OK. Thanks for the input.

I will think start from the library first to make sure the API is
consistent.

On Mon, Dec 6, 2021 at 7:25 PM Nicholas Nezis <[email protected]>
wrote:

> One library that might be worth investigating is Micrometer Core. It's an
> Apache library that provides vendor neutral API over various
> instrumentation clients. This could be a great tool to add as a Metrics
> Sink. https://micrometer.io/docs
>
> But I think we still should update the Heron metrics API to match this
> functionality. Heron currently provides the metrics aggregation from the
> various bolts/spouts into a central location. I think this mechanism is
> still important. But once it's aggregated we can leverage Heron's pluggable
> metrics sink capability to export to whatever metrics provider is desired.
>
> But I think we should update Heron to support the concept of labels so that
> the information can make it from Spout/Bolt to the Metrics Manager. So this
> might entail an update to the Heron API and also the Metrics Manager
> itself. But I agree with Ning, the existing built in Heron metrics logic
> will complicate adding the feature.
>
>
> On Mon, Dec 6, 2021 at 8:25 AM Nicholas Nezis <[email protected]>
> wrote:
>
> > I don't think we should jump to an external solution. We should add
> > support in our API, out at least extend with an optional set of
> additional
> > calls.
> >
> > We should look at Flink's API and implementation as reference. Also we
> > should look at Prometheus.
> >
> > In both of these, you usually define the keys when creating the counter,
> > and then provide the values when submitting the new value (or increment
> > call).
> >
> > I'll share some links soon.
> >
> >
> >
> > On Mon, Dec 6, 2021, 4:45 AM Ning Wang <[email protected]> wrote:
> >
> >> Hello,
> >>
> >> Nick and I have been talking about supporting tags in the Heron metrics
> >> library. I was thinking about something like DataDog so you just need to
> >> provide a list of tags in the `incr()` functions. However, one tricky
> >> thing
> >> about Heron's library is that aggregation happens in the library.
> >>
> >> I am curious if anyone has used DataDog (or StatsD) with Heron and how
> do
> >> you hook it up. For DataDog, I think the aggregation happens in the
> agent
> >> instead of the library.
> >>
> >> Thanks in advance.
> >> --ning
> >>
> >
>

Reply via email to