Decode reports from a byte buffer

[bboreham]

This is speculative: empirically it goes faster but theoretically it could make things worse.

So I would like to try it out by merging into master hence into dev, and if we don't like it we can revert.

[rade]

So I would like to try it out by merging into master hence into dev, and if we don't like it we can revert.

The cleaner way to do this is to de-automate the relevant services in flux, and then deploy an image from this branch. cc @2opremio who has done this a few times.

[2opremio]

Yep, I would rather do that before merging. @bboreham mind giving it a try? I can guide you. It basically entails:

1. Modifying flux's config in dev to stop deploying scope automatically
2. Deploying you image manually.

[rade]

Feeding 30s of prod reports to the benchmark with

go test -tags 'netgo unsafe' -run=x -cpu=2 -bench=ReportUnmarshal --bench-report-path=/tmp/prod-20171129/

cdbc01e shows

BenchmarkReportUnmarshal-2   	       1	23049267656 ns/op	1239968760 B/op	21422077 allocs/op

and the final commit

BenchmarkReportUnmarshal-2   	       1	11998462779 ns/op	2679079128 B/op	19529968 allocs/op

That's a massive improvement!