A low-overhead memory profiler.
Project description
mprofile
A low-overhead sampling memory profiler for Python, derived from heapprof, with an interface similar to tracemalloc. mprofile attempts to give results comparable to tracemalloc, but uses statistical sampling to lower memory and CPU overhead. The sampling algorithm is the one used by tcmalloc and Golang heap profilers.
Installation & usage
-
Install the profiler package using PyPI:
pip3 install mprofile
-
Enable the profiler in your application, get a snapshot of (sampled) memory usage:
import mprofile mprofile.start(sample_rate=128 * 1024) snap = mprofile.take_snapshot()
See the tracemalloc for API documentation. The API and objects returned by mprofile are compatible.
Compatibility
mprofile is compatible with Python >= 3.4. It can also be used with earlier versions of Python, but you must build CPython from source and apply the pytracemalloc patches.
Benchmarks
We are primarily interested in profiling the memory usage of webservers, so used the tornado_http
benchmark from pyperformance to estimate overhead.
mprofile has similar performance to tracemalloc when comprehensively tracing all allocations, but when statistical sampling is used, the overhead is significantly reduced.
In addition, mprofile interns call stacks in a tree data structure that reduces memory overhead of storing the traces.
With the recommended setting of sample_rate=128kB
, we observe ~5% slow down in the tornado_http
benchmark.
TODO: Run the full pyperformance suite of benchmarks.
Baseline
Python 2.7.16, no profiling:
tornado_http: Mean +- std dev: 664 ms +- 30 ms
Maximum resident set size (kbytes): 39176
tracemalloc
Python 2.7.16, tracemallocframes=128:
tornado_http: Mean +- std dev: 1.74 sec +- 0.04 sec
Maximum resident set size (kbytes): 43752
# Saving only one frame in each stack trace rather than full call stacks.
Python 2.7.16, tracemallocframes=1:
tornado_http: Mean +- std dev: 960 ms +- 30 ms
Maximum resident set size (kbytes): 40000
mprofile
Python 2.7.16, mprofileframes=128, mprofilerate=1 (i.e. tracemalloc):
tornado_http: Mean +- std dev: 1.78 sec +- 0.05 sec
Maximum resident set size (kbytes): 40588
Python 2.7.16, mprofileframes=128, mprofilerate=1024:
tornado_http: Mean +- std dev: 888 ms +- 28 ms
Maximum resident set size (kbytes): 39752
Python 2.7.16, mprofileframes=128, mprofilerate=128 * 1024:
tornado_http: Mean +- std dev: 700 ms +- 26 ms
Maximum resident set size (kbytes): 39388
# Saving only one frame in each stack trace rather than full call stacks.
Python 2.7.16, mprofileframes=1, mprofilerate=1 (i.e. tracemalloc):
tornado_http: Mean +- std dev: 890 ms +- 19 ms
Maximum resident set size (kbytes): 40152
Python 2.7.16, mprofileframes=1, mprofilerate=1024:
tornado_http: Mean +- std dev: 738 ms +- 24 ms
Maximum resident set size (kbytes): 39568
Python 2.7.16, mprofileframes=1, mprofilerate=128 * 1024:
tornado_http: Mean +- std dev: 678 ms +- 22 ms
Maximum resident set size (kbytes): 39328
Developer notes
Run the unit tests:
bazel test --test_output=streamed //src:profiler_test
Run the benchmarks:
bazel test -c opt --test_output=streamed //src:profiler_bench
Run the end-to-end (Python) tests:
bazel test --config asan --test_output=streamed //test:*
Run tests with ASAN and UBSAN:
bazel test --config asan --test_output=streamed //src:* //test:*
Contributing
Pull requests and issues are welcomed!
License
mprofile is released under the MIT License and incorporates code from heapprof, which is also released under the MIT license.
Project details
Release history Release notifications | RSS feed
Download files
Download the file for your platform. If you're not sure which to choose, learn more about installing packages.
Source Distribution
Built Distributions
File details
Details for the file mprofile-0.0.15.tar.gz
.
File metadata
- Download URL: mprofile-0.0.15.tar.gz
- Upload date:
- Size: 126.1 kB
- Tags: Source
- Uploaded using Trusted Publishing? No
- Uploaded via: twine/4.0.0 CPython/3.10.7
File hashes
Algorithm | Hash digest | |
---|---|---|
SHA256 | fb21d99dd99bc4672f057ac53bc6bd2ddc2139a18b83bc4565a0303ebd5b7efd |
|
MD5 | 8ccfa9c642d8648adf30aa7fd3d2f9bd |
|
BLAKE2b-256 | f906a2a99bd3caf8daa6ce17154a19a2e05412496379f49cf28f654033072d4c |
Provenance
File details
Details for the file mprofile-0.0.15-cp312-cp312-manylinux_2_17_x86_64.manylinux2014_x86_64.whl
.
File metadata
- Download URL: mprofile-0.0.15-cp312-cp312-manylinux_2_17_x86_64.manylinux2014_x86_64.whl
- Upload date:
- Size: 837.4 kB
- Tags: CPython 3.12, manylinux: glibc 2.17+ x86-64
- Uploaded using Trusted Publishing? No
- Uploaded via: twine/4.0.2 CPython/3.11.4
File hashes
Algorithm | Hash digest | |
---|---|---|
SHA256 | 6ad8ca31cc9b923be9e2ef35db83b0a6f2df14289d604577a9d09293b09957ae |
|
MD5 | 00496e41f41c5674055c08bfc4dd5fa5 |
|
BLAKE2b-256 | a4610aa0d3a9c454b9b7b880173706091e273634aa17a63f0e723eb0af4b69f5 |
Provenance
File details
Details for the file mprofile-0.0.15-cp311-cp311-manylinux_2_17_x86_64.manylinux2014_x86_64.whl
.
File metadata
- Download URL: mprofile-0.0.15-cp311-cp311-manylinux_2_17_x86_64.manylinux2014_x86_64.whl
- Upload date:
- Size: 835.4 kB
- Tags: CPython 3.11, manylinux: glibc 2.17+ x86-64
- Uploaded using Trusted Publishing? No
- Uploaded via: twine/4.0.0 CPython/3.10.7
File hashes
Algorithm | Hash digest | |
---|---|---|
SHA256 | e33edf1d469723d33184e14a757aa3b202f54afb991da1d0b7f7fba5c49aef27 |
|
MD5 | 6344b2925219c21b02e62af286b385b2 |
|
BLAKE2b-256 | 6950e859b298acda9e01fe018826d8961d6ac3b4a32b499ded1c9cc49793c131 |
Provenance
File details
Details for the file mprofile-0.0.15-cp310-cp310-manylinux_2_17_x86_64.manylinux2014_x86_64.whl
.
File metadata
- Download URL: mprofile-0.0.15-cp310-cp310-manylinux_2_17_x86_64.manylinux2014_x86_64.whl
- Upload date:
- Size: 833.8 kB
- Tags: CPython 3.10, manylinux: glibc 2.17+ x86-64
- Uploaded using Trusted Publishing? No
- Uploaded via: twine/4.0.0 CPython/3.10.7
File hashes
Algorithm | Hash digest | |
---|---|---|
SHA256 | 31c893321be93910e80052d0e7dfc19c8083db153cd8b2bb31498dcd920a7da9 |
|
MD5 | 3c2d5b145b26f85cccdd9304c1c3e200 |
|
BLAKE2b-256 | 67c9909a46d1573198685be8fb3590396e02bc2cb195668c5b11ea7d86bb21c9 |
Provenance
File details
Details for the file mprofile-0.0.15-cp39-cp39-manylinux_2_17_x86_64.manylinux2014_x86_64.whl
.
File metadata
- Download URL: mprofile-0.0.15-cp39-cp39-manylinux_2_17_x86_64.manylinux2014_x86_64.whl
- Upload date:
- Size: 833.2 kB
- Tags: CPython 3.9, manylinux: glibc 2.17+ x86-64
- Uploaded using Trusted Publishing? No
- Uploaded via: twine/4.0.0 CPython/3.10.7
File hashes
Algorithm | Hash digest | |
---|---|---|
SHA256 | 6f5d4d7fad4005abae0f16f76ee234cffdad7cb9d763a103672b5ea6cd3f732d |
|
MD5 | 7832a6c7083a830eb821bc0f0eeb0d44 |
|
BLAKE2b-256 | 3779d3cf2183fcae42197b576ce4fecfbe29226e3862318d2471f8b42e08ada2 |
Provenance
File details
Details for the file mprofile-0.0.15-cp38-cp38-manylinux_2_17_x86_64.manylinux2014_x86_64.whl
.
File metadata
- Download URL: mprofile-0.0.15-cp38-cp38-manylinux_2_17_x86_64.manylinux2014_x86_64.whl
- Upload date:
- Size: 833.6 kB
- Tags: CPython 3.8, manylinux: glibc 2.17+ x86-64
- Uploaded using Trusted Publishing? No
- Uploaded via: twine/4.0.0 CPython/3.10.7
File hashes
Algorithm | Hash digest | |
---|---|---|
SHA256 | 88218635ec951f1e449e23e96e0bf40b43a2ae3ca3b5c9f2506eb992515fca52 |
|
MD5 | 677471ab74ba6356733fc8d097516a0e |
|
BLAKE2b-256 | 71b51db1430c5aeb79aae47927cb2a82f551e5d12ddbd7e5f23b95a93aa8bf95 |
Provenance
File details
Details for the file mprofile-0.0.15-cp37-cp37m-manylinux_2_17_x86_64.manylinux2014_x86_64.whl
.
File metadata
- Download URL: mprofile-0.0.15-cp37-cp37m-manylinux_2_17_x86_64.manylinux2014_x86_64.whl
- Upload date:
- Size: 832.7 kB
- Tags: CPython 3.7m, manylinux: glibc 2.17+ x86-64
- Uploaded using Trusted Publishing? No
- Uploaded via: twine/4.0.0 CPython/3.10.7
File hashes
Algorithm | Hash digest | |
---|---|---|
SHA256 | 2bd8efc8d898d28181c404bfe304f13fd87ab8a439b6034235b5ea0f25670127 |
|
MD5 | cf8f4e62dc0e61af1a6fdaf433774a90 |
|
BLAKE2b-256 | 6a2f3ea66c1fda13de4efbaab2d8522a7700112595e2103d4be22d303502ede4 |
Provenance
File details
Details for the file mprofile-0.0.15-cp36-cp36m-manylinux_2_17_x86_64.manylinux2014_x86_64.whl
.
File metadata
- Download URL: mprofile-0.0.15-cp36-cp36m-manylinux_2_17_x86_64.manylinux2014_x86_64.whl
- Upload date:
- Size: 832.8 kB
- Tags: CPython 3.6m, manylinux: glibc 2.17+ x86-64
- Uploaded using Trusted Publishing? No
- Uploaded via: twine/4.0.0 CPython/3.10.7
File hashes
Algorithm | Hash digest | |
---|---|---|
SHA256 | c544275c4d1373696987bedb516a36e56600ef43ab0920cf3fb53f18344190a6 |
|
MD5 | 03e2c3b16171f62080d0e81a4170afbf |
|
BLAKE2b-256 | 57c5a82e119d6f3fe374a224c1dfc39081402fb78bc1ae95dd1d779162961c12 |