fix: allow update cancellation to restore the original guest state

AppImage/scripts/lxc_update_jobs.py

@@ -35,6 +35,9 @@ PARAMS = {
 RUN_ID = re.compile(r"^[A-Za-z0-9_-]{1,128}$")
 INTERACTION_ID = re.compile(r"^[A-Za-z0-9_-]{1,64}$")
 MAX_RUNTIME = 4 * 3600
+# The shared runner restores a guest it started in its EXIT trap. Its package
+# observation and `pct shutdown --timeout 60` need time after SIGTERM, too.
+STOP_GRACE = 300
 RESULT_TTL = 6 * 3600
 MAX_OUTPUT = 256 * 1024
 MAX_JOBS = 32
@@ -187,7 +190,7 @@ class UpdateJob:
                             break
                     if time.monotonic() - self.started >= MAX_RUNTIME:
                         self._signal("timeout")
-                    if running and self.stop_at and now - self.stop_at >= 2:
+                    if running and self.stop_at and now - self.stop_at >= STOP_GRACE:
                         try:
                             os.killpg(self.process.pid, signal.SIGKILL)
                         except ProcessLookupError:
@@ -213,7 +216,7 @@ class UpdateJob:
             self._append(f"\r\nUpdate viewer failed: {type(exc).__name__}\r\n")
             self._signal("viewer failure")
             try:
-                self.process.wait(timeout=2)
+                self.process.wait(timeout=STOP_GRACE)
             except subprocess.TimeoutExpired:
                 os.killpg(self.process.pid, signal.SIGKILL)
                 self.process.wait()

tests/test_lxc_update_jobs.py

@@ -187,6 +187,31 @@ class JobTests(unittest.TestCase):
         self.assertEqual(job.exit_code, 124)
         self.callback.assert_called_once()
 
+    def test_cancel_waits_for_the_runners_original_state_cleanup(self):
+        # apply_updates.sh's EXIT trap may need a 60s package probe followed
+        # by `pct shutdown --timeout 60`. A two-second kill truncates that trap.
+        self.script.write_text(
+            f"trap 'sleep 2.3; echo restored >> {shlex.quote(str(self.events))}; exit 0' TERM\n"
+            + self.script.read_text())
+        job = self.start()
+        wait_for(lambda: self.events.exists())
+        job.cancel()
+        wait_for(lambda: job.state == 'complete')
+        self.assertEqual(job.exit_code, 130)
+        self.assertEqual(self.events.read_text().splitlines(), ['backup', 'restored'])
+        self.assertNotIn('update', self.events.read_text())
+        self.callback.assert_called_once()
+        self.assertGreater(jobs.STOP_GRACE, 120)
+
+    def test_timeout_still_reaps_a_process_that_ignores_termination(self):
+        self.script.write_text("trap '' TERM\necho ready\nwhile :; do sleep .02; done\n")
+        with patch.object(jobs, 'MAX_RUNTIME', .15), patch.object(jobs, 'STOP_GRACE', .1):
+            job = self.start()
+            wait_for(lambda: job.state == 'complete')
+        self.assertEqual(job.exit_code, 124)
+        self.assertIsNotNone(job.process.poll())
+        self.callback.assert_called_once()
+
     def test_output_is_bounded_and_cursor_replay_is_read_only(self):
         job = self.start()
         wait_for(lambda: self.events.exists())