DCPerf mini: Shrinking Feedsim even further (#241)

Summary:

The diff introduces two key optimizations to reduce execution time:

### 1. **Dynamic Result Polling**

*   Replaced fixed wait times with intelligent polling that checks for QPS results every 100ms
*   Makes feedsim 10x more responsive by detecting results immediately when available
*   Maintains the same maximum wait time but exits early when results are ready
*   Eliminates unnecessary waiting periods

### 2. **Configurable Queue Drain Time**

*   Added new `-D` command-line option to configure post-experiment queue drain time
*   Previously hardcoded at 5 seconds, now user-adjustable (e.g., `-D 10` for 10 seconds)
*   Allows optimization for different workload requirements and benchmark scenarios

Differential Revision: D83279537

Author: marziehlenjaniMeta

benchpress/config/jobs.yml

@@ -476,17 +476,19 @@
     - '-x {max_warmup_iterations}'
     - '-S {graph_store_path}'
     - '-L {graph_load_path}'
+    - '-D {queue_drain_time}'
     - '-N'
     - '{extra_args}'
   vars:
     - 'num_instances=-1'
     - 'fixed_qps=100000'
-    - 'fixed_qps_duration=10'
+    - 'fixed_qps_duration=5'
     - 'warmup_time=4'
     - 'qps_percentage_threshold=-1'
     - 'max_warmup_iterations=-1'
     - 'graph_store_path=default_do_not_store'
     - 'graph_load_path=default_do_not_load'
+    - 'queue_drain_time=1'
     - 'extra_args='
   hooks:
     - hook: copymove

packages/feedsim/run.sh

@@ -72,6 +72,7 @@ Usage: ${0##*/} [OPTION]...
        QPS increase is less than this threshold percentage of the previous QPS.
     -x Maximum number of warmup iterations when using QPS threshold. Default: 10
     -N No retry mode. Skip sleep and PID checking in load test startup, break immediately without retrying.
+    -D Drain time in seconds. Time to wait for queue to drain after experiments. Default: 5
 EOF
 }
 
@@ -148,6 +149,9 @@ main() {
     local no_retry_mode
     no_retry_mode=""
 
+    local queue_drain_time
+    queue_drain_time="5"
+
     if [ -z "$IS_AUTOSCALE_RUN" ]; then
        echo > $BREPS_LFILE
     fi
@@ -214,6 +218,9 @@ main() {
             -N)
                 no_retry_mode="1"
                 ;;
+            -D)
+                queue_drain_time="$2"
+                ;;
             -h|--help)
                 show_help >&2
                 exit 1
@@ -224,7 +231,7 @@ main() {
         esac
 
         case $1 in
-            -t|-c|-s|-d|-p|-q|-o|-w|-i|-l|-S|-L|-r|-x)
+            -t|-c|-s|-d|-p|-q|-o|-w|-i|-l|-S|-L|-r|-x|-D)
                 if [ -z "$2" ]; then
                     echo "Invalid option: '$1' requires an argument" 1>&2
                     exit 1
@@ -350,7 +357,7 @@ main() {
         benchreps_tell_state "after fixed_qps_exp"
     fi
 
-    sleep 5 # wait for queue to drain
+    sleep "$queue_drain_time" # wait for queue to drain
     kill -SIGINT $LEAF_PID || true > /dev/null # SIGINT so exits cleanly
 }
 

packages/feedsim/third_party/src/scripts/search_qps.sh

@@ -148,8 +148,20 @@ run_loadtest() {
   kill -SIGINT $LOADTEST_PID
 
   # wait for results to show up and queries to drain
-  sleep $wait_time
-  benchreps_tell_state "after sleeping for wait_time=${wait_time} seconds"
+  iteration=0
+  while [ $iteration -lt $((wait_time * 10)) ]; do
+      sleep 0.1
+      iteration=$((iteration + 1))
+
+      # Check if QPS results are available in the output file
+      if grep -q "#: [0-9]\+.\([0-9]\+\)\? QPS" $tmp_file; then
+          actual_wait_time=$(echo "scale=1; ${iteration} * 0.1" | bc)
+          benchreps_tell_state "Results available after ${actual_wait_time} seconds"
+          break
+      fi
+  done
+  actual_wait_time=$(echo "scale=1; ${iteration} * 0.1" | bc)
+  benchreps_tell_state "after waiting for results (waited ${actual_wait_time} seconds, max wait_time=${wait_time})"
 
   # check file for QPS
   if grep -q "#: [0-9]\+.\([0-9]\+\)\? QPS" $tmp_file; then