Add new cli options to control the benchmark iterations and warmup's

README.md

@@ -110,6 +110,17 @@ While benchmarking an indexing side change, you might want to recreate the index
 python src/python/localrun.py -source wikimediumall -r
 ```
 
+For quick patch testing, you can control the number of JVM iterations and query repetitions to speed up the benchmark:
+```bash
+# Quick test: 5 JVM iterations, 10 query repetitions per JVM
+python src/python/localrun.py -source wikimediumall -iterations 5 -warmups 10
+
+# Full benchmark (default): 20 JVM iterations, 20 query repetitions per JVM
+python src/python/localrun.py -source wikimediumall -iterations 20 -warmups 20
+```
+
+**Note:** The `-iterations` parameter controls how many separate JVM processes are launched (default: 20), and `-warmups` controls how many times each query runs within a single JVM (default: 20). Running with default settings (20×20) provides the most statistically reliable results and recommended for benchmarks testing to get a complete picture. For quick patch validation, reducing these values significantly speeds up testing.
+
 For details on all the available options, use the `-h` or `--help` parameter.
 
 # Running the geo benchmark

src/python/example.py

@@ -29,12 +29,14 @@
   parser.add_argument("-b", "--baseline", default=os.environ.get("BASELINE") or "lucene_baseline", help="Path to lucene repo to be used for baseline")
   parser.add_argument("-c", "--candidate", default=os.environ.get("CANDIDATE") or "lucene_candidate", help="Path to lucene repo to be used for candidate")
   parser.add_argument("-r", "--reindex", action="store_true", help="Reindex data for candidate run")
+  parser.add_argument("-iterations", "--iterations", default=20, type=int, help="Number of JVM iterations (separate JVM processes, default: 20)")
+  parser.add_argument("-warmups", "--warmups", default=20, type=int, help="Number of times each query runs within a single JVM for warmup (default: 20)")
   args = parser.parse_args()
   print("Running benchmarks with the following args: %s" % args)
 
   sourceData = competition.sourceData(args.source)
   countsAreCorrect = args.searchConcurrency != 0
-  comp = competition.Competition(verifyCounts=not countsAreCorrect)
+  comp = competition.Competition(verifyCounts=not countsAreCorrect, jvmCount=args.iterations, taskRepeatCount=args.warmups)
 
   index = comp.newIndex(
     args.baseline,

src/python/nightlyBench.py

@@ -151,7 +151,7 @@ def buildIndex(r, runLogDir, desc, index, logFile):
   for tool in ("vmstat", "top"):
     logFileName = f"{constants.LOGS_DIR}/nightly.{tool}.log"
     if os.path.exists(logFileName):
-      print(f'remove pre-existing log file {logFileName}')
+      print(f"remove pre-existing log file {logFileName}")
       os.remove(logFileName)
 
   # aggregate at multiple stack depths so we can see patterns like "new BytesRef() is costly regardless of context", for example: