Increase escapeMetricName performance

[bakins]

While investigating something else,I did some profiling and found that escapeMetricName took a large amount of processing time. The regex was the majority of this time. escapeMetricName is called on every event processed.

The valid set of characters for a metric name is fairly small, so we can optimize this a bit. I replaced the regex in escapeMetricName with a loop over the runes. I originally used a strings.Builder to build the output, but building the string manually provided to be much faster.

This passes tests, but I am sure I am missing something obvious. This is borderline unreadable as well and needs some comments. I did this on a sick day, so I do not trust myself on this.

For benchmarks, see https://gist.github.com/bakins/f4d342d1cc93f4911ffd4efa9d93ecab

[SpencerMalone]

I've been poking at this a bit, I think it's doing the right thing, and I tried playing with some alternate implementations (such as using unicode.Is() and a custom unicode.RangeTable), but nothing topped this in performance.

I think with some comments this would be great, and I personally think it's worth keeping the // Replace all illegal metric chars with underscores. (or similar) comment in place

[matthiasr]

This looks good! I agree, a little commenting would be nice, but I don't think it's less readable than the previous solution.

Why not add the benchmark code (only for the version that is actually there)?

[bakins]

Added some comments and benchmark. Output on my laptop:

$ go test -bench=BenchmarkEscapeMetricName -test.benchmem
goos: darwin
goarch: amd64
pkg: github.com/prometheus/statsd_exporter
BenchmarkEscapeMetricName/clean-12         	50000000	        30.4 ns/op	      10 B/op	       2 allocs/op
BenchmarkEscapeMetricName/0starts_with_digit-12         	20000000	        80.1 ns/op	      64 B/op	       2 allocs/op
BenchmarkEscapeMetricName/with_underscore-12            	30000000	        49.4 ns/op	      32 B/op	       2 allocs/op
BenchmarkEscapeMetricName/with.dot-12                   	50000000	        35.1 ns/op	      16 B/op	       2 allocs/op
BenchmarkEscapeMetricName/with😱emoji-12                 	30000000	        50.8 ns/op	      32 B/op	       2 allocs/op
BenchmarkEscapeMetricName/with.*.multiple-12            	30000000	        51.4 ns/op	      32 B/op	       2 allocs/op
BenchmarkEscapeMetricName/test.web-server.foo.bar-12    	20000000	        65.2 ns/op	      64 B/op	       2 allocs/op
BenchmarkEscapeMetricName/#00-12                        	200000000	         7.72 ns/op	       0 B/op	       0 allocs/op
PASS
ok  	github.com/prometheus/statsd_exporter	13.494s

[matthiasr]

awesome, thank you!