option=orjson.OPT_NON_STR_KEYS | orjson.OPT_OMIT_MICROSECONDS | orjson.OPT_SERIALIZE_NUMPY))
def loads(obj):
return orjson.loads(zstd.decompress(obj))
with booklet.open('test.blt', 'n', value_serializer=OrjsonZstd, key_serializer='str') as db:
db['big_test'] = list(range(1000000))
with booklet.open('test.blt', 'r', value_serializer=OrjsonZstd) as db:
big_test_data = db['big_test']
If you use a custom serializer, then you’ll always need to pass it to booklet.open for additional reading and writing.
The open flag follows the standard dbm options:
Value |
Meaning |
---|---|
'r' |
Open existing database for reading only (default) |
'w' |
Open existing database for reading and writing |
'c' |
Open database for reading and writing, creating it if it doesn’t exist |
'n' |
Always create a new, empty database, open for reading and writing |
Limitations
Adding keys/values above 100,000 will incur a mild slow down in insertion. For example, inserting 100,000 will take ~3 seconds, while 1,000,000 will take ~1 minute (not 30 seconds). The max viable is probably 10,000,0000 currently.
Benchmarks
From my initial tests, the performance is comparable to other very fast key-value databases (e.g. gdbm, lmdb). Proper benchmarks will be coming soon…
Project details
Release history Release notifications | RSS feed
Download files
Download the file for your platform. If you're not sure which to choose, learn more about installing packages.
Source Distribution
Built Distribution
Hashes for booklet-0.4.0-py2.py3-none-any.whl
Algorithm | Hash digest | |
---|---|---|
SHA256 | 18c31e269e26c27b876a3cfdfa99a6a150a950b3e080c3ebe06a04cc7e5247d8 |
|
MD5 | ae585c27299a40a3c4f4fb2741330e57 |
|
BLAKE2b-256 | 9b9c8cd401c747db25a40527c9d1765f22e601ffc39b4b38f79352255133e7d8 |