From 998f6afa5326ce872322b4a1caab3c235e40dbc5 Mon Sep 17 00:00:00 2001
From: Craig Bryant <craig.bryant@hpe.com>
Date: Tue, 5 Dec 2017 08:47:56 -0700
Subject: [PATCH] Retry publish once on failures

On failure to publish, clear the topic metadata then retry, in
case the IP Addresses have changed.  This can occur when
Monasca is run in Kubernetes and the Kafka pod is restarted.
Restarting the Kafka pod can happen often enough the
API should be able handle it without losing a message.

Change-Id: If48971c40883b5be10adec947562cdda7e82d77c
Story: 2001386
Task: 5963
---
 monasca_common/kafka/producer.py | 35 ++++++++++++++++++++++++--------
 1 file changed, 27 insertions(+), 8 deletions(-)

diff --git a/monasca_common/kafka/producer.py b/monasca_common/kafka/producer.py
index b454f950..236ced95 100644
--- a/monasca_common/kafka/producer.py
+++ b/monasca_common/kafka/producer.py
@@ -1,4 +1,4 @@
-# Copyright (c) 2015 Hewlett-Packard Development Company, L.P.
+# (C) Copyright 2015, 2017 Hewlett Packard Enterprise Development LP
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
@@ -47,10 +47,29 @@ class KafkaProducer(object):
         if not isinstance(messages, list):
             messages = [messages]
 
-        try:
-            if key is None:
-                key = int(time.time() * 1000)
-            self._producer.send_messages(topic, str(key), *messages)
-        except Exception:
-            log.exception('Error publishing to {} topic.'.format(topic))
-            raise
+        first = True
+        success = False
+        while not success:
+            try:
+                if key is None:
+                    key = int(time.time() * 1000)
+                self._producer.send_messages(topic, str(key), *messages)
+                success = True
+            except Exception:
+                if first:
+                    # This is a warning because of all the other warning and
+                    # error messages that are logged in this case. This way
+                    # someone looking at the log file can see the retry
+                    log.warn("Failed send on topic {}, clear metadata and retry"
+                             .format(topic))
+
+                    # If Kafka is running in Kubernetes, the cached metadata
+                    # contains the IP Address of the Kafka pod. If the Kafka
+                    # pod has restarted, the IP Address will have changed
+                    # which would have caused the first publish to fail. So,
+                    # clear the cached metadata and retry the publish
+                    self._kafka.reset_topic_metadata(topic)
+                    first = False
+                    continue
+                log.exception('Error publishing to {} topic.'.format(topic))
+                raise