pickledir 0.3.5

File-based key-value storage for pickle-serializable keys and values.

pickledir

File-based key-value storage.

Keys and values are serialized with pickle. Data is kept in files in the specified directory.

CI-tested with Python 3.8-3.9 on macOS, Ubuntu and Windows.

---

The storage has zero initialization time, fast random access, fast reads and writes.

Unlike shelve, the data saved by PickleDir is cross-platform: you can write it on Linux and read the same files on Windows. Unlike most database-based caching solutions (including the shelve), the PickleDir does not require the "open" and "close" storage. It's always open since it's just a directory in the file system.

PickleDir is better for casual data storage. Database-based solutions are preferred when your storage has many elements (3 thousand or more). They will also be faster when working with a predictably high load in terms of reading and writing.

Install

$ pip3 install pickledir

Use

Create

from pickledir import PickleDir

cache = PickleDir('path/to/my_cache_dir')

Write

Keys do not need to be hashable. They only need to be serializable with pickle .

When you assign a value, the data is literally written to a file.

cache['key'] = 'hello, user!'
cache[5] = 23
cache[{'a', 'b', 'c'}] = 'abc'

Read

print(cache['key'])
print(cache[5])
print(cache[{'a', 'b', 'c'}])

Read all values

for key, value in cache.items():
    print(key, value)

Delete item

del cache['key']

Type hints

# declaring PickleDir with string keys and integer values:

cache: PickleDir[str, int] = PickleDir('path/to/my_cache_dir')

Set expiration time on writing

The expired items will be removed from the storage.

cache.set('a', 1000, max_age = datetime.timedelta(seconds=1))
print(cache.get('a'))  # 1000
time.sleep(2)     
print(cache.get('a'))  # None (and removed from storage)

Set expiration time on reading

The expired items will not be returned, but kept in the storage.

cache['b'] = 1000
time.sleep(2)
cache.get('b' max_age = datetime.timedelta(seconds=1)) # None
cache.get('b' max_age = datetime.timedelta(seconds=9)) # 1000

Set data version

Setting the data version makes it easy to mark old data as obsolete.

For example, you cached the result of a function, and then changed the implementation of that function. In this case, there is no need to delete old files from the cache. Just change the version number.

cache = PickleDir('path/to/dir', version=1)
cache['a'] = 'some_data'

You can read all stored data while the version value is 1.

cache = PickleDir('path/to/dir', version=1)
print(cache.get('a'))  # 'some_data'

If you decide that all the data in the cache is out of date, just pass the constructor a version number that you haven't used before.

cache = PickleDir('path/to/dir', version=2)
print(cache.get('a'))  # None

Now all that is saved with version 2 is actual data. Any other version is considered obsolete and will be gradually removed.

Do not create the PickleDir with an old version number. It will make the data unpredictable.

cacheV1 = PickleDir('path/to/dir', version=1)  # ok
cacheV1['a'] = 'old A'
cacheV1['b'] = 'old B'

cacheV2 = PickleDir('path/to/dir', version=2)  # ok
cacheV2['a'] = 'new A'

cacheV1 = PickleDir('path/to/dir', version=1)  # don't do this
print(cacheV1.get('b'))  # Schrödinger's data ('old B' or None)

Benchmarks

Casually saving 10 items and reading them again:

for i in range(10):
    cache[str(i)] = {"data": i, "other": None}
for i in range(10):
    _ = cache[str(i)]

Storage	Time
PickleDir	0.42
shelve	6.68
diskcache.Cache	1.09

Measured on macOS, Python 3.8, SATA HDD (not SSD), Journaled HFS+.

See sources in benchmark dir.

The main advantage of pickledir is the lack of time required to create a database or initialize tables. If we did not save 10 items, but 1000 in a row, shelve and diskcache would be faster than pickledir.

Under the hood

Serialized data is stored inside files in the same directory. Each file contains one or more items. The maximum number of files is limited to 4096. The values are uniformly distributed between the files.

Reading is slower when a file contains more than one item. Therefore, the PickleDir is better suited for cases with the number of items within a few thousand.