Fix handling of multiple Kafka hosts when creating alarm tree

[smarsching]

When multiple Kafka hosts are specified as part of the alarm URI, the URI cannot be parsed as a server-based URI, only as a registry-based URI. This means that the getHosts(), getPort(), and getUserInfo() methods return null. Therefore, we have to use getAuthority() instead when cloning the URI.

This PR closes #3071.

[kasemir]

This change looks benign.

I wonder how you're using this. Do you have a cluster with several kafka nodes? Do you still get one compacted topic for an alarm configuration, or is the kafka log split across the nodes?

[smarsching]

I wonder how you're using this. Do you have a cluster with several kafka nodes? Do you still get one compacted topic for an alarm configuration, or is the kafka log split across the nodes?

We use this for high availability, so that we can restart Kafka nodes without affecting operation of the alarm system.

We have three Kafka nodes (using two does not make any sense when running KRaft mode because Kafka needs a quorum of controller nodes). As we have them anyway and all our nodes act as both controller and broker nodes, we replicate all topics across all three nodes.

This way, the number of in-sync replicas can be kept at two, even when one node fails (see https://stackoverflow.com/a/58909999/2326830 for a slightly more detailed explanation). This follows recommendations for production setups from the Kakfa documentation.

Adding additional broker nodes and splitting topics IMO does not make any sense in the context of the alarm system: This method is mainly employed in order to get a higher throughput in case of massively parallel requests, and this scenario does not really apply to the alarm system.

[shroffk]

@kasemir should we reference this somewhere in the alarm server installation/deployment documentation?