Metric for Network partition detected

[Sallaxer]

Is your feature request related to a problem? Please describe.

Cluster status - cluster has experienced a network partition

Describe the solution you'd like

Metric that shows the status of the cluster: Network partition detected

Describe alternatives you've considered

A third party exporter has a metric rabbitmq_partitions that can be used to set up alerts for this cluster state

Additional context

I am using RabbitMQ version 3.12.12 and Erlang 26.2.1. I see that there is a metric rabbitmq_unreachable_cluster_peers_count. But its value is 0 while in the web interface: Network partition detected
Mnesia reports that this RabbitMQ cluster has experienced a network partition. There is a risk of losing data
This behavior can be easily reproduced, for example, if you block on one of the nodes the other nodes in the cluster (using iptables) and then unblock them again. Metrics in Prometheus will be available from all nodes, but if a third-party exporter shows a problem using the rabbitmq_partitions metric, the built-in one does not transmit such metrics and how to know that the state of the cluster is out of order is not clear (what kind of alert can be configured?).

[lukebakken]

Hello, and thanks for using RabbitMQ.

rabbitmq_unreachable_cluster_peers_count was introduced here:

• Prometheus: expose a metric for network partition observed on the node #2508
• Expose number of unreachable cluster peers via Prometheus #9465

If there is an issue that can be reproduced, please provide EXACT steps for doing so.

if you block on one of the nodes the other nodes in the cluster (using iptables) and then unblock them again

This might seem easy but anything you can do (like provide the exact commands you ran) would help us focus on the issue rather than wasting time trying to reproduce it. Even better would be a docker compose project demonstrating the issue.

If we find an actual bug, an issue can be filed.

[michaelklishin]

The idea behind the above metric was exactly the same as what you have in mind.

However, this metric will work fine in the world of 4.x/Khepri where most “partition handling” states will go away entirely since the recovery algorithm for everything will be that of Raft.

Therefore introducing a different version of these metrics does not make much sense.

This metric also matches the Prometheus take on metric scraping while cluster state metrics generally do not.

[Sallaxer]

Thanks for the quick reply )
Ok, am I understanding correctly that right now it is not possible to track this cluster state by the rabbitmq_unreachable_cluster_peers_count metric or indirectly any other metric? In future 4.x it will respond correctly to this state? And of the options, for now only using a third-party exporter?