Micrometer polling on management interface

[Fabricio Gregorio]

After WildFly 28, micrometer only works with pushing and no longer works with polling. I  couldn't find a reason for this. Is polling going to be added later?

For those migrating applications that use microprofile-metrics, it would be preferable to poll at first (opt-int).

Thank you

[Jason Lee]

The goal was to unify observability signal aggregation (metrics + traces). There is, however, a feature request to implement a Prometheus registry for Micrometer here: https://issues.redhat.com/browse/WFLY-18582

Jason Lee

Principal Software Engineer

Red Hat JBoss EAP

--
You received this message because you are subscribed to the Google Groups "WildFly" group.
To unsubscribe from this group and stop receiving emails from it, send an email to wildfly+u...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/wildfly/57e879ea-b898-4153-bb2b-e7686b23fe20n%40googlegroups.com.

[Fabricio Gregorio]

Thank you Jason!

[jvi...@gmail.com]

I also want to express my disappointment on how this was handled.

First of all, wildfly is great, using it since the old jboss days. In my eyes one of the biggest plus of  java ee/jakarta ee platform application servers and java as a whole is its long term stability of the api.

Until now, if you stuck to the standards of jakartaee and microprofile, you almost had the guarantee that you could port an application to the latest version with no or only very minor changes.

Whats even more frustrating, that microprofile metrics was replaced by a half-arsed version of a micrometer implementation. Only OLTP export, no direct exposure why the metric management interface, no annotations support,...

It seems that the decision to move to micrometer makes sense as a whole, but why not do it right before dropping support for something many people were very happy with ? I don't think maintaining microprofile metrics for a few more versions until the micrometer stuff is more complete would have been a lot of work. Your current implementation of micrometer is "experimental" at best...

That it is possible to do a migration to micrometer in a much more elegant way was shown by the quarkus team:
https://quarkus.io/guides/telemetry-micrometer
They even have a comptibility layer for microprofile metrics.

Overall replacing good working stuff with half arsed new solutions seems to be a concerning trend within the wilfly team.
Another example seems to be the Keycloack adapter, which was dropped for OpenID connect, before OpenID connect was ready to also support more complex security setups (for example identity propagation)

This is really a really  concerning trend which makes me re-think if wildfly is still the stable foundation we can build on for the future.