Aggregation

Describes how to combine multiple time series to provide a different view of the data. Aggregation of time series is done in two steps. First, each time series in the set is aligned to the same time interval boundaries, then the set of time series is optionally reduced in number.

Alignment consists of applying the perSeriesAligner operation to each time series after its data has been divided into regular alignmentPeriod time intervals. This process takes all of the data points in an alignment period, applies a mathematical transformation such as averaging, minimum, maximum, delta, etc., and converts them into a single data point per period.

Reduction is when the aligned and transformed time series can optionally be combined, reducing the number of time series through similar mathematical transformations. Reduction involves applying a crossSeriesReducer to all the time series, optionally sorting the time series into subsets with groupByFields, and applying the reducer to each subset.

The raw time series data can contain a huge amount of information from multiple sources. Alignment and reduction transforms this mass of data into a more manageable and representative collection of data, for example "the 95% latency across the average of all tasks in a cluster". This representative data can be more easily graphed and comprehended, and the individual time series data is still available for later drilldown. For more details, see Filtering and aggregation.

JSON representation 
{
  "alignmentPeriod": string,
  "perSeriesAligner": enum (Aligner),
  "crossSeriesReducer": enum (Reducer),
  "groupByFields": [
    string
  ]
}
Fields
alignmentPeriod

string (Duration format)

The alignmentPeriod specifies a time interval, in seconds, that is used to divide the data in all the time series into consistent blocks of time. This will be done before the per-series aligner can be applied to the data.

The value must be at least 60 seconds. If a per-series aligner other than ALIGN_NONE is specified, this field is required or an error is returned. If no per-series aligner is specified, or the aligner ALIGN_NONE is specified, then this field is ignored.

The maximum value of the alignmentPeriod is 104 weeks (2 years) for charts, and 90,000 seconds (25 hours) for alerting policies.
perSeriesAligner

enum (Aligner)

An Aligner describes how to bring the data points in a single time series into temporal alignment. Except for ALIGN_NONE, all alignments cause all the data points in an alignmentPeriod to be mathematically grouped together, resulting in a single data point for each alignmentPeriod with end timestamp at the end of the period.

Not all alignment operations may be applied to all time series. The valid choices depend on the metricKind and valueType of the original time series. Alignment can change the metricKind or the valueType of the time series.

Time series data must be aligned in order to perform cross-time series reduction. If crossSeriesReducer is specified, then perSeriesAligner must be specified and not equal to ALIGN_NONE and alignmentPeriod must be specified; otherwise, an error is returned.
crossSeriesReducer

enum (Reducer)

The reduction operation to be used to combine time series into a single time series, where the value of each data point in the resulting series is a function of all the already aligned values in the input time series.

Not all reducer operations can be applied to all time series. The valid choices depend on the metricKind and the valueType of the original time series. Reduction can yield a time series with a different metricKind or valueType than the input time series.

Time series data must first be aligned (see perSeriesAligner) in order to perform cross-time series reduction. If crossSeriesReducer is specified, then perSeriesAligner must be specified, and must not be ALIGN_NONE. An alignmentPeriod must also be specified; otherwise, an error is returned.
groupByFields[]

string

The set of fields to preserve when crossSeriesReducer is specified. The groupByFields determine how the time series are partitioned into subsets prior to applying the aggregation operation. Each subset contains time series that have the same value for each of the grouping fields. Each individual time series is a member of exactly one subset. The crossSeriesReducer is applied to each subset of time series. It is not possible to reduce across different resource types, so this field implicitly contains resource.type. Fields not specified in groupByFields are aggregated away. If groupByFields is not specified and all the time series have the same resource type, then the time series are aggregated into a single output time series. If crossSeriesReducer is not defined, this field is ignored.

Aligner

The Aligner specifies the operation that will be applied to the data points in each alignment period in a time series. Except for ALIGN_NONE, which specifies that no operation be applied, each alignment operation replaces the set of data values in each alignment period with a single value: the result of applying the operation to the data values. An aligned time series has a single data value at the end of each alignmentPeriod.

An alignment operation can change the data type of the values, too. For example, if you apply a counting operation to boolean values, the data valueType in the original time series is BOOLEAN, but the valueType in the aligned result is INT64.

Enums
ALIGN_NONE No alignment. Raw data is returned. Not valid if cross-series reduction is requested. The valueType of the result is the same as the valueType of the input.
ALIGN_DELTA

Align and convert to DELTA. The output is delta = y1 - y0.

This alignment is valid for CUMULATIVE and DELTA metrics. If the selected alignment period results in periods with no data, then the aligned value for such a period is created by interpolation. The valueType of the aligned result is the same as the valueType of the input.
ALIGN_RATE

Align and convert to a rate. The result is computed as rate = (y1 - y0)/(t1 - t0), or "delta over time". Think of this aligner as providing the slope of the line that passes through the value at the start and at the end of the alignmentPeriod.

This aligner is valid for CUMULATIVE and DELTA metrics with numeric values. If the selected alignment period results in periods with no data, then the aligned value for such a period is created by interpolation. The output is a GAUGE metric with valueType DOUBLE.

If, by "rate", you mean "percentage change", see the ALIGN_PERCENT_CHANGE aligner instead.
ALIGN_INTERPOLATE Align by interpolating between adjacent points around the alignment period boundary. This aligner is valid for GAUGE metrics with numeric values. The valueType of the aligned result is the same as the valueType of the input.
ALIGN_NEXT_OLDER Align by moving the most recent data point before the end of the alignment period to the boundary at the end of the alignment period. This aligner is valid for GAUGE metrics. The valueType of the aligned result is the same as the valueType of the input.
ALIGN_MIN Align the time series by returning the minimum value in each alignment period. This aligner is valid for GAUGE and DELTA metrics with numeric values. The valueType of the aligned result is the same as the valueType of the input.
ALIGN_MAX Align the time series by returning the maximum value in each alignment period. This aligner is valid for GAUGE and DELTA metrics with numeric values. The valueType of the aligned result is the same as the valueType of the input.
ALIGN_MEAN Align the time series by returning the mean value in each alignment period. This aligner is valid for GAUGE and DELTA metrics with numeric values. The valueType of the aligned result is DOUBLE.
ALIGN_COUNT Align the time series by returning the number of values in each alignment period. This aligner is valid for GAUGE and DELTA metrics with numeric or Boolean values. The valueType of the aligned result is INT64.
ALIGN_SUM Align the time series by returning the sum of the values in each alignment period. This aligner is valid for GAUGE and DELTA metrics with numeric and distribution values. The valueType of the aligned result is the same as the valueType of the input.
ALIGN_STDDEV Align the time series by returning the standard deviation of the values in each alignment period. This aligner is valid for GAUGE and DELTA metrics with numeric values. The valueType of the output is DOUBLE.
ALIGN_COUNT_TRUE Align the time series by returning the number of True values in each alignment period. This aligner is valid for GAUGE metrics with Boolean values. The valueType of the output is INT64.
ALIGN_COUNT_FALSE Align the time series by returning the number of False values in each alignment period. This aligner is valid for GAUGE metrics with Boolean values. The valueType of the output is INT64.
ALIGN_FRACTION_TRUE Align the time series by returning the ratio of the number of True values to the total number of values in each alignment period. This aligner is valid for GAUGE metrics with Boolean values. The output value is in the range [0.0, 1.0] and has valueType DOUBLE.
ALIGN_PERCENTILE_99 Align the time series by using percentile aggregation. The resulting data point in each alignment period is the 99th percentile of all data points in the period. This aligner is valid for GAUGE and DELTA metrics with distribution values. The output is a GAUGE metric with valueType DOUBLE.
ALIGN_PERCENTILE_95 Align the time series by using percentile aggregation. The resulting data point in each alignment period is the 95th percentile of all data points in the period. This aligner is valid for GAUGE and DELTA metrics with distribution values. The output is a GAUGE metric with valueType DOUBLE.
ALIGN_PERCENTILE_50 Align the time series by using percentile aggregation. The resulting data point in each alignment period is the 50th percentile of all data points in the period. This aligner is valid for GAUGE and DELTA metrics with distribution values. The output is a GAUGE metric with valueType DOUBLE.
ALIGN_PERCENTILE_05 Align the time series by using percentile aggregation. The resulting data point in each alignment period is the 5th percentile of all data points in the period. This aligner is valid for GAUGE and DELTA metrics with distribution values. The output is a GAUGE metric with valueType DOUBLE.
ALIGN_PERCENT_CHANGE

Align and convert to a percentage change. This aligner is valid for GAUGE and DELTA metrics with numeric values. This alignment returns ((current - previous)/previous) * 100, where the value of previous is determined based on the alignmentPeriod.

If the values of current and previous are both 0, then the returned value is 0. If only previous is 0, the returned value is infinity.

A 10-minute moving mean is computed at each point of the alignment period prior to the above calculation to smooth the metric and prevent false positives from very short-lived spikes. The moving mean is only applicable for data whose values are >= 0. Any values < 0 are treated as a missing datapoint, and are ignored. While DELTA metrics are accepted by this alignment, special care should be taken that the values for the metric will always be positive. The output is a GAUGE metric with valueType DOUBLE.

Reducer

A Reducer operation describes how to aggregate data points from multiple time series into a single time series, where the value of each data point in the resulting series is a function of all the already aligned values in the input time series.

Enums
REDUCE_NONE No cross-time series reduction. The output of the Aligner is returned.
REDUCE_MEAN Reduce by computing the mean value across time series for each alignment period. This reducer is valid for DELTA and GAUGE metrics with numeric or distribution values. The valueType of the output is DOUBLE.
REDUCE_MIN Reduce by computing the minimum value across time series for each alignment period. This reducer is valid for DELTA and GAUGE metrics with numeric values. The valueType of the output is the same as the valueType of the input.
REDUCE_MAX Reduce by computing the maximum value across time series for each alignment period. This reducer is valid for DELTA and GAUGE metrics with numeric values. The valueType of the output is the same as the valueType of the input.
REDUCE_SUM Reduce by computing the sum across time series for each alignment period. This reducer is valid for DELTA and GAUGE metrics with numeric and distribution values. The valueType of the output is the same as the valueType of the input.
REDUCE_STDDEV Reduce by computing the standard deviation across time series for each alignment period. This reducer is valid for DELTA and GAUGE metrics with numeric or distribution values. The valueType of the output is DOUBLE.
REDUCE_COUNT Reduce by computing the number of data points across time series for each alignment period. This reducer is valid for DELTA and GAUGE metrics of numeric, Boolean, distribution, and string valueType. The valueType of the output is INT64.
REDUCE_COUNT_TRUE Reduce by computing the number of True-valued data points across time series for each alignment period. This reducer is valid for DELTA and GAUGE metrics of Boolean valueType. The valueType of the output is INT64.
REDUCE_COUNT_FALSE Reduce by computing the number of False-valued data points across time series for each alignment period. This reducer is valid for DELTA and GAUGE metrics of Boolean valueType. The valueType of the output is INT64.
REDUCE_FRACTION_TRUE Reduce by computing the ratio of the number of True-valued data points to the total number of data points for each alignment period. This reducer is valid for DELTA and GAUGE metrics of Boolean valueType. The output value is in the range [0.0, 1.0] and has valueType DOUBLE.
REDUCE_PERCENTILE_99 Reduce by computing the 99th percentile of data points across time series for each alignment period. This reducer is valid for GAUGE and DELTA metrics of numeric and distribution type. The value of the output is DOUBLE.
REDUCE_PERCENTILE_95 Reduce by computing the 95th percentile of data points across time series for each alignment period. This reducer is valid for GAUGE and DELTA metrics of numeric and distribution type. The value of the output is DOUBLE.
REDUCE_PERCENTILE_50 Reduce by computing the 50th percentile of data points across time series for each alignment period. This reducer is valid for GAUGE and DELTA metrics of numeric and distribution type. The value of the output is DOUBLE.
REDUCE_PERCENTILE_05 Reduce by computing the 5th percentile of data points across time series for each alignment period. This reducer is valid for GAUGE and DELTA metrics of numeric and distribution type. The value of the output is DOUBLE.