Runner - Add signal handler in runner (#530)

Add signal handler in runner to gracefully exit when receiving SIGINT
(<kbd>Ctrl</kbd>+<kbd>C</kbd>) or SIGTERM during benchmark execution.

setup.py

@@ -198,6 +198,7 @@ setup(
                 'types-pkg_resources',
                 'types-pyyaml',
                 'typing-extensions>=3.10',
+                'urllib3<2.0',
                 'vcrpy>=4.1.1',
                 'yapf==0.31.0',
             ],

superbench/runner/ansible.py

@@ -59,11 +59,12 @@ class AnsibleClient():
                 self._config['cmdline'] += ' --ask-pass --ask-become-pass'
         logger.info(self._config)
 
-    def run(self, ansible_config, sudo=False):    # pragma: no cover
+    def run(self, ansible_config, cancel_callback=None, sudo=False):    # pragma: no cover
         """Run Ansible runner.
 
         Args:
             ansible_config (dict): Ansible config dict.
+            cancel_callback (Callable): Ansible runner cancel callback.
             sudo (bool): Run as sudo or not. Defaults to False.
 
         Returns:
@@ -73,7 +74,7 @@ class AnsibleClient():
             logger.info('Run as sudo ...')
             ansible_config['cmdline'] += ' --become'
         with tempfile.TemporaryDirectory(prefix='ansible') as tmpdir:
-            r = ansible_runner.run(private_data_dir=tmpdir, **ansible_config)
+            r = ansible_runner.run(private_data_dir=tmpdir, cancel_callback=cancel_callback, **ansible_config)
             logger.debug(r.stats)
         if r.rc == 0:
             logger.info('Run succeed, return code {}.'.format(r.rc))

superbench/runner/runner.py

@@ -4,8 +4,10 @@
 """SuperBench Runner."""
 
 import os
+import sys
 import json
 import random
+import signal
 from pathlib import Path
 from pprint import pformat
 from collections import defaultdict
@@ -233,6 +235,18 @@ class SuperBenchRunner():
             )
         )
 
+    def __signal_handler(self, signum, frame):
+        """Signal handler for runner.
+
+        Args:
+            signum (int): Signal number.
+            frame (FrameType): Timeout frame.
+        """
+        if signum == signal.SIGINT or signum == signal.SIGTERM:
+            logger.info('Killed by %s, exiting ...', signal.Signals(signum).name)
+            self.cleanup()
+            sys.exit(128 + signum)
+
     def __create_results_summary(self):    # pragma: no cover
         """Create the result summary file of all nodes."""
         all_results = list()
@@ -438,12 +452,17 @@ class SuperBenchRunner():
             # we do not expect timeout in ansible unless subprocess hangs
             ansible_runner_config['timeout'] = timeout + 60
 
-        rc = self._ansible_client.run(ansible_runner_config, sudo=(not self._docker_config.skip))
+        # overwrite ansible runner's default signal handler with main process's
+        rc = self._ansible_client.run(
+            ansible_runner_config, cancel_callback=lambda: None, sudo=(not self._docker_config.skip)
+        )
         return rc
 
     def run(self):
         """Run the SuperBench benchmarks distributedly."""
         self.check_env()
+        signal.signal(signal.SIGINT, self.__signal_handler)
+        signal.signal(signal.SIGTERM, self.__signal_handler)
         for benchmark_name in self._sb_benchmarks:
             if benchmark_name not in self._sb_enabled_benchmarks:
                 continue