Description of problem: Based on some performance tests: ~~~ The installer will configure much less metricd processing workers than you will likely require for large scale with the defaults. You may need to tune metric_processing_delay at the expense of system resources to obtain your aggregated measures closer to realtime. Due to the higher system usage (cpu usage) on the machines hosting metricd, I would advise moving Gnocchi API and metricd to a separate machine. Ceph needs pgs tuned for metrics pool (In my case I used another yaml in tripleo to bump the number of PGs to 512 (36 total OSDs)). ... In order to get the 1,000 instances all on the low archival-policy setting with default metrics collected I had to adjust two settings to gnocchi.conf <---------------------- workers = 48 <---------------------- metric_processing_delay = 30 <---------------------- ~~~ We recommended to a customer to tweak the number of workers and metric_processing_delay. The results were promising but as we don't see any metric_processing_delay parameter in the gnocchi puppet manifests [1], we would like to request them added. If there's a better way to configure/update those via TripleO, please move the BZ to the appropriate BZ Component and or advise on it. The goal is simple: Configure gnocchi.conf at install/setup time through the Director. [1] https://github.com/openstack/puppet-gnocchi/tree/stable/newton/manifests
The pupet patches have landed. I'll check the builds and move to modified.
What about changeable number of workers? Did we drop that feature?
The workers support is already in place: https://github.com/openstack/tripleo-heat-templates/blob/master/puppet/services/gnocchi-metricd.yaml#L32-L35 hence not in the patch listed.
The delay support: https://github.com/openstack/tripleo-heat-templates/blob/master/puppet/services/gnocchi-base.yaml
openstack-tripleo-heat-templates-7.0.3-11.el7ost
There are still remainings of the same code in /usr/share/openstack-tripleo-heat-templates/puppet/services/gnocchi-base.yaml: MetricProcessingDelay: default: 30 description: Delay between processing metrics. gnocchi::storage::metric_processing_delay: {get_param: MetricProcessingDelay} Don't we need to clean it from there?
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
The needinfo request[s] on this closed bug have been removed as they have been unresolved for 1000 days