CHANGELOG.md

Unreleased

5.7.0 - 2026-03-10

Added

• Ensure tooltip background is present so that its text remains legible. (#697)

Fixed

• Bump metriken-query to 0.2.0 which fixes histogram_quantile() semantics. (#699)
• Disable "Reset Zoom" button when already zoomed out. (#696)
• Fix charts shifting on hover.

5.6.0 - 2026-02-26

Added

• GPU GPM metrics for Hopper+ GPUs. (#689)
• Socket group and socket mode configuration for external metrics socket. (#687)
• Network ethtool sampler for ENA allowance counters. (#683)
• Unix domain socket external metrics ingestion. (#677)
• SparseCounterGroup for high-cardinality metrics. (#676)
• Per-task CPU usage metrics. (#675)
• GPU metrics dashboard section in viewer. (#673)
• CUPTI PM sampling for GPU metrics. (#671)

Changed

• Viewer uses metriken-query crate for TSDB and PromQL. (#678)
• Viewer style improvements. (#668 #674)

5.5.0 - 2026-01-23

Added

• Memory section and network errors added to the viewer. (#637)
• Use stability analysis to improve anomaly detection in mcp server. (#639)
• Macos GPU sampler. (#642)
• Hindsight HTTP endpoint for snapshotting. (#652)
• Perf counters for branch prediction and DTLB. (#655)
• Improved support for CPU L3 sampler. (#664)

5.4.0 - 2025-09-30

Fixed

• Fixes CPU IPNS, Frequency, and complex user queries in the viewer. (#629)

Added

• Homebrew and Debian Trixie packaging jobs (#595 #612)
• Installation script and documentation (#613)
• MCP server for LLM guided analysis of recordings. (#614 #617 #618 #622)
• Support for external BTF. (#620)

5.3.1 - 2025-08-21

Fixed

• Fixed RPM packaging to enable publishing to GCP Artifact Registry. (#609)

5.3.0 - 2025-08-20

Added

• Improved cgroup exploration and a query explorer for Rezolus Viewer. (#604)

5.2.3 - 2025-08-04

Fixes

• Fixes logging from libbpf so it is correctly directed to the debug log. (#597)

5.2.2 - 2025-07-28

Fixes

• Fixes cpu usage sampler reporting incorrect values on Ubuntu 24 (noble) when installed from our deb packages, or when built on systems with clang-18. (#585)

5.2.1 - 2025-07-02

Fixes

• Fixes scheduler runqueue sampler failing the BPF verifier on some systems since 5.2.0 (#576)

5.2.0 - 2025-07-01

Added

• Adds per-cgroup context switch, runqueue wait, and off-cpu time. Adds runqueue waiting time on a per-CPU basis. (#568)

Fixes

• Fixes CPU usage accounting bug when PIDs are re-used. (#571)

5.1.0 - 2025-06-24

Changed

• Replaced network_interfaces sampler with a BPF based sampler. Results in changes to some network metrics. Improves sampling latency. (#565)

Added

• Snapshot caching with TTL added to the agent. (#553)
• Rezolus viewer has more charts. (#554 #561)
• Realtime scheduling for the agent to improve sampling latency. (#559)
• Virtualization aware optimization for perf counter collection. (#548)
• Additional cgroup charts in Rezolus viewer. (#564)

Fixed

• Softirq time added to system time when running in idle task context. (#556)

5.0.1 - 2025-06-13

Fixed

• Rezolus Viewer served a 404 instead of using embedded index html. (#550)

5.0.0 - 2025-06-11

Changed

• Many metrics have been renamed to use labels to track different categories instead of encoding that in the metric name.
• Removed any system-wide total metrics when summing across some metric would give the same result.
• BlockIO latency and size distributions split by operation. (#380 #381)
• CPU performance (instructions, cycles) tracking changed to allow using the same perf counters for tracking these per-cgroup. May result in slight skew since these are no-longer a grouped read. (#383)
• Removed TCP connection state sampler due to high CPU usage. (#456)
• Prometheus exposition moved out into a separate service. (#470)

Added

• Recorder extended with additional options. (#387 #389)
• Hindsight recorder for always-on ringbuffer recording. (#409 #436 )
• Per-cgroup syscalls. (#419)
• Per-cgroup cpu usage. (#420 #425)
• SoftIRQ time tracked by IRQ kind. (#462)
• Sampler for TLB flush events. (#467)
• Softnet sampler for understanding kernel packet processing. (#480)
• Additional syscall groups. (#482 #503)
• CPU CFS throttling metrics for cgroups. (#488)
• CPU L3 Cache Hit/Access (AMD Zen only). (#487)
• CPU Migrations. (#491)
• Rezolus Viewer for opening parquet files. (#530)
• JSON endpoint for Rezolus Agent. (#532)
• BPF program stats. (#533)
• Per-sampler latency logging. (#535)

Fixed

• Set open file limit for agent sufficiently high for large systems. (#511)
• Changed recorder missed tick behavior from burst to skip. (#513)

4.1.2 - 2024-11-25

Fixed

• Fixed VM size for RPM packaging. (#376)
• Various dependencies updated.

4.1.1 - 2024-10-24

Fixed

• Fixed packaging workflow so DEB and RPM artifacts are published. (#365)

4.1.0 - 2024-10-23

Added

• rezolus-recorder which can connect to a running Rezolus instance and record metrics into a parquet file. (#356)

Fixed

• Configured listen address was ignored and a default was used. (#362)

4.0.0 - 2024-09-27

Changed

• All samplers are now async and run on-demand when a metrics endpoint is hit. Percentiles must now be calculated from the free-running histograms. (#334)
• BPF is now non-optional on Linux. (#336)
• Makes HTTP compression dependent on client accept headers, removes the need for separate configuration of compression. (#337)
• Removes histograms for counter and gauge metrics. (#338)
• Histogram grouping power reduced from 7 (0.781% error) to 3 (12.5% error). This reduces the memory footprint for each histogram. (#339)

Added

• Syscall samplers now track yield syscall. (#310)
• TCP active open (connect) latency now tracked. (#342)

3.18.1 - 2024-08-09

Changed

• Ubuntu Bionic dropped from packaging job. (#306)

3.18.0 - 2024-08-09

Fixed

• CPU per core cycle and instruction counters were incorrect. The system totals reflected the appropriate value. (#304)

Added

• Allow setting the metric snapshot interval to better address cases where metrics are not scraped on a per-second basis. (#286)
• Network carrier change metric. (#290)

Changed

• Explicitly use relaxed atomics in BPF samplers. (#276)
• Split syscall sampler into syscall_latency and syscall_counts to allow more fine-grained configuration and allow for reduced overhead. (#298)
• Split block_io latency sampler into block_io_latency and block_io_requests to allow more fine-grained configuration and allow for reduced overhead. (#299)

3.17.0 - 2024-06-27

Fixed

• Fixes sampler configuration inheritance for partially defined sampler config sections (#273)
• Linux BPF CPU usage sampler had idle time metric counting backwards (#281)

Added

• Fallback for linux network traffic sampler to get metrics from sysfs when BPF is not enabled or fails to initialize (#274)
• JSON stats endpoints (#277)

3.16.0 - 2024-05-14

Fixed

• Listen address could not be changed (#264)

Added

• CPU busy time metric for non-Prometheus endpoints (#256)

3.15.0 - 2024-05-09

Fixed

• CPU usage sampler could not be disabled (#259)
• Example config out-of-date (#260)

Added

• Example Grafana/Prometheus dashboard (#239)
• Off-CPU time added to the scheduler runqueue sampler (#240)
• Filesystem sampler added and tracks the number of open file descriptors (#242)
• Log level can now be set in the config (#246)
• Network interface statistics for tx/rx errors (drop, crc, ...) (#247)
• BPF can be enabled/disabled per-sampler or globally. (#258)

3.14.2 - 2024-04-18

Fixed

• CPU usage for soft and hard irq was incorrectly reported. (#236)

3.14.1 - 2024-04-16

Fixed

• CPU usage reporting via BPF would report CPU as always idle on some systems. (#233)

3.14.0 - 2024-04-03

Changed

• metriken crates updated which changes the msgpack output. (#224)

Fixed

• Dependency updates to address RUSTSEC-2024-0332.

3.13.0 - 2024-04-01

Changed

• Memory sampler was reporting memory usage stats in KiB, but with bytes for the unit metadata. This change corrects the sampler to report memory usage in bytes. This fix is disruptive as it will cause the memory stats to change. (#222)

3.12.0 - 2024-03-28

Added

• MacOS cpu usage sampling. (#203)
• Metric unit annotations are added and exposed as metadata.
• Logs version number on startup. (#213)

Fixed

• Incorrect summary stats (percentiles) were reported in version 3.10.2, 3.10.3, and 3.11.0. (#216)

3.11.0 - 2024-03-25

Changed

• Refactored the scheduler and syscall BPF samplers to reduce overheads. (#193 #195)

Added

• BlockIO thoughput and operation metrics using BPF. (#198)
• Network throughput and packet metrics using BPF. (#200)

Fixed

• Online CPU detection for CPU usage sampler needed a trimmed string. (#194)

3.10.3 - 2024-03-20

Fixed

• Fixes an incorrect calculation of the number of online CPUs in the BPF-based CPU usage sampler.

3.10.2 - 2024-03-20

Fixed

• Fixes a panic in the CPU perf event sampler due to a divide-by-zero. This occurs when there are no active perf event groups. (#185)

3.10.1 - 2024-03-20

Fixed

• Fixes per-CPU idle time accounting in the BPF-based sampler. Starting in release 3.9.0 these metrics incorrectly report no idle time. (#181)

3.10.0 - 2024-03-19

Added

• Additional system information fields including kernel version, CPU frequency details, network queues, and IRQ affinity. (#100)

Fixed

• Fixes a panic on some systems when perf counter initialization has failed. This bug was introduced in 3.9.0. (#175)
• Fixes CPU idle time accounting in the BPF-based sampler. In 3.9.0 the sampler incorrectly reports no idle time. (#176)

3.9.0 - 2024-03-15

Added

• CPU usage metrics are now collected via BPF when available. (#165)
• Perf event sampler can now initialize when only some counters are available. (#168)

3.8.0 - 2024-03-04

Added

• Allows Rezolus to run on MacOS though sampler support is limited.
• Provides msgpack exposition format as a more efficient exposition format.

Fixed

• Updates of various direct dependencies.

3.7.0 - 2023-12-21

Added

• Optional compression for HTTP exposition. (#128)
• Additional GPU metrics for utilization and energy consumption. (#138)

Fixed

• Duplicate metric name in Rezolus sampler. (#134)

3.6.1 - 2023-11-30

Fixed

• Fixed incorrect type annotation for CPU metrics (frequency, ipkc, ipus). (#98)
• Fixed under-reported TCP retransmits. (#121)
• Fixed TCP segment metrics. (#123)

3.6.0 - 2023-10-26

Added

• Allow configuration of individual samplers in the config file. This allows each sampler to be individually enabled/disabled and have its collection intervals adjusted.
• TCP connection state sampler which tracks the number of tcp connections in each state.
• Rezolus sampler which monitors resource utilization of Rezolus itself.
• Optional exposition of histogram buckets on the Prometheus/OpenTelemetry endpoint.
• Track latencies for each group of syscalls to help understand the breakdown of total syscall latency.

Fixed

• Corrected a length check of the mmap'd histogram regions. This fix enables the fast path for reading histogram data into userspace.

3.5.0 - 2023-10-16

Changed

• Updated metriken and replaced heatmaps with histograms. This reduces runtime resource utilization.

3.4.0 - 2023-10-10

Changed

• Moved to fetching multiple percentiles at once to reduce overhead.
• Refactor of the hardware info sampler into a separate crate to allow reuse and make improvements to that sampler.

Fixed

• Update warp to address RUSTSEC-2023-0065.

3.3.3 - 2023-08-08

Added

• Packaging support for aarch64

Fixed

• Updated dependencies to pull-in fixes and improvements.

3.3.2 - 2023-08-08

Fixed

• Fixed hardware info and cpu samplers on platforms which do not expose either die or node information in the topology, which may happen on ARM.
• Fixed BPF program generation to restore compatibility with clang 11.

3.3.1 - 2023-08-07

Fixed

• Fixed path inconsistency in Debian packaging.

3.3.0 - 2023-08-02

Added

• Added BTF type definitions for aarch64 target architecture.

Fixed

• Update dependencies to reduce overhead and pull-in bugfixes.
• Documentation improvements.

3.2.0 - 2023-07-26

Added

• Added a TCP packet latency sampler to measure the latency from packet being received to being processed by the userspace application.
• Added per-device metrics for GPU sampler.

3.1.0 - 2023-07-26

Added

• Added per-CPU metrics for usage, frequency, and perf counters.
• Added BPF to the set of default features.

3.0.0 - 2023-07-25

Changed

• Rewritten implementation of Rezolus using libbpf-rs and perf-event2 to provide a more modern approach to BPF and Perf Event instrumentation.