In elastic/elasticsearch#136813 we introduce one more configuration option in the downsample API that allows a user to choose the sampling method between:

• aggregate: when supported, this will downsample a metric by summarising its values in a aggregate document. This remains the default
• last_value: this will keep the last value and discard the rest.

POST /my-time-series-index/_downsample/my-downsampled-time-series-index
{
  "fixed_interval": "1d",
  "sampling_method": "last_value"
}

The issue here is that the sampling method will be introduced in the data stream lifecycle in a different way, see elastic/elasticsearch#137023:

PUT _data_stream/my-ds/_lifecycle
{
  "data_retention": "10d",
  "downsampling_method": "last_value",
  "downsampling": [
     {
       "after": "1d",
       "fixed_interval": "5m
     }
  ]
}

In the elasticsearch specification we used the same type to capture the fixed_interval in both, but now we see that these have diverged, so we need to split them into two types in order to preserve correctness.

This is an alternative to #5496.