Generic database utilities, including connection clustering and routing so you can scale like a pro.
The following example creates a Redis connection cluster which will distribute reads and writes based on a simple modulus lookup of the hashed key:
.. code:: python
from nydus.db import create_cluster
redis = create_cluster({
'backend': 'nydus.db.backends.redis.Redis',
'router': 'nydus.db.routers.keyvalue.PartitionRouter',
'hosts': {
0: {'db': 0},
1: {'db': 1},
2: {'db': 2},
}
})
res = redis.incr('foo')
assert res == 1
The create_cluster function is a simple helper to configure a Cluster
based on a simple dict config.
There are two required keyword arguments:
backend
: full path to the backend class, which should extend nydus.db.backends.BaseConnection
.
hosts
: a dictionary, where the key is an ordered numeric value, and the result is a dict of connection options.
(the keys are numeric values simply for readability in configuration)
defaults
: a dictionary of default settings for all hosts to inherit from.
Optionally, you may also specify a value for router
, which is the full path to the router class,
which must extend nydus.db.routers.BaseRouter
.
In some cases you may want to execute a query on many nodes (in parallel). Nydus has built-in support for this when any routing function returns a cluster of nodes:
.. code:: python
from nydus.db import create_cluster
>>>
# by not specifying a router all queries are executed on all hosts
redis = create_cluster({
'backend': 'nydus.db.backends.redis.Redis',
'hosts': {
0: {'db': 0},
1: {'db': 1},
2: {'db': 2},
}
})
>>>
# map the query over all connections returned by the default router
res = redis.incr('foo')
>>>
assert type(res) == list
assert len(res) == 3
You can also map many queries (utilizing an internal queue) over connections (again, returned by the router):
.. code:: python
with redis.map() as conn:
results = [conn.incr(k) for k in keys]
As of release 0.5.0, the map() function now supports pipelines, and the included Redis backend will pipeline commands wherever possible.
Nydus was originally designed as a toolkit to expand on the usage of Redis at DISQUS. While it does provide a framework for building clusters that are not Redis, much of the support has gone into providing utilities for routing and querying on Redis clusters.
You can configure the Redis client for a connection by specifying it's full path:
.. code:: python
redis = create_cluster({
'backend': 'nydus.db.backends.redis.Redis',
'hosts': {
0: {'db': 0},
},
})
The available host options are:
The Redis client also supports pipelines via the map command. This means that all commands will hit servers at most as of once:
.. code:: python
with redis.map() as conn:
conn.set('a', 1)
conn.incr('a')
results = [conn.get('a'), conn.get('b')]
results['a'] == 2
results['b'] == None
Simple Partition Router
There are also several options for builtin routing. The easiest is a simple partition router, which is just a simple
hash on the key:
.. code:: python
redis = create_cluster({
'backend': 'nydus.db.backends.redis.Redis',
'router': 'nydus.db.routers.keyvalue.PartitionRouter',
'hosts': {
0: {'db': 0},
},
})
Consistent Hashing Router
An improvement upon hashing, Nydus provides a Ketama-based consistent hashing router:
.. code:: python
redis = create_cluster({
'backend': 'nydus.db.backends.redis.Redis',
'router': 'nydus.db.routers.keyvalue.ConsistentHashingRouter',
'hosts': {
0: {'db': 0},
},
})
Round Robin Router
An additional option for distributing queries is the round robin router:
.. code:: python
redis = create_cluster({
'backend': 'nydus.db.backends.redis.Redis',
'router': 'nydus.db.routers.RoundRobinRouter',
'hosts': {
0: {'db': 0},
},
})
Pycassa
-------
Basic connection management for pycassa (Cassandra) clusters is supported, but we use a non-standard syntax
for creating clusters as routing behavior and per-connection options are not useful in this context:
.. code:: python
pycassa = create_cluster({
'backend': 'nydus.db.backends.pycassa.Pycassa',
'hosts': ['localhost'],
'keyspace': 'test',
})
.. note:: Pycassa handles routing of hosts internally, which means things like ``map`` have no affect.