From 6238b65f2d8a64ffc25cecc718d4a419af4cefc9 Mon Sep 17 00:00:00 2001
From: Igor Chorazewicz <igor.chorazewicz@intel.com>
Date: Thu, 29 Aug 2024 17:02:16 +0000
Subject: [PATCH] [Benchmarks] skip failed benchmarks by default

---
 scripts/benchmarks/main.py | 62 +++++++++++++++++++++++---------------
 1 file changed, 38 insertions(+), 24 deletions(-)

diff --git a/scripts/benchmarks/main.py b/scripts/benchmarks/main.py
index a9850dfc67..d2b7ef8cd0 100755
--- a/scripts/benchmarks/main.py
+++ b/scripts/benchmarks/main.py
@@ -52,34 +52,46 @@ def main(directory, additional_env_vars, save_name, compare_names, filter):
         benchmarks = [benchmark for benchmark in benchmarks if filter.search(benchmark.name())]
 
     for benchmark in benchmarks:
-        print(f"setting up {benchmark.name()}... ", end='', flush=True)
-        benchmark.setup()
-        print("complete.")
+        try:
+            print(f"setting up {benchmark.name()}... ", end='', flush=True)
+            benchmark.setup()
+            print("complete.")
+        except Exception as e:
+            if options.exit_on_failure:
+                raise e
+            else:
+                print(f"failed: {e}")
 
     results = []
     for benchmark in benchmarks:
-        merged_env_vars = {**additional_env_vars}
-        iteration_results = []
-        for iter in range(options.iterations):
-            print(f"running {benchmark.name()}, iteration {iter}... ", end='', flush=True)
-            bench_results = benchmark.run(merged_env_vars)
-            if bench_results is not None:
-                print(f"complete ({bench_results.value} {benchmark.unit()}).")
-                iteration_results.append(bench_results)
+        try:
+            merged_env_vars = {**additional_env_vars}
+            iteration_results = []
+            for iter in range(options.iterations):
+                print(f"running {benchmark.name()}, iteration {iter}... ", end='', flush=True)
+                bench_results = benchmark.run(merged_env_vars)
+                if bench_results is not None:
+                    print(f"complete ({bench_results.value} {benchmark.unit()}).")
+                    iteration_results.append(bench_results)
+                else:
+                    print(f"did not finish.")
+
+            if len(iteration_results) == 0:
+                continue
+
+            iteration_results.sort(key=lambda res: res.value)
+            median_index = len(iteration_results) // 2
+            median_result = iteration_results[median_index]
+
+            median_result.unit = benchmark.unit()
+            median_result.name = benchmark.name()
+
+            results.append(median_result)
+        except Exception as e:
+            if options.exit_on_failure:
+                raise e
             else:
-                print(f"did not finish.")
-
-        if len(iteration_results) == 0:
-            continue
-
-        iteration_results.sort(key=lambda res: res.value)
-        median_index = len(iteration_results) // 2
-        median_result = iteration_results[median_index]
-
-        median_result.unit = benchmark.unit()
-        median_result.name = benchmark.name()
-
-        results.append(median_result)
+                print(f"failed: {e}")
 
     for benchmark in benchmarks:
         print(f"tearing down {benchmark.name()}... ", end='', flush=True)
@@ -126,6 +138,7 @@ def validate_and_parse_env_args(env_args):
     parser.add_argument("--timeout", type=int, help='Timeout for individual benchmarks in seconds.', default=600)
     parser.add_argument("--filter", type=str, help='Regex pattern to filter benchmarks by name.', default=None)
     parser.add_argument("--verbose", help='Print output of all the commands.', action="store_true")
+    parser.add_argument("--exit_on_failure", help='Exit on first failure.', action="store_true")
 
     args = parser.parse_args()
     additional_env_vars = validate_and_parse_env_args(args.env)
@@ -137,6 +150,7 @@ def validate_and_parse_env_args(env_args):
     options.timeout = args.timeout
     options.ur_dir = args.ur_dir
     options.ur_adapter_name = args.ur_adapter_name
+    options.exit_on_failure = args.exit_on_failure
 
     benchmark_filter = re.compile(args.filter) if args.filter else None