FactoriOTel - Factorio with OpenTelemetr...

[fallen quarry]

Very cool! I think the Clusterio people may be interested in this. https://github.com/clusterio/clusterio

I'm not familiar with how Clusterio works but my understanding is they link up several Factorio servers into one game. I can imagine instrumentation like this could be useful for them. You could ask them what instrumentation they use and either get some ideas on it or make FactoriOTel compatable with Clusterio.

[sharp sapphire]

We have our own prometheus exporter that lives here, nice to have a solution that is easy to use standalone though https://github.com/clusterio/clusterio/tree/master/plugins/statistics_exporter

[desert rapids]

Nice to see that this is a common problem ! In your experience, what metrics are important to track besides the flows that you already export ?
I also export data about the power production/consumption but I fear that this one might not scale well.

[sharp sapphire]

Power consumption is definitely one of the more useful ones - very hard to track though because multiple power networks. I suppose just exporting a different label for each network would allow the user to sum them pretty easily?

Other than that, anything that allows for early detection of griefing is very useful for servers. Being able to look at the graph and see where a lot of stuff was deconstructed makes it a lot faster to roll back. There are also different types of grief though - belt rotations, recipe changes, train schedule changes especially. We don't track those.

[desert rapids]

That's interesting, not sure how these kind of actions could be detected in metrics.
Maybe sudden drops in consumption/production that would not be correlated with a natural cause such as ore patch depletion or biter attack. But that would be hard to distinguish from player mistakes

[desert rapids]

that's the route I chose yes, plus tracking the number of entities consuming/producing electricity in the networks to compute theoretical max consumption and production.
But the runtime API doesn't provide any nice way to retrieve this data. The networks retrieved by querying all electric poles and the entities by querying all the entities by name. AFAIK there is no "network created", "network destroyed" or "entity connected to network" event that would save us from scanning all the entities periodically