Running benchmarks

0.9.117

Experimental: This feature is not yet stable

• Enable with: Add experimental = ["benchmarks"] to .config/nextest.toml, or set NEXTEST_EXPERIMENTAL_BENCHMARKS=1 in the environment
• Tracking issue: #2874

Nextest supports running benchmarks with the cargo nextest bench command. Using cargo nextest bench can be helpful in case your benchmarks require setup or wrapper scripts.

Supported benchmark harnesses include:

• Criterion.rs
• The default libtest benchmark runner, in nightly Rust.
• Any other benchmark harnesses that follow the custom test harness protocol.

Note

This page is about running benchmarks to measure performance. Benchmarks can also be run in test mode as regular tests via cargo nextest run, without requiring an experimental feature flag. For more, see Criterion benchmarks.

Benchmark-specific settings

Since benchmarks typically take longer than tests to run, nextest applies a different set of slow and global timeouts for them. Access these via the bench.slow-timeout and bench.global-timeout settings, respectively.

Setting timeouts for benchmarks

[profile.default]
# Set a global timeout of 2 hours for benchmarks.
bench.global-timeout = "2h"

[profile.default.overrides]
# Terminate this benchmark after 10 minutes.
filter = "test(bench_commands)"
bench.slow-timeout = { period = "60s", terminate-after = 10 }

The regular slow-timeout and global-timeout settings are ignored for benchmarks.

Options and arguments

TODO