osdir.com


[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

[Telemetry][TC]Telemetry status


https://github.com/gnocchixyz/gnocchi/issues/1049#issuecomment-555768072

On 12/18/19 10:15 AM, Tobias Urdin wrote:
> As an operator I second pretty much everything Samsaid, using
> ceilometer-api never really worked without hand holding all the time.
> We migrated over to Gnocchi as that was "the way forward" from the
> Telemetry team and it has worked great.
>
> Gnocchi has a learning curve but after that it has been running
> flawlessly even at a larger scale, just introduced more workers and
> everything is set.
>
> I think the long term plan from the Telemetry team was to move out any
> storage abstraction and cultivate ceilometer to a single focus area around
> collecting metrics. Moving any API, transformations, storage etc to
> another service.
>
> I think it's said to see Gnocchi, the actual solutions to the problem,
> being unmaintained and out of the OpenStack developer ecosystem. I
> assume there
> is a cost to bringing it back in after it was moved out but maybe it's
> something that is needed?
>
> While I don't have a deep understand in Gnocchi I would have no choice
> but to try to spend more time learning it and fixing any issues that we
> might
> see since at this point we can't live without it, as our billing
> providers supports the Gnocchi API, we using Heat with Gnocchi and Aodh
> to autoscale etc.
>
> As a final note; thanks for bringing back the cpu_util metric, means I
> can drop the ugly customized code that was required to bring that metric
> back while
> it was removed :)
>
> Best regards
> Tobias
>
> On 12/18/19 5:39 AM, Sam Morrison wrote:
>>> On 17 Dec 2019, at 10:14 pm, Thierry Carrez <thierry at openstack.org> wrote:
>>>
>>> Zane Bitter wrote:
>>>> On 15/12/19 10:20 pm, Rong Zhu wrote:
>>>>> 1.Add Ceilometer API back
>>>>>        Since Gnocchi is out of OpenStack and is unmaintained, we need to add Ceilometer API back again.
>>>> This is concerning because even the people who wrote it don't consider it adequate to the job. That inadequacy has been the source of significant reputational damage to OpenStack in the past, and many folks (me included) are anxious to avoid a repeat.
>>> Yes this concerns me too, and even if we workaround the issue by adding Ceilo API back, I'd like to have a long-term plan to solve this issue. It seems there are several options on the table (including integrating Monasca and Ceilometer into a single stack, and other big-bang replacements) but it's never a one-for-one comparison as the solutions seem to address slightly disjoint problem spaces.
>>>
>>> I'd like to hear from more Ceilometer users. What are they using Ceilometer for, and what long-term plans would be acceptable. There is a trade-off between adopting short-term workarounds that reintroduce performance issues vs. undergoing a complex migration to the "right" way of fixing this. Like for example there is little point in pushing Monasca/Ceilometer stack integration if most users say, like Catalyst Cloud seems to say, that they would rather have a slow performing Ceilometer API back.
>> Nectar Cloud has been a ceilometer user from the early days. Well we tried to be and couldnâ??t use it as ceilometer api and mongo db just didnâ??t work at our scale. Gnocchi solved all these issues for us and we use ceilometer/aodh/gnocchi happily in production for several years now.
>> If telemetry project is going down the path of the old days it will mean we will either drop ceilometer all together and look at alternative solutions like monasca or Prometheus etc. I just canâ??t see how the old architecture of ceilometer is ever going to be usable.
>>
>> If there is some confidence that gnocchi publisher will be supported in the future we would keep using gnocchi and just maintain it ourselves. Itâ??s an open source project and I was hoping the openstack community could keep it going. Weâ??d be happy to help maintain it at least.
>>
>> We use ceilometer/gnocchi to collect and store all metrics from openstack services. We have also written some custom pollsters and gnocchi is quite flexible here to allow this. With these metrics we build reports for our users, our operators, our funders (the government)
>>
>>
>> Please reconsider your direction much like adding cpu_util back in (thank you for this!)
>>
>> Cheers,
>> Sam
>>
>>
>>
>>>> Telemetry is part of the TC "Approved Release" that is eligible for the trademark program; I think at a minimum the TC will want to remove the resurrected Ceilometer API from the "designated sections" that users are required to run to participate in any trademark program that includes the functionality in question. But I think that we should explore other ways of reducing the chance of anyone confusing this for a viable way of building a cloud, including possibly changing the name (Antediluvian API?) and having this part of the stack live outside of the official OpenStack project.
>>> Legacy API?
>>>
>>> -- 
>>> Thierry Carrez (ttx)
>>>
>>
>
>