feat: improve performance of getMetricAsPrometheusString

[shappir]

Use Array.prototype.join instead string concatenation for constructing Prometheus metric string. This results in improved performance and reduced memory consumption, especially when there's a large amount of dimensions (labels and label values). In addition uses array iteration methods instead of explicit loops.

[SimenB]

Thanks! Do you have any benchmarks before/after?

https://github.com/siimon/prom-client/tree/master/benchmarks

[shappir]

This change primarily impacts the performance of metrics that have a large number of rows in the Prometheus metrics output, e.g. histograms with several labels. In our application I saw metric generation time reduced by ~50%. This is not surprising since the number of times string values are copied as result of concatination is the same order as the number of rows.

In the benchmark test I got 50% faster and 50% "acceptably slower". Every once in a while I would get one entry or another as slower, but it was inconsistent.

[SimenB]

Cool, thanks!

[SimenB]

Hey @shappir! I just released this in 14.2.0.

However, I'm having some issues trying to rebase next on master on top of #544. You wouldn't by any chance be able to send a forward-porting PR to next applying the changes in this PR?

[shappir]

Is this still an issue? Do you need me to do anything?

[SimenB]

If it's not too much trouble, a PR against next applying the same optimizations would be very helpful 🙂

[shappir]

Created a pr for next: #548