ab7c43a34b
The default used to be -1 which translated to 'ha-mode: all'.
That means that every queue is mirrored on all controllers.
By setting this to zero we are moving to 'ha-mode: exactly' with
the number of queues to be set to CEIL(N/2):
$nr_ha_queues = CEIL(nr_of_rabbit_nodes/2)
set_policy='ha-all ^(?!amq\\.).* {\"ha-mode\":\"exactly\",\"ha-params\":${nr_ha_queues}}'
This way the queues will not be copied around all servers,
but just to a subset of them. This still provides the necessary
resilience in case of a controller crash, but is less demanding
in terms of performance (and likely triggers fewer bugs in rabbit)
Co-Authored-By: John Eckersberg <jeckersb@redhat.com>
Change-Id: If04a7bf24a5c94ec15726843a8a4d2504b656cb7
11 lines
541 B
YAML
11 lines
541 B
YAML
---
|
|
other:
|
|
- |
|
|
We now set the default number of rabbitmq queues to CEIL(N/2). (Where N is
|
|
the number of rabbitmq nodes). Previously this was set to N by default
|
|
which translated to having all queues mirrored to all controllers. By
|
|
changing the default to CEIL(N/2) the queues will not be copied around all
|
|
servers, but just to a subset of them. This still provides the necessary
|
|
resilience in case of a controller crash, but is less demanding in terms of
|
|
performance (and likely triggers fewer bugs in rabbit)
|