*: types refactor

[manasachi]

Modified struct definitions and removed test case for PercentileLatencies

[fuweid]

I think we need to do some refactors before introducing this feature.

1. Rethink ResponseStats.

We need to rethink about ResponseStats. I made a mistake on ResponseMetric.Gather().

kperf/metrics/request.go

Line 20 in 636d5a4

Gather() (latencies []float64, percentileLatencies map[float64]float64, failureList []error, bytes int64)

This function should just return ResponseStats (raw data) instead of percentile latencies.
The caller should take responsibility to make summary on raw data as it wants.

So, I think ResponseStats should be like

// ResponseStats is the report about benchmark result.
type ResponseStats struct {
	// Total represents total number of requests.
	Total int
	// List of failures
	FailureList []error
        // All the observed latencies
	Latencies []float64
	// total bytes read from apiserver
	TotalReceivedBytes int64
}

// ResponseMetric is a measurement related to http response.
type ResponseMetric interface {
	...
	// Gather returns the summary.
	Gather() ResponseStats
}

I remove the Duration from ResponseStats because metric collector doesn't know when it starts and when it ends.
So, the Duration should be filled by Schedule.

kperf/request/schedule.go

Line 20 in 636d5a4

func Schedule(ctx context.Context, spec *types.LoadProfileSpec, restCli []rest.Interface) (*types.ResponseStats, error) {

The Schedule should return their own type instead of ResponseStats.

// in request pkg

type Result struct {
       types.ResponseStats
       // Duration means the time of benchmark.
       Duration time.Duration
}

func Schedule(ctx context.Context, spec *types.LoadProfileSpec, restCli []rest.Interface) (*Result, error) {}

2. Introduce RunnerMetricReport type for command output

For command output, it's client facing. It could be changed or updated frequently.
So, we should introduce new type RunnerMetricReport for render purposes.
If there is no new raw metric, we don't need to touch ResponseStats.
All the aggregated features can be handled by updated RunnerMetricReport type.

// in api/types pkg

type RunnerMetricReport struct {
      // Total represents total number of requests.
	Total int  
	// List of failures
	FailureList []error
        // Duration means the time of benchmark.
        Duration time.Duration
        // All the observed latencies
	Latencies []float64
	// total bytes read from apiserver
	TotalReceivedBytes int64
        // PercentileLatencies represents the latency distribution in seconds.
        PercentileLatencies [][2]float64 // [2]float64{percentile, value}
}

3. Introduce --format=json and --raw-data flags together

Change current format into json. No need to maintain fmt.Fprintf() method.
If --raw-data is set, use json.NewEncoder() to render. If we don't have --raw-data, set Latencies field to nil and render it into json.

What do you think?

[fuweid]

LGTM