Am I using the correct library?

[genglinxiao]

Hi there,

This should probably go to a forum instead of issues, but I cannot find more suitable and active place for it:

I'm developing a dotnet application that extract data from a device and then feed it into a monitoring backend that uses Prometheus. I was pointed to this library and I was able to get it going, except that there are more default metrics than my own metrics. I was searching for ways to disable the default metrics and then saw this thread: 80

If I get it correctly, the user of this library is supposed to append rather than change metrics, which makes me wondering, am I using the correct library? If not, what is the proper way of developing a customized prometheus client using dotnet/C# on Linux platform?

Thanks!

[sandersaares]

Does the chapter in the readme on Suppressing default metrics provide what you need?

[genglinxiao]

Does the chapter in the readme on Suppressing default metrics provide what you need?

Hi @sandersaares , yes, suppressing default metrics would work for me. However, my question is more on the design side. Maybe, this library is designed for scenarios quite different from mine and suppressing default metrics is a kind of hack that force this library into my application?

[sandersaares]

In general, if you can think of metrics as "observing a variable in my code" then this library is likely well suited as long as your observations can be modeled using the Prometheus APIs of counter, gauge and histogram.

Where people sometimes experience difficulties is when they want to do ad-hoc transformation from arbitrary data into Prometheus data. For a simple example, people sometimes ask to make a counter decrease in value. In a Prometheus context, that is not a valid operation, so this library does not allow that - you cannot go from "There have been 9 events observed" to "There have been 8 events observed" (unless you apply a hack such as deleting and re-creating the metric). Sometimes people want this because they are just trying to translate some other system to the Prometheus format. This library is not suitable for that - it only allows operations that make sense in a Prometheus context.

[genglinxiao]

In general, if you can think of metrics as "observing a variable in my code" then this library is likely well suited as long as your observations can be modeled using the Prometheus APIs of counter, gauge and histogram.

Where people sometimes experience difficulties is when they want to do ad-hoc transformation from arbitrary data into Prometheus data. For a simple example, people sometimes ask to make a counter decrease in value. In a Prometheus context, that is not a valid operation, so this library does not allow that - you cannot go from "There have been 9 events observed" to "There have been 8 events observed" (unless you apply a hack such as deleting and re-creating the metric). Sometimes people want this because they are just trying to translate some other system to the Prometheus format. This library is not suitable for that - it only allows operations that make sense in a Prometheus context.

Thanks! That's understandable. How about monitoring multiple clients? I'm sorry if this is too obvious, but I may have to get data from multiple devices, how can I differentiate the items from different devices? All these devices are connected to the same computer that runs my code. Do I need a different library to support this?