fix(ci): Balance splits across benchmarking CI jobs according to the number of CPU cores (#5099)

ddyurchenko · web-flow · commit 435109b97c41 · 2025-01-13T16:49:31.000+01:00
* Fix how balancing over CPU cores works, so now tasks are split by variants instead of only directories
* Fix bug with taskset trying to assign to non-existing core
* Fix bug with off by 1 error in benchmarks count
* Fail job with exit code 1 if sirun could not start it
* Fail CI job when GROUP_SIZE is larger than number of CPU cores
diff --git a/benchmark/sirun/runall.sh b/benchmark/sirun/runall.sh
@@ -1,5 +1,7 @@
 #!/bin/bash
 
+set -e
+
 # Temporary until merged to master
 wget -O sirun.tar.gz https://github.com/DataDog/sirun/releases/download/v0.1.10/sirun-v0.1.10-x86_64-unknown-linux-musl.tar.gz \
 	&& tar -xzf sirun.tar.gz \
@@ -34,46 +36,49 @@ echo "using Node.js ${VERSION}"
 CPU_AFFINITY="${CPU_START_ID:-24}" # reset for each node.js version
 SPLITS=${SPLITS:-1}
 GROUP=${GROUP:-1}
-BENCH_COUNT=0
 
+BENCH_COUNT=0
 for D in *; do
   if [ -d "${D}" ]; then
-    BENCH_COUNT=$(($BENCH_COUNT+1))
+    cd "${D}"
+    variants="$(node ../get-variants.js)"
+    for V in $variants; do BENCH_COUNT=$(($BENCH_COUNT+1)); done
+    cd ..
   fi
 done
 
-# over count so that it can be divided by bash as an integer
-BENCH_COUNT=$(($BENCH_COUNT+$BENCH_COUNT%$SPLITS))
-GROUP_SIZE=$(($BENCH_COUNT/$SPLITS))
-
-run_all_variants () {
-  local variants="$(node ../get-variants.js)"
-
-  node ../squash-affinity.js
-
-  for V in $variants; do
-    echo "running ${1}/${V} in background, pinned to core ${CPU_AFFINITY}..."
-
-    export SIRUN_VARIANT=$V
-
-    (time node ../run-one-variant.js >> ../results.ndjson && echo "${1}/${V} finished.") &
-    ((CPU_AFFINITY=CPU_AFFINITY+1))
-  done
-}
+GROUP_SIZE=$(($(($BENCH_COUNT+$SPLITS-1))/$SPLITS)) # round up
 
 BENCH_INDEX=0
 BENCH_END=$(($GROUP_SIZE*$GROUP))
 BENCH_START=$(($BENCH_END-$GROUP_SIZE))
 
+if [[ ${GROUP_SIZE} -gt 24 ]]; then
+  echo "Group size ${GROUP_SIZE} is larger than available number of CPU cores on Benchmarking Platform machines (24 cores)"
+  exit 1
+fi
+
 for D in *; do
   if [ -d "${D}" ]; then
-    if [[ ${BENCH_INDEX} -ge ${BENCH_START} && ${BENCH_INDEX} -lt ${BENCH_END} ]]; then
-      cd "${D}"
-      run_all_variants $D
-      cd ..
-    fi
+    cd "${D}"
+    variants="$(node ../get-variants.js)"
+
+    node ../squash-affinity.js
+
+    for V in $variants; do
+      if [[ ${BENCH_INDEX} -ge ${BENCH_START} && ${BENCH_INDEX} -lt ${BENCH_END} ]]; then
+        echo "running $((BENCH_INDEX+1)) out of ${BENCH_COUNT}, ${D}/${V} in background, pinned to core ${CPU_AFFINITY}..."
+
+        export SIRUN_VARIANT=$V
+
+        (time node ../run-one-variant.js >> ../results.ndjson && echo "${D}/${V} finished.") &
+        ((CPU_AFFINITY=CPU_AFFINITY+1))
+      fi
+
+      BENCH_INDEX=$(($BENCH_INDEX+1))
+    done
 
-    BENCH_INDEX=$(($BENCH_INDEX+1))
+    cd ..
   fi
 done
 
