contest-hw: hwksft: always grab worker's journal

Signed-off-by: Jakub Kicinski <kuba@kernel.org>

Author: kuba-moo

contest/hw/hwksft.py

@@ -24,7 +24,7 @@
 from lib.mc_client import MCClient, resolve_machines, resolve_nic_id  # noqa: E402
 from lib.deployer import (build_kernel, build_ksft, deploy_artifacts,  # noqa: E402
                           kexec_machine, wait_for_results, fetch_results,
-                          set_log_file, WaitResult)
+                          set_log_file, WaitResult, grab_hw_worker_journal)
 
 # Config:
 #
@@ -168,10 +168,12 @@ def test(binfo, rinfo, cbarg):  # pylint: disable=unused-argument
 
         # 7. Wait for hw-worker with crash monitoring
         wait_result = wait_for_results(config, mc, reservation_id,
-                                       machine_ids, machine_ips,
-                                       results_path=results_path)
+                                       machine_ids, machine_ips)
 
-        # 8. Copy back results
+        # 8. Grab hw-worker journal for debugging
+        grab_hw_worker_journal(machine_ips[0], results_path)
+
+        # 9. Copy back results
         if wait_result.ok:
             cases = fetch_results(config, machine_ips, reservation_id, rinfo)
         else:
@@ -187,7 +189,7 @@ def test(binfo, rinfo, cbarg):  # pylint: disable=unused-argument
             }]
     finally:
         set_log_file(None)
-        # 9. Release reservation
+        # 10. Release reservation
         try:
             mc.reservation_close(reservation_id)
         except Exception as e:

contest/hw/lib/deployer.py

@@ -207,8 +207,18 @@ def _refresh():
         print(f"kexec: {ipaddr} is back")
 
 
-def wait_for_results(config, mc, reservation_id, machine_ids, machine_ips,
-                     results_path=None):
+def grab_hw_worker_journal(ipaddr, results_path):
+    """Fetch hw-worker journal from the test machine and save locally."""
+    journal = _ssh(ipaddr,
+                   'journalctl -u nipa-hw-worker.service -n 250 --no-pager',
+                   check=False)
+    if journal:
+        journal_file = os.path.join(results_path, 'hw-worker-journal')
+        with open(journal_file, 'w', encoding='utf-8') as fp:
+            fp.write(journal)
+
+
+def wait_for_results(config, mc, reservation_id, machine_ids, machine_ips):
     """Main wait loop with crash monitoring.
 
     Polls SOL logs via mc.get_sol_logs() to detect crashes.
@@ -271,14 +281,6 @@ def wait_for_results(config, mc, reservation_id, machine_ids, machine_ips,
                 return WaitResult(ok=True)
 
             # Service finished and no results.json — something went wrong.
-            # Grab journalctl output for debugging.
-            journal = _ssh(primary_ip,
-                           'journalctl -u nipa-hw-worker.service -n 100 --no-pager',
-                           check=False)
-            if journal and results_path:
-                journal_file = os.path.join(results_path, 'hw-worker-journal')
-                with open(journal_file, 'w', encoding='utf-8') as fp:
-                    fp.write(journal)
             msg = f"hw-worker exited without results (state={state})"
             print(f"wait_for_results: {msg}")
             return WaitResult(ok=False, error=msg)
@@ -300,10 +302,6 @@ def wait_for_results(config, mc, reservation_id, machine_ids, machine_ips,
                                                            'unreferenced object 0x'))]
                     for cl in crash_lines:
                         print(f"wait_for_results: crash on machine {mid}: {cl.strip()}")
-                    if results_path:
-                        crash_file = os.path.join(results_path, f'crash-machine-{mid}')
-                        with open(crash_file, 'a', encoding='utf-8') as fp:
-                            fp.write(sol_text + '\n')
 
             if mid in crash_detected_at:
                 if _has_reboot(sol_text):

contest/hw/tests/test_hwksft.py

@@ -154,6 +154,23 @@ def test_reserve_409_returns_json(self, mock_post):
 
 
 class TestDeployer(unittest.TestCase):
+    def test_grab_hw_worker_journal(self):
+        """Verify journal is fetched and saved to results dir."""
+        from lib.deployer import grab_hw_worker_journal
+
+        with tempfile.TemporaryDirectory() as tmpdir:
+            with mock.patch('lib.deployer._ssh',
+                             return_value='Mar 14 hw-worker[1]: test log\n') as mock_ssh:
+                grab_hw_worker_journal('10.0.0.1', tmpdir)
+
+            mock_ssh.assert_called_once()
+            self.assertIn('journalctl', mock_ssh.call_args[0][1])
+            self.assertIn('-n 250', mock_ssh.call_args[0][1])
+            journal_file = os.path.join(tmpdir, 'hw-worker-journal')
+            self.assertTrue(os.path.exists(journal_file))
+            with open(journal_file) as fp:
+                self.assertIn('test log', fp.read())
+
     @mock.patch('subprocess.run')
     def test_deploy_artifacts(self, mock_run):
         mock_run.return_value = mock.Mock(returncode=0, stdout=b'', stderr=b'')
@@ -378,36 +395,20 @@ def ssh_retcode_side_effect(ip, cmd, timeout=30):
                 return 1  # no results.json
             return 0
 
-        ssh_call_count = {'n': 0}
-
         def ssh_side_effect(ip, cmd, check=True, timeout=30):
-            ssh_call_count['n'] += 1
             if 'systemctl show' in cmd:
                 return 'failed\n'
-            if 'journalctl' in cmd:
-                return 'Mar 14 hw-worker[123]: some log\n'
             return ''
 
-        with tempfile.TemporaryDirectory() as tmpdir:
-            with mock.patch('lib.deployer._ssh_retcode',
-                             side_effect=ssh_retcode_side_effect):
-                with mock.patch('lib.deployer._ssh',
-                                 side_effect=ssh_side_effect) as mock_ssh:
-                    result = wait_for_results(config, mc, 42, [1], ['10.0.0.1'],
-                                              results_path=tmpdir)
-
-            self.assertIsInstance(result, WaitResult)
-            self.assertFalse(result.ok)
-            self.assertIn('hw-worker exited without results', result.error)
-
-            # Verify journalctl output was fetched and saved
-            journal_calls = [c for c in mock_ssh.call_args_list
-                             if 'journalctl' in c[0][1]]
-            self.assertEqual(len(journal_calls), 1)
-            journal_file = os.path.join(tmpdir, 'hw-worker-journal')
-            self.assertTrue(os.path.exists(journal_file))
-            with open(journal_file) as fp:
-                self.assertIn('some log', fp.read())
+        with mock.patch('lib.deployer._ssh_retcode',
+                         side_effect=ssh_retcode_side_effect):
+            with mock.patch('lib.deployer._ssh',
+                             side_effect=ssh_side_effect):
+                result = wait_for_results(config, mc, 42, [1], ['10.0.0.1'])
+
+        self.assertIsInstance(result, WaitResult)
+        self.assertFalse(result.ok)
+        self.assertIn('hw-worker exited without results', result.error)
 
     @mock.patch('subprocess.run')
     @mock.patch('time.monotonic')