This introduce an optimization of top level term aggregation on field with a low cardinality.
We then use a Vec as the underlying map.
In addition, we buffer subaggregations.
---------
Co-authored-by: Pascal Seitz <pascal.seitz@datadoghq.com>
Co-authored-by: Paul Masurel <paul@quickwit.io>
* change AggregationLimits behavior
This fixes an issue encountered with the current behaviour of
AggregationLimits.
Previously we had AggregationLimits and RessourceLimitGuard, which both
track the memory, but only RessourceLimitGuard released memory when
dropped, while AggregationLimits did not.
This PR changes AggregationLimits to be a guard itself and removes the
RessourceLimitGuard.
* rename AggregationLimits to AggregationLimitsGuard
PR https://github.com/quickwit-oss/quickwit/pull/4962 fixes an issue
where the AggregationLimits are not passed correctly. Since the
AggregationLimits are shared properly we run into contention issues.
This PR includes some straightforward improvement to reduce contention,
by only calling if the memory changed and avoiding the second read.
We probably need some sharding with multiple counters or local caching before updating the
global after some threshold.
* Fix DateHistogram bucket gap
Fixes a computation issue of the number of buckets needed in the
DateHistogram.
This is due to a missing normalization from request values (ms) to fast field
values (ns), when converting an intermediate result to the final result.
This results in a wrong computation by a factor 1_000_000.
The Histogram normalizes values to nanoseconds, to make the user input like
extended_bounds (ms precision) and the values from the fast field (ns precision for date type) compatible.
This normalization happens only for date type fields, as other field types don't have precision settings.
The normalization does not happen due a missing `column_type`, which is not
correctly passed after merging an empty aggregation (which does not have a `column_type` set), with a regular aggregation.
Another related issue is an empty aggregation, which will not have
`column_type` set, will not convert the result to human readable format.
This PR fixes the issue by:
- Limit the allowed field types of DateHistogram to DateType
- Instead of passing the column_type, which is only available on the segment level, we flag the aggregation as `is_date_agg`.
- Fix the merge logic
Add a flag to to normalization only once. This is not an issue
currently, but it could become easily one.
closes https://github.com/quickwit-oss/quickwit/issues/3837
* use older nightly for time crate (breaks build)
* clear memory consumption in AggregationLimits
clear memory consumption in AggregationLimits at the end of segment collection
* switch to ResourceLimitGuard
* unduplicate code
* merge methods
* Apply suggestions from code review
Co-authored-by: Paul Masurel <paul@quickwit.io>
---------
Co-authored-by: Paul Masurel <paul@quickwit.io>
* add memory limit for aggregations
introduce AggregationLimits to set memory consumption limit and bucket limits
memory limit is checked during aggregation, bucket limit is checked before returning the aggregation request.
* Apply suggestions from code review
Co-authored-by: Paul Masurel <paul@quickwit.io>
* add ByteCount with human readable format
---------
Co-authored-by: Paul Masurel <paul@quickwit.io>
