Use _stop_event consistently across all threads as the way to signal that it should stop. Add thread joins with timeouts to allow the program to exit cleanly

providers/spot/dxcluster.py

@@ -1,8 +1,7 @@
 import logging
 import re
 from datetime import datetime
-from threading import Thread
-from time import sleep
+from threading import Event, Thread
 
 import pytz
 import telnetlib3
@@ -41,23 +40,26 @@ class DXCluster(SpotProvider):
             self._LINE_PATTERN_ALLOW_RBN if self._allow_rbn_spots else self._LINE_PATTERN_EXCLUDE_RBN
         )
         self._telnet = None
-        self._thread = Thread(target=self._handle, name=f"DXClusterSpotProvider-{self.name}")
-        self._thread.daemon = True
-        self._running = True
+        self._thread = None
+        self._stop_event = Event()
 
     def start(self):
+        self._thread = Thread(target=self._handle, name=f"DXClusterSpotProvider-{self.name}", daemon=True)
         self._thread.start()
 
     def stop(self):
-        self._running = False
+        self._stop_event.set()
         if self._telnet:
             self._telnet.close()
-        self._thread.join()
+        if self._thread:
+            self._thread.join(timeout=15)
+            if self._thread.is_alive():
+                logger.warning(f"DX Cluster {self._hostname} worker thread did not exit on time and will be killed.")
 
     def _handle(self):
-        while self._running:
+        while not self._stop_event.is_set():
             connected = False
-            while not connected and self._running:
+            while not connected and not self._stop_event.is_set():
                 try:
                     self.status = "Connecting"
                     logger.info(f"DX Cluster {self._hostname} connecting...")
@@ -69,14 +71,14 @@ class DXCluster(SpotProvider):
                 except ConnectionRefusedError:
                     self.status = "Error"
                     logger.warning(f"Connection refused to DX cluster {self._hostname}")
-                    sleep(300)
+                    self._stop_event.wait(timeout=300)
                 except Exception:
                     self.status = "Error"
                     logger.exception(f"Exception while connecting to DX Cluster Provider ({self._hostname}).")
-                    sleep(5)
+                    self._stop_event.wait(timeout=5)
 
             self.status = "Waiting for Data"
-            while connected and self._running:
+            while connected and not self._stop_event.is_set():
                 try:
                     # Check new telnet info against regular expression
                     telnet_output = self._telnet.read_until("\n".encode("latin-1"))
@@ -106,19 +108,19 @@ class DXCluster(SpotProvider):
 
                 except EOFError:
                     connected = False
-                    if self._running:
+                    if not self._stop_event.is_set():
                         self.status = "Restarting"
                         logger.warning(f"Disconnected from DX Cluster {self._hostname}. Reconnecting...")
-                        sleep(5)
+                        self._stop_event.wait(timeout=5)
                     else:
                         logger.info(f"DX Cluster {self._hostname} shutting down...")
                         self.status = "Shutting down"
                 except Exception:
                     connected = False
-                    if self._running:
+                    if not self._stop_event.is_set():
                         self.status = "Error"
                         logger.exception(f"Exception in DX Cluster Provider ({self._hostname})")
-                        sleep(5)
+                        self._stop_event.wait(timeout=5)
                     else:
                         logger.info(f"DX Cluster {self._hostname} shutting down...")
                         self.status = "Shutting down"