If the host agent goes offline, timeout and reset. (#255)

Author: Peter Giacomo Lombardo

instana/agent/host.py

@@ -4,8 +4,10 @@
 """
 from __future__ import absolute_import
 
-import json
 import os
+import json
+import urllib3
+import requests
 from datetime import datetime
 
 from ..log import logger
@@ -230,6 +232,10 @@ def report_data_payload(self, payload):
                     # The host agent returned something indicating that is has a request for us that we
                     # need to process.
                     self.handle_agent_tasks(json.loads(response.content)[0])
+        except requests.exceptions.ConnectionError:
+            pass
+        except urllib3.exceptions.MaxRetryError:
+            pass
         except Exception as exc:
             logger.debug("report_data_payload: Instana host agent connection error (%s)", type(exc), exc_info=True)
         return response

instana/collector/host.py

@@ -42,6 +42,10 @@ def prepare_and_report_data(self):
                         self.agent.machine.fsm.ready()
                 else:
                     return
+
+            if self.agent.machine.fsm.current == "good2go" and self.agent.is_timed_out():
+                logger.info("The Instana host agent has gone offline or is no longer reachable for > 1 min.  Will retry periodically.")
+                self.agent.reset()
         except Exception:
             logger.debug('Harmless state machine thread disagreement.  Will self-correct on next timer cycle.')