monitoring: TimeSeriesData_PointData behavior vs documentation

[ilevinpanw]

Client

Monitoring/MonitoringPB

Environment

N/A

Go Environment

go version go1.21.8 darwin/arm64

Code

Code taken directly from metric.pb.go:

// A point's value columns and time interval. Each point has one or more
// point values corresponding to the entries in `point_descriptors` field in
// the TimeSeriesDescriptor associated with this object.
type TimeSeriesData_PointData struct {
  state         protoimpl.MessageState
  sizeCache     protoimpl.SizeCache
  unknownFields protoimpl.UnknownFields
  
  // The values that make up the point.
  Values []*TypedValue `protobuf:"bytes,1,rep,name=values,proto3" json:"values,omitempty"`
  // The time interval associated with the point.
  TimeInterval *TimeInterval `protobuf:"bytes,2,opt,name=time_interval,json=timeInterval,proto3" json:"time_interval,omitempty"`
}

Sample code:

func materializeTimeSeriesData[T number](iter *monitoring.TimeSeriesDataIterator) ([]T, error) {
    res := make([]T, 0)
    
    resp, err := iter.Next()
    if errors.Is(err, iterator.Done) {
      return []T{}, nil
    }
    
    if err != nil {
      return nil, err
    }
    
    for _, p := range resp.GetPointData() {
      values := p.GetValues()
      pointData := values[0].GetValue()
      ...
    }
}

Expected behavior

It is expected according to the documentation that Values will always have a length of 1 and above.

Actual behavior

Due to some temporary issue with the GCP Metrics service, length of Values was zero. It also has the omitempty tag which is confusing.

I'd appreciate a clarification of what the intended behavior is. Thanks.