Updated timeline for removing management metrics collection retaining non-stats info in management API?

[bording]

Community Support Policy

• I have read RabbitMQ's Community Support Policy

RabbitMQ version used

4.2.x

How is RabbitMQ deployed?

Community Docker image

Steps to reproduce the behavior in question

The management metrics collection has been marked as deprecated, and it was announced that it would be removed in 4.0, but that release has come and gone and the feature has not been removed yet. Is there an updated timeline for when it's planned to be removed?

The reason I'm asking is that, currently, when you set management_agent.disable_metrics_collector = true or use disable_stats=true in an API query string, there is far more removed from the response than what I would consider "stats", things that I feel need to be included to make the management API useful.

The biggest example is policy information. When stats are disabled, it is no longer possible to get the names of the policy and operator policy applied to an entity. The effective policy definition info is also gone.

The delivery limit value on quorum queues is also removed when stats are disabled.

It's hard to see how any of these things qualify as "stats", and removing access to these would completely break the NServiceBus RabbitMQ transport, where we need to inspect and set policies around the delivery limit.

Is there a plan to move things like this out of the metrics collector before removing it? If there isn't, I think there very much needs to be.

[michaelklishin]

It is much more likely that the stats collection in the management plugin will remain at least in 4.x.

We have learned the hard way that some users cannot provision any other monitoring solution. Don't ask why, they just can't, so our management UI is the only source of monitoring data they'd ever have.

We still encourage everyone who can adopt Prometheus and Grafana (or another Prometheus format-based tool). That's a better option, and it includes certain dashboards (Raft metrics, Erlang distribution metrics) that the management UI does not provide https://grafana.com/orgs/rabbitmq.

[bording]

It is much more likely that the stats collection in the management plugin will remain at least in 4.x.

We have learned the hard way that some users cannot provision any other monitoring solution. Don't ask why, they just can't, so our management UI is the only source of monitoring data they'd ever have.

It's good to hear that we won't need to be worrying about changes to the management plugin any time soon.

We still encourage everyone who can adopt Prometheus and Grafana (or another Prometheus format-based tool). That's a better option, and it includes certain dashboards (Raft metrics, Erlang distribution metrics) that the management UI does not provide https://grafana.com/orgs/rabbitmq.

I agree that it makes sense for metrics data be be accessed through dedicated monitoring tools, but the reason I opened this discussion is about all of the non-metrics data that also disappears when when stats collection is disabled. That's what I really want to hear a response on. There is a lot of useful management information in the API that currently requires stats collection to be enabled to be able to access, and there is no replacement for it in the emitted Prometheus metrics. I'd like to get a confirmation from the RabbitMQ team that you're aware of this and would fix this before removing stats collection from the plugin in the future.

[michaelklishin]

@bording you need to list them, at least a few examples. It can be that some potentially can be "re-classified", or that there are technical reasons for why they are not available.

We have to many metrics to guess.

[lukebakken]

From @bording's comment:

The biggest example is policy information. When stats are disabled, it is no longer possible to get the names of the policy and operator policy applied to an entity. The effective policy definition info is also gone.
The delivery limit value on quorum queues is also removed when stats are disabled.

I'll look at this tomorrow @bording 👍 Feels like just an oversight, hopefully!

[michaelklishin]

The reason can be fundamental: the stats database is not available at all when the stats are disabled. That's just a hypothesis.

[lukebakken]

I can reproduce what @bording reports and have opened the following issue: #15182