bloomberg
Bus Error on macOS When Using Memray with XGBoost (memray::intercept::calloc recursion crash)
Is there an existing issue for this?
- [x] I have searched the existing issues
Current Behavior
Thanks for all the great work on this tool.
When profiling an xgboost training call on macOS using memray, the process crashes with a bus error. After investigation, it appears to be caused by infinite recursion between Memray’s calloc interceptor and macOS’s thread-local variable resolution (dyld::ThreadLocalVariables).
Expected Behavior
A memray_profile.bin file created, without crashing.
Steps To Reproduce
Environment OS: macOS (Apple Silicon, M2)
Python: 3.12 (via uv and .venv)
Memray: 1.16.0
XGBoost: 2.0.3 (ARM64 native binary)
Allocator: pymalloc (default)
Repro Steps Here’s a minimal example that reproduces the issue:
import xgboost as xgb
from sklearn.datasets import load_breast_cancer
from sklearn.model_selection import train_test_split
X, y = load_breast_cancer(return_X_y=True)
X_train, X_test, y_train, y_test = train_test_split(X, y)
model = xgb.XGBClassifier(n_jobs=1)
model.fit(X_train, y_train) # ← this line causes the crash when Memray is tracking
When run with:
memray run --output memray-profile.bin example.py
...the process crashes with:
zsh: bus error memray run --output memray-profile.bin example.py
Memray Version
1.16.0
Python Version
3.12
Operating System
macOS
Anything else?
Summary: When profiling an xgboost training call on macOS using memray, the process crashes with a bus error. After investigation, it appears to be caused by infinite recursion between Memray’s calloc interceptor and macOS’s thread-local variable resolution (dyld::ThreadLocalVariables).
Environment OS: macOS (Apple Silicon, M2)
Python: 3.12 (via uv and .venv)
Memray: 1.16.0
XGBoost: 2.0.3 (ARM64 native binary)
Allocator: pymalloc (default)
Command:
memray run --output memray-profile.bin example.py
Repro Steps Here’s a minimal example that reproduces the issue:
import xgboost as xgb
from sklearn.datasets import load_breast_cancer
from sklearn.model_selection import train_test_split
X, y = load_breast_cancer(return_X_y=True)
X_train, X_test, y_train, y_test = train_test_split(X, y)
model = xgb.XGBClassifier(n_jobs=1)
model.fit(X_train, y_train) # ← this line causes the crash when Memray is tracking
When run with:
memray run --output memray-profile.bin example.py
...the process crashes with:
zsh: bus error memray run --output memray-profile.bin example.py
Diagnosis The crash trace shows repeated recursion between:
memray::intercept::calloc → _tlv_get_addr → memray::intercept::calloc
…ending in a fatal bus error.
From crash logs:
memray::intercept::calloc(unsigned long, unsigned long)
dyld::ThreadLocalVariables::instantiateVariable(...)
libdyld.dylib::_tlv_get_addr
... (repeats) This appears to be a recursion between:
Memray's calloc interception and macOS’s thread-local variable handling (triggered when xgboost uses threads or libomp)
Which re-enters calloc, again invoking Memray
Workarounds Works fine on Linux using the same script and versions.
Avoiding memray run and using memray.Tracker() manually still crashes.
Works with scalene or memory_profiler on macOS.
