diff --git a/rllib/env/policy_server_input.py b/rllib/env/policy_server_input.py
index f09fbf975..6bc2d7267 100644
--- a/rllib/env/policy_server_input.py
+++ b/rllib/env/policy_server_input.py
@@ -90,6 +90,7 @@ class PolicyServerInput(ThreadingMixIn, HTTPServer, InputReader):
         logger.info("Starting connector server at {}:{}".format(address, port))
         logger.info("")
         thread = threading.Thread(name="server", target=self.serve_forever)
+        thread.daemon = True
         thread.start()
 
     @override(InputReader)
diff --git a/rllib/evaluation/sample_batch_builder.py b/rllib/evaluation/sample_batch_builder.py
index e858d31c0..0978712c0 100644
--- a/rllib/evaluation/sample_batch_builder.py
+++ b/rllib/evaluation/sample_batch_builder.py
@@ -162,8 +162,6 @@ class MultiAgentSampleBatchBuilder:
 
         # Append into policy batches and reset
         for agent_id, post_batch in sorted(post_batches.items()):
-            self.policy_builders[self.agent_to_policy[agent_id]].add_batch(
-                post_batch)
             if self.postp_callback:
                 self.postp_callback({
                     "episode": episode,
@@ -172,6 +170,8 @@ class MultiAgentSampleBatchBuilder:
                     "post_batch": post_batch,
                     "all_pre_batches": pre_batches,
                 })
+            self.policy_builders[self.agent_to_policy[agent_id]].add_batch(
+                post_batch)
 
         self.agent_builders.clear()
         self.agent_to_policy.clear()