PC-10133: Move SLO into separate package

[nieomylnieja]

This PR introduces:

• Almost-complete rewrite of SLO validation.
• Full test coverage of all the validated cases.
• Moves SLO to its own package.
• Major changes to validation library, namely:
  • New builtins
  • Intelligent pointer handling with ForPointer
  • More flexibility to errors handling, rules creation and Validator behaviour
• Added ability to record tests, run:

  NOBL9_SDK_TEST_RECORD_FILE=/home/mh/test-output make test

  Make sure the path is absolute!

The error handling after these changes solidifies with the following inter object dependencies:

The following benchmark was run to asses the performance of the solution:

var inputSLOs = func() []SLO {
	rec, err := os.Open("/home/mh/nobl9/recorded-tests.json")
	if err != nil {
		panic(err)
	}
	var slos []SLO
	if err = json.NewDecoder(rec).Decode(&slos); err != nil {
		panic(err)
	}
	return slos
}()

func BenchmarkValidate(b *testing.B) {
	for i := 0; i < b.N; i++ {
		for i := range inputSLOs {
			_ = validator.Check(inputSLOs[i])
		}
	}
}

With the following results for erroneous objects (around 240 of them):

• old code:

  $ go test -bench=Validate -benchtime=10s -benchmem
  goos: linux
  goarch: amd64
  pkg: github.com/nobl9/nobl9-go/manifest/v1alpha
  cpu: 11th Gen Intel(R) Core(TM) i5-1145G7 @ 2.60GHz
  BenchmarkValidate-8         7450           1640675 ns/op         2093472 B/op      22081 allocs/op
  PASS
  ok      github.com/nobl9/nobl9-go/manifest/v1alpha      12.411s

• new code:

  $ go test -bench=Validate -benchtime=10s -benchmem
  goos: linux
  goarch: amd64
  pkg: github.com/nobl9/nobl9-go/manifest/v1alpha/slo
  cpu: 11th Gen Intel(R) Core(TM) i5-1145G7 @ 2.60GHz
  BenchmarkValidate-8         2427           4949896 ns/op         1571118 B/op      35522 allocs/op
  PASS
  ok      github.com/nobl9/nobl9-go/manifest/v1alpha/slo  12.559s

This is expected, as there's much more function calls and details provided to errors.
Let's compare clean SLO:

• old code:

  $ go test -bench=Validate -benchtime=10s -benchmem
  goos: linux
  goarch: amd64
  pkg: github.com/nobl9/nobl9-go/manifest/v1alpha
  cpu: 11th Gen Intel(R) Core(TM) i5-1145G7 @ 2.60GHz
  BenchmarkValidate-8       534976             21294 ns/op           11483 B/op         48 allocs/op
  PASS
  ok      github.com/nobl9/nobl9-go/manifest/v1alpha      11.637s

• new code:

  $ go test -bench=Validate -benchtime=10s -benchmem
  goos: linux
  goarch: amd64
  pkg: github.com/nobl9/nobl9-go/manifest/v1alpha/slo
  cpu: 11th Gen Intel(R) Core(TM) i5-1145G7 @ 2.60GHz
  BenchmarkValidate-8       470991             25588 ns/op            9036 B/op         20 allocs/op
  PASS
  ok      github.com/nobl9/nobl9-go/manifest/v1alpha/slo  12.345s

Here, the difference is not as noticeable anymore, more so, the new code now allocates much less, but that also depends on an integration, I've run it only for Instana as it has a fairly complex validation.

FAQ:

Question might arise: how to do cross field checks?
Here a second question may appear: what is a cross field check?
Let's say we want to compare a value of a deeply nested field to a value of another field which resides in a different place on a struct properties tree. Say we want to make sure spec.indicator.metricSource.project is equal to metadata.project. The proposed validation DOES NOT allow reverse access to properties, you can only access your children. Hence the only way to achieve that is to simply create a rule on a higher level (in the example, on the top SLO level), which has the access to the properties in question and use WithName method to associate it with the downstream path of spec.indicator.metricSource.project, alternatively you can also return PropertyError from your custom top level validation rule.

Due to rebasing shenanigans the commits history is botched...

[kskitek]

Uh Oh! that's another big refactor

I like the new validation. It's very verbose but also very readable and easily adjustable (and testable!)

I also agree with the structure of Rules and Validators.

I didn't check the PR in details though and left just a few comments/questions.

[skrolikiewicz]