Embed micrometer

[z0mb1ek]

HI! Micrometer is a standard in the industry, it would be great if we had consumer metrics similar to spring kafka.

[scottf]

Can you enumerate things that might benefit from observation? You said consumer metrics, but I'm not sure what you mean by this. Does the io.nats.client.Statistics have any information you want?

[z0mb1ek]

I mean statistics per consumer, for example, counting the number of messages per second, delays in processing one message, the number of bytes per message, and so on.

[scottf]

We definitely do not want to enforce one specific framework - you notice that SLF4J is not even used. So even though Micrometer may be an equivalent to SLF4j, it's still a framework. This is the same reason we use a hand coded json parser, to prevent inclusion of libraries that are not standard to the JVM.
Literally the one exception to this is the fact that we use BouncyCastle.

Also, considering that the user is already receiving messages, there is no reason they cannot add their own collection of metric data using any method they want.

So maybe it's me, but I'm not seeing the value here. For instance, if I'm in production, I should already understand the size of my messages and also understand the volume of messages. On the server side, both these pieces of information are needed to tune the number of servers, the size of cluster, storage, networking etc.

On the client side, this information is needed to properly design the architecture of number of connections and how to horizontally scale consumers.

This is something we see done with benchmarking in an environment trying to replicate production.

[scottf]

@z0mb1ek Can you please tell us something about your company and your use case for NATS?

[z0mb1ek]

I've been using nats for various projects for several years. This is usually event sourcing or some kind of data processing. At the moment, for example, I have rolled back my falling instance version 2.12.2 to 2.12.0 and am looking forward to further memory-eating fixes.