memray

Profile memory usage

$ memray run [script.py]

$ memray run -o [output.bin] [script.py]

$ memray flamegraph [output.bin]

$ memray summary [output.bin]

$ memray run --live [script.py]

$ memray run --native [script.py]

$ memray tree [output.bin]

$ memray attach [pid]

memray [run] [flamegraph] [summary] [tree] [attach] [options] [args]

memray profiles Python memory allocations, tracking where memory is allocated and by which code. It helps identify memory leaks, high-water marks, and allocation patterns.
The tool tracks every allocation and deallocation, recording stack traces. Post-run analysis generates visualizations showing memory usage over time and by location.
Flame graphs show allocation call stacks, with width representing allocation size. They quickly identify which functions allocate most memory.
Live mode provides real-time visualization during execution. It's useful for understanding memory patterns in long-running applications.
Native tracking includes C extension allocations, not just Python objects. This reveals memory used by numpy, pandas, and other libraries with native components.
Process attachment profiles already-running applications, useful for production debugging without restart.

run SCRIPT

Profile script execution.

Generate flame graph HTML.

Show allocation summary.

Show tree view of allocations.

Show statistics.

Show table of allocations.

Attach to running process.

Output file for profile data.

Show live memory view.

Track native (C/C++) allocations.

Follow child processes.

Overwrite existing output.

Tracking overhead slows execution. Large profiles consume disk space. Native tracking requires debug symbols for readable output. Some allocations may be missed in highly optimized code.

memray was developed by Bloomberg and released as open source in 2022. It was designed for debugging memory issues in large-scale Python applications used in financial services.

py-spy(1), tracemalloc(3), valgrind(1), heaptrack(1)