Retry with shorter 5sec timeout when trying to open() fixtures

- this is intended to reduce flapping tests caused by intermittent
- fixture startup issues on travis-ci

test/fixtures.py

@@ -126,8 +126,11 @@ class ZookeeperFixture(Fixture):
 
         # Party!
         self.out("Starting...")
-        self.child.start()
+        while True:
-        self.child.wait_for(r"binding to port")
+            self.child.start()
+            if self.child.wait_for(r"binding to port", timeout=5):
+                break
+            self.child.stop()
         self.out("Done!")
 
     def close(self):
@@ -222,8 +225,11 @@ class KafkaFixture(Fixture):
         self.out("Done!")
 
         self.out("Starting...")
-        self.child.start()
+        while True:
-        self.child.wait_for(r"\[Kafka Server %d\], Started" % self.broker_id)
+            self.child.start()
+            if self.child.wait_for(r"\[Kafka Server %d\], Started" % self.broker_id, timeout=5):
+                break
+            self.child.stop()
         self.out("Done!")
         self.running = True
 

test/service.py

@@ -110,14 +110,15 @@ class SpawnedService(threading.Thread):
                     log.exception("Received exception when killing child process")
                 self.dump_logs()
 
-                raise RuntimeError("Waiting for %r timed out after %d seconds" % (pattern, timeout))
+                log.error("Waiting for %r timed out after %d seconds", pattern, timeout)
+                return False
 
             if re.search(pattern, '\n'.join(self.captured_stdout), re.IGNORECASE) is not None:
                 log.info("Found pattern %r in %d seconds via stdout", pattern, (t2 - t1))
-                return
+                return True
             if re.search(pattern, '\n'.join(self.captured_stderr), re.IGNORECASE) is not None:
                 log.info("Found pattern %r in %d seconds via stderr", pattern, (t2 - t1))
-                return
+                return True
             time.sleep(0.1)
 
     def start(self):