Add OTLP Benchmark to measure push performance

[SungJin1212]

This PR adds a benchmark to measure OTLP push performance.

The Benchmark results are:

goos: darwin
goarch: arm64
pkg: github.com/cortexproject/cortex/pkg/util/push
cpu: Apple M1 Max
BenchmarkOTLPWriteHandlerPush
BenchmarkOTLPWriteHandlerPush/numSeries:1,_samplesPerSeries:10,_numHistograms:1
BenchmarkOTLPWriteHandlerPush/numSeries:1,_samplesPerSeries:10,_numHistograms:1-10         	   26020	     46032 ns/op	   86703 B/op	     534 allocs/op
BenchmarkOTLPWriteHandlerPush/numSeries:1,_samplesPerSeries:100,_numHistograms:1
BenchmarkOTLPWriteHandlerPush/numSeries:1,_samplesPerSeries:100,_numHistograms:1-10        	    6900	    179708 ns/op	  232596 B/op	    2638 allocs/op
BenchmarkOTLPWriteHandlerPush/numSeries:1,_samplesPerSeries:1000,_numHistograms:1
BenchmarkOTLPWriteHandlerPush/numSeries:1,_samplesPerSeries:1000,_numHistograms:1-10       	     855	   1391819 ns/op	 1707477 B/op	   23364 allocs/op
BenchmarkOTLPWriteHandlerPush/numSeries:1,_samplesPerSeries:1,_numHistograms:10
BenchmarkOTLPWriteHandlerPush/numSeries:1,_samplesPerSeries:1,_numHistograms:10-10         	    7507	    154604 ns/op	  265258 B/op	    2129 allocs/op
BenchmarkOTLPWriteHandlerPush/numSeries:1,_samplesPerSeries:1,_numHistograms:100
BenchmarkOTLPWriteHandlerPush/numSeries:1,_samplesPerSeries:1,_numHistograms:100-10        	     871	   1377683 ns/op	 2248653 B/op	   20196 allocs/op
BenchmarkOTLPWriteHandlerPush/numSeries:1,_samplesPerSeries:1,_numHistograms:1000
BenchmarkOTLPWriteHandlerPush/numSeries:1,_samplesPerSeries:1,_numHistograms:1000-10       	      73	  14949837 ns/op	22246728 B/op	  200325 allocs/op
BenchmarkOTLPWriteHandlerPush/numSeries:10,_samplesPerSeries:1,_numHistograms:1
BenchmarkOTLPWriteHandlerPush/numSeries:10,_samplesPerSeries:1,_numHistograms:1-10         	    7959	    153782 ns/op	  231220 B/op	    2070 allocs/op
BenchmarkOTLPWriteHandlerPush/numSeries:100,_samplesPerSeries:1,_numHistograms:1
BenchmarkOTLPWriteHandlerPush/numSeries:100,_samplesPerSeries:1,_numHistograms:1-10        	     906	   1321136 ns/op	 1718853 B/op	   19181 allocs/op
BenchmarkOTLPWriteHandlerPush/numSeries:1000,_samplesPerSeries:1,_numHistograms:1
BenchmarkOTLPWriteHandlerPush/numSeries:1000,_samplesPerSeries:1,_numHistograms:1-10       	      86	  13179998 ns/op	17437157 B/op	  189383 allocs/op

Which issue(s) this PR fixes:
Fixes #

Checklist

• Tests updated
• Documentation added
• CHANGELOG.md updated - the order of entries should be [CHANGE], [FEATURE], [ENHANCEMENT], [BUGFIX]

[SungJin1212]

Currently, our OTLPHandler uses a Prometheus converter, which appears to cause significant memory allocation during the conversion to the Cortex format.

I was thinking we could manage a dedicated Cortex converter, which can directly convert OTLP to Cortex format. This would allow us to skip the intermediate conversion from Prometheus format to Cortex. If the case, the memory allocation will be reduced, but this is expected to come at the cost of tracking overhead.

@yeya24 @friedrichg
WDYT?

[yeya24]

Hey @SungJin1212, I think that makes sense. But at the same time, do we have a benchmark to understand OTLP performance? If not let's create one so that we can compare the performance impact and its baseline

[SungJin1212]

@yeya24
Yeah, let me add a Benchmark to show current memory allocations.

[SungJin1212]

@yeya24
I added a Benchmark to evaluate push performance.

[yeya24]

Thanks!