k8ika0s opened a new issue, #49280:
URL: https://github.com/apache/arrow/issues/49280

   ### Describe the bug, including details regarding any error messages, 
version, and platform.
   
   ### Describe the bug
   
   Two C++ unit-test paths use synthetic "impossibly large" allocations to 
assert `OutOfMemory`:
   
   - `TypedTestBuffer.ResizeOOM` in 
`/Users/kaitlyndavis/dev/github.com/k8ika0s/arrow_s390x/cpp/src/arrow/buffer_test.cc`
   - `TestMemoryPoolBase::TestOOM` (used by `arrow-misc-test`) in 
`/Users/kaitlyndavis/dev/github.com/k8ika0s/arrow_s390x/cpp/src/arrow/memory_pool_test.h`
   
   On some allocator/platform combinations (notably with `mimalloc` enabled), 
the near-`max` request can trigger allocator fatal behavior before Arrow can 
surface `Status::OutOfMemory`. This makes the test behavior allocator-dependent 
and can cause instability in CI/dev runs.
   
   ### To Reproduce
   
   Build C++ tests with mimalloc enabled and run OOM-focused cases, for example:
   
   ```bash
   cmake -S cpp -B build -DARROW_BUILD_TESTS=ON -DARROW_MIMALLOC=ON 
-DARROW_JEMALLOC=OFF
   cmake --build build --target arrow-buffer-test arrow-misc-test -j
   ./build/debug/arrow-buffer-test --gtest_filter='*ResizeOOM*'
   ./build/debug/arrow-misc-test --gtest_filter='*OOM*'
   ```
   
   ### Expected behavior
   
   Synthetic OOM tests should be deterministic and should not depend on 
allocator internals/fatal paths.
   
   ### Actual behavior
   
   With mimalloc, the "max-size" synthetic OOM request may not reliably produce 
Arrow OutOfMemory; behavior can depend on allocator failure mode.
   
   ### Proposed fix
   - Keep synthetic OOM tests, but clamp requests to a still-impossible size 
(e.g. 1 << 48, 256 TB) instead of raw near-max bounds.
   - Skip these synthetic OOM checks under mimalloc backend where allocator 
fatal path can bypass Arrow error propagation.
   
   This is test-only hardening (no production runtime code changes).
   
   ### Component(s)
   
   C++


-- 
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.

To unsubscribe, e-mail: [email protected]

For queries about this service, please contact Infrastructure at:
[email protected]

Reply via email to