feat(benchmark): Add CPU logging to benchmark script and enhance JVM thread settings

Author: tae898

bindings/python/examples/benchmark-vector/benchmark_arcadedb_msmarco.py

@@ -27,6 +27,7 @@
 import json
 import re
 import shutil
+import threading
 import time
 from pathlib import Path
 from typing import Dict, List, Tuple
@@ -66,6 +67,27 @@ def timed_section(name: str, fn):
     return result, dur, start_rss, end_rss
 
 
+def start_cpu_logger(interval_sec: int = 2):
+    """Log process CPU% and RSS periodically without blocking main work."""
+
+    if not psutil:
+        return None
+
+    proc = psutil.Process()
+    proc.cpu_percent(None)  # prime
+    stop_event = threading.Event()
+
+    def _loop():
+        while not stop_event.wait(interval_sec):
+            cpu = proc.cpu_percent(None)
+            rss = proc.memory_info().rss / (1024 * 1024)
+            print(f"[cpu] {cpu:5.1f}% | rss={rss:8.1f} MB")
+
+    t = threading.Thread(target=_loop, daemon=True)
+    t.start()
+    return stop_event
+
+
 # -------------------------
 # dataset helpers (reuse MSMARCO shard layout)
 # -------------------------
@@ -435,6 +457,8 @@ def main():
     )
     args = ap.parse_args()
 
+    stop_cpu = start_cpu_logger(2)
+
     np.random.seed(args.seed)
     eval_k = 50
 
@@ -670,6 +694,8 @@ def record(name: str, result, dur, rss_start, rss_end):
                 record("close_db_final", {}, dur, r0, r1)
             except Exception:
                 pass
+        if stop_cpu:
+            stop_cpu.set()
 
     db_size = dir_size_mb(db_path)
     rss_after_vals = [

bindings/python/examples/benchmark-vector/run_arcadedb_sweep.sh

@@ -45,8 +45,9 @@ fi
 THREAD_ENV=""
 if [[ -n "$THREADS_PER_TASK" ]]; then
     THREAD_ENV="OMP_NUM_THREADS=${THREADS_PER_TASK} MKL_NUM_THREADS=${THREADS_PER_TASK} OPENBLAS_NUM_THREADS=${THREADS_PER_TASK} VECLIB_MAXIMUM_THREADS=${THREADS_PER_TASK} BLIS_NUM_THREADS=${THREADS_PER_TASK} NUMEXPR_NUM_THREADS=${THREADS_PER_TASK}"
-    # Restrict JVM ForkJoinPool for JVector parallelism
-    export ARCADEDB_JVM_ARGS="${ARCADEDB_JVM_ARGS:-} -Djava.util.concurrent.ForkJoinPool.common.parallelism=${THREADS_PER_TASK}"
+    # Restrict JVM pools for JVector: common pool, physical core count, and reported processors
+    JVM_THREAD_FLAGS="-Djava.util.concurrent.ForkJoinPool.common.parallelism=${THREADS_PER_TASK} -Djvector.physical_core_count=${THREADS_PER_TASK} -XX:ActiveProcessorCount=${THREADS_PER_TASK}"
+    export ARCADEDB_JVM_ARGS="${ARCADEDB_JVM_ARGS:-} ${JVM_THREAD_FLAGS}"
 fi
 
 BASE="${THREAD_ENV} python \"${BENCH_PY}\" --dataset-dir \"${DATASET_DIR}\" --db-root \"${DB_ROOT}\""

bindings/python/jar_exclusions.txt

@@ -6,5 +6,10 @@
 # Format: One pattern per line, comments start with #
 # Empty lines are ignored
 
-# gRPC wire protocol (~40MB) - not commonly needed for Python use cases
 arcadedb-grpcw-*.jar
+js-language-*.jar
+icu4j-*.jar
+truffle-*.jar
+regex-*.jar
+arcadedb-metrics-*.jar
+arcadedb-mongodbw-*.jar