Less locking for metric vectors.

[robx]

Joint work with @alicebob. @beorn7, you seem to know the vector stuff well?

• Inline the label hashing to remove the need to keep around and lock a shared buffer and hash.
• Try looking up metrics with a read lock before falling back to a write-locked "get or create".

This should improve lock contention issues when using variable labels in a concurrent setting. (We're currently using the hacky https://github.com/realzeitmedia/promvec, which avoids locking altogether.)

The benchmark results look quite good. BenchmarkHandler varies a lot between runs, that delta is not always positive.

$ benchcmp old.txt new.txt 
benchmark                                       old ns/op     new ns/op     delta
BenchmarkCounterWithLabelValues-4               462           144           -68.83%
BenchmarkCounterWithLabelValuesConcurrent-4     561           91.7          -83.65%
BenchmarkCounterWithMappedLabels-4              2360          760           -67.80%
BenchmarkCounterWithPreparedMappedLabels-4      763           178           -76.67%
BenchmarkCounterNoLabels-4                      26.7          22.8          -14.61%
BenchmarkGaugeWithLabelValues-4                 407           129           -68.30%
BenchmarkGaugeNoLabels-4                        15.7          13.7          -12.74%
BenchmarkSummaryWithLabelValues-4               1672          1550          -7.30%
BenchmarkSummaryNoLabels-4                      1671          1408          -15.74%
BenchmarkHistogramWithLabelValues-4             550           191           -65.27%
BenchmarkHistogramNoLabels-4                    72.5          62.4          -13.93%
BenchmarkHistogramObserve1-4                    53.3          45.7          -14.26%
BenchmarkHistogramObserve2-4                    171           200           +16.96%
BenchmarkHistogramObserve4-4                    433           530           +22.40%
BenchmarkHistogramObserve8-4                    987           1070          +8.41%
BenchmarkHistogramWrite1-4                      5718          4103          -28.24%
BenchmarkHistogramWrite2-4                      6783          5710          -15.82%
BenchmarkHistogramWrite4-4                      12864         12178         -5.33%
BenchmarkHistogramWrite8-4                      25073         22742         -9.30%
BenchmarkHandler-4                              12246782      20103907      +64.16%
BenchmarkSummaryObserve1-4                      1917          1950          +1.72%
BenchmarkSummaryObserve2-4                      5591          4818          -13.83%
BenchmarkSummaryObserve4-4                      12727         12346         -2.99%
BenchmarkSummaryObserve8-4                      26542         25127         -5.33%
BenchmarkSummaryWrite1-4                        23767         20058         -15.61%
BenchmarkSummaryWrite2-4                        48698         46691         -4.12%
BenchmarkSummaryWrite4-4                        92121         92998         +0.95%
BenchmarkSummaryWrite8-4                        196476        182895        -6.91%

benchmark                                       old allocs     new allocs     delta
BenchmarkCounterWithLabelValues-4               0              0              +0.00%
BenchmarkCounterWithLabelValuesConcurrent-4     0              0              +0.00%
BenchmarkCounterWithMappedLabels-4              3              2              -33.33%
BenchmarkCounterWithPreparedMappedLabels-4      1              0              -100.00%
BenchmarkCounterNoLabels-4                      0              0              +0.00%
BenchmarkGaugeWithLabelValues-4                 0              0              +0.00%
BenchmarkGaugeNoLabels-4                        0              0              +0.00%
BenchmarkSummaryWithLabelValues-4               0              0              +0.00%
BenchmarkSummaryNoLabels-4                      0              0              +0.00%
BenchmarkHistogramWithLabelValues-4             0              0              +0.00%
BenchmarkHistogramNoLabels-4                    0              0              +0.00%

benchmark                                       old bytes     new bytes     delta
BenchmarkCounterWithLabelValues-4               0             0             +0.00%
BenchmarkCounterWithLabelValuesConcurrent-4     0             0             +0.00%
BenchmarkCounterWithMappedLabels-4              384           336           -12.50%
BenchmarkCounterWithPreparedMappedLabels-4      48            0             -100.00%
BenchmarkCounterNoLabels-4                      0             0             +0.00%
BenchmarkGaugeWithLabelValues-4                 0             0             +0.00%
BenchmarkGaugeNoLabels-4                        0             0             +0.00%
BenchmarkSummaryWithLabelValues-4               0             0             +0.00%
BenchmarkSummaryNoLabels-4                      0             0             +0.00%
BenchmarkHistogramWithLabelValues-4             0             0             +0.00%
BenchmarkHistogramNoLabels-4                    0             0             +0.00%

[beorn7]

Thanks for your contribution.

I like the general idea. I'll have a closer look ASAP.

[beorn7]

👍 Looks perfect to me. Merging...

And thanks again. I guess we can apply the same idea to a few other performance critical paths...