With the existence of LabelMap, there are simpler mechanisms available to give multiple callers a set of labels to apply to all metrics calls.
Problem areas:
- copy constructor
- reset - I already have a TODO in the code about how reset is poorly implemented, and this will need to be fixed.
- metrics gathering - a derived metric has already had its stats reported and should return empty set
In the bargain however, this one feature addresses #777, #407, and #598 all without introducing performance regressions.
Additional Concerns
Based on conversation in #813, it would be useful for consistency with the other client implementations if, when all of the labels have been captured (full coverage versus #407's currying suggestion) that the derived gauge be initialized to 0 if it is not already in the store.
With the existence of LabelMap, there are simpler mechanisms available to give multiple callers a set of labels to apply to all metrics calls.
Problem areas:
In the bargain however, this one feature addresses #777, #407, and #598 all without introducing performance regressions.
Additional Concerns
Based on conversation in #813, it would be useful for consistency with the other client implementations if, when all of the labels have been captured (full coverage versus #407's currying suggestion) that the derived gauge be initialized to 0 if it is not already in the store.