Summary

New release added after a significant time since the previous one (02/2024), hence bumping to 0.5.0.
While the full changelog is fairly large the major changes are in performance counter support in the hbt/ part of the project.

Details

Following is not an exhaustive list but highlights some key PRs.
Dynolog

1. Core improvements on Dynolog to add scheduling of monitors using Ticks (experimental) #346
2. Dyno CPU time monitor from /proc/stat #347
3. Support rocky/CentOS7 in docker file #245

Hbt/aka performance counter library

1. ARM: perf event discovery from sysfs (#249 , #343)
2. AMD Zen5 support (#295 , #336 )
3. Fixes (#285 and more)

Full Changelog: v0.3.2...v0.5.0

Release files

• dynolog_0.5.0-0-amd64.deb : x86 debian binary
• dynolog_0.5.0_prometheus-0-amd64.deb : x86 debian binary with Prometheus support enabled in the build

Summary

This incremental release adds all of the networking metrics to Prometheus.

Details

1. Adding network metrics to prometheus logging #224
2. BPerf related updates to hbt library (#196 , #197 , #198)
3. Compiler warning fixes in hbt library code

Full Changelog: v0.3.1...v0.3.2

Summary

Includes iterative improvement to Prometheus logging support and updates to performance monitoring events library (hbt) for SPR and ARM.
PS: Packages with "_ods" suffix support logging to Meta ODS data store.

Release notes

dynolog

• Add more metrics and smooth out prometheus implementation (#181, #156 )
• Adds docker image for demo (#135)
• Improve dyno gputrace's error message (#178)

hbt (performance events library)

• Introduce core counters into BuiltinMetrics and respin generated events.(#123 , #124 , #126).
• Fix hostname read to correctly truncate on large hostnames (#175)
• Implement SW counter collection (#172)
• Support to identify ARM CPU family ( , #162) and refactor code generation

Full Changelog: v0.3.0...v0.3.1

Summary

Includes resource limits for systemd process, updates and minor fixes to hearbeat for AMD, ARM (avoid breaking ARM builds).
RPM and Debian packages are attached.
PS: Packages with "_ods" suffix support logging to Meta ODS data store.

Release notes

v0.2.3...v0.3.0

dynolog

• Add resource limits to avoid dynolog taking too high memory or cpu on a system (#125).
• Adds docker building flow to facilitate deployment without requirement to specific architecture (#135)

hbt (performance events library)

• Fix and add AMD L3 counters for Zen4 and Milan.(#123 , #124 , #126).
• Fixes for ARM builds (#127 , #130 , #131).
• Add thread level performance count reader (#128) and reducer capability (#133).

FBRelayLogger is now able to publish metrics from multiple devices, for instance in the gpu_monitor_loop we log metrics from potentially multiple GPUs.

Add logrotate based logging rotation to manage log file

• Added pause/resume functionality for DCGM module.