Hi Greg! On 9/25/24 17:42, Waines, Greg wrote:
HOWEVER ... today (Caracal) ... - I think Gnocchi is still alive and well ... https://github.com/gnocchixyz/gnocchi * but maintained outside of OpenStack
Yeah, but it's still well maintained. The fact it's outside of OpenStack isn't a problem.
- I think OpenStack Panko was deprecated in ~ Wallaby ? * and don't think(?) it is maintained outside of OpenStack ?
Panko is dead, forget about it.
- The OpenStack Landscape at https://www.openstack.org/software/ * shows only Ceilometer and Aodh
Indeed, since Gnocchi isn't OpenStack maintained. There's also cloudkitty in the mix though.
MY QUESTIONs : ----------------------- - so what are the current recommendations for OpenStack Telemetry ? - is it just Ceilometer and Aodh ? * Does Ceilometer provide DBs ? (like it used to a long time ago)
Ceilometer used to have a mongodb backend, but his was a long time ago. This didn't scale and it was removed. So for Ceilometer, there's no client, and no API, and no db either. Ceilometer just plugs on Gnocchi. Gnocchi itself has a few backends. Best is influxdb as a timeseries backend for Gnocchi, though this doesn't support clustering in the free-software version (one has to pay for supporting clustering). So best is to use either galera or postgress for its indexing, and ceph for its timeseries blobs. That's what we use in production, with a dedicated Ceph (so it is always not too busy, and doesn't interfere with the production cluster).
- is Gnocchi (although maintained outside of OpenStack) still supported in the OpenStack environment as a DB option ?
Yes.
- what is used for event DB ? ... now that panko is gone ... Gnocchi ? Ceilometer ?
aodh can do what you want whenever an event on the metrics happen. You have no choice but to use aodh only, and forget about Panko.
- are there actually multiple options/solutions for OpenStack Telemetry ? e.g. * Ceilometer / Gnocchi / Aodh * Ceilometer / External Elastic Solution / Aodh * Others ?
Our setup for our public cloud consist of: - 3 nodes dedicated to rabbitmq for the notification bus (separated from the rest of the messages), and also running gnocchi-api (which can become very busy). - 3 nodes running galera (also separated from the "main" services dbs) - a ceph cluster with 3 mons, and 6 cephosd nodes - an elastic cluster for Cloudkitty to send billing data Note that we used to use the storage v1 for cloudkitty, and this doesn't scale. When you reach 5k+ VMs on your cluster, cloudkitty-processor can't cope with the load, and goes slower than recording metrics. BTW, I was wondering: since windriver is moving toward using Debian everywhere with its derivative, is StarlingX also moving to Debian? Will that use Debian packages for OpenStack? I hope the above helps, Cheers, Thomas Goirand (zigo)