Re: Sum all metrics based on one label

[Brian Candler]

I'm afraid I don't understand the question.

You showed a graph with one metric (actually zero metrics where there are gaps).  But then you said "combine 3 metrics above".  Which 3 metrics?  Can you show examples of the metrics with their labels?

If you mean the three sections of that line graph, then the legend shows that those line segments are all from the *same* timeseries.  A timeseries is defined as the same metric name and the same set of labels.  If there were multiple timeseries then prometheus would have shown them in the legend, and given each timeseries a different colour.

There could be multiple underlying timeseries which the sum() combines: that is, the subexpression function_calls_total{function="accelbyte-deleteusingsdk-fauzantest-1a00"} could select multiple timeseries - but since all those timeseries have the same value for the "function" label, sum by (function) (...) returns only a single timeseries.

As for the value resetting: you'll need to explain what this metric represents, and how it is generated.

if it's a counter, then resets are going to happen occasionally when your exporter restarts (unless it has a way to persist the counter when restarting).  Functions like rate() can be used to calculate rates of increase, which will skip the period where the counter resets, and increase() will extrapolate the rate over the given time period.  It makes no sense to sum() a counter with itself.

If it's a gauge, then something has gone down, and that's a matter of truth that the graph is showing you.

An example of the first case is if the value is incremented every time a function is called.  An example of the second case is a value showing you the number of workers which are available to process a given function.