Bug 1976826

+++ This bug was initially created as a clone of Bug #1975790 +++

Description of problem:

When using a load balancer in ACTIVE_STANDBY topology, the haproxy instance that runs in the amphora is prone to memory allocation errors, because each haproxy worker consumes a lot of memory, and multiple workers are running at the same time after a configuration update.

The visible side effects in the octavia worker are some exceptions thrown when calling the amphora-agent:

ERROR oslo_messaging.rpc.server octavia.amphorae.drivers.haproxy.exceptions.InternalServerError: Internal Server Error

ERROR octavia.amphorae.drivers.haproxy.exceptions [XXX - XXX - - -] Amphora agent returned unexpected result code 500 with response {'message': 'Error reloading haproxy', 'details': 'Redirecting to /bin/systemctl reload haproxy-XXX.service\nJob for haproxy-XXX.service canceled.\n'}


Version-Release number of selected component (if applicable):
16.1

How reproducible:
100%

Steps to Reproduce:
1. Create a LB in Active standby
2. Create a listener and a pool
3. Create many members

Actual results:
After each new member creation, there is one more haproxy processes that should be cleaned up, memory consumption can be important depending on the listener configuration


Expected results:
A configuration change should not reduce the available memory in the amphora


Additional info:

Detailed report can be found in the upstream story https://storyboard.openstack.org/#!/story/2009005