Change some default values for Rabbit driver
It was observed that for the rabbitmq-server version 3.8.2 default
values during failover of one of the rabbit nodes cause rapid connection
recreation. In some cases it leads to the creation of broken exchanges
and hangs of OpenStack operations. Changing rabbit_retry_interval to 5,
rabbit_retry_backoff to 10 and kombu_reconnect_delay to 5.0 fix an
issue.
Related-Issue: PROD-34332
Change-Id: I51f358328833ff93de98eed90bdd4de456a6bfec
diff --git a/oslo_templates/files/queens/oslo/messaging/_rabbit.conf b/oslo_templates/files/queens/oslo/messaging/_rabbit.conf
index 3a12be1..f04ff9d 100644
--- a/oslo_templates/files/queens/oslo/messaging/_rabbit.conf
+++ b/oslo_templates/files/queens/oslo/messaging/_rabbit.conf
@@ -53,6 +53,10 @@
# How long to wait before reconnecting in response to an AMQP consumer
# cancel notification. (floating point value)
#kombu_reconnect_delay = 1.0
+# NOTE(pas-ha) default value is problematic with RMQ 3.8, see PROD-34322
+# recreating queues on a secondary broker immediately after primary broker
+# has gone down leads to these queues being non-functional.
+kombu_reconnect_delay = 5.0
# EXPERIMENTAL: Possible values are: gzip, bz2. If not set compression
# will not be used. This option may not be available in future
@@ -122,15 +126,25 @@
# How frequently to retry connecting with RabbitMQ. (integer value)
#rabbit_retry_interval = 1
+# NOTE(pas-ha) default value is problematic with RMQ 3.8, see PROD-34322
+# recreating queues on a secondary broker immediately after primary broker
+# has gone down leads to these queues being non-functional.
{%- if _data.rabbit_retry_interval is defined %}
rabbit_retry_interval = {{ _data.rabbit_retry_interval }}
+{%- else %}
+rabbit_retry_interval = 5
{%- endif %}
# How long to backoff for between retries when connecting to RabbitMQ.
# (integer value)
#rabbit_retry_backoff = 2
+# NOTE(pas-ha) default value is problematic with RMQ 3.8, see PROD-34322
+# recreating queues on a secondary broker immediately after primary broker
+# has gone down leads to these queues being non-functional.
{%- if _data.rabbit_retry_backoff is defined %}
rabbit_retry_backoff = {{ _data.rabbit_retry_backoff }}
+{%- else %}
+rabbit_retry_backoff = 10
{%- endif %}
# Maximum interval of RabbitMQ connection retries. Default is 30