Performance/Sysbench: baseline gating, MB/sec normalization, clock sanity + docs/YAML updates by smuppand · Pull Request #244 · qualcomm-linux/qcom-linux-testkit

smuppand · 2026-01-02T16:54:37Z

This PR refreshes the Sysbench_Performance suite end-to-end (runner, library helpers, config/docs, and CI job wiring).

Included changes:

Sysbench_Performance/run.sh: multi-thread runs via THREADS_LIST, per-iteration capture + AVG summary, baseline gating with configurable delta, CSV export, safer FileIO dataset handling, and early clock sanity check (ensure_reasonable_clock) to avoid epoch-timestamp issues.
Runner/utils/lib_performance.sh: add MB/sec normalization helper (sysbench_mib_to_mb) and sysbench parsing/gating helpers used by the runner for consistent KPI units.
Sysbench baseline config: align baseline keys/metrics for t1/t4 gating (time_sec lower-is-better; *_mbps higher-is-better) and keep fileio KPIs in MB/sec for consistency.
YAML job updates: expose sysbench knobs (threads list, iterations, delta, baseline path, output dir, fileio dir) and ensure artifacts/logs are collected.
README updates: document usage (env/CLI), baseline file format/keys, gating rules, and troubleshooting (clock sanity + timesync diagnostics).

Add sysbench parsers for time_sec and mem_mbps metrics. Add values helpers (append/avg) for iteration aggregation. Add fifo-based tee runner and CSV append helper for KPIs. Signed-off-by: Srikanth Muppandam <smuppand@qti.qualcomm.com>

Runner/suites/Performance/Sysbench_Performance/run.sh

Runner/suites/Performance/Sysbench_Performance/Sysbench_Performance.yaml

vnarapar · 2026-01-08T19:08:20Z

please re order commits, i would suggest this way
7aa3348 - perf(lib_performance): add sysbench helper APIs
d5f5ec3 - perf(sysbench): add baseline KPI config
10fa4bd - functestlib: expand ensure_reasonable_clock for no-network systems
cc987f2 - perf(sysbench): add Sysbench_Performance runner
10de338 - docs(sysbench): add usage and KPI guidance
5503f55 - lava(sysbench): add Sysbench_Performance job template

Add baseline KPIs for CPU/Memory/Threads/Mutex averages. Document thread-key naming used by the gate evaluator. Enable PASS/FAIL comparisons using delta threshold in run.sh. Signed-off-by: Srikanth Muppandam <smuppand@qti.qualcomm.com>

- Log timedatectl status/show-timesync/timesync-status outputs when epoch detected - Log systemctl status for systemd-timesyncd for debug - Try RTC restore (hwclock / rtc0 since_epoch) before falling back - Seed time from kernel build timestamp (uname -v / /proc/version) if available - Continue gracefully if clock remains invalid Signed-off-by: Srikanth Muppandam <smuppand@qti.qualcomm.com>

Run CPU/Memory/Threads/Mutex workloads with iterations support. Capture per-iteration logs while streaming output to console. Compute KPI averages and optionally gate against baseline+delta. Signed-off-by: Srikanth Muppandam <smuppand@qti.qualcomm.com>

Document CLI/env options, output artifacts, and examples. Explain baseline file format and recommended delta threshold. Describe CSV/KPI summary fields for CI and local runs. Signed-off-by: Srikanth Muppandam <smuppand@qti.qualcomm.com>

smuppand · 2026-01-09T06:48:51Z

please re order commits, i would suggest this way 7aa3348 - perf(lib_performance): add sysbench helper APIs d5f5ec3 - perf(sysbench): add baseline KPI config 10fa4bd - functestlib: expand ensure_reasonable_clock for no-network systems cc987f2 - perf(sysbench): add Sysbench_Performance runner 10de338 - docs(sysbench): add usage and KPI guidance 5503f55 - lava(sysbench): add Sysbench_Performance job template

Yes, thanks for pointing out the order makes sense. Updated the order.

Provide LAVA YAML to run sysbench suite on target devices. Expose key parameters (threads, iterations, delta, taskset). Store logs and summaries as job artifacts for analysis. Signed-off-by: Srikanth Muppandam <smuppand@qti.qualcomm.com>

vnarapar

LGTM

perf(lib_performance): add sysbench helper APIs

7aa3348

Add sysbench parsers for time_sec and mem_mbps metrics. Add values helpers (append/avg) for iteration aggregation. Add fifo-based tee runner and CSV append helper for KPIs. Signed-off-by: Srikanth Muppandam <smuppand@qti.qualcomm.com>

smuppand requested review from Vkenjam, bhargav0610 and vnarapar January 2, 2026 16:54

smuppand force-pushed the sysbench branch from f986e45 to 10fa4bd Compare January 2, 2026 16:57