1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
|
import subprocess
import sys
import tempfile
import requests
TEMPLATE = """
import resource
import time
with open({path!r}, "rb") as f:
data = f.read()
initial_rss = resource.getrusage(resource.RUSAGE_SELF).ru_maxrss
{setup}
start = time.perf_counter()
for _ in range(5):
decode(data)
stop = time.perf_counter()
max_rss = resource.getrusage(resource.RUSAGE_SELF).ru_maxrss
# mem_mib = (max_rss * 1024 - len(data)) / (1024 * 1024)
mem_mib = (max_rss - initial_rss) / 1024
time_ms = ((stop - start) / 5) * 1000
print([mem_mib, time_ms])
"""
JSON = """
import json
decode = json.loads
"""
UJSON = """
import ujson
decode = ujson.loads
"""
ORJSON = """
import orjson
decode = orjson.loads
"""
RAPIDJSON = """
import rapidjson
decode = rapidjson.loads
"""
SIMDJSON = """
import simdjson
decode = simdjson.loads
"""
MSGSPEC = """
import msgspec
decode = msgspec.json.decode
"""
MSGSPEC_STRUCTS = """
import msgspec
from typing import Union
class Package(msgspec.Struct, gc=False):
build: str
build_number: int
depends: tuple[str, ...]
md5: str
name: str
sha256: str
subdir: str
version: str
license: str = ""
noarch: Union[str, bool, None] = None
size: int = 0
timestamp: int = 0
class RepoData(msgspec.Struct, gc=False):
repodata_version: int
info: dict
packages: dict[str, Package]
removed: tuple[str, ...]
decode = msgspec.json.Decoder(RepoData).decode
"""
def main():
import argparse
parser = argparse.ArgumentParser(
description="Benchmark decoding a large JSON message using various JSON libraries"
)
parser.add_argument(
"--versions",
action="store_true",
help="Output library version info, and exit immediately",
)
args = parser.parse_args()
benchmarks = [
("json", None, JSON),
("ujson", "ujson", UJSON),
("orjson", "orjson", ORJSON),
("rapidjson", "python-rapidjson", RAPIDJSON),
("simdjson", "pysimdjson", SIMDJSON),
("msgspec", "msgspec", MSGSPEC),
("msgspec structs", None, MSGSPEC_STRUCTS),
]
if args.versions:
import importlib.metadata
for _, lib, _ in benchmarks:
if lib is not None:
version = importlib.metadata.version(lib)
print(f"- {lib}: {version}")
sys.exit(0)
with tempfile.NamedTemporaryFile() as f:
# Download the repodata.json
resp = requests.get(
"https://conda.anaconda.org/conda-forge/noarch/repodata.json"
)
resp.raise_for_status()
f.write(resp.content)
# Run the benchmark for each library
results = {}
import ast
for lib, _, setup in benchmarks:
script = TEMPLATE.format(path=f.name, setup=setup)
# We execute each script in a subprocess to isolate their memory usage
output = subprocess.check_output([sys.executable, "-c", script])
results[lib] = ast.literal_eval(output.decode())
# Compose the results table
best_mem, best_time = results["msgspec structs"]
columns = (
"",
"memory (MiB)",
"vs.",
"time (ms)",
"vs.",
)
rows = [
(
f"**{lib}**",
f"{mem:.1f}",
f"{mem / best_mem:.1f}x",
f"{time:.1f}",
f"{time / best_time:.1f}x",
)
for lib, (mem, time) in results.items()
]
rows.sort(key=lambda x: float(x[1]))
widths = tuple(
max(max(map(len, x)), len(c)) for x, c in zip(zip(*rows), columns)
)
row_template = ("|" + (" %%-%ds |" * len(columns))) % widths
header = row_template % tuple(columns)
bar_underline = "+%s+" % "+".join("=" * (w + 2) for w in widths)
bar = "+%s+" % "+".join("-" * (w + 2) for w in widths)
parts = [bar, header, bar_underline]
for r in rows:
parts.append(row_template % r)
parts.append(bar)
print("\n".join(parts))
if __name__ == "__main__":
main()
|
