From 6fb8ac250ae27810516b89f273923d6b786606b5 Mon Sep 17 00:00:00 2001
From: Rabi Mishra <ramishra@redhat.com>
Date: Fri, 23 Aug 2019 10:45:28 +0530
Subject: [PATCH] Add retries when loading keystone data and fetching endpoints

We may end up loading lots of nested stacks concurrently with
convergence and those would try to discover endpoints and fetch
access info objects from keystone. This at times results in
ConnectTimeout errors from keystone. We can avoid these errors
by adding some retries.

Also adds retries to client_plugin get_endpoint() calls, which has
simillar issue.

Change-Id: I18cde971248eff5783f97c9e7a60316d7dd93431
Task: 36349
---
 heat/common/context.py               | 13 +++++++++++++
 heat/engine/clients/client_plugin.py |  2 ++
 2 files changed, 15 insertions(+)

diff --git a/heat/common/context.py b/heat/common/context.py
index 55025e9181..c2b48611ae 100644
--- a/heat/common/context.py
+++ b/heat/common/context.py
@@ -12,6 +12,7 @@
 #    under the License.
 
 from keystoneauth1 import access
+from keystoneauth1 import exceptions as ksa_exceptions
 from keystoneauth1.identity import access as access_plugin
 from keystoneauth1.identity import generic
 from keystoneauth1 import loading as ks_loading
@@ -24,6 +25,7 @@ import oslo_messaging
 from oslo_middleware import request_id as oslo_request_id
 from oslo_utils import importutils
 import six
+import tenacity
 
 from heat.common import config
 from heat.common import endpoint_utils
@@ -51,6 +53,15 @@ TRUSTEE_CONF_GROUP = 'trustee'
 ks_loading.register_auth_conf_options(cfg.CONF, TRUSTEE_CONF_GROUP)
 
 
+retry_on_connection_timeout = tenacity.retry(
+    stop=tenacity.stop_after_attempt(cfg.CONF.client_retry_limit+1),
+    wait=tenacity.wait_random(max=2),
+    retry=tenacity.retry_if_exception_type(
+        (ksa_exceptions.ConnectFailure,
+         ksa_exceptions.DiscoveryFailure)),
+    reraise=True)
+
+
 def list_opts():
     trustee_opts = ks_loading.get_auth_common_conf_options()
     trustee_opts.extend(ks_loading.get_auth_plugin_conf_options(
@@ -288,6 +299,8 @@ class RequestContext(context.RequestContext):
 
 
 class StoredContext(RequestContext):
+
+    @retry_on_connection_timeout
     def _load_keystone_data(self):
         self._keystone_loaded = True
         auth_ref = self.auth_plugin.get_access(self.keystone_session)
diff --git a/heat/engine/clients/client_plugin.py b/heat/engine/clients/client_plugin.py
index ab737da6f9..6f485d56a4 100644
--- a/heat/engine/clients/client_plugin.py
+++ b/heat/engine/clients/client_plugin.py
@@ -24,6 +24,7 @@ import requests
 import six
 
 from heat.common import config
+from heat.common import context
 from heat.common import exception as heat_exception
 
 cfg.CONF.import_opt('client_retry_limit', 'heat.common.config')
@@ -93,6 +94,7 @@ class ClientPlugin(object):
     def url_for(self, **kwargs):
         keystone_session = self.context.keystone_session
 
+        @context.retry_on_connection_timeout
         def get_endpoint():
             return keystone_session.get_endpoint(**kwargs)