Created attachment 1078671 [details]
OOM kill on overcloud-controller

Description of problem:
Puppet configuration is based on processorcount to setup each number of worker of service as nova, neutron...

But he never check amount of ram so in case of I got lot of CPU, I get lot of worker. And quickly I get OOM cause each thread increase his mem usage.

Version-Release number of selected component (if applicable): 
openstack-tripleo-puppet-elements-0.0.1-4.el7ost.noarch
openstack-puppet-modules-2015.1.8-12.el7ost.noarch

How reproducible:
Install a undercloud or overcloud controller node with 16cpu 8G RAM

Steps to Reproduce:
1. Install undercloud
2. Use it a couple of day
3.

Actual results: OOM killer kill some of process...


Expected results: No more memory

Workaround:

Change value of :
/etc/keystone/keystone.conf:public_workers=16
/etc/keystone/keystone.conf:admin_workers=16
/etc/glance/glance-api.conf:workers=16
/etc/neutron/neutron.conf:api_workers = 16
/etc/nova/nova.conf:ec2_workers=16
/etc/nova/nova.conf:osapi_compute_workers=16

Additional info:

Example for nova:
https://github.com/openstack/puppet-nova/blob/0d57b76bf6c9a8989127cd16505ee62abc049087/manifests/api.pp
# [*ec2_workers*]
#   (optional) Number of workers for EC2 service
#   Defaults to $::processorcount
#
# [*metadata_workers*]
#   (optional) Number of workers for metadata service
#   Defaults to $::processorcount