Try stop with SIGINT first, pass timeout to constructor

freqtrade/tests/test_watchdog.py

@@ -16,3 +16,24 @@ def test_watchdog_kill(caplog):
     log = ["Watchdog started", "Watchdog stopped"]
     for line in log:
         assert line in caplog.text
+
+
+def test_try_kill_failed(mocker):
+    mocker.patch("os.kill")
+    mocker.patch("os.waitpid", return_value=(0, 0))
+    watchdog = Watchdog(1, 1)
+    assert watchdog.try_kill(0) is False
+
+
+def test_try_kill_success(mocker):
+    mocker.patch("os.kill")
+    mocker.patch("os.waitpid", return_value=(0, 1))
+    watchdog = Watchdog(1, 1)
+    assert watchdog.try_kill(0) is True
+
+
+def test_try_kill_error(mocker):
+    mocker.patch("os.kill")
+    mocker.patch("os.waitpid", side_effect=OSError)
+    watchdog = Watchdog(1, 1)
+    assert watchdog.try_kill(0) is True

freqtrade/watchdog.py

@@ -7,6 +7,7 @@ from multiprocessing import Value
 logger = logging.getLogger('freqtrade.watchdog')
 
 WATCHDOG_TIMEOUT = 300
+KILL_TIMEOUT = 60
 
 
 class Watchdog:
@@ -14,8 +15,9 @@ class Watchdog:
     shared_heartbeat = Value('d', 0.0)
     kill_signal = None
 
-    def __init__(self, timeout=WATCHDOG_TIMEOUT):
+    def __init__(self, timeout=WATCHDOG_TIMEOUT, kill_timeout=KILL_TIMEOUT):
         self.timeout = timeout
+        self.kill_timeout = kill_timeout
         self.heartbeat()
 
     def heartbeat(self) -> None:
@@ -26,11 +28,27 @@ class Watchdog:
         logger.warning("Kill signal: {}".format(signum))
         self.kill_signal = signum
 
+    def try_kill(self, pid):
+        os.kill(pid, signal.SIGINT)
+        for count in range(0, self.kill_timeout):
+            try:
+                pid, err_code = os.waitpid(pid, os.WNOHANG)
+                if pid != 0 or err_code != 0:
+                    return True
+                time.sleep(1)
+            except OSError:
+                return True
+        return False
+
     def kill(self, pid):
         logger.info("Stopping pid {}".format(pid))
         if pid:
-            os.kill(pid, signal.SIGTERM)  # Better use sigint and then sigterm?
+            if self.try_kill(pid):
-            os.wait()
+                logger.info("Process finished gracefully")
+            else:
+                logger.warning("Process not responded, kill by SIGTERM")
+                os.kill(pid, signal.SIGTERM)
+                os.wait()
 
     def start(self) -> bool:
         pid = os.fork()