Perf fix for encode and decode to use bytes data type

[srikartati]

As part of Antrea benchmarking, we found that bytes.Buffer type is incurring very high overhead in terms of allocations and memory consumed. Moving to bytes data type instead of bytes.Buffer wrapper.

Fixes: #205

Before fix:

vagrantÉk8s-node-worker-1:ü/antrea/pkg/flowaggregator$ go test -test.v -run=none -test.benchmem  -bench=. -memprofile memprofile.out -cpuprofile profile.out
goos: linux
goarch: amd64
pkg: antrea.io/antrea/pkg/flowaggregator
cpu: Intel(R) Core(TM) i9-9980HK CPU É 2.40GHz
BenchmarkIntraNodeFlowRecords
    flowaggregator_perf_test.go:446: Num messages received: 1516755
BenchmarkIntraNodeFlowRecords-2   	       1	120001712897 ns/op	22902440680 B/op	431854675 allocs/op
PASS
ok  	antrea.io/antrea/pkg/flowaggregator	120.119s

After this fix:

vagrant@k8s-node-worker-1:~/antrea/pkg/flowaggregator$ go test -test.v -run=none -test.benchmem  -bench=. -memprofile memprofile.out -cpuprofile profile.out
goos: linux
goarch: amd64
pkg: antrea.io/antrea/pkg/flowaggregator
cpu: Intel(R) Core(TM) i9-9980HK CPU @ 2.40GHz
BenchmarkIntraNodeFlowRecords
    flowaggregator_perf_test.go:446: Num messages received: 1526508
BenchmarkIntraNodeFlowRecords-2   	       1	120005595596 ns/op	19760001232 B/op	356061206 allocs/op
PASS
ok  	antrea.io/antrea/pkg/flowaggregator	120.328s

[codecov]

Codecov Report

Merging #207 (72263b4) into main (6d5fafa) will increase coverage by 0.57%.
The diff coverage is 65.21%.

@@            Coverage Diff             @@
##             main     #207      +/-   ##
==========================================
+ Coverage   78.79%   79.36%   +0.57%     
==========================================
  Files          17       17              
  Lines        2155     2118      -37     
==========================================
- Hits         1698     1681      -17     
+ Misses        307      300       -7     
+ Partials      150      137      -13     

Flag	Coverage Δ
integration-tests	56.95% <52.17%> (-0.56%)	⬇️
unit-tests	79.03% <65.21%> (+0.57%)	⬆️

Flags with carried forward coverage won't be shown. Click here to find out more.

Impacted Files	Coverage Δ
pkg/entities/set.go	70.58% <50.00%> (-6.01%)	⬇️
pkg/exporter/process.go	65.86% <50.00%> (-1.00%)	⬇️
pkg/collector/process.go	87.20% <54.54%> (-2.27%)	⬇️
pkg/entities/ie.go	55.17% <64.93%> (+0.91%)	⬆️
pkg/entities/record.go	76.71% <70.00%> (+7.48%)	⬆️
pkg/intermediate/aggregate.go	73.54% <100.00%> (+0.40%)	⬆️

[zyiou]

Overall LGTM. Thanks for the changes. Encoding, decoding function and the tests looks much cleaner.

[srikartati]

Overall LGTM. Thanks for the changes. Encoding, decoding function and the tests looks much cleaner.

Thanks for the review, Yiou.
I think we can optimize further by doing pre-allocation of bytes slice in the record by using total elements size. This will improve the performance further. Will take it up in future PR.