Durabledict

Durabledict contains a collection of dictionary classes backed by a durable data store (Redis, Django models, etc) suitable for use in a distributed manner. Dictionary values are cached locally in the instance of the dictionary, and only sync with their values with their durable data stores when a value in the data store has changed.

Usage

Durabledict contains various flavors of a dictionary-like objects backed by a durable data store. All dicts classes are located in the durabledict package. At present, Durabledict offers the following dicts:

1. durabledict.redis.RedisDict - Backed by Redis.
2. durabledict.models.ModelDict - Backed by DB objects (most likely Django models).
3. durabledict.zookeeper.ZookeeperDict - Backed by Zookeeper.

Each dictionary class has a different __init__ method which take different arguments, so consult their documentation for specific usage detail.

Once you have an instance of a durabledict, just use it like you would a normal dictionary:

from durabledict import RedisDict
from redis import Redis

# Construct a new RedisDict object
settings = RedisDict('settings', Redis())

# Assign and retrieve a value from the dict
settings['foo'] = 'bar'
settings['foo']
>>> 'bar'

# Assign and retrieve another value
settings['buzz'] = 'foogle'
settings['buzz']
>>> 'foogle'

# Delete a value and access receives a KeyError
del settings['foo']
settings['foo']
>>> KeyError

All dict types pickle their objects inside their durable data store, so any object that is "pickleable" can be saved to those stores.

Notes on Persistence, Consistency and the In-Memory Cache

Nearly all methods called on a Durabledict dictionary class (i.e. RedisDict) are proxied to an internal dict object that serves as a cache for access to dict values. This cache is only updated with fresh values from durable storage if there actually has been a change in the values stored in durable storage.

To check if the data in durable storage has changed, each durabledict backend is responsible for providing a fast last_updated() method that quickly tells the dict the last time any value in the durable storage has been updated. For instance, the Durabledict constructor requires a cache object passed in as an argument, which provides implementations of cache-line interface methods for maintaining the last_updated state. A memcache client is a good candidate for this object.

Out of the box by default, all Durabledict classes will sync with their durable data store on all writes (insert, updates and deletes) as well as immediately before any read operation on the dictionary. This mode provides high read consistency of data at the expense of read speed. You can be guaranteed that any read operation on your dict, i.e. settings['cool_feature'], will always use the most up to date data. If another consumer of your durable data store has modified a value in that store since you instantiated your object, you will immediately be able to read the new data with your dict instance.

Manually Control Durable Storage Sync

As mentioned above in, the downside to syncing with durable storage before each read of dict data is it lowers your read performance. If you read 100 keys from your dictionary, that means 100 accesses to check `last_updated(). Even with a data store as fast as memecache, that adds up quite quickly.

It therefore may be advantageous for you to not sync with durable storage before every read from the dict and instead control that syncing manually. To do so, pass autosync=False when you construct the dict, i.e.:

from durabledict import RedisDict
from redis import Redis

# Construct a new RedisDict object that does not sync on reads
settings = RedisDict('settings', Redis(), autosync=False)

This causes the dictionary behave in the following way:

1. Like normal, the dictionary initializes from the durable data store upon instantiation.
2. Writes (both inserts and updates), along with deletes of values to the dictionary will still automatically sync with the data store each time the operation happens.
3. Any time a dictionary is read from, only data current in the internal cache is used. The dict will not attempt to sync with its durable data store before reads.
4. To force the dict to attempt to sync with its durable data store, you may call the sync() method on the dictionary. As with when autosync is false, if last_update says there are no changes, the dict will skip updating from durable storage.

A good use case for manual syncing is a read-heavy web application, where you're using a durabledict for settings configuration. Very few requests actually change the dictionary contents - most simply read from the dictionary. In this situation, you would perhaps only sync() at the beginning of a user's web request to make sure the dict is up to date, but then not during the request in order to push the response to the user as fast as possible.

Encoding

All durabledict implementations accept an encoding keyword argument, which defines the encoding object the dictionary should use when serializing data to and from the persistent data store. The overarching goal of the encoding is to serialize the dictionary value object into a format suitable for persisting to durable storage, and then at a later date reconstructing that object from its serialized representation into an object in memory.

By default, durabledict uses pickle as its encoding format, which allows it to serialize complex object easily at the expense of known security implications:security and other limitations. See this IBM Developerworks:devworks article for an overview of Pickle.

In addition to the built in encoding.PickleEncoding, durabledict also features encoding.JSONEncoding which encodes the data as JSON and encoding.NoOpEncoding which does not encode the data at all (suitable only for the MemoryDict implementation).

Integration with Django

If you would like to store your dict values in the database for your Django application, you should use the durabledict.models.Durabledict class. This class takes an instance of a model's manager, as well as key_col and value_col arguments which can be used to tell Durabledict which columns on your object it should use to store data.

It's also probably most advantageous to construct your dicts with autosync=False (see "Manually Control Durable Storage Sync" above) and manually call sync() before each request. This can be accomplished most easily via the request_started signal:

django.core.signals.request_started.connect(settings.sync)

Creating Your Own Durable Dict

Creating your own durable dict is easy. All you need to do is subclass durabledict.base.DurableDict and implement the following required interface methods.

1. persist(key, value) - Persist value at key to your data store.
2. depersist(key) - Delete the value at key from your data store.
3. durables() - Return a key=val dict of all keys in your data store.
4. last_updated() - A comparable value of when the data in your data store was last updated.

You may also implement a couple optional dictionary methods, which durabledict.base.DurableDict will call when the actual non-underscored version is called on the dict.

1. _pop(key[,default]) - If key is in the dictionary, remove it and return its value, else return default. If default is not given and key is not in the dictionary, a KeyError is raised.
2. _setdefault(key[,default]) - If key is in the dictionary, return its value. If not, insert key with a value of default and return default. default defaults to None.