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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256
|
#!/usr/bin/env python
"""
This is a crude script for detecting reference leaks in the C-based cbor2
implementation. It is by no means fool-proof and won't pick up all possible ref
leaks, but it is a reasonable "confidence test" that things aren't horribly
wrong. The script assumes you're in an environment with objgraph and cbor2
installed.
The script outputs a nicely formatted table of the tests run, and the number of
"extra" objects that existed after the tests (indicating a ref-leak), or "-" if
no extra objects existed. The ideal output is obviously "-" in all rows.
"""
import sys
import tracemalloc
from collections import OrderedDict, namedtuple
from datetime import datetime, timedelta, timezone
from decimal import Decimal
from fractions import Fraction
import objgraph
def import_cbor2():
# Similar hack to that used in tests/conftest to get separate C and Python
# implementations
import cbor2
import cbor2.decoder
import cbor2.encoder
import cbor2.types
class Module:
# Mock module class
pass
py_cbor2 = Module()
for source in (cbor2.types, cbor2.encoder, cbor2.decoder):
for name in dir(source):
setattr(py_cbor2, name, getattr(source, name))
return cbor2, py_cbor2
c_cbor2, py_cbor2 = import_cbor2()
UTC = timezone.utc
TEST_VALUES = [
# label, kwargs, value
("None", {}, None),
("10e0", {}, 1),
("10e12", {}, 1000000000000),
("10e29", {}, 100000000000000000000000000000),
("-10e0", {}, -1),
("-10e12", {}, -1000000000000),
("-10e29", {}, -100000000000000000000000000000),
("float1", {}, 1.0),
("float2", {}, 3.8),
("str", {}, "foo"),
("bigstr", {}, "foobarbaz " * 1000),
("bytes", {}, b"foo"),
("bigbytes", {}, b"foobarbaz\x00" * 1000),
("datetime", {"timezone": UTC}, datetime(2019, 5, 9, 22, 4, 5, 123456)),
("decimal", {}, Decimal("1.1")),
("fraction", {}, Fraction(1, 5)),
("intlist", {}, [1, 2, 3]),
("bigintlist", {}, [1, 2, 3] * 1000),
("strlist", {}, ["foo", "bar", "baz"]),
("bigstrlist", {}, ["foo", "bar", "baz"] * 1000),
("dict", {}, {"a": 1, "b": 2, "c": 3}),
("bigdict", {}, {"a" * i: i for i in range(1000)}),
("set", {}, {1, 2, 3}),
("bigset", {}, set(range(1000))),
("bigdictlist", {}, [{"a" * i: i for i in range(100)}] * 100),
(
"objectdict",
{"timezone": UTC},
{"name": "Foo", "species": "cat", "dob": datetime(2013, 5, 20), "weight": 4.1},
),
(
"objectdictlist",
{"timezone": UTC},
[{"name": "Foo", "species": "cat", "dob": datetime(2013, 5, 20), "weight": 4.1}] * 100,
),
("tag", {}, c_cbor2.CBORTag(1, 1)),
("nestedtag", {}, {c_cbor2.CBORTag(1, 1): 1}),
]
Leaks = namedtuple("Leaks", ("count", "comparison"))
Tests = namedtuple("Tests", ("objgraph", "malloc"))
Result = namedtuple("Result", ("encoding", "decoding", "roundtrip"))
peak = {}
def growth():
return objgraph.growth(limit=None, peak_stats=peak)
def test_malloc(op):
count = 0
start = datetime.now()
# NOTE: Filter pointing to the op() line in the loop below, because we're
# only interested in memory allocated by that line. Naturally, if this file
# is edited, the lineno parameter below must be adjusted!
only_op = tracemalloc.Filter(True, __file__, lineno=102, all_frames=True)
tracemalloc.start(10)
try:
# Perform a pre-run of op so that any one-time memory allocation
# (module imports, etc.) don't affect the later diffs
op()
before = tracemalloc.take_snapshot().filter_traces([only_op])
while True:
count += 1
op()
if datetime.now() - start > timedelta(seconds=0.2):
break
after = tracemalloc.take_snapshot().filter_traces([only_op])
diff = after.compare_to(before, "traceback")
diff = [entry for entry in diff if entry.size_diff > 0]
return count, diff
finally:
tracemalloc.stop()
def test_objgraph(op):
count = 0
start = datetime.now()
# See notes above
op()
growth()
while True:
count += 1
op()
if datetime.now() - start > timedelta(seconds=0.2):
break
return count, growth()
def test(op):
return Tests(Leaks(*test_objgraph(op)), Leaks(*test_malloc(op)))
def format_leaks(result):
if result.objgraph.comparison:
return "%d objs (/%d)" % (
sum(leak[-1] for leak in result.objgraph.comparison),
result.objgraph.count,
)
elif result.malloc.comparison and (
result.malloc.count < result.malloc.comparison[0].size_diff
):
# Running the loop always results in *some* memory allocation, but as
# long as the bytes allocated are less than the number of loops it's
# unlikely to be an actual leak
return "%d bytes (/%d)" % (
result.malloc.comparison[0].size_diff,
result.malloc.count,
)
else:
return "-"
def output_table(results):
# Build table content
head = ("Test", "Encoding", "Decoding", "Round-trip")
rows = [head] + [
(
label,
format_leaks(result.encoding),
format_leaks(result.decoding),
format_leaks(result.roundtrip),
)
for label, result in results.items()
]
# Format table output
cols = zip(*rows)
col_widths = [max(len(row) for row in col) for col in cols]
sep = "".join(
(
"+-",
"-+-".join("-" * width for width in col_widths),
"-+",
)
)
print(sep)
print(
"".join(
(
"| ",
" | ".join(
"{value:<{width}}".format(value=value, width=width)
for value, width in zip(head, col_widths)
),
" |",
)
)
)
print(sep)
for row in rows[1:]:
print(
"".join(
(
"| ",
" | ".join(
"{value:<{width}}".format(value=value, width=width)
for value, width in zip(row, col_widths)
),
" |",
)
)
)
print(sep)
print()
print(
"""\
There *will* be false positives in the table above. Ignore leaks involving a
tiny number of objects (e.g. 1) or a small number of bytes (e.g. < 8Kb) as such
allocations are quite normal.
In the case of a ref-leak of an object that can reference others (lists, sets,
dicts, or anything with a __dict__), expect to see 100s or 1000s of "objs"
leaked. In the case of a ref-leak of a simple object (int, str, bytes, etc.),
expect to see a few hundred Kb allocated.
If leaks occur across the board, it's likely to be in something universal like
dump/load. If it's restricted to a type, check the encoding and decoding
methods for that type.
"""
)
def main():
results = OrderedDict()
sys.stderr.write("Testing")
sys.stderr.flush()
for name, kwargs, value in TEST_VALUES:
encoded = c_cbor2.dumps(value, **kwargs)
results[name] = Result(
encoding=test(lambda: c_cbor2.dumps(value, **kwargs)),
decoding=test(lambda: c_cbor2.loads(encoded)),
roundtrip=test(lambda: c_cbor2.loads(c_cbor2.dumps(value, **kwargs))),
)
sys.stderr.write(".")
sys.stderr.flush()
sys.stderr.write("\n")
sys.stderr.write("\n")
output_table(results)
sys.stderr.write("\n")
if __name__ == "__main__":
main()
|