Fix IDSelector memory leak with SWIG, add unit tests (#3704) #3810
+39
−26
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
|
This pull request was exported from Phabricator. Differential Revision: D61992599 |
mnorris11
pushed a commit
to mnorris11/faiss
that referenced
this pull request
Aug 29, 2024
…h#3810) Summary: Pull Request resolved: facebookresearch#3810 Pull Request resolved: facebookresearch#3704 Background -- Issue: facebookresearch#2996 Prior attempted fix: https://github.com/facebookresearch/faiss/pull/3007/files#diff-d704f33c46a4ef2936c8cf2a66b05f1993e25e79ee5c19d4b63c4e0cf46b0a42 Partially copied from facebookresearch#3139 with an additional unit test. It is a confirmed and reproducible memory leak every time. There is a workaround. See the comments on facebookresearch#2996. Relevant SWIG docs: https://www.swig.org/Doc4.1/SWIGDocumentation.html#Typemaps_nn2 Current status -- `buck test faiss/tests:test_search_params -- test_ownership_2` Test prints: without SWIG fix: `[0.49, 0.82, 1.15, 1.39, 1.67, 1.88, 2.16, 2.36, 2.65, 2.85, 3.13, 3.34, 3.62, 3.82, 4.11, 4.31, 4.6, 4.8, 5.08, 5.28]` with SWIG fix: `[0.52, 0.7, 0.7, 0.7, 0.7, 0.7, 0.7, 0.7, 0.7, 0.7, 0.7, 0.7, 0.7, 0.71, 0.71, 0.71, 0.71, 0.71, 0.71, 0.71]` Note: This test is not ideal. Ideally we could enable ASAN to catch these memory leaks. But ASAN does not seem to catch these Python memory leaks. Example: ``` def test_ownership_3(self): d = 32 quantizer = faiss.IndexFlatL2(d) quantizer.this.own(False) ``` The above test produces no ASAN error, even though the `quantizer` object leaks. Differential Revision: D61992599
mnorris11
force-pushed
the
export-D61992599
branch
from
b888ed1 to
cfa5848
Compare
|
This pull request was exported from Phabricator. Differential Revision: D61992599 |
|
Looking into errors. |
|
This pull request was exported from Phabricator. Differential Revision: D61992599 |
mnorris11
force-pushed
the
export-D61992599
branch
from
cfa5848 to
299f0b3
Compare
mnorris11
pushed a commit
to mnorris11/faiss
that referenced
this pull request
Sep 12, 2024
…h#3810) Summary: Pull Request resolved: facebookresearch#3810 Pull Request resolved: facebookresearch#3704 Background -- Issue: facebookresearch#2996 Prior attempted fix: https://github.com/facebookresearch/faiss/pull/3007/files#diff-d704f33c46a4ef2936c8cf2a66b05f1993e25e79ee5c19d4b63c4e0cf46b0a42 Partially copied from facebookresearch#3139 with an additional unit test. It is a confirmed and reproducible memory leak every time. There is a workaround. See the comments on facebookresearch#2996. Relevant SWIG docs: https://www.swig.org/Doc4.1/SWIGDocumentation.html#Typemaps_nn2 Current status -- `buck test faiss/tests:test_search_params -- test_ownership_2` Test prints: without SWIG fix: `[0.49, 0.82, 1.15, 1.39, 1.67, 1.88, 2.16, 2.36, 2.65, 2.85, 3.13, 3.34, 3.62, 3.82, 4.11, 4.31, 4.6, 4.8, 5.08, 5.28]` with SWIG fix: `[0.52, 0.7, 0.7, 0.7, 0.7, 0.7, 0.7, 0.7, 0.7, 0.7, 0.7, 0.7, 0.7, 0.71, 0.71, 0.71, 0.71, 0.71, 0.71, 0.71]` Note: This test is not ideal. Ideally we could enable ASAN to catch these memory leaks. But ASAN does not seem to catch these Python memory leaks. Example: ``` def test_ownership_3(self): d = 32 quantizer = faiss.IndexFlatL2(d) quantizer.this.own(False) ``` The above test produces no ASAN error, even though the `quantizer` object leaks. Why change HNSW test? -- This fix causes the HNSW test to fail with heap-use-after-free. This is because the index.storage.get_distance_computer() somehow gets freed during clone_index, but only when reassigning to the same variable like `index.storage = clone(index.storage)`. I checked in https://fburl.com/code/qw6fznjt, and it is non-null before returning on the CPP side. After adding the temp variable, I also had to set `index.own_fields = False`, otherwise we get a heap-use-after-free again due to it being deleted already. Differential Revision: D61992599
…h#3810) Summary: Pull Request resolved: facebookresearch#3810 Pull Request resolved: facebookresearch#3704 Background -- Issue: facebookresearch#2996 Prior attempted fix: https://github.com/facebookresearch/faiss/pull/3007/files#diff-d704f33c46a4ef2936c8cf2a66b05f1993e25e79ee5c19d4b63c4e0cf46b0a42 Partially copied from facebookresearch#3139 with an additional unit test. It is a confirmed and reproducible memory leak every time. There is a workaround. See the comments on facebookresearch#2996. Relevant SWIG docs: https://www.swig.org/Doc4.1/SWIGDocumentation.html#Typemaps_nn2 Current status -- `buck test faiss/tests:test_search_params -- test_ownership_2` Test prints: without SWIG fix: `[0.49, 0.82, 1.15, 1.39, 1.67, 1.88, 2.16, 2.36, 2.65, 2.85, 3.13, 3.34, 3.62, 3.82, 4.11, 4.31, 4.6, 4.8, 5.08, 5.28]` with SWIG fix: `[0.52, 0.7, 0.7, 0.7, 0.7, 0.7, 0.7, 0.7, 0.7, 0.7, 0.7, 0.7, 0.7, 0.71, 0.71, 0.71, 0.71, 0.71, 0.71, 0.71]` Note: This test is not ideal. Ideally we could enable ASAN to catch these memory leaks. But ASAN does not seem to catch these Python memory leaks. Example: ``` def test_ownership_3(self): d = 32 quantizer = faiss.IndexFlatL2(d) quantizer.this.own(False) ``` The above test produces no ASAN error, even though the `quantizer` object leaks. Why change HNSW test? -- This fix causes the HNSW test to fail with heap-use-after-free. This is because the index.storage.get_distance_computer() somehow gets freed during clone_index, but only when reassigning to the same variable like `index.storage = clone(index.storage)`. I checked in https://fburl.com/code/qw6fznjt, and it is non-null before returning on the CPP side. After adding the temp variable, I also had to set `index.own_fields = False`, otherwise we get a heap-use-after-free again due to it being deleted already. Differential Revision: D61992599
|
This pull request was exported from Phabricator. Differential Revision: D61992599 |
mnorris11
force-pushed
the
export-D61992599
branch
from
299f0b3 to
d5d8caa
Compare
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Summary:
Background
Issue: #2996
Prior attempted fix: https://github.com/facebookresearch/faiss/pull/3007/files#diff-d704f33c46a4ef2936c8cf2a66b05f1993e25e79ee5c19d4b63c4e0cf46b0a42
Partially copied from #3139 with an additional unit test.
It is a confirmed and reproducible memory leak every time. There is a workaround. See the comments on #2996.
Relevant SWIG docs: https://www.swig.org/Doc4.1/SWIGDocumentation.html#Typemaps_nn2
Current status
buck test faiss/tests:test_search_params -- test_ownership_2Test prints:
without SWIG fix:
[0.49, 0.82, 1.15, 1.39, 1.67, 1.88, 2.16, 2.36, 2.65, 2.85, 3.13, 3.34, 3.62, 3.82, 4.11, 4.31, 4.6, 4.8, 5.08, 5.28]with SWIG fix:
[0.52, 0.7, 0.7, 0.7, 0.7, 0.7, 0.7, 0.7, 0.7, 0.7, 0.7, 0.7, 0.7, 0.71, 0.71, 0.71, 0.71, 0.71, 0.71, 0.71]Note: This test is not ideal. Ideally we could enable ASAN to catch these memory leaks. But ASAN does not seem to catch these Python memory leaks. Example:
The above test produces no ASAN error, even though the
quantizerobject leaks.Differential Revision: D61992599