Show aggregated process metrics in containers/k8s pods

[2opremio]

As a user I would like to see things such as open file descriptors (or plugin metrics), at the container level (and kubernetes pod level) instead of clicking on each process inside the container to view them.

In fact, obtaining the CPU/memory consumption of the container by aggregating the CPU/memory of processes would also fix #1133

[2opremio]

According to @tomwilkie we probably want to do this sort of aggregations through prometheus.

[alepuccetti]

Aggregation Proposal

Proposal on how to implementation data sources aggregation in scope.

Aggregation Architecture for Scope

To describe an aggregation type we need 2 attributes:

• AggregationPolicy: define how to perform the aggregation default, max, min, sum, average, median, list,...
• AggregationLevel: where to perform the aggregation container, host, pod, replication-controller,...

All the data source types that will allow aggregation must be extended to include these 2 attributes.

type AggregationType {
    AggregationPolicy   string
    AggregationLevel    string
}

type MetricTemplate struct {
    ID              string  `json:"id"`
    Label           string  `json:"label,omitempty"`
    Format          string  `json:"format,omitempty"`
    Priority        float64 `json:"priority,omitempty"`
    Aggregations    []AggregationType `json:"aggregations,omitempty"`
}

Allowing a list of AggregationType makes the aggregation more flexible, it is possible to specify many policies for the same level (or different policies for different levels) for the same data source and as result, the UI could show multiple different aggregate values accordingly with the specified values.
I think that extending the templates is sufficient and the actual data point do not need to be changed because we want to allow aggregate a specific type of measure.

List of data structure to extend:

scope/report/metadata_template.go

Line 22 in b27dc99

type MetadataTemplate struct {

scope/report/metric_template.go

Line 9 in 7b0f0cb

type MetricTemplate struct {

scope/report/table.go

Line 83 in 7b0f0cb

type TableTemplate struct {

JSON Example

This example is based on the structure of the HTTP statistics report

{
  "Process": {
    "metric_templates": {
      "http_requests_per_second": {
        "id": "http_requests_per_second",
        "label": "HTTP Req/Second",
        "priority": 0.1,
        "aggregations": [
             {
               "aggregation_policy": "sum",
               "aggregation_level": "container”
             },
             {
               "aggregation_policy": "max",
               "aggregation_level": "container”
             }
          ]
      }
    },
    "nodes": {
      "example.org;29770": {
        "metrics": {
          "http_requests_per_second": {
            "samples": [
              {
                "date": "2016-09-21T07:22:24.293175Z",
                "value": 1.0
              }
            ]
          }
        }
      }
    }
  }
}

At first we can implement the "sum" policy, but the json types are expressive enough to be able to add more later without breaking compatibility (e.g. max, min, avarage, list, ...).

/cc @fons, @alban, @tomwilkie

[rade]

AggregationLevels

That should be singular, right? i.e. the value is the name of a single level.

[alepuccetti]

That should be singular, right? i.e. the value is the name of a single level.

Yes, that is a typo. Sorry about that.

[rade]

@rndstr I believe your work in #2664 etc will address this.