We too use ceilometer and gnocchi and it works great. The old way with mongo simply didn't work. Our environment is around 10k running instances and gnocchi currently has data for 1,667,072 resources with 6,879,910 metrics. We are happy with gnocchi and so far haven't needed to change anything but would help contribute if there were things we needed. Would be great if we could just maintain gnocchi as opposed to making yet another thing. Sam On Wed, 20 Nov 2019 at 08:39, Matthias Runge <mrunge@matthias-runge.de> wrote:
On 19/11/2019 21:55, Lingxian Kong wrote:
On Wed, Nov 20, 2019 at 8:33 AM Matthias Runge <mrunge@matthias-runge.de <mailto:mrunge@matthias-runge.de>> wrote:
tbh, I am surprised to see the telemetry team trying to roll back to something, which is known to cause a lot of performance issues. There were good reasons for splitting ceilometer into several components.
I understand your concerns and the team will find a better way to deal with the rollback, the expected result is that it won't break anyone who are using Gnocchi or other storage backend but it means much to the cloud providers(like OVH and us) who are still using old version Ceilometer and relying on the API.
I assume, you haven't seen any performance issues with both Ceilometer or Gnocchi?
There were lots of good ideas and suggestions already shared in this thread. My proposal here would be to keep ceilometer as is (as data collecting agent) and to write missing glue to digest or send data
to a
time-series database, like InfluxDB or Prometheus (plus many more options, not mentioned here).
Does InfluxDB or Prometheus support to store samples that could be leveraged for auditing or billing purpose? I'm not familiar with TSDB but I got the answer NO according to the chatting with some other people
Billing is a complicated beast for multiple reasons. Let's look only at ingestion. First of all, you are right, e.g Prometheus should not be used in billing situations, see[1], also on [2] (same source as [1]).
However, the same is true for Gnocchi and also for Ceilometer. You are going to loose metrics if your collection is faster than your backend can digest it, for Gnocchi is such behaviour documented. If you are looking at speed, I'd consider prometheus to be much faster than gnocchi and ceilometer and thus better suited for handling metrics.
[1]
https://github.com/prometheus/docs/blob/master/content/docs/introduction/ove... [2] https://prometheus.io/docs/introduction/overview/#when-does-it-not-fit
If I remember correctly, MongoDB was deprecated because of the
issues it
caused and also because it was removed from Linux distributions,
since
there were the licensing issues.
I don't think the license change will affect the cloud that only uses MongoDB as internal service backend storage unless I'm missing something.
You can read a bit of story on MongoDB relicensing at [3].
The license change was regarding commercial use. You can find the full text at [4], depending on your use case, it may be valuable to ask a lawyer, especially, when in doubt.
[3]
https://hub.packtpub.com/mongodb-withdraws-controversial-server-side-public-... [4] https://www.mongodb.com/licensing/server-side-public-license
From my POV (and apparently others here in this thread as well), I'd take a step back and revisit https://storyboard.openstack.org/#!/board/205 again, in order to run into the same issues as in Newton cycle. I mean, if you'd want ceilometer from Newton, it is still there.
Matthias