with the same performance of msgpack but also support python objects, sets, tuples.
msgpack protocol is modified so that there is no supports for other languages, except using packb with compatible_mode = true.
This is modified from msgpack-python (described below).
Here is the benchmark results on my x64 server, each for 200000 times:
python performance.py datatype pymsgpack time(s) cpickle time(s) speed lifting ratio length(bytes) tuple0 :: pymsgpack: 0.1 cpickle 0.173 ratio 1.73416934768 len: 3 : 4 tuple1 :: pymsgpack: 0.239 cpickle 0.487 ratio 2.03731597438 len: 255 : 507 tuple2 :: pymsgpack: 0.045 cpickle 0.121 ratio 2.64331974967 len: 804 : 822 list0 :: pymsgpack: 0.09 cpickle 0.208 ratio 2.29779431123 len: 1 : 6 list1 :: pymsgpack: 0.237 cpickle 0.518 ratio 2.18276936665 len: 253 : 508 list11 :: pymsgpack: 0.665 cpickle 2.719 ratio 4.08486282398 len: 708 : 1373 list111:: pymsgpack: 0.206 cpickle 0.792 ratio 3.84661280845 len: 2253 : 2258 list2 :: pymsgpack: 0.083 cpickle 0.229 ratio 2.75877366213 len: 802 : 823 dict0 :: pymsgpack: 0.093 cpickle 0.205 ratio 2.18427884432 len: 1 : 6 dict1 :: pymsgpack: 0.041 cpickle 0.068 ratio 1.67492092969 len: 16 : 33 dict2 :: pymsgpack: 0.063 cpickle 0.108 ratio 1.71722933995 len: 1048 : 1070 dict3 :: pymsgpack: 0.651 cpickle 0.928 ratio 1.42595863402 len: 1669 : 1758 set0 :: pymsgpack: 0.112 cpickle 1.074 ratio 9.5092526728 len: 3 : 27 set1 :: pymsgpack: 0.084 cpickle 0.378 ratio 4.47175768422 len: 55 : 131 set2 :: pymsgpack: 0.052 cpickle 0.341 ratio 6.47297065153 len: 804 : 846 newobj1:: pymsgpack: 0.307 cpickle 2.049 ratio 6.6596654085 len: 62 : 87 oldobj1:: pymsgpack: 0.424 cpickle 1.39 ratio 3.27720864994 len: 62 : 87 newobj2:: pymsgpack: 0.778 cpickle 3.213 ratio 4.12987933922 len: 7799 : 4618 oldobj2:: pymsgpack: 0.858 cpickle 2.685 ratio 3.1270893371 len: 7796 : 4607
How to use:
- git clone [this repo] and cd into it
- make
- sudo ./setup.py install
- import pymsgpack in your script and use like msgpack
compatible_mode is available for communication with other msgpack interface.
| author: | INADA Naoki |
|---|---|
| version: | 0.4.6 |
| date: | 2015-03-13 |
MessagePack is a fast, compact binary serialization format, suitable for similar data to JSON. This package provides CPython bindings for reading and writing MessagePack data.
$ pip install msgpack-python
msgpack-python provides pure python implementation. PyPy can use this.
When you can't use binary distribution, you need to install Visual Studio or Windows SDK on Windows. Without extension, using pure python implementation on CPython runs slowly.
For Python 2.7, Microsoft Visual C++ Compiler for Python 2.7 is recommended solution.
For Python 3.5, Microsoft Visual Studio 2015 Community Edition or Express Edition can be used to build extension module.
Use packb for packing and unpackb for unpacking.
msgpack provides dumps and loads as alias for compatibility with
json and pickle.
pack and dump packs to file-like object.
unpack and load unpacks from file-like object.
>>> import msgpack
>>> msgpack.packb([1, 2, 3])
'\x93\x01\x02\x03'
>>> msgpack.unpackb(_)
[1, 2, 3]unpack unpacks msgpack's array to Python's list, but can unpack to tuple:
>>> msgpack.unpackb(b'\x93\x01\x02\x03', use_list=False)
(1, 2, 3)You should always pass the use_list keyword argument. See performance issues relating to use_list option below.
Read the docstring for other options.
Unpacker is a "streaming unpacker". It unpacks multiple objects from one
stream (or from bytes provided through its feed method).
import msgpack
from io import BytesIO
buf = BytesIO()
for i in range(100):
buf.write(msgpack.packb(range(i)))
buf.seek(0)
unpacker = msgpack.Unpacker(buf)
for unpacked in unpacker:
print unpackedIt is also possible to pack/unpack custom data types. Here is an example for
datetime.datetime.
import datetime
import msgpack
useful_dict = {
"id": 1,
"created": datetime.datetime.now(),
}
def decode_datetime(obj):
if b'__datetime__' in obj:
obj = datetime.datetime.strptime(obj["as_str"], "%Y%m%dT%H:%M:%S.%f")
return obj
def encode_datetime(obj):
if isinstance(obj, datetime.datetime):
return {'__datetime__': True, 'as_str': obj.strftime("%Y%m%dT%H:%M:%S.%f")}
return obj
packed_dict = msgpack.packb(useful_dict, default=encode_datetime)
this_dict_again = msgpack.unpackb(packed_dict, object_hook=decode_datetime)Unpacker's object_hook callback receives a dict; the
object_pairs_hook callback may instead be used to receive a list of
key-value pairs.
It is also possible to pack/unpack custom data types using the ext type.
>>> import msgpack
>>> import array
>>> def default(obj):
... if isinstance(obj, array.array) and obj.typecode == 'd':
... return msgpack.ExtType(42, obj.tostring())
... raise TypeError("Unknown type: %r" % (obj,))
...
>>> def ext_hook(code, data):
... if code == 42:
... a = array.array('d')
... a.fromstring(data)
... return a
... return ExtType(code, data)
...
>>> data = array.array('d', [1.2, 3.4])
>>> packed = msgpack.packb(data, default=default)
>>> unpacked = msgpack.unpackb(packed, ext_hook=ext_hook)
>>> data == unpacked
TrueAs an alternative to iteration, Unpacker objects provide unpack,
skip, read_array_header and read_map_header methods. The former two
read an entire message from the stream, respectively deserialising and returning
the result, or ignoring it. The latter two methods return the number of elements
in the upcoming container, so that each element in an array, or key-value pair
in a map, can be unpacked or skipped individually.
Each of these methods may optionally write the packed data it reads to a callback function:
from io import BytesIO
def distribute(unpacker, get_worker):
nelems = unpacker.read_map_header()
for i in range(nelems):
# Select a worker for the given key
key = unpacker.unpack()
worker = get_worker(key)
# Send the value as a packed message to worker
bytestream = BytesIO()
unpacker.skip(bytestream.write)
worker.send(bytestream.getvalue())In old days, msgpack doesn't distinguish string and binary types like Python 1. The type for represent string and binary types is named raw.
msgpack can distinguish string and binary type for now. But it is not like Python 2. Python 2 added unicode string. But msgpack renamed raw to str and added bin type. It is because keep compatibility with data created by old libs. raw was used for text more than binary.
Currently, while msgpack-python supports new bin type, default setting doesn't use it and decodes raw as bytes instead of unicode (str in Python 3).
You can change this by using use_bin_type=True option in Packer and encoding="utf-8" option in Unpacker.
>>> import msgpack
>>> packed = msgpack.packb([b'spam', u'egg'], use_bin_type=True)
>>> msgpack.unpackb(packed, encoding='utf-8')
['spam', u'egg']To use ext type, pass msgpack.ExtType object to packer.
>>> import msgpack
>>> packed = msgpack.packb(msgpack.ExtType(42, b'xyzzy'))
>>> msgpack.unpackb(packed)
ExtType(code=42, data='xyzzy')You can use it with default and ext_hook. See below.
The msgpack-python 0.3 have some incompatible changes.
The default value of use_list keyword argument is True from 0.3.
You should pass the argument explicitly for backward compatibility.
Unpacker.unpack() and some unpack methods now raises OutOfData instead of StopIteration. StopIteration is used for iterator protocol only.
CPython's GC starts when growing allocated object.
This means unpacking may cause useless GC.
You can use gc.disable() when unpacking large message.
List is the default sequence type of Python.
But tuple is lighter than list.
You can use use_list=False while unpacking when performance is important.
Python's dict can't use list as key and MessagePack allows array for key of mapping.
use_list=False allows unpacking such message.
Another way to unpacking such object is using object_pairs_hook.
MessagePack uses pytest for testing. Run test with following command:
$ py.test