Brian
I sent this PR with my proposal
https://github.com/apache/zookeeper/pull/840

Summary of changes:

- Refactor ServerMetrics, make it a single instance.
- Make ServerMetrics use the MetricsProvider API.
- Introduce a DefaultMetricsProvider which uses builtin metrics classes.
- Clean up ServerMetrics boot code.
- Integrate MetricsProvider with four letter words interface (monitor and reset)
- Add getBasicSummary() API to MetricsProvider to make it possile to
model AvgMinMaxCounter (Summary without percentiles)

Let's follow up on the patch

Regards
Enrico

Il giorno sab 26 gen 2019 alle ore 08:40 Enrico Olivelli
<[email protected]> ha scritto:
>
>
>
> Il giorno sab 26 gen 2019, 01:09 Brian Nixon <[email protected]> ha 
> scritto:
>>
>> Hi Enrico,
>>
>> Most of the major remaining contributions from the Facebook engineers
>> around metrics are captured in
>> https://issues.apache.org/jira/browse/ZOOKEEPER-3245 and our attention is
>> still focused on upstreaming that and other work. We'd like to see the
>> metrics provider work move forward because we'd like to be among the users
>> but don't have the bandwidth to take on the driving role for it for a few
>> months. If you choose not to wait for us to come back around to it then I'd
>> be happy to watch the feature develop and see if it works for our systems
>> internally.
>
>
> Sounds good.
> I will wait for ZOOKEEPER-3245 (and help for reviews) and then I will send my 
> proposals .
> I have to wait for Maven migration as well, there is no hurry.
>
> Thanks
> Enrico
>
>
>>
>> Thanks,
>> Brian
>>
>>
>> On Fri, Jan 18, 2019 at 1:51 PM Enrico Olivelli <[email protected]> wrote:
>>
>> > Hi fellows,
>> > I would like to start the 'real work' about Metrics providers.
>> > I am seeing new patches coming from FB friends and this is great.
>> >
>> > We have three topics here:
>> > - implementing providers (Prometheus will be the first)
>> > - packaging: I would like to separate MetricsProvider API from zookeeper
>> > server into a separate Maven module as soon as we drop Ant
>> > - instrumentation !
>> >
>> > Instrumentation is the hard part and we have to coordinate the work.
>> >
>> > IMHO a good goal would be to have all pf the existing metrics ported to the
>> > MetricsProvider API, with a Provider implementation which exposes metrics
>> > in the legacy way.
>> >
>> > Facebook friends already did a great work in introducing a common way of
>> > defining metrics, we only have to refactor a bit this system in order to
>> > move to the new API.
>> >
>> > The point is.....how can we coordinate? Is it too early to start?
>> >
>> > Given that most of the new metrics have been designed and are running in
>> > production at Facebook I think that such work should be driven by one of
>> > our 'Facebook friends', anyway I am available to do whatever is best in
>> > order to move this work forward.
>> >
>> >
>> > Cheers
>> > Enrico
>> >
>> >
>> > --
>> >
>> >
>> > -- Enrico Olivelli
>> >

Reply via email to