Install codspeed-runner 5.0.2-alpha.1

Install prebuilt binaries via shell script

curl --proto '=https' --tlsv1.2 -LsSf https://github.com/CodSpeedHQ/codspeed/releases/download/v5.0.2-alpha.1/codspeed-runner-installer.sh | sh

Download codspeed-runner 5.0.2-alpha.1

🚀 Highlights

More precise measurements by default

cycle-estimation is now enabled by default, giving more precise benchmark cycle estimation in Simulation mode. See docs.

New: allocation exclusion

Opt in to exclude-allocations to remove allocator time from benchmarks in Simulation mode and reduce variance. See docs.

samply-based profiler by default on Linux

The walltime profiler now uses samply on all OSes. Set CODSPEED_WALLTIME_PROFILER=perf to restore the previous behavior.

⚠️ Breaking changes

• cycle-estimation is now enabled by default (see above)
• The samply-based profiler is now the default on Linux; set CODSPEED_WALLTIME_PROFILER=perf to revert

Details

🚀 Features

• Load eBPF through a delegated BPF token by @GuillaumeLagrange
• Enable cycle estimation and allocation exclusion by default by @not-matthias in #461
• Make samply profiler default for all OSes (#472) by @GuillaumeLagrange in #475

🐛 Bug Fixes

• Classify mapped libraries without init-userns privilege by @GuillaumeLagrange
• Disable allocator exclusion by default (#489) by @not-matthias in #489

💼 Other

• Bump the memtrack pin to 1.4.0 by @adriencaccia in #484

🏗️ Refactor

• Hook-based CPU isolation for the walltime runner by @GuillaumeLagrange

🧪 Testing

• Run every tracking test under both BPF variants by @GuillaumeLagrange

⚙️ Internals

• Bump submodule to reduce debug info server timeouts by @GuillaumeLagrange
• Make tests run release builds of the binaries by @GuillaumeLagrange in #456
• Update submodule to fix fmt and ignore ubuntu debuginfod serv by @GuillaumeLagrange

Install codspeed-runner 5.0.1

Install prebuilt binaries via shell script

curl --proto '=https' --tlsv1.2 -LsSf https://github.com/CodSpeedHQ/codspeed/releases/download/v5.0.1/codspeed-runner-installer.sh | sh

Download codspeed-runner 5.0.1

Install codspeed-runner 4.19.2-alpha.1

Install prebuilt binaries via shell script

curl --proto '=https' --tlsv1.2 -LsSf https://github.com/CodSpeedHQ/codspeed/releases/download/v4.19.2-alpha.1/codspeed-runner-installer.sh | sh

Download codspeed-runner 4.19.2-alpha.1

Install memtrack 1.4.0

Install prebuilt binaries via shell script

curl --proto '=https' --tlsv1.2 -LsSf https://github.com/CodSpeedHQ/codspeed/releases/download/memtrack-v1.4.0/memtrack-installer.sh | sh

Download memtrack 1.4.0

Release Notes

This release ships improvements for the memory instrument.

The memory instrument now lazily attaches to allocators, which lowers the startup time to a few seconds (down from minutes on certain systems). Additionally, we shipped a series of serialization improvements, removing in some large benchmarks 3+ minutes of extra CI time.

🐛 Bug Fixes

• Track successful posix_memalign allocations (#470) by @not-matthias in #470
• Prompt sudo on interactive stdin (#459) by @not-matthias in #459
• Reject empty command before executor setup (#467) by @not-matthias in #467

💼 Other

• Bump the memtrack pin to 1.3.0 by @adriencaccia in #474

Install codspeed-runner 4.19.1

Install prebuilt binaries via shell script

curl --proto '=https' --tlsv1.2 -LsSf https://github.com/CodSpeedHQ/codspeed/releases/download/v4.19.1/codspeed-runner-installer.sh | sh

Download codspeed-runner 4.19.1

Release Notes

This release adds a new --experimental-exclude-allocations feature to remove allocator time from Simulation benchmarks.

🚀 Features

• Remove eager allocator discovery by @not-matthias
• Attach allocator probes on demand via exec-mapping watcher by @not-matthias
• Add exec-mapping watcher BPF prog and inode maps by @not-matthias
• Add experimental --exclude-allocations setting by @not-matthias

🐛 Bug Fixes

• Join attach worker thread on drop, not only on finish() by @not-matthias in #445
• Fail loud when a watcher mapping is unresolved on a live process by @not-matthias
• Canonicalize map_files dirent name for low-address mappings by @not-matthias
• Classify unprefixed jemalloc builds as jemalloc by @not-matthias
• Fail closed when tracking_enabled lookup fails by @not-matthias
• Lower zstd compression level to -5 to reduce serialization overhead by @not-matthias
• Wire MemtrackWriter serializer through BufWriter to fix per-event zstd compression bottleneck by @not-matthias

💼 Other

• Let the tracker own the poller and drain on shutdown by @not-matthias
• Return the underlying sink from MemtrackWriter::finish by @not-matthias

🏗️ Refactor

• Split monolithic eBPF C and Rust into domain files by @not-matthias
• Split memtrack into module by @not-matthias
• Attach uprobes at resolved file offsets instead of func_name by @not-matthias
• Align cycle-estimation flag with experimental naming scheme by @not-matthias in #425

📚 Documentation

• Add crate AGENTS.md guide by @not-matthias in #455
• Stop hardcoding MCP tool schemas in the optimize skill by @fargito in #451

⚡ Performance

• Drain ringbuf bursts eagerly and reserve poller cores by @not-matthias in #436
• Grow the event ring buffer to 256 MiB by @not-matthias
• Wake the ringbuf consumer only past a data watermark by @not-matthias
• Pre-size and reuse the memtrack encode window buffer by @not-matthias
• Replace drain/writer threads with the encode pipeline by @not-matthias
• Add parallel zstd encode pipeline for artifact events by @not-matthias
• Detach BPF links in parallel at teardown by @not-matthias in #440
• Resolve allocator symbol offsets in parallel by @not-matthias

🧪 Testing

• Add dlopen/on-demand allocator tests + CI matrix by @not-matthias
• Benchmark encode_events with a realistic allocation trace by @not-matthias

⚙️ Internals

• Lower on-demand attach log to debug by @not-matthias in #462
• Install rust toolchain by explicit channel for older rustup by @not-matthias
• Also run runner-shared benchmarks in walltime mode by @not-matthias
• Dedup memtrack build deps into install-bpf-deps action by @not-matthias

Install codspeed-runner 4.19.0

Install prebuilt binaries via shell script

curl --proto '=https' --tlsv1.2 -LsSf https://github.com/CodSpeedHQ/codspeed/releases/download/v4.19.0/codspeed-runner-installer.sh | sh

Download codspeed-runner 4.19.0

Install codspeed-runner 4.18.5-alpha.1

Install prebuilt binaries via shell script

curl --proto '=https' --tlsv1.2 -LsSf https://github.com/CodSpeedHQ/codspeed/releases/download/v4.18.5-alpha.1/codspeed-runner-installer.sh | sh

Download codspeed-runner 4.18.5-alpha.1

Install memtrack 1.3.0

Install prebuilt binaries via shell script

curl --proto '=https' --tlsv1.2 -LsSf https://github.com/CodSpeedHQ/codspeed/releases/download/memtrack-v1.3.0/memtrack-installer.sh | sh

Download memtrack 1.3.0

Release Notes

💼 Other

• Bump pinned valgrind-codspeed to 3.26.0-0codspeed6 (#448) by @adriencaccia in #448
  • fix: hangs on ARM64 due to atomics by @not-matthias in CodSpeedHQ/valgrind-codspeed#23
  • fix(aspacem): raise amd64 managed address ceiling to 1 TiB by @not-matthias in CodSpeedHQ/valgrind-codspeed#22

Install codspeed-runner 4.18.4

Install prebuilt binaries via shell script

curl --proto '=https' --tlsv1.2 -LsSf https://github.com/CodSpeedHQ/codspeed/releases/download/v4.18.4/codspeed-runner-installer.sh | sh

Download codspeed-runner 4.18.4

Release Notes

🚀 Features

• Bump valgrind-codspeed to 3.26.0-0codspeed5 (#444) by @adriencaccia in #444
  • Fix valgrind on ARM by @not-matthias in CodSpeedHQ/valgrind-codspeed#21

Install codspeed-runner 4.18.3

Install prebuilt binaries via shell script

curl --proto '=https' --tlsv1.2 -LsSf https://github.com/CodSpeedHQ/codspeed/releases/download/v4.18.3/codspeed-runner-installer.sh | sh

Download codspeed-runner 4.18.3