Version 2.4.4

memtier_benchmark 2.4.4 is a maintenance release on the 2.4 line.

Fixes

• Correct Ops/sec for sub-millisecond aggregate spans. run_stats::timeval_factorial_average averaged tv_sec/tv_usec independently with truncation, so aggregate durations under ~1 ms could round to 0 µs and make the summary report Ops/sec = 0 (and Hits/sec, KB/sec) even with Count > 0 and a non-zero average latency. Now normalized through int64_t microseconds. (#463, #465)
• RPM repository is publicly readable again. The YUM publish step uploaded objects with a private ACL, causing 403 AccessDenied for packages.redis.io/rpm/** (repo metadata and packages). Objects are now uploaded with public-read, matching the DEB pipeline. (#500, #501/#502)
• --monitor-input: skip HELLO lines while loading and bound the fuzz payload. (#496/#497)
• monitor-input fuzz harness: fix 5 pre-existing nightly failures. (#475/#495)

CI / test hardening

• Backport nightly fuzz/soak harness fixes; add an opt-in timeout to the soak Benchmark.run() path. (#493, #499)

Observability

• cluster-mode: report endpoint count and connection fan-out. (#489/#490)

Full changelog: 2.4.3...2.4.4