Make sure build components timeout if the initial connection fails

buildman/component/buildcomponent.py

@@ -17,6 +17,7 @@ from buildman.jobutil.workererror import WorkerError
 from data.database import BUILD_PHASE
 
 HEARTBEAT_DELTA = datetime.timedelta(seconds=30)
+BUILD_HEARTBEAT_DELAY = datetime.timedelta(seconds=30)
 HEARTBEAT_TIMEOUT = 10
 INITIAL_TIMEOUT = 25
 
@@ -76,8 +77,8 @@ class BuildComponent(BaseComponent):
   @trollius.coroutine
   def start_build(self, build_job):
     """ Starts a build. """
-    logger.debug('Starting build for component %s (worker version: %s)',
-                 self.builder_realm, self._worker_version)
+    logger.debug('Starting build for component %s (build %s, worker version: %s)',
+                 self.builder_realm, build_job.repo_build.uuid, self._worker_version)
 
     self._current_job = build_job
     self._build_status = StatusHandler(self.build_logs, build_job.repo_build.uuid)
@@ -150,6 +151,11 @@ class BuildComponent(BaseComponent):
 
     self.call("io.quay.builder.build", **build_arguments).add_done_callback(self._build_complete)
 
+    # Set the heartbeat for the future. If the builder never receives the build call,
+    # then this will cause a timeout after 30 seconds. We know the builder has registered
+    # by this point, so it makes sense to have a timeout.
+    self._last_heartbeat = datetime.datetime.utcnow() + BUILD_HEARTBEAT_DELAY
+
   @staticmethod
   def _commit_sha(build_config):
     """ Determines whether the metadata is using an old schema or not and returns the commit. """
@@ -401,8 +407,8 @@ class BuildComponent(BaseComponent):
         yield From(self._timeout())
         raise Return()
 
-      logger.debug('Heartbeat on realm %s is valid: %s.', self.builder_realm,
-                   self._last_heartbeat)
+      logger.debug('Heartbeat on realm %s is valid: %s (%s).', self.builder_realm,
+                   self._last_heartbeat, self._component_status)
 
       yield From(trollius.sleep(HEARTBEAT_TIMEOUT))