4 stable releases

new 1.2.1	Jan 11, 2026
1.2.0	Jul 20, 2025
1.1.0	Jul 20, 2025
1.0.0	Jul 20, 2025

#786 in Filesystem

MIT license

15KB
118 lines

rmtree

rmtree is a faster rm -rf.

rmtree speeds up recursive file and directory removal by performing filesystem traversals and removals in parallel.

# Install the "rmtree" command-line tool.
cargo install rmtree

rmtree --help
rmtree /large/nested/directory /trees/quickly

Documentation

Run rmtree --help for command-line usage details.

rmtree API documentation on docs.rs

Installation

Pre-built rmtree binaries

Pre-built rmtree binaries are statically linked using musl libc so that they can run on any Linux system.

Benchmarks

These timings were gathered on a 40 core Intel(R) Xeon(R) Gold 6242R CPU @ 3.10GHz. The storage was accessed over NFS against a live cluster from a Linux host.

A modestly-sized C++ project with build artifacts was used in these measurements. The test directory contains 1663 directories and 5822 files totalling 12 GB. This skews towards many small files and many directories.

Timings

All timings were performed with a warm NFS cache. As a read-only baseline, timings for find and sharkdp/fd are included.

Command	Time (s)
`fd`	0.249
`find`	0.448

The following table summarizes the timings for rm -rf from GNU coreutils 8.32 and rmtree with different thread settings.

Command	Time (s)	Speedup	Scaling	Normalized
`rm -rf`	25.697	1.000
`rmtree -t 1`	23.729	1.083	1.000	1.000
`rmtree -t 2`	11.146	2.305	2.129	1.064
`rmtree -t 3`	7.866	3.267	3.017	1.006
`rmtree -t 4`	5.476	4.693	4.333	1.083
`rmtree -t 5`	4.556	5.640	5.208	1.042
`rmtree -t 6`	4.041	6.359	5.872	0.978
`rmtree -t 7`	3.581	7.176	6.626	0.947
`rmtree -t 8`	3.039	8.456	7.808	0.976
`rmtree -t 9`	2.712	9.475	8.750	0.972
`rmtree -t 10`	2.635	9.752	9.005	0.901
`rmtree -t 11`	2.179	11.793	10.890	0.990
`rmtree -t 12`	2.039	12.603	11.638	0.970
`rmtree -t 16`	1.941	13.239	12.225	0.764
`rmtree -t 20`	1.634	15.726	14.522	0.726
`rmtree -t 24`	1.354	18.979	17.525	0.730
`rmtree -t 32`	1.655	15.527	14.338	0.448

Relative performance is the performance relative to rm -rf. Thread scaling is measured relative to rmtree -j1.

The sweet spot on the number of threads for maximum efficiency will vary depending on your specific storage and hardware. In this specific test, we can see negative consequences from adding more threads in the jump from 24 to 32 threads.

Code Status

rmtree is actively maintained and its core functionality is stable and feature-complete.

Dependencies

~2–14MB
~117K SLoC

bin+lib rmtree