You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Summary:
This diff reduce the execution time of Feedsim mini by eliminating fixed sleeps, replacing them with checking for QPS, and tunning configuration parameters.
### Key Changes:
**1. Reduced Test Durations:**
* `fixed_qps_duration`
* `warmup_time`
**2. Enhanced Warmup Control:**
* Added QPS threshold-based steady state detection (`-r` parameter)
* Added maximum warmup iterations parameter (`-x` parameter) with default of 60 iterations (configurable to 10 in jobs config)
* Dynamic warmup that stops when QPS increase falls below threshold (0.1) instead of relying solely on fixed time
**3. Improved Monitoring & Debugging:**
* Enhanced logging throughout the warmup process with iteration-by-iteration QPS and latency reporting
* Better state reporting for debugging benchmark execution
Rollback Plan:
Differential Revision: D81509229
printf"QPS steady state reached. QPS increase percentage (%.2f%%) is less than threshold (%.2f%%)\n"$qps_increase_percentage$qps_threshold>>$BREPS_LFILE
375
+
break
376
+
fi
377
+
else
378
+
# If previous_qps is 0, skip the check for this iteration
379
+
printf"Previous QPS is 0, skipping threshold check for iteration %d\n"$warmup_iteration>>$BREPS_LFILE
380
+
fi
381
+
fi
382
+
383
+
previous_qps=$current_qps
384
+
warmup_iteration=$((warmup_iteration +1))
385
+
386
+
# Wait between warmup iterations (except for the last one)
387
+
if [ $warmup_iteration-lt$max_warmup_iterations ];then
0 commit comments