Description of problem: As per the neutron.conf text , metadata_workers is recommended as half the cpu core. Whereas api_workers are supposed to be = physical cpu cores . Although the THT set them both equally using 'neutron_workers: ./puppet/controller.yaml: neutron::server::api_workers: {get_input: neutron_workers} ./puppet/controller.yaml: neutron::agents::metadata::metadata_workers: {get_input: neutron_workers} Version-Release number of selected component (if applicable): RHOS8 How reproducible: Always Steps to Reproduce: 1. 2. 3. Actual results: metadata_workers & api_workers are set equally as 0 . Expected results: Additional info: From : /etc/neutron/metadata_agent.ini ~~~ # Number of separate worker processes for metadata server (defaults to half of # the number of CPUs) (integer value) #metadata_workers = 12 ~~~ From : /etc/neutron/neutron.conf ~~~ # Number of separate API worker processes for service. If not specified, the # default is equal to the number of CPUs available for best performance. # (integer value) #api_workers = <None> ~~~
To assign to Or Idgar, once he gets a Bugzilla account :)
This bugzilla has been removed from the release and needs to be reviewed and Triaged for another Target Release.
We no longer default neutron workers to equal the number of processors since this results in very large memory and processor overhead on servers with a large number of cores. Particularly if most of the other services also do this. It is now set to half of the processor cores with a minimum of 2 and maximum of twelve. There is a bug in the upstream heat template for the metadata workers as the default of an empty string no longer results in the proper calculated default being chosen. It should follow the same pattern using heat "conditions" that's used in neutron-api.yaml.
I'm adding the values I see in osp12 RC puddle, not sure if that's good enough for verification, (adding Udi to check) cat /etc/neutron/metadata_agent.ini | less : # Number of separate worker processes for metadata server (defaults to half of # the number of CPUs) (integer value) #metadata_workers = 16 # Number of separate API worker processes for service. If not specified, the # default is equal to the number of CPUs available for best performance. # (integer value) #api_workers = <None> api_workers=2
Verified with: openstack-tripleo-heat-templates-7.0.3-3.el7ost.noarch After re-checking the values with beagles; we saw that on my 12 core machine we have the right value on neutron.conf ( checked on the controller-1) neutron.conf ( on controller1) ------------------------------- #api_workers = <None> api_workers=6 # Number of RPC worker processes for service. (integer value) #rpc_workers = 1 rpc_workers=6 # Number of RPC worker processes dedicated to state reports queue. (integer # value) #rpc_state_report_workers = 1
Since the problem described in this bug report should be resolved in a recent advisory, it has been closed with a resolution of ERRATA. For information on the advisory, and where to find the updated files, follow the link below. If the solution does not work for you, open a new bug report. https://access.redhat.com/errata/RHEA-2017:3462