only check if thread is alive after cancelling it (#279)

it appears that in some cases, especially under high load, calling
`.cancel()` will immediately stop the thread, making the `.join()` call
fail. To avoid this, we first `cancel()`, then check `is_alive()` and
only _then_ `join()` the thread.

fixes #276
closes #279

Author: beniwohli

CHANGELOG.md

@@ -9,6 +9,7 @@
  * added "authorization" to list of sensitive keywords, to ensure that "Authorization" 
    HTTP headers are properly sanitized (#275)
  * taught the Logbook handler how to handle the `stack=False` option (#278)
+ * fixed a race condition with managing the timer-send thread (#279)
 
 ## v3.0.0
 

elasticapm/base.py

@@ -318,9 +318,10 @@ def _start_send_timer(self, timeout=None):
         self._send_timer.start()
 
     def _stop_send_timer(self):
-        if self._send_timer and self._send_timer.is_alive() and not self._send_timer == threading.current_thread():
+        if self._send_timer and not self._send_timer == threading.current_thread():
             self._send_timer.cancel()
-            self._send_timer.join()
+            if self._send_timer.is_alive():
+                self._send_timer.join()
 
     def _send_remote(self, url, data, headers=None):
         if headers is None: