Data model mappings for batch load

The following discusses the schema for data model mappings and gives and example.

Data model mappings schema

The CreateBatchLoadTask request syntax and a BatchLoadTaskDescription object returned by a call to DescribeBatchLoadTask include a DataModelConfiguration object that includes the DataModel for batch loading. The DataModel defines mappings from source data that's stored in CSV format in an S3 location to a target Timestream for LiveAnalytics database and table.

The TimeColumn field indicates the source data's location for the value to be mapped to the destination table's time column in Timestream for LiveAnalytics. The TimeUnit specifies the unit for the TimeColumn, and can be one of MILLISECONDS, SECONDS, MICROSECONDS, or NANOSECONDS. There are also mappings for dimensions and measures. Dimension mappings are composed of source columns and target fields.

For more information, see DimensionMapping. The mappings for measures have two options, MixedMeasureMappings and MultiMeasureMappings.

To summarize, a DataModel contains mappings from a data source in an S3 location to a target Timestream for LiveAnalytics table for the following.

If possible, we recommend that you map measure data to multi-measure records in Timestream for LiveAnalytics. For information about the benefits of multi-measure records, see Multi-measure records.

If multiple measures in the source data are stored in one row, you can map those multiple measures to multi-measure records in Timestream for LiveAnalytics using MultiMeasureMappings. If there are values that must map to a single-measure record, you can use MixedMeasureMappings.

MixedMeasureMappings and MultiMeasureMappings both include MultiMeasureAttributeMappings. Multi-measure records are supported regardless of whether single-measure records are needed.

If only multi-measure target records are needed in Timestream for LiveAnalytics, you can define measure mappings in the following structure.

CreateBatchLoadTask
    MeasureNameColumn
    MultiMeasureMappings
        TargetMultiMeasureName
        MultiMeasureAttributeMappings array

If single-measure target records are needed in Timestream for LiveAnalytics, you can define measure mappings in the following structure.

CreateBatchLoadTask
    MeasureNameColumn
    MixedMeasureMappings array
        MixedMeasureMapping
            MeasureName
            MeasureValueType
            SourceColumn
            TargetMeasureName
            MultiMeasureAttributeMappings array

When you use MultiMeasureMappings, the MultiMeasureAttributeMappings array is always required. When you use the MixedMeasureMappings array, if the MeasureValueType is MULTI for a given MixedMeasureMapping, MultiMeasureAttributeMappings is required for that MixedMeasureMapping. Otherwise, MeasureValueType indicates the measure type for the single-measure record.

Either way, there is an array of MultiMeasureAttributeMapping available. You define the mappings to multi-measure records in each MultiMeasureAttributeMapping as follows:

Data model mappings with MultiMeasureMappings example

This example demonstrates mapping to multi-measure records, the preferred approach, which store each measure value in a dedicated column. You can download a sample CSV at sample CSV. The sample has the following headings to map to a target column in a Timestream for LiveAnalytics table.

Identify the time and measure_name columns in the CSV file. In this case these map directly to the Timestream for LiveAnalytics table columns of the same names.

When using the API, you specify time in the TimeColumn field and a supported time unit value such as MILLISECONDS in the TimeUnit field. These correspond to Source columnn name and Timestamp time input in the console. You can group or partition records using measure_name which is defined with the MeasureNameColumn key.

In the sample, region, location, and hostname are dimensions. Dimensions are mapped in an array of DimensionMapping objects.

For measures, the value TargetMultiMeasureAttributeName will become a column in the Timestream for LiveAnalytics table. You can keep the same name such as in this example. Or you can specify a new one. MeasureValueType is one of DOUBLE, BIGINT, BOOLEAN, VARCHAR, or TIMESTAMP.

{
  "TimeColumn": "time",
  "TimeUnit": "MILLISECONDS",
  "DimensionMappings": [
    {
      "SourceColumn": "region",
      "DestinationColumn": "region"
    },
    {
      "SourceColumn": "location",
      "DestinationColumn": "location"
    },
    {
      "SourceColumn": "hostname",
      "DestinationColumn": "hostname"
    }
  ],
  "MeasureNameColumn": "measure_name",
  "MultiMeasureMappings": {
    "MultiMeasureAttributeMappings": [
      {
        "SourceColumn": "memory_utilization",
        "TargetMultiMeasureAttributeName": "memory_utilization",
        "MeasureValueType": "DOUBLE"
      },
      {
        "SourceColumn": "cpu_utilization",
        "TargetMultiMeasureAttributeName": "cpu_utilization",
        "MeasureValueType": "DOUBLE"
      }
    ]
  }
}

Data model mappings with MixedMeasureMappings example

We recommend that you only use this approach when you need to map to single-measure records in Timestream for LiveAnalytics.