Optimize filters aggregation with a single filter #99202
Labels
:Analytics/Aggregations
Aggregations
>enhancement
Team:Analytics
Meta label for analytical engine team (ESQL/Aggs/Geo)
Comments
elasticsearchmachine
added
the
Team:Analytics
|
Pinging @elastic/es-analytics-geo (Team:Analytics) |
kkrik-es
added a commit
to kkrik-es/elasticsearch
that referenced
this issue
Sep 6, 2023
When FiltersAggregator has a single filter, there is no benefit in using a DisiPriorityQueue as the heap will only contain values from a single iterator. In such a case, it's preferable to use the filtering approximation iterator directly as competitive iterator. Fixes elastic#99202
kkrik-es
added a commit
that referenced
this issue
Sep 6, 2023
* Use a competitive iterator in FiltersAggregator. The iterator is used to combine filtering with querying in leaf collection. Its benefit is that rangers with docs that are filtered out by all filters are skipped from doc collection. The competitive iterator is restricted to FiltersAggregator, not used in FilterByFilterAggregator that's already optimized. It only applies to top-level filter aggregations with no "other" bucket defined; the latter leads to collecting all docs so there's no point in skipping doc ranges. Fixes #97544 * Fix function name. * Advance iterator on two-phase mismatch. * Restore docid tracking. * Fix failing tests. * Fix failing test. * Fix more tests. * Update docs/changelog/98360.yaml * More test fixes. * Update docs/changelog/98360.yaml * Skip checking useCompetitiveIterator in collect * Find approximate matches in CompetitiveIterator * Use DisiPriorityQueue to simplify FiltersAggregator * Skip competitive iterator when all docs match. * Check for empty priority queue. * Skip DisiPriorityQueue on single filter agg. When FiltersAggregator has a single filter, there is no benefit in using a DisiPriorityQueue as the heap will only contain values from a single iterator. In such a case, it's preferable to use the filtering approximation iterator directly as competitive iterator. Fixes #99202 * Update docs/changelog/99215.yaml * Use FilterMatchingDisiWrapper in leaf collectors.
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Description
Follow-up of #98360: when there is a single filter, the collector could save the overhead of the priority queue (both in
collect()andcompetitiveIterator()), which would likely result in a speedup.The text was updated successfully, but these errors were encountered: