v5.0.0-alpha.5

Changed

Removed the JSON metrics endpoint. Prometheus / OTel compatible endpoint should be
the primary way of collecting metrics from Rezolus.
Many metrics have had name changes to move attributes into labels.
Removed _total metrics in favor of aggregating across labels.
BlockIO latency and size histograms split by operation.
Syscall latency histogram split by syscall family.
Rezolus recorder moved into the main binary.
Replaced dynamic metrics with preallocated grouped metrics.
Removed dependency on openssl.
Metric time reflects start of sampling refresh.
Removed TCP connection state sampler.

Added

Timed collection in the rezolus record subcommand.
Output format selection in rezolus record subcommand.
cgroup cpu usage and performance counter metrics.
cgroup syscall counter metrics.
flight-recorder subcommand and systemd service.
TLB flush metrics.

Full Changelog: v4.1.2...v5.0.0-alpha.5

v5.0.0-alpha.4

Changed

• Removed the JSON metrics endpoint. Prometheus / OTel compatible endpoint should be
  the primary way of collecting metrics from Rezolus.
• Many metrics have had name changes to move attributes into labels.
• Removed _total metrics in favor of aggregating across labels.
• BlockIO latency and size histograms split by operation.
• Syscall latency histogram split by syscall family.
• Rezolus recorder moved into the main binary.
• Replaced dynamic metrics with preallocated grouped metrics.
• Removed dependency on openssl.
• Metric time reflects start of sampling refresh.
• Removed TCP connection state sampler.

Added

• Timed collection in the rezolus record subcommand.
• Output format selection in rezolus record subcommand.
• cgroup cpu usage and performance counter metrics.
• cgroup syscall counter metrics.
• flight-recorder subcommand and systemd service.
• TLB flush metrics.

Full Changelog: v4.1.2...v5.0.0-alpha.4

v4.1.2

Fixed

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

v4.1.1

Fixed

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

v4.1.0

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)

v4.0.0

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)

v3.18.1

Changed

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

v3.18.0

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 secondly 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)

v3.17.0

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)

v3.16.0

Fixed

• Listen address could not be changed (#264)

Added

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