exporter/collector: avoid using req.String() to check for empty request

[ridwanmsharif]

This change is motivated by some CPU profiling that showed that its responsible for over 20% of CPU time in my environment (running avalanche, OTel with Prom receiver and the gmp exporter). Sending this change because its logically equivalent to before but avoids the serialization to the string.

Flamegraph: pprof/?id=f2daddf0bdf127bfa136618996b263c2&tab=flame&path=1ygnz740h91y6o

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 62.99%. Comparing base (4caace7) to head (15099b9).
Report is 40 commits behind head on main.

Additional details and impacted files

@@            Coverage Diff             @@
##             main     #889      +/-   ##
==========================================
+ Coverage   61.03%   62.99%   +1.96%     
==========================================
  Files          56       57       +1     
  Lines        5903     6013     +110     
==========================================
+ Hits         3603     3788     +185     
+ Misses       2143     2063      -80     
- Partials      157      162       +5     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[bwplotka]

LGTM, amazing work! Sounds like our benchmarks already has been proven to be useful.

The best optimizations are one-liners 💪🏽

I guess it would be nice to run some microbenchmark here to double check, but in theory this be faster vs stringifying

[ridwanmsharif]

Added some microbenchmarks, here are some savings:

➜  collector git:(ridwanmsharif/check-empty) ✗ go test -bench . -benchmem -count=10 > orig.txt
... applied the fix ...
➜  collector git:(ridwanmsharif/check-empty) ✗ go test -bench . -benchmem -count=10 > fix.txt 
➜  collector git:(ridwanmsharif/check-empty) ✗ benchstat orig.txt fix.txt            
goos: linux
goarch: amd64
pkg: github.com/GoogleCloudPlatform/opentelemetry-operations-go/exporter/collector
cpu: Intel(R) Xeon(R) CPU @ 2.20GHz
                         │  orig.txt   │               fix.txt               │
                         │   sec/op    │   sec/op     vs base                │
_TestExport-16             3.194µ ± 7%   1.910µ ± 7%  -40.21% (p=0.000 n=10)
_TestReadWALAndExport-16   7.585m ± 3%   6.797m ± 3%  -10.39% (p=0.000 n=10)
geomean                    155.6µ        113.9µ       -26.80%

                         │   orig.txt   │               fix.txt                │
                         │     B/op     │     B/op      vs base                │
_TestExport-16               953.0 ± 0%     656.0 ± 0%  -31.16% (p=0.000 n=10)
_TestReadWALAndExport-16   8.060Ki ± 1%   7.466Ki ± 0%   -7.36% (p=0.000 n=10)
geomean                    2.739Ki        2.187Ki       -20.14%

                         │  orig.txt  │              fix.txt               │
                         │ allocs/op  │ allocs/op   vs base                │
_TestExport-16             20.00 ± 0%   10.00 ± 0%  -50.00% (p=0.000 n=10)
_TestReadWALAndExport-16   128.0 ± 0%   108.0 ± 0%  -15.62% (p=0.000 n=10)
geomean                    50.60        32.86       -35.05%

When WAL is disabled (default) the savings are higher, but with WAL this still has significant savings.

Note that these numbers are pretty fudged since the exporting logic is omitted, but some significant improvement is visible nonetheless

[damemi]

/gcbrun