producer: increase performance of sampling rate lookup

[vincentbernat]

Formatting a string with version and domain when looking up sampling rate hurts performance. Instead, use a dedicated structure for key. It also reduces the pressure on the GC.

Before:

BenchmarkDecodeEncodeNetflow/with_encoding-12             158208              8611 ns/op            8178 B/op        130 allocs/op
BenchmarkDecodeEncodeNetflow/without_encoding-12          162832              7559 ns/op            8178 B/op        130 allocs/op

After:

BenchmarkDecodeEncodeNetflow/with_encoding-12             138242              8132 ns/op            8176 B/op        129 allocs/op
BenchmarkDecodeEncodeNetflow/without_encoding-12          162312              7306 ns/op            8176 B/op        129 allocs/op

[lspgn]

Thank you! Could you add the benchmarks?

Not seeing a major difference in perfs though:

BenchmarkTemplatesAdd
    templates_test.go:15: Creating 1 templates
    templates_test.go:15: Creating 100 templates
    templates_test.go:15: Creating 10000 templates
    templates_test.go:15: Creating 1000000 templates
    templates_test.go:15: Creating 6573939 templates
BenchmarkTemplatesAdd-4      	 6573939	       171.8 ns/op
BenchmarkTemplatesAddGet
    templates_test.go:22: Adding 1000 templates
    templates_test.go:24: Getting 1 templates
    templates_test.go:22: Adding 1000 templates
    templates_test.go:24: Getting 100 templates
    templates_test.go:22: Adding 1000 templates
    templates_test.go:24: Getting 10000 templates
    templates_test.go:22: Adding 1000 templates
    templates_test.go:24: Getting 1000000 templates
    templates_test.go:22: Adding 1000 templates
    templates_test.go:24: Getting 24474410 templates
BenchmarkTemplatesAddGet-4   	24474410	        47.66 ns/op

Vs

BenchmarkTemplatesAdd
    templates_test.go:15: Creating 1 templates
    templates_test.go:15: Creating 100 templates
    templates_test.go:15: Creating 10000 templates
    templates_test.go:15: Creating 1000000 templates
    templates_test.go:15: Creating 4606766 templates
    templates_test.go:15: Creating 7184278 templates
BenchmarkTemplatesAdd-4      	 7184278	       168.3 ns/op
BenchmarkTemplatesAddGet
    templates_test.go:22: Adding 1000 templates
    templates_test.go:24: Getting 1 templates
    templates_test.go:22: Adding 1000 templates
    templates_test.go:24: Getting 100 templates
    templates_test.go:22: Adding 1000 templates
    templates_test.go:24: Getting 10000 templates
    templates_test.go:22: Adding 1000 templates
    templates_test.go:24: Getting 1000000 templates
    templates_test.go:22: Adding 1000 templates
    templates_test.go:24: Getting 24551028 templates
BenchmarkTemplatesAddGet-4   	24551028	        46.35 ns/op

[vincentbernat]

Well, you get the performance increase too (~2%). Moreover, because there is an allocation less, this reduces GC pressure.

[lspgn]

Oh yes, it's a good improvement. Was just hoping to dive into the benchmarks to understand, if easy to add? Otherwise I'll just merge.

[vincentbernat]

That was an end-to-end benchmark I have in Akvorado.