Repository navigation
Highlights
One runner binary
codspeed exec and the memory instrument no longer download a separate tool on their first run: exec-harness and memtrack now ship inside the runner (#531).
Configurable walltime unwinding
--walltime-unwinding-mode and --walltime-stack-sample-size (or CODSPEED_WALLTIME_UNWINDING_MODE and CODSPEED_WALLTIME_STACK_SAMPLE_SIZE) set how walltime profiles unwind stacks. --perf-unwinding-mode keeps working as a deprecated alias (#586).
Experimental
- Simulation:
--experimental-rep-bulk-cost(orCODSPEED_EXPERIMENTAL_REP_BULK_COST), combined with--cycle-estimation, prices amd64rep movsandrep stoscopies from measured bulk costs instead of per element (#620, #60). --mode telemetrycollects the OpenTelemetry spans and logs your benchmarks export and uploads them with the results. Combine it with a measuring mode, e.g.--mode telemetry,walltime(#589).
Fixes
- Simulation: report the correct main thread stack size to programs (#62).
- Walltime: Go and Java benchmarks on Linux use frame-pointer unwinding whichever profiler runs (#603).
- Walltime and Memory: glibc frames keep the same name across runs, so differential flamegraphs no longer flag them as new or dropped (#601).
- Exec: Node.js analysis runs use the same V8 flags as the codspeed-node integration (#595).
- Run:
codspeed run -- <command> <args>keeps each argument as your shell parsed it, so a quoteda|bis no longer split into a pipe. Shell syntax spread across several arguments must now be one quoted string (#622). - Security: bump rustls-webpki to address RUSTSEC-2026-0382 and RUSTSEC-2026-0383 (#607).
Release Notes
🚀 Features
- Add --experimental-rep-bulk-cost flag (#620) by @not-matthias in #620
- Add codspeed-write-benchmarks skill by @not-matthias
- Collect the OTLP data exported by the benchmarks (#589) by @lvaroqui in #589
- Declare benchmarks run in another process by @lvaroqui in #531
- Bundle memtrack into the runner binary by @GuillaumeLagrange
- Bundle exec-harness into the runner binary by @GuillaumeLagrange
- Track subprocesses instead of preloading a library by @GuillaumeLagrange
- Add --walltime-unwinding-mode with DWARF stack size by @not-matthias
🐛 Bug Fixes
- Quote each argument of a multi-argument command (#622) by @not-matthias in #622
- Compile node analysis workloads eagerly with sparkplug (#595) by @not-matthias in #595
- Infer Go and Java unwinding for both profilers by @not-matthias in #603
- Bump merged sampler and align unwinding arguments by @not-matthias in #586
- Pass stack sample size as --unwinding-stack-size by @not-matthias
- Install libc6-dbg during memory and walltime setup (#601) by @not-matthias in #601
💼 Other
- Inline aarch64 atomics in musl C code instead of linking libgcc by @adriencaccia
- Support building against musl by @GuillaumeLagrange
- Bump rustls-webpki and webpki root certs (#607) by @GuillaumeLagrange in #607
- Bump j178/prek-action from 2.0.4 to 3.0.0 (#495) by @dependabot[bot] in #495
🏗️ Refactor
- Move the CLI into the library by @GuillaumeLagrange
- Scope the rolling buffer display to the run environment (#506) by @GuillaumeLagrange in #506
📚 Documentation
- Mention benchmark writing in the plugin description by @not-matthias in #621
- Point optimize at codspeed-write-benchmarks by @not-matthias
- Point setup-harness at codspeed-write-benchmarks by @not-matthias
⚡ Performance
- Read the memtrack event stream through a 1 MiB buffer (#616) by @not-matthias in #616
🧪 Testing
- Run the executor tests in release by @GuillaumeLagrange
⚙️ Internals
- Bump pinned valgrind-codspeed to 3.26.0-0codspeed14 (#626) by @adriencaccia in #626
- fix(callgrind): model REP copies with calibrated bulk costs by @not-matthias in CodSpeedHQ/valgrind-codspeed#60
- Release the runner with dist's defaults (#625) by @adriencaccia in #625
- Bump cargo-dist to 0.34.0 by @GuillaumeLagrange in #623
- Build the macOS release on macos-latest by @adriencaccia
- Run the release build on pull requests by @adriencaccia
- Bump pinned valgrind-codspeed to 3.26.0-0codspeed13 (#618) by @adriencaccia in #618
- fix(initimg): keep glibc's reported main stack size near RLIMIT_STACK by @not-matthias in CodSpeedHQ/valgrind-codspeed#62
- Prime one shared rust-cache for the ubuntu-26.04 x64 jobs by @GuillaumeLagrange in #608
- Share one rust-cache entry per os across bpf-tests shards by @GuillaumeLagrange
Install codspeed-runner 5.7.0
Install prebuilt binaries via shell script
curl --proto '=https' --tlsv1.2 -LsSf https://github.com/CodSpeedHQ/codspeed/releases/download/v5.7.0/codspeed-runner-installer.sh | shDownload codspeed-runner 5.7.0
| File | Platform | Checksum |
|---|---|---|
| codspeed-runner-aarch64-apple-darwin.tar.gz | Apple Silicon macOS | checksum |
| codspeed-runner-aarch64-unknown-linux-musl.tar.gz | ARM64 MUSL Linux | checksum |
| codspeed-runner-x86_64-unknown-linux-musl.tar.gz | x64 MUSL Linux | checksum |