tracemalloc —- 跟踪内存分配

3.4 新版功能.

源代码: Lib/tracemalloc.py


The tracemalloc module is a debug tool to trace memory blocks allocated by Python. It provides the following information:

  • Traceback where an object was allocated

  • Statistics on allocated memory blocks per filename and per line number: total size, number and average size of allocated memory blocks

  • Compute the differences between two snapshots to detect memory leaks

To trace most memory blocks allocated by Python, the module should be started as early as possible by setting the PYTHONTRACEMALLOC environment variable to 1, or by using -X tracemalloc command line option. The tracemalloc.start() function can be called at runtime to start tracing Python memory allocations.

By default, a trace of an allocated memory block only stores the most recent frame (1 frame). To store 25 frames at startup: set the PYTHONTRACEMALLOC environment variable to 25, or use the -X tracemalloc=25 command line option.

示例

显示前10项

显示内存分配最多的10个文件:

  1. import tracemalloc
  2. tracemalloc.start()
  3. # ... run your application ...
  4. snapshot = tracemalloc.take_snapshot()
  5. top_stats = snapshot.statistics('lineno')
  6. print("[ Top 10 ]")
  7. for stat in top_stats[:10]:
  8. print(stat)

Python测试套件的输出示例:

  1. [ Top 10 ]
  2. <frozen importlib._bootstrap>:716: size=4855 KiB, count=39328, average=126 B
  3. <frozen importlib._bootstrap>:284: size=521 KiB, count=3199, average=167 B
  4. /usr/lib/python3.4/collections/__init__.py:368: size=244 KiB, count=2315, average=108 B
  5. /usr/lib/python3.4/unittest/case.py:381: size=185 KiB, count=779, average=243 B
  6. /usr/lib/python3.4/unittest/case.py:402: size=154 KiB, count=378, average=416 B
  7. /usr/lib/python3.4/abc.py:133: size=88.7 KiB, count=347, average=262 B
  8. <frozen importlib._bootstrap>:1446: size=70.4 KiB, count=911, average=79 B
  9. <frozen importlib._bootstrap>:1454: size=52.0 KiB, count=25, average=2131 B
  10. <string>:5: size=49.7 KiB, count=148, average=344 B
  11. /usr/lib/python3.4/sysconfig.py:411: size=48.0 KiB, count=1, average=48.0 KiB

We can see that Python loaded 4855 KiB data (bytecode and constants) from modules and that the collections module allocated 244 KiB to build namedtuple types.

See Snapshot.statistics() for more options.

计算差异

获取两个快照并显示差异:

  1. import tracemalloc
  2. tracemalloc.start()
  3. # ... start your application ...
  4. snapshot1 = tracemalloc.take_snapshot()
  5. # ... call the function leaking memory ...
  6. snapshot2 = tracemalloc.take_snapshot()
  7. top_stats = snapshot2.compare_to(snapshot1, 'lineno')
  8. print("[ Top 10 differences ]")
  9. for stat in top_stats[:10]:
  10. print(stat)

Example of output before/after running some tests of the Python test suite:

  1. [ Top 10 differences ]
  2. <frozen importlib._bootstrap>:716: size=8173 KiB (+4428 KiB), count=71332 (+39369), average=117 B
  3. /usr/lib/python3.4/linecache.py:127: size=940 KiB (+940 KiB), count=8106 (+8106), average=119 B
  4. /usr/lib/python3.4/unittest/case.py:571: size=298 KiB (+298 KiB), count=589 (+589), average=519 B
  5. <frozen importlib._bootstrap>:284: size=1005 KiB (+166 KiB), count=7423 (+1526), average=139 B
  6. /usr/lib/python3.4/mimetypes.py:217: size=112 KiB (+112 KiB), count=1334 (+1334), average=86 B
  7. /usr/lib/python3.4/http/server.py:848: size=96.0 KiB (+96.0 KiB), count=1 (+1), average=96.0 KiB
  8. /usr/lib/python3.4/inspect.py:1465: size=83.5 KiB (+83.5 KiB), count=109 (+109), average=784 B
  9. /usr/lib/python3.4/unittest/mock.py:491: size=77.7 KiB (+77.7 KiB), count=143 (+143), average=557 B
  10. /usr/lib/python3.4/urllib/parse.py:476: size=71.8 KiB (+71.8 KiB), count=969 (+969), average=76 B
  11. /usr/lib/python3.4/contextlib.py:38: size=67.2 KiB (+67.2 KiB), count=126 (+126), average=546 B

We can see that Python has loaded 8173 KiB of module data (bytecode and constants), and that this is 4428 KiB more than had been loaded before the tests, when the previous snapshot was taken. Similarly, the linecache module has cached 940 KiB of Python source code to format tracebacks, all of it since the previous snapshot.

If the system has little free memory, snapshots can be written on disk using the Snapshot.dump() method to analyze the snapshot offline. Then use the Snapshot.load() method reload the snapshot.

Get the traceback of a memory block

Code to display the traceback of the biggest memory block:

  1. import tracemalloc
  2. # Store 25 frames
  3. tracemalloc.start(25)
  4. # ... run your application ...
  5. snapshot = tracemalloc.take_snapshot()
  6. top_stats = snapshot.statistics('traceback')
  7. # pick the biggest memory block
  8. stat = top_stats[0]
  9. print("%s memory blocks: %.1f KiB" % (stat.count, stat.size / 1024))
  10. for line in stat.traceback.format():
  11. print(line)

Example of output of the Python test suite (traceback limited to 25 frames):

  1. 903 memory blocks: 870.1 KiB
  2. File "<frozen importlib._bootstrap>", line 716
  3. File "<frozen importlib._bootstrap>", line 1036
  4. File "<frozen importlib._bootstrap>", line 934
  5. File "<frozen importlib._bootstrap>", line 1068
  6. File "<frozen importlib._bootstrap>", line 619
  7. File "<frozen importlib._bootstrap>", line 1581
  8. File "<frozen importlib._bootstrap>", line 1614
  9. File "/usr/lib/python3.4/doctest.py", line 101
  10. import pdb
  11. File "<frozen importlib._bootstrap>", line 284
  12. File "<frozen importlib._bootstrap>", line 938
  13. File "<frozen importlib._bootstrap>", line 1068
  14. File "<frozen importlib._bootstrap>", line 619
  15. File "<frozen importlib._bootstrap>", line 1581
  16. File "<frozen importlib._bootstrap>", line 1614
  17. File "/usr/lib/python3.4/test/support/__init__.py", line 1728
  18. import doctest
  19. File "/usr/lib/python3.4/test/test_pickletools.py", line 21
  20. support.run_doctest(pickletools)
  21. File "/usr/lib/python3.4/test/regrtest.py", line 1276
  22. test_runner()
  23. File "/usr/lib/python3.4/test/regrtest.py", line 976
  24. display_failure=not verbose)
  25. File "/usr/lib/python3.4/test/regrtest.py", line 761
  26. match_tests=ns.match_tests)
  27. File "/usr/lib/python3.4/test/regrtest.py", line 1563
  28. main()
  29. File "/usr/lib/python3.4/test/__main__.py", line 3
  30. regrtest.main_in_temp_cwd()
  31. File "/usr/lib/python3.4/runpy.py", line 73
  32. exec(code, run_globals)
  33. File "/usr/lib/python3.4/runpy.py", line 160
  34. "__main__", fname, loader, pkg_name)

We can see that the most memory was allocated in the importlib module to load data (bytecode and constants) from modules: 870.1 KiB. The traceback is where the importlib loaded data most recently: on the import pdb line of the doctest module. The traceback may change if a new module is loaded.

Pretty top

Code to display the 10 lines allocating the most memory with a pretty output, ignoring <frozen importlib._bootstrap> and <unknown> files:

  1. import linecache
  2. import os
  3. import tracemalloc
  4. def display_top(snapshot, key_type='lineno', limit=10):
  5. snapshot = snapshot.filter_traces((
  6. tracemalloc.Filter(False, "<frozen importlib._bootstrap>"),
  7. tracemalloc.Filter(False, "<unknown>"),
  8. ))
  9. top_stats = snapshot.statistics(key_type)
  10. print("Top %s lines" % limit)
  11. for index, stat in enumerate(top_stats[:limit], 1):
  12. frame = stat.traceback[0]
  13. print("#%s: %s:%s: %.1f KiB"
  14. % (index, frame.filename, frame.lineno, stat.size / 1024))
  15. line = linecache.getline(frame.filename, frame.lineno).strip()
  16. if line:
  17. print(' %s' % line)
  18. other = top_stats[limit:]
  19. if other:
  20. size = sum(stat.size for stat in other)
  21. print("%s other: %.1f KiB" % (len(other), size / 1024))
  22. total = sum(stat.size for stat in top_stats)
  23. print("Total allocated size: %.1f KiB" % (total / 1024))
  24. tracemalloc.start()
  25. # ... run your application ...
  26. snapshot = tracemalloc.take_snapshot()
  27. display_top(snapshot)

Python测试套件的输出示例:

  1. Top 10 lines
  2. #1: Lib/base64.py:414: 419.8 KiB
  3. _b85chars2 = [(a + b) for a in _b85chars for b in _b85chars]
  4. #2: Lib/base64.py:306: 419.8 KiB
  5. _a85chars2 = [(a + b) for a in _a85chars for b in _a85chars]
  6. #3: collections/__init__.py:368: 293.6 KiB
  7. exec(class_definition, namespace)
  8. #4: Lib/abc.py:133: 115.2 KiB
  9. cls = super().__new__(mcls, name, bases, namespace)
  10. #5: unittest/case.py:574: 103.1 KiB
  11. testMethod()
  12. #6: Lib/linecache.py:127: 95.4 KiB
  13. lines = fp.readlines()
  14. #7: urllib/parse.py:476: 71.8 KiB
  15. for a in _hexdig for b in _hexdig}
  16. #8: <string>:5: 62.0 KiB
  17. #9: Lib/_weakrefset.py:37: 60.0 KiB
  18. self.data = set()
  19. #10: Lib/base64.py:142: 59.8 KiB
  20. _b32tab2 = [a + b for a in _b32tab for b in _b32tab]
  21. 6220 other: 3602.8 KiB
  22. Total allocated size: 5303.1 KiB

See Snapshot.statistics() for more options.

Record the current and peak size of all traced memory blocks

The following code computes two sums like 0 + 1 + 2 + ... inefficiently, by creating a list of those numbers. This list consumes a lot of memory temporarily. We can use get_traced_memory() and reset_peak() to observe the small memory usage after the sum is computed as well as the peak memory usage during the computations:

  1. import tracemalloc
  2. tracemalloc.start()
  3. # Example code: compute a sum with a large temporary list
  4. large_sum = sum(list(range(100000)))
  5. first_size, first_peak = tracemalloc.get_traced_memory()
  6. tracemalloc.reset_peak()
  7. # Example code: compute a sum with a small temporary list
  8. small_sum = sum(list(range(1000)))
  9. second_size, second_peak = tracemalloc.get_traced_memory()
  10. print(f"{first_size=}, {first_peak=}")
  11. print(f"{second_size=}, {second_peak=}")

输出:

  1. first_size=664, first_peak=3592984
  2. second_size=804, second_peak=29704

Using reset_peak() ensured we could accurately record the peak during the computation of small_sum, even though it is much smaller than the overall peak size of memory blocks since the start() call. Without the call to reset_peak(), second_peak would still be the peak from the computation large_sum (that is, equal to first_peak). In this case, both peaks are much higher than the final memory usage, and which suggests we could optimise (by removing the unnecessary call to list, and writing sum(range(...))).

API

函数

tracemalloc.clear_traces()

Clear traces of memory blocks allocated by Python.

See also stop().

tracemalloc.get_object_traceback(obj)

Get the traceback where the Python object obj was allocated. Return a Traceback instance, or None if the tracemalloc module is not tracing memory allocations or did not trace the allocation of the object.

See also gc.get_referrers() and sys.getsizeof() functions.

tracemalloc.get_traceback_limit()

Get the maximum number of frames stored in the traceback of a trace.

The tracemalloc module must be tracing memory allocations to get the limit, otherwise an exception is raised.

The limit is set by the start() function.

tracemalloc.get_traced_memory()

Get the current size and peak size of memory blocks traced by the tracemalloc module as a tuple: (current: int, peak: int).

tracemalloc.reset_peak()

Set the peak size of memory blocks traced by the tracemalloc module to the current size.

Do nothing if the tracemalloc module is not tracing memory allocations.

This function only modifies the recorded peak size, and does not modify or clear any traces, unlike clear_traces(). Snapshots taken with take_snapshot() before a call to reset_peak() can be meaningfully compared to snapshots taken after the call.

See also get_traced_memory().

3.9 新版功能.

tracemalloc.get_tracemalloc_memory()

Get the memory usage in bytes of the tracemalloc module used to store traces of memory blocks. Return an int.

tracemalloc.is_tracing()

True if the tracemalloc module is tracing Python memory allocations, False otherwise.

See also start() and stop() functions.

tracemalloc.start(nframe: int=1)

Start tracing Python memory allocations: install hooks on Python memory allocators. Collected tracebacks of traces will be limited to nframe frames. By default, a trace of a memory block only stores the most recent frame: the limit is 1. nframe must be greater or equal to 1.

You can still read the original number of total frames that composed the traceback by looking at the Traceback.total_nframe attribute.

Storing more than 1 frame is only useful to compute statistics grouped by 'traceback' or to compute cumulative statistics: see the Snapshot.compare_to() and Snapshot.statistics() methods.

Storing more frames increases the memory and CPU overhead of the tracemalloc module. Use the get_tracemalloc_memory() function to measure how much memory is used by the tracemalloc module.

The PYTHONTRACEMALLOC environment variable (PYTHONTRACEMALLOC=NFRAME) and the -X tracemalloc=NFRAME command line option can be used to start tracing at startup.

See also stop(), is_tracing() and get_traceback_limit() functions.

tracemalloc.stop()

Stop tracing Python memory allocations: uninstall hooks on Python memory allocators. Also clears all previously collected traces of memory blocks allocated by Python.

Call take_snapshot() function to take a snapshot of traces before clearing them.

See also start(), is_tracing() and clear_traces() functions.

tracemalloc.take_snapshot()

Take a snapshot of traces of memory blocks allocated by Python. Return a new Snapshot instance.

The snapshot does not include memory blocks allocated before the tracemalloc module started to trace memory allocations.

Tracebacks of traces are limited to get_traceback_limit() frames. Use the nframe parameter of the start() function to store more frames.

The tracemalloc module must be tracing memory allocations to take a snapshot, see the start() function.

See also the get_object_traceback() function.

域过滤器

class tracemalloc.DomainFilter(inclusive: bool, domain: int)

Filter traces of memory blocks by their address space (domain).

3.6 新版功能.

  • inclusive

    If inclusive is True (include), match memory blocks allocated in the address space domain.

    If inclusive is False (exclude), match memory blocks not allocated in the address space domain.

  • domain

    Address space of a memory block (int). Read-only property.

过滤器

class tracemalloc.Filter(inclusive: bool, filename_pattern: str, lineno: int=None, all_frames: bool=False, domain: int=None)

对内存块的跟踪进行筛选。

See the fnmatch.fnmatch() function for the syntax of filename_pattern. The '.pyc' file extension is replaced with '.py'.

示例:

  • Filter(True, subprocess.__file__) only includes traces of the subprocess module

  • Filter(False, tracemalloc.__file__) excludes traces of the tracemalloc module

  • Filter(False, "<unknown>") excludes empty tracebacks

在 3.5 版更改: The '.pyo' file extension is no longer replaced with '.py'.

在 3.6 版更改: Added the domain attribute.

  • domain

    Address space of a memory block (int or None).

    tracemalloc uses the domain 0 to trace memory allocations made by Python. C extensions can use other domains to trace other resources.

  • inclusive

    If inclusive is True (include), only match memory blocks allocated in a file with a name matching filename_pattern at line number lineno.

    If inclusive is False (exclude), ignore memory blocks allocated in a file with a name matching filename_pattern at line number lineno.

  • lineno

    Line number (int) of the filter. If lineno is None, the filter matches any line number.

  • filename_pattern

    Filename pattern of the filter (str). Read-only property.

  • all_frames

    If all_frames is True, all frames of the traceback are checked. If all_frames is False, only the most recent frame is checked.

    This attribute has no effect if the traceback limit is 1. See the get_traceback_limit() function and Snapshot.traceback_limit attribute.

Frame

class tracemalloc.Frame

Frame of a traceback.

The Traceback class is a sequence of Frame instances.

  • filename

    文件名(字符串

  • lineno

    行号(整形

快照

class tracemalloc.Snapshot

Snapshot of traces of memory blocks allocated by Python.

The take_snapshot() function creates a snapshot instance.

  • compare_to(old_snapshot: Snapshot, key_type: str, cumulative: bool=False)

    Compute the differences with an old snapshot. Get statistics as a sorted list of StatisticDiff instances grouped by key_type.

    See the Snapshot.statistics() method for key_type and cumulative parameters.

    The result is sorted from the biggest to the smallest by: absolute value of StatisticDiff.size_diff, StatisticDiff.size, absolute value of StatisticDiff.count_diff, Statistic.count and then by StatisticDiff.traceback.

  • dump(filename)

    将快照写入文件

    使用 load() 重载快照。

  • filter_traces(filters)

    Create a new Snapshot instance with a filtered traces sequence, filters is a list of DomainFilter and Filter instances. If filters is an empty list, return a new Snapshot instance with a copy of the traces.

    All inclusive filters are applied at once, a trace is ignored if no inclusive filters match it. A trace is ignored if at least one exclusive filter matches it.

    在 3.6 版更改: DomainFilter instances are now also accepted in filters.

  • classmethod load(filename)

    从文件载入快照。

    另见 dump().

  • statistics(key_type: str, cumulative: bool=False)

    获取 Statistic 信息列表,按 key_type 分组排序:

    key_type

    描述

    ‘filename’

    文件名

    ‘lineno’

    文件名和行号

    ‘traceback’

    回溯

    If cumulative is True, cumulate size and count of memory blocks of all frames of the traceback of a trace, not only the most recent frame. The cumulative mode can only be used with key_type equals to 'filename' and 'lineno'.

    The result is sorted from the biggest to the smallest by: Statistic.size, Statistic.count and then by Statistic.traceback.

  • traceback_limit

    Maximum number of frames stored in the traceback of traces: result of the get_traceback_limit() when the snapshot was taken.

  • traces

    Traces of all memory blocks allocated by Python: sequence of Trace instances.

    The sequence has an undefined order. Use the Snapshot.statistics() method to get a sorted list of statistics.

统计

class tracemalloc.Statistic

统计内存分配

Snapshot.statistics() 返回 Statistic 实例的列表。.

参见 StatisticDiff 类。

  • count

    内存块数(整形)。

  • size

    Total size of memory blocks in bytes (int).

  • traceback

    Traceback where the memory block was allocated, Traceback instance.

StatisticDiff

class tracemalloc.StatisticDiff

Statistic difference on memory allocations between an old and a new Snapshot instance.

Snapshot.compare_to() returns a list of StatisticDiff instances. See also the Statistic class.

  • count

    Number of memory blocks in the new snapshot (int): 0 if the memory blocks have been released in the new snapshot.

  • count_diff

    Difference of number of memory blocks between the old and the new snapshots (int): 0 if the memory blocks have been allocated in the new snapshot.

  • size

    Total size of memory blocks in bytes in the new snapshot (int): 0 if the memory blocks have been released in the new snapshot.

  • size_diff

    Difference of total size of memory blocks in bytes between the old and the new snapshots (int): 0 if the memory blocks have been allocated in the new snapshot.

  • traceback

    Traceback where the memory blocks were allocated, Traceback instance.

跟踪

class tracemalloc.Trace

Trace of a memory block.

The Snapshot.traces attribute is a sequence of Trace instances.

在 3.6 版更改: Added the domain attribute.

  • domain

    Address space of a memory block (int). Read-only property.

    tracemalloc uses the domain 0 to trace memory allocations made by Python. C extensions can use other domains to trace other resources.

  • size

    Size of the memory block in bytes (int).

  • traceback

    Traceback where the memory block was allocated, Traceback instance.

回溯

class tracemalloc.Traceback

Sequence of Frame instances sorted from the oldest frame to the most recent frame.

A traceback contains at least 1 frame. If the tracemalloc module failed to get a frame, the filename "<unknown>" at line number 0 is used.

When a snapshot is taken, tracebacks of traces are limited to get_traceback_limit() frames. See the take_snapshot() function. The original number of frames of the traceback is stored in the Traceback.total_nframe attribute. That allows to know if a traceback has been truncated by the traceback limit.

The Trace.traceback attribute is an instance of Traceback instance.

在 3.7 版更改: Frames are now sorted from the oldest to the most recent, instead of most recent to oldest.

  • total_nframe

    Total number of frames that composed the traceback before truncation. This attribute can be set to None if the information is not available.

在 3.9 版更改: The Traceback.total_nframe attribute was added.

  • format(limit=None, most_recent_first=False)

    Format the traceback as a list of lines with newlines. Use the linecache module to retrieve lines from the source code. If limit is set, format the limit most recent frames if limit is positive. Otherwise, format the abs(limit) oldest frames. If most_recent_first is True, the order of the formatted frames is reversed, returning the most recent frame first instead of last.

    Similar to the traceback.format_tb() function, except that format() does not include newlines.

    示例:

    1. print("Traceback (most recent call first):")
    2. for line in traceback:
    3. print(line)

    输出:

    1. Traceback (most recent call first):
    2. File "test.py", line 9
    3. obj = Object()
    4. File "test.py", line 12
    5. tb = tracemalloc.get_object_traceback(f())