Auto date histogram aggregations use BestBucketsDeferringCollector which causes TimeSeriesAggregator not to be used at all if a time series aggregation is nested as a child of an auto date histogram. Writing one such aggregation at the moment produces a result but that result is incorrect and includes always the same time series bucket as a child of each auto data histogram bucket. This happens because there is no TimeSeriesIndexSearcher which fetches, sorts and advances a (doc value) iterator for time series documents. As a result, a typical aggregation would include always the same (first or last depending on sorting) time series bucket with potential issues caused by a (very) large number of buckets returned as a result of the auto date histogram selecting a fine-grain bucket size (depending on the dataset).
We need to remove queries from the Rally dataset which use an auto date histogram with a nested time series aggregation.
Auto date histogram aggregations use
BestBucketsDeferringCollector
which causesTimeSeriesAggregator
not to be used at all if a time series aggregation is nested as a child of an auto date histogram. Writing one such aggregation at the moment produces a result but that result is incorrect and includes always the same time series bucket as a child of each auto data histogram bucket. This happens because there is noTimeSeriesIndexSearcher
which fetches, sorts and advances a (doc value) iterator for time series documents. As a result, a typical aggregation would include always the same (first or last depending on sorting) time series bucket with potential issues caused by a (very) large number of buckets returned as a result of the auto date histogram selecting a fine-grain bucket size (depending on the dataset).We need to remove queries from the Rally dataset which use an auto date histogram with a nested time series aggregation.