rust/tantivy
1
0
Fork 0
mirror of https://github.com/quickwit-oss/tantivy.git synced 2026-01-06 01:02:55 +00:00
Code Issues Packages Projects Releases Wiki Activity

Compare commits

base: rust:trinity.pointard/opti-sstable-sorted-ords-to-term
Branches Tags
rust:main rust:bucket_id_agg rust:paul.masurel/skip-first-doc-on-intersection rust:stuhood.lazy-scorers-blocks-upstream rust:paul.masurel/seek_into_the_danger_zone_result rust:paul.masurel/search_into_danger_zone_fastfield rust:stuhood.lazy-scorers-blocks rust:paradedb/fix-agg-validation rust:flatheadmill-geo rust:paradedb-paradedb/fix-overflow-issue rust:low_card_optimization rust:paul.masurel/clippy rust:qw-airmail-20250916 rust:unit-test-aggregation-sum-other rust:clippy-and-cleanups rust:trinity.pointard/opti-sstable-sorted-ords-to-term rust:release_0.24 rust:dependabot/cargo/rand-0.9.1 rust:dependabot/cargo/lru-0.14.0 rust:dependabot/cargo/fs4-0.13.1 rust:tagged-user-input-ast rust:1686a/parse-agg-res rust:dependabot/cargo/rand_distr-0.5.1 rust:issue/2411-adding-panic-handler-to-rayon-merge-thread-pool rust:dependabot/cargo/nom-8 rust:upgrade_deps rust:test_parser rust:dependabot/github_actions/codecov/codecov-action-5 rust:binggan-0.12.0 rust:agg_format rust:path_collector rust:test_order_bug rust:test_columnar_multi rust:columnar_optionalmask_in_multivalues rust:prefix-phrase-query-optim rust:tantivy-with-oneshot rust:update_example rust:store-reader-enumerate rust:document-deserialize-seed rust:check_doc_mem rust:experimental-do-not-merge rust:expdonotmerge rust:cherry-pick-vec-instead-of-btreemap rust:validate_length rust:raphael_bench_avx_filter rust:raphael_optim_memset2 rust:query-field-name-avoid-alloc rust:check_bug rust:missing_test rust:list_fields rust:block-cache-control rust:trinity--multilayer-sstable rust:tracing-tantivy2 rust:tracing-tantivy rust:index_writer_err rust:default_fast_field_tokenizer rust:fmassot/add-preserve-original-ascii-folding rust:numeric_with_str_col rust:straightforward-simplify rust:fmassot/fake-test-branch rust:fmassot/dynamic-tokenizer-poc-follow-up rust:fmassot/add-box-token-filter-and-refactor rust:dynamic-tokenizer-proof-of-concept rust:fmassot/add-box-tokenizer-and-token-filter rust:tokenizer-api-0.1.1 rust:addconversion rust:removed-extra-enum-alternative-for-datetime-precision rust:issue/1981-reproduce-bm25-bug rust:tokenizer_docs rust:merge-proptests rust:columnar-todo rust:quickwit-0.5 rust:remove-byteorder rust:remove_dyn rust:pr/alexcole/1855-inject-bm25-statistics rust:evance-expose-build-query-from-user-input-ast rust:missing-specialized-method-arcdyncolumn rust:fast-u64-range-query rust:0.19.2 rust:columnar-cli2 rust:trinity--pub-term-as-slice rust:typed-column3 rust:columnar-merge3 rust:columnar-merge-exploring-fn rust:use_columnar rust:typed-column2 rust:term_agg_bucket_limit rust:range rust:typed-column rust:columnar-main rust:forced-types rust:u128-columnar rust:use_stats rust:truncation_comment rust:0.19.1 rust:trinity/yoke-documents rust:column-handle rust:columnar-cli rust:columnar rust:revert-1711-sparse_dense_index rust:commit-changes rust:sparse_dense_index_headroom rust:optional_column rust:no-column-for-multivalued rust:fmassot--bench-hdfs-with-array rust:bugfix-position-broken rust:float rust:simplify rust:debug_time rust:sparse_codecs rust:fix_estimate rust:column-reader rust:column-reader-b rust:columnreader-simple rust:col-trait-refact rust:ip_fastfield rust:column-trait-2 rust:refact-dyn-column rust:column-trait rust:refact-codec-trait rust:pr/PSeitz/1485-1 rust:fastfieldcodec rust:termmap_per_field rust:fastfield-codecs-tests rust:rayon-indexing-writers rust:issue/1251 rust:index-bench rust:fix_open_bytes_index rust:warming rust:revert-1222-issue/1195 rust:add-random-bench-for-fast-field rust:wasm-friendly-exploratory rust:revert-1159-patch-1 rust:debug-position-panic-0.15.3 rust:fix-issue-1111 rust:remove_rand rust:unused-deps rust:hfb/phrase_match_slop rust:0.15.3 rust:expose-min-max-on-multi-fast-field-reader rust:revert-1094-expose-min-max-on-multi-fast-field-reader rust:hotfix-1088 rust:owned-bytes-instead-of-rawdoc rust:rihardsk-date-ranges rust:issue/997 rust:0.13.3 rust:fail-fast-store-checkpoint rust:debugging-store rust:issue/reproduce-897 rust:issue/969-reproduce rust:troublescooter-static rust:intermediate-mergeable-troublescooter rust:barrotsteindev-filter-collector-tpredicatevalue rust:issue/938b rust:issue/938 rust:barrotsteindev-filter-collector rust:issue/922 rust:issue/896 rust:quickfix-explain rust:slog rust:limit-random-seeks rust:0.13.1 rust:issue/866b rust:bugfix-unittest-macos rust:githubactions rust:fieldnorm_readers rust:blockwand rust:nodeffeatfails rust:removedaliveiterator rust:robyoung-introduce-offset-to-top-docs rust:specialization-of-foreach-for-union-scorer rust:elshize-bmw rust:cutt rust:nrt rust:storewriter-out rust:segment_writer_in_ram2 rust:polling rust:audunhalland-query-boost rust:poll-meta rust:segment_writer_in_ram rust:bundle rust:petr-tik-n662_footer_is_incompatible_impl rust:race-condition-in-tests rust:hotfix-0.10.3-issue681 rust:remove-tru rust:kkoziara-pre-tokenized-text rust:issue/681 rust:issue/680 rust:refactoring/field rust:updating/uuid rust:fix-bench rust:hotfix-656 rust:readd-macos-ci rust:incremental-search rust:crate-division rust:streamer-with-state rust:criterion rust:python-binding-changes rust:bump-version rust:gh-pages rust:issue/554 rust:perf/experimental-union-for_each rust:issue/526b rust:softcommits rust:0.9.1-hotfix rust:revert-521-fix-non-english-stemmers rust:issue/500-hotfix rust:hotfix/0.8.2 rust:issue/483 rust:bug/merge-deted rust:issue/407 rust:issue/457 rust:segment_fruits rust:doc rust:dds/lenient rust:issue/webasm rust:issue/396 rust:broken_collector_with_result rust:dss/automaton-weight-suggestion rust:wasm rust:common-crawl rust:tantivy-imhotep
rust:0.25.0 rust:0.22.1 rust:0.24.2 rust:0.24.1 rust:tantivy-query-grammar-v0.24.0 rust:tantivy-columnar-v0.5.0 rust:tantivy-bitpacker-v0.8.0 rust:ownedbytes-v0.9.0 rust:tantivy-stacker-v0.5.0 rust:tantivy-sstable-v0.5.0 rust:tantivy-common-v0.9.0 rust:tantivy-tokenizer-api-v0.5.0 rust:0.24 rust:qw-airmail-20250224-hotfix-merge-panic rust:qw-airmail-20250219-hotfix-merge-panic rust:qw-airmail-202406 rust:qw-airmail-20240611 rust:qw-airmail-20240606 rust:quickwit-rev rust:0.22.0 rust:tantivy-stacker-v0.3.0 rust:ownedbytes-v0.7.0 rust:tantivy-tokenizer-api-v0.3.0 rust:tantivy-columnar-v0.3.0 rust:tantivy-common-v0.7.0 rust:tantivy-query-grammar-v0.22.0 rust:tantivy-sstable-v0.3.0 rust:tantivy-bitpacker-v0.6.0 rust:0.21.1 rust:0.21 rust:0.20.2 rust:0.20.1 rust:0.20 rust:quickwit-0.5-rev rust:0.19.2 rust:0.19.1 rust:0.19 rust:0.18.1 rust:0.18 rust:0.17 rust:0.16.1 rust:0.15.3 rust:0.15.2 rust:0.15.1 rust:0.15 rust:0.14 rust:0.13.3 rust:0.13.2 rust:0.13.1 rust:0.13 rust:0.12 rust:0.11.3 rust:0.11.1 rust:0.11 rust:0.10.3 rust:0.1 rust:0.10.2 rust:0.10.1 rust:0.10.0 rust:0.9.1 rust:0.9 rust:0.8.2 rust:0.8.2b rust:0.8.1 rust:0.8.0 rust:0.7.2 rust:0.7.1 rust:0.7.0 rust:0.6.1 rust:0.6.0 rust:0.5.2 rust:0.5.1 rust:0.5.0 rust:0.4.3 rust:0.4.2 rust:0.4.0 rust:0.3.1 rust:0.3.0 rust:0.2.0
..
compare: rust:low_card_optimization
Branches Tags
rust:bucket_id_agg rust:paul.masurel/skip-first-doc-on-intersection rust:main rust:stuhood.lazy-scorers-blocks-upstream rust:paul.masurel/seek_into_the_danger_zone_result rust:paul.masurel/search_into_danger_zone_fastfield rust:stuhood.lazy-scorers-blocks rust:paradedb/fix-agg-validation rust:flatheadmill-geo rust:paradedb-paradedb/fix-overflow-issue rust:low_card_optimization rust:paul.masurel/clippy rust:qw-airmail-20250916 rust:unit-test-aggregation-sum-other rust:clippy-and-cleanups rust:trinity.pointard/opti-sstable-sorted-ords-to-term rust:release_0.24 rust:dependabot/cargo/rand-0.9.1 rust:dependabot/cargo/lru-0.14.0 rust:dependabot/cargo/fs4-0.13.1 rust:tagged-user-input-ast rust:1686a/parse-agg-res rust:dependabot/cargo/rand_distr-0.5.1 rust:issue/2411-adding-panic-handler-to-rayon-merge-thread-pool rust:dependabot/cargo/nom-8 rust:upgrade_deps rust:test_parser rust:dependabot/github_actions/codecov/codecov-action-5 rust:binggan-0.12.0 rust:agg_format rust:path_collector rust:test_order_bug rust:test_columnar_multi rust:columnar_optionalmask_in_multivalues rust:prefix-phrase-query-optim rust:tantivy-with-oneshot rust:update_example rust:store-reader-enumerate rust:document-deserialize-seed rust:check_doc_mem rust:experimental-do-not-merge rust:expdonotmerge rust:cherry-pick-vec-instead-of-btreemap rust:validate_length rust:raphael_bench_avx_filter rust:raphael_optim_memset2 rust:query-field-name-avoid-alloc rust:check_bug rust:missing_test rust:list_fields rust:block-cache-control rust:trinity--multilayer-sstable rust:tracing-tantivy2 rust:tracing-tantivy rust:index_writer_err rust:default_fast_field_tokenizer rust:fmassot/add-preserve-original-ascii-folding rust:numeric_with_str_col rust:straightforward-simplify rust:fmassot/fake-test-branch rust:fmassot/dynamic-tokenizer-poc-follow-up rust:fmassot/add-box-token-filter-and-refactor rust:dynamic-tokenizer-proof-of-concept rust:fmassot/add-box-tokenizer-and-token-filter rust:tokenizer-api-0.1.1 rust:addconversion rust:removed-extra-enum-alternative-for-datetime-precision rust:issue/1981-reproduce-bm25-bug rust:tokenizer_docs rust:merge-proptests rust:columnar-todo rust:quickwit-0.5 rust:remove-byteorder rust:remove_dyn rust:pr/alexcole/1855-inject-bm25-statistics rust:evance-expose-build-query-from-user-input-ast rust:missing-specialized-method-arcdyncolumn rust:fast-u64-range-query rust:0.19.2 rust:columnar-cli2 rust:trinity--pub-term-as-slice rust:typed-column3 rust:columnar-merge3 rust:columnar-merge-exploring-fn rust:use_columnar rust:typed-column2 rust:term_agg_bucket_limit rust:range rust:typed-column rust:columnar-main rust:forced-types rust:u128-columnar rust:use_stats rust:truncation_comment rust:0.19.1 rust:trinity/yoke-documents rust:column-handle rust:columnar-cli rust:columnar rust:revert-1711-sparse_dense_index rust:commit-changes rust:sparse_dense_index_headroom rust:optional_column rust:no-column-for-multivalued rust:fmassot--bench-hdfs-with-array rust:bugfix-position-broken rust:float rust:simplify rust:debug_time rust:sparse_codecs rust:fix_estimate rust:column-reader rust:column-reader-b rust:columnreader-simple rust:col-trait-refact rust:ip_fastfield rust:column-trait-2 rust:refact-dyn-column rust:column-trait rust:refact-codec-trait rust:pr/PSeitz/1485-1 rust:fastfieldcodec rust:termmap_per_field rust:fastfield-codecs-tests rust:rayon-indexing-writers rust:issue/1251 rust:index-bench rust:fix_open_bytes_index rust:warming rust:revert-1222-issue/1195 rust:add-random-bench-for-fast-field rust:wasm-friendly-exploratory rust:revert-1159-patch-1 rust:debug-position-panic-0.15.3 rust:fix-issue-1111 rust:remove_rand rust:unused-deps rust:hfb/phrase_match_slop rust:0.15.3 rust:expose-min-max-on-multi-fast-field-reader rust:revert-1094-expose-min-max-on-multi-fast-field-reader rust:hotfix-1088 rust:owned-bytes-instead-of-rawdoc rust:rihardsk-date-ranges rust:issue/997 rust:0.13.3 rust:fail-fast-store-checkpoint rust:debugging-store rust:issue/reproduce-897 rust:issue/969-reproduce rust:troublescooter-static rust:intermediate-mergeable-troublescooter rust:barrotsteindev-filter-collector-tpredicatevalue rust:issue/938b rust:issue/938 rust:barrotsteindev-filter-collector rust:issue/922 rust:issue/896 rust:quickfix-explain rust:slog rust:limit-random-seeks rust:0.13.1 rust:issue/866b rust:bugfix-unittest-macos rust:githubactions rust:fieldnorm_readers rust:blockwand rust:nodeffeatfails rust:removedaliveiterator rust:robyoung-introduce-offset-to-top-docs rust:specialization-of-foreach-for-union-scorer rust:elshize-bmw rust:cutt rust:nrt rust:storewriter-out rust:segment_writer_in_ram2 rust:polling rust:audunhalland-query-boost rust:poll-meta rust:segment_writer_in_ram rust:bundle rust:petr-tik-n662_footer_is_incompatible_impl rust:race-condition-in-tests rust:hotfix-0.10.3-issue681 rust:remove-tru rust:kkoziara-pre-tokenized-text rust:issue/681 rust:issue/680 rust:refactoring/field rust:updating/uuid rust:fix-bench rust:hotfix-656 rust:readd-macos-ci rust:incremental-search rust:crate-division rust:streamer-with-state rust:criterion rust:python-binding-changes rust:bump-version rust:gh-pages rust:issue/554 rust:perf/experimental-union-for_each rust:issue/526b rust:softcommits rust:0.9.1-hotfix rust:revert-521-fix-non-english-stemmers rust:issue/500-hotfix rust:hotfix/0.8.2 rust:issue/483 rust:bug/merge-deted rust:issue/407 rust:issue/457 rust:segment_fruits rust:doc rust:dds/lenient rust:issue/webasm rust:issue/396 rust:broken_collector_with_result rust:dss/automaton-weight-suggestion rust:wasm rust:common-crawl rust:tantivy-imhotep
rust:0.25.0 rust:0.22.1 rust:0.24.2 rust:0.24.1 rust:tantivy-query-grammar-v0.24.0 rust:tantivy-columnar-v0.5.0 rust:tantivy-bitpacker-v0.8.0 rust:ownedbytes-v0.9.0 rust:tantivy-stacker-v0.5.0 rust:tantivy-sstable-v0.5.0 rust:tantivy-common-v0.9.0 rust:tantivy-tokenizer-api-v0.5.0 rust:0.24 rust:qw-airmail-20250224-hotfix-merge-panic rust:qw-airmail-20250219-hotfix-merge-panic rust:qw-airmail-202406 rust:qw-airmail-20240611 rust:qw-airmail-20240606 rust:quickwit-rev rust:0.22.0 rust:tantivy-stacker-v0.3.0 rust:ownedbytes-v0.7.0 rust:tantivy-tokenizer-api-v0.3.0 rust:tantivy-columnar-v0.3.0 rust:tantivy-common-v0.7.0 rust:tantivy-query-grammar-v0.22.0 rust:tantivy-sstable-v0.3.0 rust:tantivy-bitpacker-v0.6.0 rust:0.21.1 rust:0.21 rust:0.20.2 rust:0.20.1 rust:0.20 rust:quickwit-0.5-rev rust:0.19.2 rust:0.19.1 rust:0.19 rust:0.18.1 rust:0.18 rust:0.17 rust:0.16.1 rust:0.15.3 rust:0.15.2 rust:0.15.1 rust:0.15 rust:0.14 rust:0.13.3 rust:0.13.2 rust:0.13.1 rust:0.13 rust:0.12 rust:0.11.3 rust:0.11.1 rust:0.11 rust:0.10.3 rust:0.1 rust:0.10.2 rust:0.10.1 rust:0.10.0 rust:0.9.1 rust:0.9 rust:0.8.2 rust:0.8.2b rust:0.8.1 rust:0.8.0 rust:0.7.2 rust:0.7.1 rust:0.7.0 rust:0.6.1 rust:0.6.0 rust:0.5.2 rust:0.5.1 rust:0.5.0 rust:0.4.3 rust:0.4.2 rust:0.4.0 rust:0.3.1 rust:0.3.0 rust:0.2.0

48 Commits
trinity.po ... low_card_o

Author SHA1 Message Date
Paul Masurel
b2573a3b16 low cardinality optimisation 2025-11-19 18:41:10 +01:00
Moe
70e591e230 feat: added filter aggregation (#2711) 
* Initial impl

* Added `Filter` impl in `build_single_agg_segment_collector_with_reader` + Added tests

* Added `Filter(FilterBucketResult)` + Made tests work.

* Fixed type issues.

* Fixed a test.

* 8a7a73a: Pass `segment_reader`

* Added more tests.

* Improved parsing + tests

* refactoring

* Added more tests.

* refactoring: moved parsing code under QueryParser

* Use Tantivy syntax instead of ES

* Added a sanity check test.

* Simplified impl + tests

* Added back tests in a more maintable way

* nitz.

* nitz

* implemented very simple fast-path

* improved a comment

* implemented fast field support

* Used `BoundsRange`

* Improved fast field impl + tests

* Simplified execution.

* Fixed exports + nitz

* Improved the tests to check to the expected result.

* Improved test by checking the whole result JSON

* Removed brittle perf checks.

* Added efficiency verification tests.

* Added one more efficiency check test.

* Improved the efficiency tests.

* Removed unnecessary parsing code + added direct Query obj

* Fixed tests.

* Improved tests

* Fixed code structure

* Fixed lint issues

* nitz.

* nitz

* nitz.

* nitz.

* nitz.

* Added an example

* Fixed PR comments.

* Applied PR comments + nitz

* nitz.

* Improved the code.

* Fixed a perf issue.

* Added batch processing.

* Made the example more interesting

* Fixed bucket count

* Renamed Direct to CustomQuery

* Fixed lint issues.

* No need for scorer to be an `Option`

* nitz

* Used BitSet

* Added an optimization for AllQuery

* Fixed merge issues.

* Fixed lint issues.

* Added benchmark for FILTER

* Removed the Option wrapper.

* nitz.

* Applied PR comments.

* Fixed the AllQuery optimization

* Applied PR comments.

* feat: used `erased_serde` to allow filter query to be serialized

* further improved a comment

* Added back tests.

* removed an unused method

* removed an unused method

* Added documentation

* nitz.

* Added query builder.

* Fixed a comment.

* Applied PR comments.

* Fixed doctest issues.

* Added ser/de

* Removed bench in test

* Fixed a lint issue.
 2025-11-18 20:54:31 +01:00
Arthur
5277367cb0 remove duplicated call to index_writer.commit() in example (#2732) 2025-11-12 14:52:44 +01:00
Paul Masurel
8b02bff9b8 Removing obsolete benchmark screenshot (#2730) 
Co-authored-by: Paul Masurel <paul.masurel@datadoghq.com>
 2025-11-05 09:55:13 +01:00
PSeitz
60225bdd45 cleanup (#2724) 
Co-authored-by: Pascal Seitz <pascal.seitz@datadoghq.com>
 2025-10-23 10:23:34 +02:00
PSeitz
938bfec8b7 use FxHashMap for Aggregations Request (#2722) 
Co-authored-by: Pascal Seitz <pascal.seitz@datadoghq.com>
 2025-10-21 15:59:18 +02:00
PSeitz
dabcaa5809 fix merge intermediate aggregation results (#2719) 
Previously the merging relied on the order of the results, which is invalid since https://github.com/quickwit-oss/tantivy/pull/2035.
This bug is only hit in specific scenarios, when the aggregation collectors are built in a different order on different segments.

Co-authored-by: Pascal Seitz <pascal.seitz@datadoghq.com>
 2025-10-17 12:41:31 +02:00
PSeitz
d410a3b0c0 Add Filtering for Term Aggregations (#2717) 
* Add Filtering for Term Aggregations

Closes #2702

* add AggregationsSegmentCtx memory consumption

---------

Co-authored-by: Pascal Seitz <pascal.seitz@datadoghq.com>
 2025-10-15 17:39:53 +02:00
Remi
fc93391d0e Minor clarifications on the AggregationsWithAccessor refacto (#2716) 2025-10-14 19:59:33 +02:00
PSeitz
f8e79271ab Replace AggregationsWithAccessor (#2715) 
* add nested histogram-termagg benchmark

* Replace AggregationsWithAccessor with AggData

With AggregationsWithAccessor pre-computation and caching was done on the collector level.
If you have 10000 sub collectors (e.g. a term aggregation with sub aggregations) this is very inefficient.
`AggData` instead moves the data from the collector to a node which reflects the cardinality of the request tree instead of the cardinality of the segment collector.
It also moves the global struct shared with all aggregations in to aggregation specific structs. So each aggregation has its own space to store cached data and aggregation specific information.

This also breaks up the dependency to the elastic search aggregation structure somewhat.

Due to lifetime issues, we move the agg request specific object out of `AggData` during the collection and move it back at the end (for now). That's some unnecessary work, which costs CPU.

This allows better caching and will also pave the way for another potential optimization, by separating the collector and its storage. Currently we allocate a new collector for each sub aggregation bucket (for nested aggregations), but ideally we would have just one collector instance.

* renames

* move request data to agg request files

---------

Co-authored-by: Pascal Seitz <pascal.seitz@datadoghq.com>
 2025-10-14 09:22:11 +02:00
PSeitz
33835b6a01 Add DocSet::cost() (#2707) 
* query: add DocSet cost hint and use it for intersection ordering

- Add DocSet::cost()
- Use cost() instead of size_hint() to order scorers in intersect_scorers

This isolates cost-related changes without the new seek APIs from
PR #2538

* add comments

---------

Co-authored-by: Pascal Seitz <pascal.seitz@datadoghq.com>
 2025-10-13 16:25:49 +02:00
PSeitz
270ca5123c refactor postings (#2709) 
rename shallow_seek to seek_block
remove full_block from public postings API

This is as preparation to optionally handle Bitsets in the postings
 2025-10-08 16:55:25 +02:00
Mustafa S. Moiz
714366d3b9 docs: correct grammar (#2704) 
Correct phrasing for a single line in the docs (`one documents` -> `a document`).
 2025-10-08 16:47:09 +02:00
PSeitz-dd
40659d4d07 improve naming in buffered_union (#2705) 2025-09-24 10:58:46 +02:00
PSeitz
e1e131a804 add and/or queries benchmark (#2701) 2025-09-22 16:32:49 +02:00
PSeitz-dd
70da310b2d perf: deduplicate queries (#2698) 
* deduplicate queries

Deduplicate queries in the UserInputAst after parsing queries

* add return type
 2025-09-22 12:16:58 +02:00
PSeitz
85010b589a clippy (#2700) 
* clippy

* clippy

* clippy

* clippy + fmt

---------

Co-authored-by: Pascal Seitz <pascal.seitz@datadoghq.com>
 2025-09-19 18:04:25 +02:00
PSeitz-dd
2340dca628 fix compiler warnings (#2699) 
* fix compiler warnings

* fix import
 2025-09-19 15:55:04 +02:00
Remi
71a26d5b24 Fix CI with rust 1.90 (#2696) 
* Empty commit

* Fix dead code lint error
 2025-09-18 23:06:33 +02:00
PSeitz-dd
203751f2fe Optimize ExistsQuery for a high number of dynamic columns (#2694) 
* Optimize ExistsQuery for a high number of dynamic columns

The previous algorithm checked _each_ doc in _each_ column for
existence. This causes huge cost on JSON fields with e.g. 100k columns.
Compute a bitset instead if we have more than one column.

add `iter_docs` to the multivalued_index

* add benchmark

subfields=1
exists_json_union    Memory: 89.3 KB (+2.01%)    Avg: 0.4865ms (-26.03%)    Median: 0.4865ms (-26.03%)    [0.4865ms .. 0.4865ms]
subfields=2
exists_json_union    Memory: 68.1 KB     Avg: 1.7048ms (-0.46%)    Median: 1.7048ms (-0.46%)    [1.7048ms .. 1.7048ms]
subfields=3
exists_json_union    Memory: 61.8 KB     Avg: 2.0742ms (-2.22%)    Median: 2.0742ms (-2.22%)    [2.0742ms .. 2.0742ms]
subfields=4
exists_json_union    Memory: 119.8 KB (+103.44%)    Avg: 3.9500ms (+42.62%)    Median: 3.9500ms (+42.62%)    [3.9500ms .. 3.9500ms]
subfields=5
exists_json_union    Memory: 120.4 KB (+107.65%)    Avg: 3.9610ms (+20.65%)    Median: 3.9610ms (+20.65%)    [3.9610ms .. 3.9610ms]
subfields=6
exists_json_union    Memory: 120.6 KB (+107.49%)    Avg: 3.8903ms (+3.11%)    Median: 3.8903ms (+3.11%)    [3.8903ms .. 3.8903ms]
subfields=7
exists_json_union    Memory: 120.9 KB (+106.93%)    Avg: 3.6220ms (-16.22%)    Median: 3.6220ms (-16.22%)    [3.6220ms .. 3.6220ms]
subfields=8
exists_json_union    Memory: 121.3 KB (+106.23%)    Avg: 4.0981ms (-15.97%)    Median: 4.0981ms (-15.97%)    [4.0981ms .. 4.0981ms]
subfields=16
exists_json_union    Memory: 123.1 KB (+103.09%)    Avg: 4.3483ms (-92.26%)    Median: 4.3483ms (-92.26%)    [4.3483ms .. 4.3483ms]
subfields=256
exists_json_union    Memory: 204.6 KB (+19.85%)    Avg: 3.8874ms (-99.01%)    Median: 3.8874ms (-99.01%)    [3.8874ms .. 3.8874ms]
subfields=4096
exists_json_union    Memory: 2.0 MB     Avg: 3.5571ms (-99.90%)    Median: 3.5571ms (-99.90%)    [3.5571ms .. 3.5571ms]
subfields=65536
exists_json_union    Memory: 28.3 MB     Avg: 14.4417ms (-99.97%)    Median: 14.4417ms (-99.97%)    [14.4417ms .. 14.4417ms]
subfields=262144
exists_json_union    Memory: 113.3 MB     Avg: 66.2860ms (-99.95%)    Median: 66.2860ms (-99.95%)    [66.2860ms .. 66.2860ms]

* rename methods
 2025-09-16 18:21:03 +02:00
PSeitz-dd
7963b0b4aa Add fast field fallback for term query if not indexed (#2693) 
* Add fast field fallback for term query if not indexed

* only fallback without scores
 2025-09-12 14:58:21 +02:00
Paul Masurel
d5eefca11d Merge pull request #2692 from quickwit-oss/paul.masurel/coerce-floats-too-in-search-too 
This PR changes the logic used on the ingestion of floats.
 2025-09-10 09:46:54 +02:00
Paul Masurel
5d6c8de23e Align search float search logic to the columnar coercion rules 
It applies the same logic on floats as for u64 or i64.
In all case, the idea is (for the inverted index) to coerce number
to their canonical representation, before indexing and before searching.

That way a document with the float 1.0 will be searchable when the user
searches for 1.

Note that contrary to the columnar, we do not attempt to coerce all of the
terms associated to a given json path to a single numerical type.
We simply rely on this "point-wise" canonicalization.
 2025-09-09 19:28:17 +02:00
PSeitz
a06365f39f Update CHANGELOG.md for bugfixes (#2674) 
* Update CHANGELOG.md

* Update CHANGELOG.md
 2025-09-04 11:51:00 +02:00
Raphaël Cohen
f4b374110f feat: Regex query grammar (#2677) 
* feat: Regex query grammar

* feat: Disable regexes by default

* chore: Apply formatting
 2025-09-03 10:07:04 +02:00
PSeitz-dd
c37af9c1ff update release instructions (#2687) 2025-08-22 07:57:48 +08:00
PSeitz
33794a114c chore: Release (#2686) 
Co-authored-by: Pascal Seitz <pascal.seitz@datadoghq.com>
 2025-08-20 18:29:37 +08:00
PSeitz-dd
8676a1f57b prepare release: update Changelog (#2685) 2025-08-20 16:07:53 +08:00
PSeitz-dd
021ff2ad63 move bench to binggan (#2684) 2025-08-14 17:02:44 +08:00
Paul Masurel
39e027667b per field size details (#2679) 
* Added per-field size details.

This also does a bunch of refactoring.

merging field metadata does not silently asserts that arguments should be sorted.
merging does not set `stored`.

We do not rely on a hashmap to group fields, but instead rely on the fact that
the term dictionary is sorted.

The inverted level method that exposes field metadata is not exposed
as public anymore.

* CR comment

---------

Co-authored-by: Paul Masurel <paul.masurel@datadoghq.com>
 2025-08-13 13:12:22 +02:00
PSeitz-dd
a1d65c3df3 test stable ordering with pagination (#2683) 2025-08-13 15:36:28 +08:00
trinity-1686a
2e4615c2d3 Merge pull request #2678 from Darkheir/feat/query_grammar_space_between_field_and_value 
feat: Support spaces between field name and value
 2025-08-11 09:57:23 +02:00
Darkheir
610091e2c4 feat: Applies PR review suggestion 2025-08-04 10:12:51 +02:00
trinity-1686a
c301e7b1c4 Merge pull request #2673 from paradedb/stuhood.fix-order-by-dup-string 
Fix `TopDocs::order_by_string_fast_field` for duplicates
 2025-07-30 18:25:03 +02:00
Stu Hood
d9eb093368 Attempt to clarify sorted_ords_to_term_cb. 2025-07-29 21:56:31 -07:00
Darkheir
d4b090124c feat: Support spaces between field name and value 2025-07-23 11:12:13 +02:00
PSeitz-dd
811c68cdb2 fix field_names in top_hits aggregation (#2675) 2025-07-21 12:19:30 +08:00
trinity-1686a
bc1c789897 Merge pull request #2676 from quickwit-oss/trinity.pointard/allow-partial-default-field-success 
ignore failure to parse query when other default field suceeded
 2025-07-18 14:20:41 +02:00
trinity Pointard
e7c8c331bd ignore failure to parse query when other default field suceeded 2025-07-17 14:47:28 +02:00
Eric Ridge
2f01152a3c adjust Dictionary::sorted_ords_to_term_cb() to allow duplicates 2025-07-16 13:38:43 -07:00
PSeitz
4e84c70387 Fix TopNComputer for reverse order (#2672) 
Co-authored-by: Pascal Seitz <pascal.seitz@datadoghq.com>
 2025-07-16 21:44:04 +08:00
Paul M.
f2c77f06c5 Update fs4 to latest (0.13.1) (#2654) 
- One change was needed to handle the `Result<bool>` that now returns from `try_lock_exclusive`

Co-authored-by: Paul M. <prov223@tutanota.com>
 2025-07-14 11:26:19 +08:00
MassimilianoBaglioni
74334f9c9a Fixed typo in documentation (#2629) 
Co-authored-by: Massimiliano Baglioni <massimilianobaglioni@MacBook-Air-di-Massimiliano.local>
 2025-07-11 14:45:59 +08:00
Parth
cc4beb61ba update CHANGELOG (#2670) 
* update CHANGELOG

* Update CHANGELOG.md

Co-authored-by: PSeitz <PSeitz@users.noreply.github.com>

* Update CHANGELOG.md

---------

Co-authored-by: PSeitz <PSeitz@users.noreply.github.com>
 2025-07-11 11:33:11 +08:00
Dale Seo
6742e5981b fix a typo in the comment (#2668) 2025-07-10 07:14:57 +02:00
Philippe Noël
b128299976 Update ParadeDB logo (#2669) 2025-07-10 07:14:35 +02:00
PSeitz
945af922d1 clippy (#2661) 
* clippy

* use readable version

---------

Co-authored-by: Pascal Seitz <pascal.seitz@datadoghq.com>
 2025-07-02 11:25:03 +02:00
PSeitz-dd
295d07e55c fix union performance regression (#2663) 
closes https://github.com/quickwit-oss/tantivy/issues/2656
 2025-07-01 20:32:25 +02:00
145 changed files with 8107 additions and 2876 deletions
Expand all files Collapse all files

36
CHANGELOG.md
View File

@@ -1,3 +1,31 @@
Tantivy 0.25
================================
## Bugfixes
- fix union performance regression in tantivy 0.24 [#2663](https://github.com/quickwit-oss/tantivy/pull/2663)(@PSeitz)
- make zstd optional in sstable [#2633](https://github.com/quickwit-oss/tantivy/pull/2633)(@Parth)
- Fix TopDocs::order_by_string_fast_field for asc order [#2672](https://github.com/quickwit-oss/tantivy/pull/2672)(@stuhood @PSeitz)
## Features/Improvements
- add docs/example and Vec<u32> values to sstable [#2660](https://github.com/quickwit-oss/tantivy/pull/2660)(@PSeitz)
- Add string fast field support to `TopDocs`. [#2642](https://github.com/quickwit-oss/tantivy/pull/2642)(@stuhood)
- update edition to 2024 [#2620](https://github.com/quickwit-oss/tantivy/pull/2620)(@PSeitz)
- Allow optional spaces between the field name and the value in the query parser [#2678](https://github.com/quickwit-oss/tantivy/pull/2678)(@Darkheir)
- Support mixed field types in query parser [#2676](https://github.com/quickwit-oss/tantivy/pull/2676)(@trinity-1686a)
- Add per-field size details [#2679](https://github.com/quickwit-oss/tantivy/pull/2679)(@fulmicoton)
Tantivy 0.24.2
================================
- Fix TopNComputer for reverse order. [#2672](https://github.com/quickwit-oss/tantivy/pull/2672)(@stuhood @PSeitz)
Affected queries are [order_by_fast_field](https://docs.rs/tantivy/latest/tantivy/collector/struct.TopDocs.html#method.order_by_fast_field) and
[order_by_u64_field](https://docs.rs/tantivy/latest/tantivy/collector/struct.TopDocs.html#method.order_by_u64_field)
for `Order::Asc`
Tantivy 0.24.1
================================
- Fix: bump required rust version to 1.81
Tantivy 0.24
================================
Tantivy 0.24 will be backwards compatible with indices created with v0.22 and v0.21. The new minimum rust version will be 1.75. Tantivy 0.23 will be skipped.
@@ -80,6 +108,14 @@ This will slightly increase space and access time. [#2439](https://github.com/qu
- Fix trait bound of StoreReader::iter [#2360](https://github.com/quickwit-oss/tantivy/pull/2360)(@adamreichold)
- remove read_postings_no_deletes [#2526](https://github.com/quickwit-oss/tantivy/pull/2526)(@PSeitz)
Tantivy 0.22.1
================================
- Fix TopNComputer for reverse order. [#2672](https://github.com/quickwit-oss/tantivy/pull/2672)(@stuhood @PSeitz)
Affected queries are [order_by_fast_field](https://docs.rs/tantivy/latest/tantivy/collector/struct.TopDocs.html#method.order_by_fast_field) and
[order_by_u64_field](https://docs.rs/tantivy/latest/tantivy/collector/struct.TopDocs.html#method.order_by_u64_field)
for `Order::Asc`
Tantivy 0.22
================================

29
Cargo.toml
View File

@@ -1,6 +1,6 @@
[package]
name = "tantivy"
version = "0.24.0"
version = "0.25.0"
authors = ["Paul Masurel <paul.masurel@gmail.com>"]
license = "MIT"
categories = ["database-implementations", "data-structures"]
@@ -33,7 +33,7 @@ tempfile = { version = "3.12.0", optional = true }
log = "0.4.16"
serde = { version = "1.0.219", features = ["derive"] }
serde_json = "1.0.140"
fs4 = { version = "0.8.0", optional = true }
fs4 = { version = "0.13.1", optional = true }
levenshtein_automata = "0.2.1"
uuid = { version = "1.0.0", features = ["v4", "serde"] }
crossbeam-channel = "0.5.4"
@@ -57,18 +57,19 @@ measure_time = "0.9.0"
arc-swap = "1.5.0"
bon = "3.3.1"
columnar = { version = "0.5", path = "./columnar", package = "tantivy-columnar" }
sstable = { version = "0.5", path = "./sstable", package = "tantivy-sstable", optional = true }
stacker = { version = "0.5", path = "./stacker", package = "tantivy-stacker" }
query-grammar = { version = "0.24.0", path = "./query-grammar", package = "tantivy-query-grammar" }
tantivy-bitpacker = { version = "0.8", path = "./bitpacker" }
common = { version = "0.9", path = "./common/", package = "tantivy-common" }
tokenizer-api = { version = "0.5", path = "./tokenizer-api", package = "tantivy-tokenizer-api" }
columnar = { version = "0.6", path = "./columnar", package = "tantivy-columnar" }
sstable = { version = "0.6", path = "./sstable", package = "tantivy-sstable", optional = true }
stacker = { version = "0.6", path = "./stacker", package = "tantivy-stacker" }
query-grammar = { version = "0.25.0", path = "./query-grammar", package = "tantivy-query-grammar" }
tantivy-bitpacker = { version = "0.9", path = "./bitpacker" }
common = { version = "0.10", path = "./common/", package = "tantivy-common" }
tokenizer-api = { version = "0.6", path = "./tokenizer-api", package = "tantivy-tokenizer-api" }
sketches-ddsketch = { version = "0.3.0", features = ["use_serde"] }
hyperloglogplus = { version = "0.4.1", features = ["const-loop"] }
futures-util = { version = "0.3.28", optional = true }
futures-channel = { version = "0.3.28", optional = true }
fnv = "1.0.7"
typetag = "0.2.21"
[target.'cfg(windows)'.dependencies]
winapi = "0.3.9"
@@ -87,7 +88,7 @@ more-asserts = "0.3.1"
rand_distr = "0.4.3"
time = { version = "0.3.10", features = ["serde-well-known", "macros"] }
postcard = { version = "1.0.4", features = [
"use-std",
"use-std",
], default-features = false }
[target.'cfg(not(windows))'.dev-dependencies]
@@ -167,3 +168,11 @@ harness = false
[[bench]]
name = "agg_bench"
harness = false
[[bench]]
name = "exists_json"
harness = false
[[bench]]
name = "and_or_queries"
harness = false

2
README.md
View File

@@ -23,8 +23,6 @@ performance for different types of queries/collections.
Your mileage WILL vary depending on the nature of queries and their load.
<img src="doc/assets/images/searchbenchmark.png">
Details about the benchmark can be found at this [repository](https://github.com/quickwit-oss/search-benchmark-game).
## Features

27
RELEASE.md
View File

@@ -1,4 +1,4 @@
# Release a new Tantivy Version
# Releasing a new Tantivy Version
## Steps
@@ -10,12 +10,29 @@
6. Set git tag with new version
In conjucation with `cargo-release` Steps 1-4 (I'm not sure if the change detection works):
Set new packages to version 0.0.0
[`cargo-release`](https://github.com/crate-ci/cargo-release) will help us with steps 1-5:
Replace prev-tag-name
```bash
cargo release --workspace --no-publish -v --prev-tag-name 0.19 --push-remote origin minor --no-tag --execute
cargo release --workspace --no-publish -v --prev-tag-name 0.24 --push-remote origin minor --no-tag
```
no-tag or it will create tags for all the subpackages
`no-tag` or it will create tags for all the subpackages
cargo release will _not_ ignore unchanged packages, but it will print warnings for them.
e.g. "warning: updating ownedbytes to 0.10.0 despite no changes made since tag 0.24"
We need to manually ignore these unchanged packages
```bash
cargo release --workspace --no-publish -v --prev-tag-name 0.24 --push-remote origin minor --no-tag --exclude tokenizer-api
```
Add `--execute` to actually publish the packages, otherwise it will only print the commands that would be run.
### Tag Version
```bash
git tag 0.25.0
git push upstream tag 0.25.0
```

76
benches/agg_bench.rs
View File

@@ -71,8 +71,15 @@ fn bench_agg(mut group: InputGroup<Index>) {
register!(group, histogram);
register!(group, histogram_hard_bounds);
register!(group, histogram_with_avg_sub_agg);
register!(group, histogram_with_term_agg_few);
register!(group, avg_and_range_with_avg_sub_agg);
// Filter aggregation benchmarks
register!(group, filter_agg_all_query_count_agg);
register!(group, filter_agg_term_query_count_agg);
register!(group, filter_agg_all_query_with_sub_aggs);
register!(group, filter_agg_term_query_with_sub_aggs);
group.run();
}
@@ -339,6 +346,17 @@ fn histogram_with_avg_sub_agg(index: &Index) {
});
execute_agg(index, agg_req);
}
fn histogram_with_term_agg_few(index: &Index) {
let agg_req = json!({
"rangef64": {
"histogram": { "field": "score_f64", "interval": 10 },
"aggs": {
"my_texts": { "terms": { "field": "text_few_terms" } }
}
}
});
execute_agg(index, agg_req);
}
fn avg_and_range_with_avg_sub_agg(index: &Index) {
let agg_req = json!({
"rangef64": {
@@ -460,3 +478,61 @@ fn get_test_index_bench(cardinality: Cardinality) -> tantivy::Result<Index> {
Ok(index)
}
// Filter aggregation benchmarks
fn filter_agg_all_query_count_agg(index: &Index) {
let agg_req = json!({
"filtered": {
"filter": "*",
"aggs": {
"count": { "value_count": { "field": "score" } }
}
}
});
execute_agg(index, agg_req);
}
fn filter_agg_term_query_count_agg(index: &Index) {
let agg_req = json!({
"filtered": {
"filter": "text:cool",
"aggs": {
"count": { "value_count": { "field": "score" } }
}
}
});
execute_agg(index, agg_req);
}
fn filter_agg_all_query_with_sub_aggs(index: &Index) {
let agg_req = json!({
"filtered": {
"filter": "*",
"aggs": {
"avg_score": { "avg": { "field": "score" } },
"stats_score": { "stats": { "field": "score_f64" } },
"terms_text": {
"terms": { "field": "text_few_terms" }
}
}
}
});
execute_agg(index, agg_req);
}
fn filter_agg_term_query_with_sub_aggs(index: &Index) {
let agg_req = json!({
"filtered": {
"filter": "text:cool",
"aggs": {
"avg_score": { "avg": { "field": "score" } },
"stats_score": { "stats": { "field": "score_f64" } },
"terms_text": {
"terms": { "field": "text_few_terms" }
}
}
}
});
execute_agg(index, agg_req);
}

224
benches/and_or_queries.rs Normal file
View File

@@ -0,0 +1,224 @@
// Benchmarks boolean conjunction queries using binggan.
//
// What’s measured:
// - Or and And queries with varying selectivity (only `Term` queries for now on leafs)
// - Nested AND/OR combinations (on multiple fields)
// - No-scoring path using the Count collector (focus on iterator/skip performance)
// - Top-K retrieval (k=10) using the TopDocs collector
//
// Corpus model:
// - Synthetic docs; each token a/b/c is independently included per doc
// - If none of a/b/c are included, emit a neutral filler token to keep doc length similar
//
// Notes:
// - After optimization, when scoring is disabled Tantivy reads doc-only postings
// (IndexRecordOption::Basic), avoiding frequency decoding overhead.
// - This bench isolates boolean iteration speed and intersection/union cost.
// - Use `cargo bench --bench boolean_conjunction` to run.
use binggan::{black_box, BenchRunner};
use rand::prelude::*;
use rand::rngs::StdRng;
use rand::SeedableRng;
use tantivy::collector::{Count, TopDocs};
use tantivy::query::QueryParser;
use tantivy::schema::{Schema, TEXT};
use tantivy::{doc, Index, ReloadPolicy, Searcher};
#[derive(Clone)]
struct BenchIndex {
#[allow(dead_code)]
index: Index,
searcher: Searcher,
query_parser: QueryParser,
}
impl BenchIndex {
#[inline(always)]
fn count_query(&self, query_str: &str) -> usize {
let query = self.query_parser.parse_query(query_str).unwrap();
self.searcher.search(&query, &Count).unwrap()
}
#[inline(always)]
fn topk_len(&self, query_str: &str, k: usize) -> usize {
let query = self.query_parser.parse_query(query_str).unwrap();
self.searcher
.search(&query, &TopDocs::with_limit(k))
.unwrap()
.len()
}
}
/// Build a single index containing both fields (title, body) and
/// return two BenchIndex views:
/// - single_field: QueryParser defaults to only "body"
/// - multi_field: QueryParser defaults to ["title", "body"]
fn build_shared_indices(num_docs: usize, p_a: f32, p_b: f32, p_c: f32) -> (BenchIndex, BenchIndex) {
// Unified schema (two text fields)
let mut schema_builder = Schema::builder();
let f_title = schema_builder.add_text_field("title", TEXT);
let f_body = schema_builder.add_text_field("body", TEXT);
let schema = schema_builder.build();
let index = Index::create_in_ram(schema.clone());
// Populate index with stable RNG for reproducibility.
let mut rng = StdRng::from_seed([7u8; 32]);
// Populate: spread each present token 90/10 to body/title
{
let mut writer = index.writer(500_000_000).unwrap();
for _ in 0..num_docs {
let has_a = rng.gen_bool(p_a as f64);
let has_b = rng.gen_bool(p_b as f64);
let has_c = rng.gen_bool(p_c as f64);
let mut title_tokens: Vec<&str> = Vec::new();
let mut body_tokens: Vec<&str> = Vec::new();
if has_a {
if rng.gen_bool(0.1) {
title_tokens.push("a");
} else {
body_tokens.push("a");
}
}
if has_b {
if rng.gen_bool(0.1) {
title_tokens.push("b");
} else {
body_tokens.push("b");
}
}
if has_c {
if rng.gen_bool(0.1) {
title_tokens.push("c");
} else {
body_tokens.push("c");
}
}
if title_tokens.is_empty() && body_tokens.is_empty() {
body_tokens.push("z");
}
writer
.add_document(doc!(
f_title=>title_tokens.join(" "),
f_body=>body_tokens.join(" ")
))
.unwrap();
}
writer.commit().unwrap();
}
// Prepare reader/searcher once.
let reader = index
.reader_builder()
.reload_policy(ReloadPolicy::Manual)
.try_into()
.unwrap();
let searcher = reader.searcher();
// Build two query parsers with different default fields.
let qp_single = QueryParser::for_index(&index, vec![f_body]);
let qp_multi = QueryParser::for_index(&index, vec![f_title, f_body]);
let single_view = BenchIndex {
index: index.clone(),
searcher: searcher.clone(),
query_parser: qp_single,
};
let multi_view = BenchIndex {
index,
searcher,
query_parser: qp_multi,
};
(single_view, multi_view)
}
fn main() {
// Prepare corpora with varying selectivity. Build one index per corpus
// and derive two views (single-field vs multi-field) from it.
let scenarios = vec![
(
"N=1M, p(a)=5%, p(b)=1%, p(c)=15%".to_string(),
1_000_000,
0.05,
0.01,
0.15,
),
(
"N=1M, p(a)=1%, p(b)=1%, p(c)=15%".to_string(),
1_000_000,
0.01,
0.01,
0.15,
),
];
let mut runner = BenchRunner::new();
for (label, n, pa, pb, pc) in scenarios {
let (single_view, multi_view) = build_shared_indices(n, pa, pb, pc);
// Single-field group: default field is body only
{
let mut group = runner.new_group();
group.set_name(format!("single_field — {}", label));
group.register_with_input("+a_+b_count", &single_view, |benv: &BenchIndex| {
black_box(benv.count_query("+a +b"))
});
group.register_with_input("+a_+b_+c_count", &single_view, |benv: &BenchIndex| {
black_box(benv.count_query("+a +b +c"))
});
group.register_with_input("+a_+b_top10", &single_view, |benv: &BenchIndex| {
black_box(benv.topk_len("+a +b", 10))
});
group.register_with_input("+a_+b_+c_top10", &single_view, |benv: &BenchIndex| {
black_box(benv.topk_len("+a +b +c", 10))
});
// OR queries
group.register_with_input("a_OR_b_count", &single_view, |benv: &BenchIndex| {
black_box(benv.count_query("a OR b"))
});
group.register_with_input("a_OR_b_OR_c_count", &single_view, |benv: &BenchIndex| {
black_box(benv.count_query("a OR b OR c"))
});
group.register_with_input("a_OR_b_top10", &single_view, |benv: &BenchIndex| {
black_box(benv.topk_len("a OR b", 10))
});
group.register_with_input("a_OR_b_OR_c_top10", &single_view, |benv: &BenchIndex| {
black_box(benv.topk_len("a OR b OR c", 10))
});
group.run();
}
// Multi-field group: default fields are [title, body]
{
let mut group = runner.new_group();
group.set_name(format!("multi_field — {}", label));
group.register_with_input("+a_+b_count", &multi_view, |benv: &BenchIndex| {
black_box(benv.count_query("+a +b"))
});
group.register_with_input("+a_+b_+c_count", &multi_view, |benv: &BenchIndex| {
black_box(benv.count_query("+a +b +c"))
});
group.register_with_input("+a_+b_top10", &multi_view, |benv: &BenchIndex| {
black_box(benv.topk_len("+a +b", 10))
});
group.register_with_input("+a_+b_+c_top10", &multi_view, |benv: &BenchIndex| {
black_box(benv.topk_len("+a +b +c", 10))
});
// OR queries
group.register_with_input("a_OR_b_count", &multi_view, |benv: &BenchIndex| {
black_box(benv.count_query("a OR b"))
});
group.register_with_input("a_OR_b_OR_c_count", &multi_view, |benv: &BenchIndex| {
black_box(benv.count_query("a OR b OR c"))
});
group.register_with_input("a_OR_b_top10", &multi_view, |benv: &BenchIndex| {
black_box(benv.topk_len("a OR b", 10))
});
group.register_with_input("a_OR_b_OR_c_top10", &multi_view, |benv: &BenchIndex| {
black_box(benv.topk_len("a OR b OR c", 10))
});
group.run();
}
}
}

69
benches/exists_json.rs Normal file
View File

@@ -0,0 +1,69 @@
use binggan::plugins::PeakMemAllocPlugin;
use binggan::{black_box, InputGroup, PeakMemAlloc, INSTRUMENTED_SYSTEM};
use serde_json::json;
use tantivy::collector::Count;
use tantivy::query::ExistsQuery;
use tantivy::schema::{Schema, FAST, TEXT};
use tantivy::{doc, Index};
#[global_allocator]
pub static GLOBAL: &PeakMemAlloc<std::alloc::System> = &INSTRUMENTED_SYSTEM;
fn main() {
let doc_count: usize = 500_000;
let subfield_counts: &[usize] = &[1, 2, 3, 4, 5, 6, 7, 8, 16, 256, 4096, 65536, 262144];
let indices: Vec<(String, Index)> = subfield_counts
.iter()
.map(|&sub_fields| {
(
format!("subfields={sub_fields}"),
build_index_with_json_subfields(doc_count, sub_fields),
)
})
.collect();
let mut group = InputGroup::new_with_inputs(indices);
group.add_plugin(PeakMemAllocPlugin::new(GLOBAL));
group.config().num_iter_group = Some(1);
group.config().num_iter_bench = Some(1);
group.register("exists_json", exists_json_union);
group.run();
}
fn exists_json_union(index: &Index) {
let reader = index.reader().expect("reader");
let searcher = reader.searcher();
let query = ExistsQuery::new("json".to_string(), true);
let count = searcher.search(&query, &Count).expect("exists search");
// Prevents optimizer from eliding the search
black_box(count);
}
fn build_index_with_json_subfields(num_docs: usize, num_subfields: usize) -> Index {
// Schema: single JSON field stored as FAST to support ExistsQuery.
let mut schema_builder = Schema::builder();
let json_field = schema_builder.add_json_field("json", TEXT | FAST);
let schema = schema_builder.build();
let index = Index::create_from_tempdir(schema).expect("create index");
{
let mut index_writer = index
.writer_with_num_threads(1, 200_000_000)
.expect("writer");
for i in 0..num_docs {
let sub = i % num_subfields;
// Only one subpath set per document; rotate subpaths so that
// no single subpath is full, but the union covers all docs.
let v = json!({ format!("field_{sub}"): i as u64 });
index_writer
.add_document(doc!(json_field => v))
.expect("add_document");
}
index_writer.commit().expect("commit");
}
index
}

2
bitpacker/Cargo.toml
View File

@@ -1,6 +1,6 @@
[package]
name = "tantivy-bitpacker"
version = "0.8.0"
version = "0.9.0"
edition = "2024"
authors = ["Paul Masurel <paul.masurel@gmail.com>"]
license = "MIT"

4
bitpacker/src/bitpacker.rs
View File

@@ -48,7 +48,7 @@ impl BitPacker {
pub fn flush<TWrite: io::Write + ?Sized>(&mut self, output: &mut TWrite) -> io::Result<()> {
if self.mini_buffer_written > 0 {
let num_bytes = (self.mini_buffer_written + 7) / 8;
let num_bytes = self.mini_buffer_written.div_ceil(8);
let bytes = self.mini_buffer.to_le_bytes();
output.write_all(&bytes[..num_bytes])?;
self.mini_buffer_written = 0;
@@ -138,7 +138,7 @@ impl BitUnpacker {
// We use `usize` here to avoid overflow issues.
let end_bit_read = (end_idx as usize) * self.num_bits;
let end_byte_read = (end_bit_read + 7) / 8;
let end_byte_read = end_bit_read.div_ceil(8);
assert!(
end_byte_read <= data.len(),
"Requested index is out of bounds."

6
bitpacker/src/blocked_bitpacker.rs
View File

@@ -140,10 +140,10 @@ impl BlockedBitpacker {
pub fn iter(&self) -> impl Iterator<Item = u64> + '_ {
// todo performance: we could decompress a whole block and cache it instead
let bitpacked_elems = self.offset_and_bits.len() * BLOCK_SIZE;
let iter = (0..bitpacked_elems)
(0..bitpacked_elems)
.map(move |idx| self.get(idx))
.chain(self.buffer.iter().cloned());
iter
.chain(self.buffer.iter().cloned())
}
}

35
columnar/Cargo.toml
View File

@@ -1,6 +1,6 @@
[package]
name = "tantivy-columnar"
version = "0.5.0"
version = "0.6.0"
edition = "2024"
license = "MIT"
homepage = "https://github.com/quickwit-oss/tantivy"
@@ -12,10 +12,10 @@ categories = ["database-implementations", "data-structures", "compression"]
itertools = "0.14.0"
fastdivide = "0.4.0"
stacker = { version= "0.5", path = "../stacker", package="tantivy-stacker"}
sstable = { version= "0.5", path = "../sstable", package = "tantivy-sstable" }
common = { version= "0.9", path = "../common", package = "tantivy-common" }
tantivy-bitpacker = { version= "0.8", path = "../bitpacker/" }
stacker = { version= "0.6", path = "../stacker", package="tantivy-stacker"}
sstable = { version= "0.6", path = "../sstable", package = "tantivy-sstable" }
common = { version= "0.10", path = "../common", package = "tantivy-common" }
tantivy-bitpacker = { version= "0.9", path = "../bitpacker/" }
serde = "1.0.152"
downcast-rs = "2.0.1"
@@ -33,6 +33,29 @@ harness = false
name = "bench_access"
harness = false
[[bench]]
name = "bench_first_vals"
harness = false
[[bench]]
name = "bench_values_u64"
harness = false
[[bench]]
name = "bench_values_u128"
harness = false
[[bench]]
name = "bench_create_column_values"
harness = false
[[bench]]
name = "bench_column_values_get"
harness = false
[[bench]]
name = "bench_optional_index"
harness = false
[features]
unstable = []
zstd-compression = ["sstable/zstd-compression"]

3
columnar/benches/bench_access.rs
View File

@@ -19,7 +19,7 @@ fn main() {
let mut add_card = |card1: Card| {
inputs.push((
format!("{card1}"),
card1.to_string(),
generate_columnar_and_open(card1, NUM_DOCS),
));
};
@@ -50,6 +50,7 @@ fn bench_group(mut runner: InputGroup<Column>) {
let mut buffer = vec![None; BLOCK_SIZE];
for i in (0..NUM_DOCS).step_by(BLOCK_SIZE) {
// fill docs
#[allow(clippy::needless_range_loop)]
for idx in 0..BLOCK_SIZE {
docs[idx] = idx as u32 + i;
}

61
columnar/benches/bench_column_values_get.rs Normal file
View File

@@ -0,0 +1,61 @@
use std::sync::Arc;
use binggan::{InputGroup, black_box};
use rand::rngs::StdRng;
use rand::{Rng, SeedableRng};
use tantivy_columnar::ColumnValues;
use tantivy_columnar::column_values::{CodecType, serialize_and_load_u64_based_column_values};
fn get_data() -> Vec<u64> {
let mut rng = StdRng::seed_from_u64(2u64);
let mut data: Vec<_> = (100..55_000_u64)
.map(|num| num + rng.r#gen::<u8>() as u64)
.collect();
data.push(99_000);
data.insert(1000, 2000);
data.insert(2000, 100);
data.insert(3000, 4100);
data.insert(4000, 100);
data.insert(5000, 800);
data
}
#[inline(never)]
fn value_iter() -> impl Iterator<Item = u64> {
0..20_000
}
type Col = Arc<dyn ColumnValues<u64>>;
fn main() {
let data = get_data();
let inputs: Vec<(String, Col)> = vec![
(
"bitpacked".to_string(),
serialize_and_load_u64_based_column_values(&data.as_slice(), &[CodecType::Bitpacked]),
),
(
"linear".to_string(),
serialize_and_load_u64_based_column_values(&data.as_slice(), &[CodecType::Linear]),
),
(
"blockwise_linear".to_string(),
serialize_and_load_u64_based_column_values(
&data.as_slice(),
&[CodecType::BlockwiseLinear],
),
),
];
let mut group: InputGroup<Col> = InputGroup::new_with_inputs(inputs);
group.register("fastfield_get", |col: &Col| {
let mut sum = 0u64;
for pos in value_iter() {
sum = sum.wrapping_add(col.get_val(pos as u32));
}
black_box(sum);
});
group.run();
}

44
columnar/benches/bench_create_column_values.rs Normal file
View File

@@ -0,0 +1,44 @@
use binggan::{InputGroup, black_box};
use rand::rngs::StdRng;
use rand::{Rng, SeedableRng};
use tantivy_columnar::column_values::{CodecType, serialize_u64_based_column_values};
fn get_data() -> Vec<u64> {
let mut rng = StdRng::seed_from_u64(2u64);
let mut data: Vec<_> = (100..55_000_u64)
.map(|num| num + rng.r#gen::<u8>() as u64)
.collect();
data.push(99_000);
data.insert(1000, 2000);
data.insert(2000, 100);
data.insert(3000, 4100);
data.insert(4000, 100);
data.insert(5000, 800);
data
}
fn main() {
let data = get_data();
let mut group: InputGroup<(CodecType, Vec<u64>)> = InputGroup::new_with_inputs(vec![
(
"bitpacked codec".to_string(),
(CodecType::Bitpacked, data.clone()),
),
(
"linear codec".to_string(),
(CodecType::Linear, data.clone()),
),
(
"blockwise linear codec".to_string(),
(CodecType::BlockwiseLinear, data.clone()),
),
]);
group.register("serialize column_values", |data| {
let mut buffer = Vec::new();
serialize_u64_based_column_values(&data.1.as_slice(), &[data.0], &mut buffer).unwrap();
black_box(buffer.len());
});
group.run();
}

106
columnar/benches/bench_first_vals.rs
View File

@@ -1,12 +1,9 @@
#![feature(test)]
extern crate test;
use std::sync::Arc;
use binggan::{InputGroup, black_box};
use rand::prelude::*;
use tantivy_columnar::column_values::{CodecType, serialize_and_load_u64_based_column_values};
use tantivy_columnar::*;
use test::{Bencher, black_box};
struct Columns {
pub optional: Column,
@@ -68,88 +65,45 @@ pub fn serialize_and_load(column: &[u64], codec_type: CodecType) -> Arc<dyn Colu
serialize_and_load_u64_based_column_values(&column, &[codec_type])
}
fn run_bench_on_column_full_scan(b: &mut Bencher, column: Column) {
let num_iter = black_box(NUM_VALUES);
b.iter(|| {
fn main() {
let Columns {
optional,
full,
multi,
} = get_test_columns();
let inputs = vec![
("full".to_string(), full),
("optional".to_string(), optional),
("multi".to_string(), multi),
];
let mut group = InputGroup::new_with_inputs(inputs);
group.register("first_full_scan", |column| {
let mut sum = 0u64;
for i in 0..num_iter as u32 {
for i in 0..NUM_VALUES as u32 {
let val = column.first(i);
sum += val.unwrap_or(0);
}
sum
black_box(sum);
});
}
fn run_bench_on_column_block_fetch(b: &mut Bencher, column: Column) {
let mut block: Vec<Option<u64>> = vec![None; 64];
let fetch_docids = (0..64).collect::<Vec<_>>();
b.iter(move || {
group.register("first_block_fetch", |column| {
let mut block: Vec<Option<u64>> = vec![None; 64];
let fetch_docids = (0..64).collect::<Vec<_>>();
column.first_vals(&fetch_docids, &mut block);
block[0]
black_box(block[0]);
});
}
fn run_bench_on_column_block_single_calls(b: &mut Bencher, column: Column) {
let mut block: Vec<Option<u64>> = vec![None; 64];
let fetch_docids = (0..64).collect::<Vec<_>>();
b.iter(move || {
group.register("first_block_single_calls", |column| {
let mut block: Vec<Option<u64>> = vec![None; 64];
let fetch_docids = (0..64).collect::<Vec<_>>();
for i in 0..fetch_docids.len() {
block[i] = column.first(fetch_docids[i]);
}
block[0]
black_box(block[0]);
});
}
/// Column first method
#[bench]
fn bench_get_first_on_full_column_full_scan(b: &mut Bencher) {
let column = get_test_columns().full;
run_bench_on_column_full_scan(b, column);
}
#[bench]
fn bench_get_first_on_optional_column_full_scan(b: &mut Bencher) {
let column = get_test_columns().optional;
run_bench_on_column_full_scan(b, column);
}
#[bench]
fn bench_get_first_on_multi_column_full_scan(b: &mut Bencher) {
let column = get_test_columns().multi;
run_bench_on_column_full_scan(b, column);
}
/// Block fetch column accessor
#[bench]
fn bench_get_block_first_on_optional_column(b: &mut Bencher) {
let column = get_test_columns().optional;
run_bench_on_column_block_fetch(b, column);
}
#[bench]
fn bench_get_block_first_on_multi_column(b: &mut Bencher) {
let column = get_test_columns().multi;
run_bench_on_column_block_fetch(b, column);
}
#[bench]
fn bench_get_block_first_on_full_column(b: &mut Bencher) {
let column = get_test_columns().full;
run_bench_on_column_block_fetch(b, column);
}
#[bench]
fn bench_get_block_first_on_optional_column_single_calls(b: &mut Bencher) {
let column = get_test_columns().optional;
run_bench_on_column_block_single_calls(b, column);
}
#[bench]
fn bench_get_block_first_on_multi_column_single_calls(b: &mut Bencher) {
let column = get_test_columns().multi;
run_bench_on_column_block_single_calls(b, column);
}
#[bench]
fn bench_get_block_first_on_full_column_single_calls(b: &mut Bencher) {
let column = get_test_columns().full;
run_bench_on_column_block_single_calls(b, column);
group.run();
}

106
columnar/benches/bench_optional_index.rs Normal file
View File

@@ -0,0 +1,106 @@
use binggan::{InputGroup, black_box};
use rand::rngs::StdRng;
use rand::{Rng, SeedableRng};
use tantivy_columnar::column_index::{OptionalIndex, Set};
const TOTAL_NUM_VALUES: u32 = 1_000_000;
fn gen_optional_index(fill_ratio: f64) -> OptionalIndex {
let mut rng: StdRng = StdRng::from_seed([1u8; 32]);
let vals: Vec<u32> = (0..TOTAL_NUM_VALUES)
.map(|_| rng.gen_bool(fill_ratio))
.enumerate()
.filter(|(_pos, val)| *val)
.map(|(pos, _)| pos as u32)
.collect();
OptionalIndex::for_test(TOTAL_NUM_VALUES, &vals)
}
fn random_range_iterator(
start: u32,
end: u32,
avg_step_size: u32,
avg_deviation: u32,
) -> impl Iterator<Item = u32> {
let mut rng: StdRng = StdRng::from_seed([1u8; 32]);
let mut current = start;
std::iter::from_fn(move || {
current += rng.gen_range(avg_step_size - avg_deviation..=avg_step_size + avg_deviation);
if current >= end { None } else { Some(current) }
})
}
fn n_percent_step_iterator(percent: f32, num_values: u32) -> impl Iterator<Item = u32> {
let ratio = percent / 100.0;
let step_size = (1f32 / ratio) as u32;
let deviation = step_size - 1;
random_range_iterator(0, num_values, step_size, deviation)
}
fn walk_over_data(codec: &OptionalIndex, avg_step_size: u32) -> Option<u32> {
walk_over_data_from_positions(
codec,
random_range_iterator(0, TOTAL_NUM_VALUES, avg_step_size, 0),
)
}
fn walk_over_data_from_positions(
codec: &OptionalIndex,
positions: impl Iterator<Item = u32>,
) -> Option<u32> {
let mut dense_idx: Option<u32> = None;
for idx in positions {
dense_idx = dense_idx.or(codec.rank_if_exists(idx));
}
dense_idx
}
fn main() {
// Build separate inputs for each fill ratio.
let inputs: Vec<(String, OptionalIndex)> = vec![
("fill=1%".to_string(), gen_optional_index(0.01)),
("fill=5%".to_string(), gen_optional_index(0.05)),
("fill=10%".to_string(), gen_optional_index(0.10)),
("fill=50%".to_string(), gen_optional_index(0.50)),
("fill=90%".to_string(), gen_optional_index(0.90)),
];
let mut group: InputGroup<OptionalIndex> = InputGroup::new_with_inputs(inputs);
// Translate orig->codec (rank_if_exists) with sampling
group.register("orig_to_codec_10pct_hit", |codec: &OptionalIndex| {
black_box(walk_over_data(codec, 100));
});
group.register("orig_to_codec_1pct_hit", |codec: &OptionalIndex| {
black_box(walk_over_data(codec, 1000));
});
group.register("orig_to_codec_full_scan", |codec: &OptionalIndex| {
black_box(walk_over_data_from_positions(codec, 0..TOTAL_NUM_VALUES));
});
// Translate codec->orig (select/select_batch) on sampled ranks
fn bench_translate_codec_to_orig_util(codec: &OptionalIndex, percent_hit: f32) {
let num_non_nulls = codec.num_non_nulls();
let idxs: Vec<u32> = if percent_hit == 100.0f32 {
(0..num_non_nulls).collect()
} else {
n_percent_step_iterator(percent_hit, num_non_nulls).collect()
};
let mut output = vec![0u32; idxs.len()];
output.copy_from_slice(&idxs[..]);
codec.select_batch(&mut output);
black_box(output);
}
group.register("codec_to_orig_0.005pct_hit", |codec: &OptionalIndex| {
bench_translate_codec_to_orig_util(codec, 0.005);
});
group.register("codec_to_orig_10pct_hit", |codec: &OptionalIndex| {
bench_translate_codec_to_orig_util(codec, 10.0);
});
group.register("codec_to_orig_full_scan", |codec: &OptionalIndex| {
bench_translate_codec_to_orig_util(codec, 100.0);
});
group.run();
}

110
columnar/benches/bench_values_u128.rs
View File

@@ -1,15 +1,12 @@
#![feature(test)]
use std::ops::RangeInclusive;
use std::sync::Arc;
use binggan::{InputGroup, black_box};
use common::OwnedBytes;
use rand::rngs::StdRng;
use rand::seq::SliceRandom;
use rand::{Rng, SeedableRng, random};
use tantivy_columnar::ColumnValues;
use test::Bencher;
extern crate test;
// TODO does this make sense for IPv6 ?
fn generate_random() -> Vec<u64> {
@@ -47,78 +44,77 @@ fn get_data_50percent_item() -> Vec<u128> {
}
data.push(SINGLE_ITEM);
data.shuffle(&mut rng);
let data = data.iter().map(|el| *el as u128).collect::<Vec<_>>();
data
data.iter().map(|el| *el as u128).collect::<Vec<_>>()
}
#[bench]
fn bench_intfastfield_getrange_u128_50percent_hit(b: &mut Bencher) {
fn main() {
let data = get_data_50percent_item();
let column = get_u128_column_from_data(&data);
let column_range = get_u128_column_from_data(&data);
let column_random = get_u128_column_random();
b.iter(|| {
struct Inputs {
data: Vec<u128>,
column_range: Arc<dyn ColumnValues<u128>>,
column_random: Arc<dyn ColumnValues<u128>>,
}
let inputs = Inputs {
data,
column_range,
column_random,
};
let mut group: InputGroup<Inputs> =
InputGroup::new_with_inputs(vec![("u128 benches".to_string(), inputs)]);
group.register(
"intfastfield_getrange_u128_50percent_hit",
|inp: &Inputs| {
let mut positions = Vec::new();
inp.column_range.get_row_ids_for_value_range(
*FIFTY_PERCENT_RANGE.start() as u128..=*FIFTY_PERCENT_RANGE.end() as u128,
0..inp.data.len() as u32,
&mut positions,
);
black_box(positions.len());
},
);
group.register("intfastfield_getrange_u128_single_hit", |inp: &Inputs| {
let mut positions = Vec::new();
column.get_row_ids_for_value_range(
*FIFTY_PERCENT_RANGE.start() as u128..=*FIFTY_PERCENT_RANGE.end() as u128,
0..data.len() as u32,
&mut positions,
);
positions
});
}
#[bench]
fn bench_intfastfield_getrange_u128_single_hit(b: &mut Bencher) {
let data = get_data_50percent_item();
let column = get_u128_column_from_data(&data);
b.iter(|| {
let mut positions = Vec::new();
column.get_row_ids_for_value_range(
inp.column_range.get_row_ids_for_value_range(
*SINGLE_ITEM_RANGE.start() as u128..=*SINGLE_ITEM_RANGE.end() as u128,
0..data.len() as u32,
0..inp.data.len() as u32,
&mut positions,
);
positions
black_box(positions.len());
});
}
#[bench]
fn bench_intfastfield_getrange_u128_hit_all(b: &mut Bencher) {
let data = get_data_50percent_item();
let column = get_u128_column_from_data(&data);
b.iter(|| {
group.register("intfastfield_getrange_u128_hit_all", |inp: &Inputs| {
let mut positions = Vec::new();
column.get_row_ids_for_value_range(0..=u128::MAX, 0..data.len() as u32, &mut positions);
positions
inp.column_range.get_row_ids_for_value_range(
0..=u128::MAX,
0..inp.data.len() as u32,
&mut positions,
);
black_box(positions.len());
});
}
// U128 RANGE END
#[bench]
fn bench_intfastfield_scan_all_fflookup_u128(b: &mut Bencher) {
let column = get_u128_column_random();
b.iter(|| {
group.register("intfastfield_scan_all_fflookup_u128", |inp: &Inputs| {
let mut a = 0u128;
for i in 0u64..column.num_vals() as u64 {
a += column.get_val(i as u32);
for i in 0u64..inp.column_random.num_vals() as u64 {
a += inp.column_random.get_val(i as u32);
}
a
black_box(a);
});
}
#[bench]
fn bench_intfastfield_jumpy_stride5_u128(b: &mut Bencher) {
let column = get_u128_column_random();
b.iter(|| {
let n = column.num_vals();
group.register("intfastfield_jumpy_stride5_u128", |inp: &Inputs| {
let n = inp.column_random.num_vals();
let mut a = 0u128;
for i in (0..n / 5).map(|val| val * 5) {
a += column.get_val(i);
a += inp.column_random.get_val(i);
}
a
black_box(a);
});
group.run();
}

244
columnar/benches/bench_values_u64.rs
View File

@@ -1,13 +1,10 @@
#![feature(test)]
extern crate test;
use std::ops::RangeInclusive;
use std::sync::Arc;
use binggan::{InputGroup, black_box};
use rand::prelude::*;
use tantivy_columnar::column_values::{CodecType, serialize_and_load_u64_based_column_values};
use tantivy_columnar::*;
use test::Bencher;
// Warning: this generates the same permutation at each call
fn generate_permutation() -> Vec<u64> {
@@ -27,37 +24,11 @@ pub fn serialize_and_load(column: &[u64], codec_type: CodecType) -> Arc<dyn Colu
serialize_and_load_u64_based_column_values(&column, &[codec_type])
}
#[bench]
fn bench_intfastfield_jumpy_veclookup(b: &mut Bencher) {
let permutation = generate_permutation();
let n = permutation.len();
b.iter(|| {
let mut a = 0u64;
for _ in 0..n {
a = permutation[a as usize];
}
a
});
}
#[bench]
fn bench_intfastfield_jumpy_fflookup_bitpacked(b: &mut Bencher) {
let permutation = generate_permutation();
let n = permutation.len();
let column: Arc<dyn ColumnValues<u64>> = serialize_and_load(&permutation, CodecType::Bitpacked);
b.iter(|| {
let mut a = 0u64;
for _ in 0..n {
a = column.get_val(a as u32);
}
a
});
}
const FIFTY_PERCENT_RANGE: RangeInclusive<u64> = 1..=50;
const SINGLE_ITEM: u64 = 90;
const SINGLE_ITEM_RANGE: RangeInclusive<u64> = 90..=90;
const ONE_PERCENT_ITEM_RANGE: RangeInclusive<u64> = 49..=49;
fn get_data_50percent_item() -> Vec<u128> {
let mut rng = StdRng::from_seed([1u8; 32]);
@@ -69,135 +40,122 @@ fn get_data_50percent_item() -> Vec<u128> {
data.push(SINGLE_ITEM);
data.shuffle(&mut rng);
let data = data.iter().map(|el| *el as u128).collect::<Vec<_>>();
data
data.iter().map(|el| *el as u128).collect::<Vec<_>>()
}
// U64 RANGE START
#[bench]
fn bench_intfastfield_getrange_u64_50percent_hit(b: &mut Bencher) {
let data = get_data_50percent_item();
let data = data.iter().map(|el| *el as u64).collect::<Vec<_>>();
let column: Arc<dyn ColumnValues<u64>> = serialize_and_load(&data, CodecType::Bitpacked);
b.iter(|| {
let mut positions = Vec::new();
column.get_row_ids_for_value_range(
FIFTY_PERCENT_RANGE,
0..data.len() as u32,
&mut positions,
);
positions
});
}
type VecCol = (Vec<u64>, Arc<dyn ColumnValues<u64>>);
#[bench]
fn bench_intfastfield_getrange_u64_1percent_hit(b: &mut Bencher) {
let data = get_data_50percent_item();
let data = data.iter().map(|el| *el as u64).collect::<Vec<_>>();
let column: Arc<dyn ColumnValues<u64>> = serialize_and_load(&data, CodecType::Bitpacked);
b.iter(|| {
let mut positions = Vec::new();
column.get_row_ids_for_value_range(
ONE_PERCENT_ITEM_RANGE,
0..data.len() as u32,
&mut positions,
);
positions
});
}
#[bench]
fn bench_intfastfield_getrange_u64_single_hit(b: &mut Bencher) {
let data = get_data_50percent_item();
let data = data.iter().map(|el| *el as u64).collect::<Vec<_>>();
let column: Arc<dyn ColumnValues<u64>> = serialize_and_load(&data, CodecType::Bitpacked);
b.iter(|| {
let mut positions = Vec::new();
column.get_row_ids_for_value_range(SINGLE_ITEM_RANGE, 0..data.len() as u32, &mut positions);
positions
});
}
#[bench]
fn bench_intfastfield_getrange_u64_hit_all(b: &mut Bencher) {
let data = get_data_50percent_item();
let data = data.iter().map(|el| *el as u64).collect::<Vec<_>>();
let column: Arc<dyn ColumnValues<u64>> = serialize_and_load(&data, CodecType::Bitpacked);
b.iter(|| {
let mut positions = Vec::new();
column.get_row_ids_for_value_range(0..=u64::MAX, 0..data.len() as u32, &mut positions);
positions
});
}
// U64 RANGE END
#[bench]
fn bench_intfastfield_stride7_vec(b: &mut Bencher) {
fn bench_access() {
let permutation = generate_permutation();
let n = permutation.len();
b.iter(|| {
let column_perm: Arc<dyn ColumnValues<u64>> =
serialize_and_load(&permutation, CodecType::Bitpacked);
let permutation_gcd = generate_permutation_gcd();
let column_perm_gcd: Arc<dyn ColumnValues<u64>> =
serialize_and_load(&permutation_gcd, CodecType::Bitpacked);
let mut group: InputGroup<VecCol> = InputGroup::new_with_inputs(vec![
(
"access".to_string(),
(permutation.clone(), column_perm.clone()),
),
(
"access_gcd".to_string(),
(permutation_gcd.clone(), column_perm_gcd.clone()),
),
]);
group.register("stride7_vec", |inp: &VecCol| {
let n = inp.0.len();
let mut a = 0u64;
for i in (0..n / 7).map(|val| val * 7) {
a += permutation[i as usize];
a += inp.0[i];
}
a
black_box(a);
});
}
#[bench]
fn bench_intfastfield_stride7_fflookup(b: &mut Bencher) {
let permutation = generate_permutation();
let n = permutation.len();
let column: Arc<dyn ColumnValues<u64>> = serialize_and_load(&permutation, CodecType::Bitpacked);
b.iter(|| {
let mut a = 0;
group.register("fullscan_vec", |inp: &VecCol| {
let mut a = 0u64;
for i in 0..inp.0.len() {
a += inp.0[i];
}
black_box(a);
});
group.register("stride7_column_values", |inp: &VecCol| {
let n = inp.1.num_vals() as usize;
let mut a = 0u64;
for i in (0..n / 7).map(|val| val * 7) {
a += column.get_val(i as u32);
a += inp.1.get_val(i as u32);
}
a
black_box(a);
});
}
#[bench]
fn bench_intfastfield_scan_all_fflookup(b: &mut Bencher) {
let permutation = generate_permutation();
let n = permutation.len();
let column: Arc<dyn ColumnValues<u64>> = serialize_and_load(&permutation, CodecType::Bitpacked);
let column_ref = column.as_ref();
b.iter(|| {
let mut a = 0u64;
for i in 0u32..n as u32 {
a += column_ref.get_val(i);
}
a
});
}
#[bench]
fn bench_intfastfield_scan_all_fflookup_gcd(b: &mut Bencher) {
let permutation = generate_permutation_gcd();
let n = permutation.len();
let column: Arc<dyn ColumnValues<u64>> = serialize_and_load(&permutation, CodecType::Bitpacked);
b.iter(|| {
group.register("fullscan_column_values", |inp: &VecCol| {
let mut a = 0u64;
let n = inp.1.num_vals() as usize;
for i in 0..n {
a += column.get_val(i as u32);
a += inp.1.get_val(i as u32);
}
a
black_box(a);
});
group.run();
}
#[bench]
fn bench_intfastfield_scan_all_vec(b: &mut Bencher) {
let permutation = generate_permutation();
b.iter(|| {
let mut a = 0u64;
for i in 0..permutation.len() {
a += permutation[i as usize] as u64;
}
a
});
fn bench_range() {
let data_50 = get_data_50percent_item();
let data_u64 = data_50.iter().map(|el| *el as u64).collect::<Vec<_>>();
let column_data: Arc<dyn ColumnValues<u64>> =
serialize_and_load(&data_u64, CodecType::Bitpacked);
let mut group: InputGroup<Arc<dyn ColumnValues<u64>>> =
InputGroup::new_with_inputs(vec![("dist_50pct_item".to_string(), column_data.clone())]);
group.register(
"fastfield_getrange_u64_50percent_hit",
|col: &Arc<dyn ColumnValues<u64>>| {
let mut positions = Vec::new();
col.get_row_ids_for_value_range(FIFTY_PERCENT_RANGE, 0..col.num_vals(), &mut positions);
black_box(positions.len());
},
);
group.register(
"fastfield_getrange_u64_1percent_hit",
|col: &Arc<dyn ColumnValues<u64>>| {
let mut positions = Vec::new();
col.get_row_ids_for_value_range(
ONE_PERCENT_ITEM_RANGE,
0..col.num_vals(),
&mut positions,
);
black_box(positions.len());
},
);
group.register(
"fastfield_getrange_u64_single_hit",
|col: &Arc<dyn ColumnValues<u64>>| {
let mut positions = Vec::new();
col.get_row_ids_for_value_range(SINGLE_ITEM_RANGE, 0..col.num_vals(), &mut positions);
black_box(positions.len());
},
);
group.register(
"fastfield_getrange_u64_hit_all",
|col: &Arc<dyn ColumnValues<u64>>| {
let mut positions = Vec::new();
col.get_row_ids_for_value_range(0..=u64::MAX, 0..col.num_vals(), &mut positions);
black_box(positions.len());
},
);
group.run();
}
fn main() {
bench_access();
bench_range();
}

2
columnar/src/column/mod.rs
View File

@@ -114,7 +114,7 @@ impl<T: PartialOrd + Copy + Debug + Send + Sync + 'static> Column<T> {
}
}
/// Translates a block of docis to row_ids.
/// Translates a block of docids to row_ids.
///
/// returns the row_ids and the matching docids on the same index
/// e.g.

15
columnar/src/column_index/merge/stacked.rs
View File

@@ -56,7 +56,7 @@ fn get_doc_ids_with_values<'a>(
ColumnIndex::Full => Box::new(doc_range),
ColumnIndex::Optional(optional_index) => Box::new(
optional_index
.iter_docs()
.iter_non_null_docs()
.map(move |row| row + doc_range.start),
),
ColumnIndex::Multivalued(multivalued_index) => match multivalued_index {
@@ -73,7 +73,7 @@ fn get_doc_ids_with_values<'a>(
MultiValueIndex::MultiValueIndexV2(multivalued_index) => Box::new(
multivalued_index
.optional_index
.iter_docs()
.iter_non_null_docs()
.map(move |row| row + doc_range.start),
),
},
@@ -105,10 +105,11 @@ fn get_num_values_iterator<'a>(
) -> Box<dyn Iterator<Item = u32> + 'a> {
match column_index {
ColumnIndex::Empty { .. } => Box::new(std::iter::empty()),
ColumnIndex::Full => Box::new(std::iter::repeat(1u32).take(num_docs as usize)),
ColumnIndex::Optional(optional_index) => {
Box::new(std::iter::repeat(1u32).take(optional_index.num_non_nulls() as usize))
}
ColumnIndex::Full => Box::new(std::iter::repeat_n(1u32, num_docs as usize)),
ColumnIndex::Optional(optional_index) => Box::new(std::iter::repeat_n(
1u32,
optional_index.num_non_nulls() as usize,
)),
ColumnIndex::Multivalued(multivalued_index) => Box::new(
multivalued_index
.get_start_index_column()
@@ -177,7 +178,7 @@ impl<'a> Iterable<RowId> for StackedOptionalIndex<'a> {
ColumnIndex::Full => Box::new(columnar_row_range),
ColumnIndex::Optional(optional_index) => Box::new(
optional_index
.iter_docs()
.iter_non_null_docs()
.map(move |row_id: RowId| columnar_row_range.start + row_id),
),
ColumnIndex::Multivalued(_) => {

26
columnar/src/column_index/multivalued_index.rs
View File

@@ -215,6 +215,32 @@ impl MultiValueIndex {
}
}
/// Returns an iterator over document ids that have at least one value.
pub fn iter_non_null_docs(&self) -> Box<dyn Iterator<Item = DocId> + '_> {
match self {
MultiValueIndex::MultiValueIndexV1(idx) => {
let mut doc: DocId = 0u32;
let num_docs = idx.num_docs();
Box::new(std::iter::from_fn(move || {
// This is not the most efficient way to do this, but it's legacy code.
while doc < num_docs {
let cur = doc;
doc += 1;
let start = idx.start_index_column.get_val(cur);
let end = idx.start_index_column.get_val(cur + 1);
if end > start {
return Some(cur);
}
}
None
}))
}
MultiValueIndex::MultiValueIndexV2(idx) => {
Box::new(idx.optional_index.iter_non_null_docs())
}
}
}
/// Converts a list of ranks (row ids of values) in a 1:n index to the corresponding list of
/// docids. Positions are converted inplace to docids.
///

43
columnar/src/column_index/optional_index/mod.rs
View File

@@ -1,4 +1,4 @@
use std::io::{self, Write};
use std::io;
use std::sync::Arc;
mod set;
@@ -11,7 +11,7 @@ use set_block::{
};
use crate::iterable::Iterable;
use crate::{DocId, InvalidData, RowId};
use crate::{DocId, RowId};
/// The threshold for for number of elements after which we switch to dense block encoding.
///
@@ -88,7 +88,7 @@ pub struct OptionalIndex {
impl Iterable<u32> for &OptionalIndex {
fn boxed_iter(&self) -> Box<dyn Iterator<Item = u32> + '_> {
Box::new(self.iter_docs())
Box::new(self.iter_non_null_docs())
}
}
@@ -280,8 +280,9 @@ impl OptionalIndex {
self.num_non_null_docs
}
pub fn iter_docs(&self) -> impl Iterator<Item = RowId> + '_ {
// TODO optimize
pub fn iter_non_null_docs(&self) -> impl Iterator<Item = RowId> + '_ {
// TODO optimize. We could iterate over the blocks directly.
// We use the dense value ids and retrieve the doc ids via select.
let mut select_batch = self.select_cursor();
(0..self.num_non_null_docs).map(move |rank| select_batch.select(rank))
}
@@ -334,38 +335,6 @@ enum Block<'a> {
Sparse(SparseBlock<'a>),
}
#[derive(Debug, Copy, Clone)]
enum OptionalIndexCodec {
Dense = 0,
Sparse = 1,
}
impl OptionalIndexCodec {
fn to_code(self) -> u8 {
self as u8
}
fn try_from_code(code: u8) -> Result<Self, InvalidData> {
match code {
0 => Ok(Self::Dense),
1 => Ok(Self::Sparse),
_ => Err(InvalidData),
}
}
}
impl BinarySerializable for OptionalIndexCodec {
fn serialize<W: Write + ?Sized>(&self, writer: &mut W) -> io::Result<()> {
writer.write_all(&[self.to_code()])
}
fn deserialize<R: io::Read>(reader: &mut R) -> io::Result<Self> {
let optional_codec_code = u8::deserialize(reader)?;
let optional_codec = Self::try_from_code(optional_codec_code)?;
Ok(optional_codec)
}
}
fn serialize_optional_index_block(block_els: &[u16], out: &mut impl io::Write) -> io::Result<()> {
let is_sparse = is_sparse(block_els.len() as u32);
if is_sparse {

173
columnar/src/column_index/optional_index/tests.rs
View File

@@ -164,7 +164,11 @@ fn test_optional_index_large() {
fn test_optional_index_iter_aux(row_ids: &[RowId], num_rows: RowId) {
let optional_index = OptionalIndex::for_test(num_rows, row_ids);
assert_eq!(optional_index.num_docs(), num_rows);
assert!(optional_index.iter_docs().eq(row_ids.iter().copied()));
assert!(
optional_index
.iter_non_null_docs()
.eq(row_ids.iter().copied())
);
}
#[test]
@@ -219,170 +223,3 @@ fn test_optional_index_for_tests() {
assert!(!optional_index.contains(3));
assert_eq!(optional_index.num_docs(), 4);
}
#[cfg(all(test, feature = "unstable"))]
mod bench {
use rand::rngs::StdRng;
use rand::{Rng, SeedableRng};
use test::Bencher;
use super::*;
const TOTAL_NUM_VALUES: u32 = 1_000_000;
fn gen_bools(fill_ratio: f64) -> OptionalIndex {
let mut out = Vec::new();
let mut rng: StdRng = StdRng::from_seed([1u8; 32]);
let vals: Vec<RowId> = (0..TOTAL_NUM_VALUES)
.map(|_| rng.gen_bool(fill_ratio))
.enumerate()
.filter(|(_pos, val)| *val)
.map(|(pos, _)| pos as RowId)
.collect();
serialize_optional_index(&&vals[..], TOTAL_NUM_VALUES, &mut out).unwrap();
open_optional_index(OwnedBytes::new(out)).unwrap()
}
fn random_range_iterator(
start: u32,
end: u32,
avg_step_size: u32,
avg_deviation: u32,
) -> impl Iterator<Item = u32> {
let mut rng: StdRng = StdRng::from_seed([1u8; 32]);
let mut current = start;
std::iter::from_fn(move || {
current += rng.gen_range(avg_step_size - avg_deviation..=avg_step_size + avg_deviation);
if current >= end { None } else { Some(current) }
})
}
fn n_percent_step_iterator(percent: f32, num_values: u32) -> impl Iterator<Item = u32> {
let ratio = percent / 100.0;
let step_size = (1f32 / ratio) as u32;
let deviation = step_size - 1;
random_range_iterator(0, num_values, step_size, deviation)
}
fn walk_over_data(codec: &OptionalIndex, avg_step_size: u32) -> Option<u32> {
walk_over_data_from_positions(
codec,
random_range_iterator(0, TOTAL_NUM_VALUES, avg_step_size, 0),
)
}
fn walk_over_data_from_positions(
codec: &OptionalIndex,
positions: impl Iterator<Item = u32>,
) -> Option<u32> {
let mut dense_idx: Option<u32> = None;
for idx in positions {
dense_idx = dense_idx.or(codec.rank_if_exists(idx));
}
dense_idx
}
#[bench]
fn bench_translate_orig_to_codec_1percent_filled_10percent_hit(bench: &mut Bencher) {
let codec = gen_bools(0.01f64);
bench.iter(|| walk_over_data(&codec, 100));
}
#[bench]
fn bench_translate_orig_to_codec_5percent_filled_10percent_hit(bench: &mut Bencher) {
let codec = gen_bools(0.05f64);
bench.iter(|| walk_over_data(&codec, 100));
}
#[bench]
fn bench_translate_orig_to_codec_5percent_filled_1percent_hit(bench: &mut Bencher) {
let codec = gen_bools(0.05f64);
bench.iter(|| walk_over_data(&codec, 1000));
}
#[bench]
fn bench_translate_orig_to_codec_full_scan_1percent_filled(bench: &mut Bencher) {
let codec = gen_bools(0.01f64);
bench.iter(|| walk_over_data_from_positions(&codec, 0..TOTAL_NUM_VALUES));
}
#[bench]
fn bench_translate_orig_to_codec_full_scan_10percent_filled(bench: &mut Bencher) {
let codec = gen_bools(0.1f64);
bench.iter(|| walk_over_data_from_positions(&codec, 0..TOTAL_NUM_VALUES));
}
#[bench]
fn bench_translate_orig_to_codec_full_scan_90percent_filled(bench: &mut Bencher) {
let codec = gen_bools(0.9f64);
bench.iter(|| walk_over_data_from_positions(&codec, 0..TOTAL_NUM_VALUES));
}
#[bench]
fn bench_translate_orig_to_codec_10percent_filled_1percent_hit(bench: &mut Bencher) {
let codec = gen_bools(0.1f64);
bench.iter(|| walk_over_data(&codec, 100));
}
#[bench]
fn bench_translate_orig_to_codec_50percent_filled_1percent_hit(bench: &mut Bencher) {
let codec = gen_bools(0.5f64);
bench.iter(|| walk_over_data(&codec, 100));
}
#[bench]
fn bench_translate_orig_to_codec_90percent_filled_1percent_hit(bench: &mut Bencher) {
let codec = gen_bools(0.9f64);
bench.iter(|| walk_over_data(&codec, 100));
}
#[bench]
fn bench_translate_codec_to_orig_1percent_filled_0comma005percent_hit(bench: &mut Bencher) {
bench_translate_codec_to_orig_util(0.01f64, 0.005f32, bench);
}
#[bench]
fn bench_translate_codec_to_orig_10percent_filled_0comma005percent_hit(bench: &mut Bencher) {
bench_translate_codec_to_orig_util(0.1f64, 0.005f32, bench);
}
#[bench]
fn bench_translate_codec_to_orig_1percent_filled_10percent_hit(bench: &mut Bencher) {
bench_translate_codec_to_orig_util(0.01f64, 10f32, bench);
}
#[bench]
fn bench_translate_codec_to_orig_1percent_filled_full_scan(bench: &mut Bencher) {
bench_translate_codec_to_orig_util(0.01f64, 100f32, bench);
}
fn bench_translate_codec_to_orig_util(
percent_filled: f64,
percent_hit: f32,
bench: &mut Bencher,
) {
let codec = gen_bools(percent_filled);
let num_non_nulls = codec.num_non_nulls();
let idxs: Vec<u32> = if percent_hit == 100.0f32 {
(0..num_non_nulls).collect()
} else {
n_percent_step_iterator(percent_hit, num_non_nulls).collect()
};
let mut output = vec![0u32; idxs.len()];
bench.iter(|| {
output.copy_from_slice(&idxs[..]);
codec.select_batch(&mut output);
});
}
#[bench]
fn bench_translate_codec_to_orig_90percent_filled_0comma005percent_hit(bench: &mut Bencher) {
bench_translate_codec_to_orig_util(0.9f64, 0.005, bench);
}
#[bench]
fn bench_translate_codec_to_orig_90percent_filled_full_scan(bench: &mut Bencher) {
bench_translate_codec_to_orig_util(0.9f64, 100.0f32, bench);
}
}

139
columnar/src/column_values/bench.rs
View File

@@ -1,139 +0,0 @@
use std::sync::Arc;
use common::OwnedBytes;
use rand::rngs::StdRng;
use rand::{Rng, SeedableRng};
use test::{self, Bencher};
use super::*;
use crate::column_values::u64_based::*;
fn get_data() -> Vec<u64> {
let mut rng = StdRng::seed_from_u64(2u64);
let mut data: Vec<_> = (100..55000_u64)
.map(|num| num + rng.r#gen::<u8>() as u64)
.collect();
data.push(99_000);
data.insert(1000, 2000);
data.insert(2000, 100);
data.insert(3000, 4100);
data.insert(4000, 100);
data.insert(5000, 800);
data
}
fn compute_stats(vals: impl Iterator<Item = u64>) -> ColumnStats {
let mut stats_collector = StatsCollector::default();
for val in vals {
stats_collector.collect(val);
}
stats_collector.stats()
}
#[inline(never)]
fn value_iter() -> impl Iterator<Item = u64> {
0..20_000
}
fn get_reader_for_bench<Codec: ColumnCodec>(data: &[u64]) -> Codec::ColumnValues {
let mut bytes = Vec::new();
let stats = compute_stats(data.iter().cloned());
let mut codec_serializer = Codec::estimator();
for val in data {
codec_serializer.collect(*val);
}
codec_serializer
.serialize(&stats, Box::new(data.iter().copied()).as_mut(), &mut bytes)
.unwrap();
Codec::load(OwnedBytes::new(bytes)).unwrap()
}
fn bench_get<Codec: ColumnCodec>(b: &mut Bencher, data: &[u64]) {
let col = get_reader_for_bench::<Codec>(data);
b.iter(|| {
let mut sum = 0u64;
for pos in value_iter() {
let val = col.get_val(pos as u32);
sum = sum.wrapping_add(val);
}
sum
});
}
#[inline(never)]
fn bench_get_dynamic_helper(b: &mut Bencher, col: Arc<dyn ColumnValues>) {
b.iter(|| {
let mut sum = 0u64;
for pos in value_iter() {
let val = col.get_val(pos as u32);
sum = sum.wrapping_add(val);
}
sum
});
}
fn bench_get_dynamic<Codec: ColumnCodec>(b: &mut Bencher, data: &[u64]) {
let col = Arc::new(get_reader_for_bench::<Codec>(data));
bench_get_dynamic_helper(b, col);
}
fn bench_create<Codec: ColumnCodec>(b: &mut Bencher, data: &[u64]) {
let stats = compute_stats(data.iter().cloned());
let mut bytes = Vec::new();
b.iter(|| {
bytes.clear();
let mut codec_serializer = Codec::estimator();
for val in data.iter().take(1024) {
codec_serializer.collect(*val);
}
codec_serializer.serialize(&stats, Box::new(data.iter().copied()).as_mut(), &mut bytes)
});
}
#[bench]
fn bench_fastfield_bitpack_create(b: &mut Bencher) {
let data: Vec<_> = get_data();
bench_create::<BitpackedCodec>(b, &data);
}
#[bench]
fn bench_fastfield_linearinterpol_create(b: &mut Bencher) {
let data: Vec<_> = get_data();
bench_create::<LinearCodec>(b, &data);
}
#[bench]
fn bench_fastfield_multilinearinterpol_create(b: &mut Bencher) {
let data: Vec<_> = get_data();
bench_create::<BlockwiseLinearCodec>(b, &data);
}
#[bench]
fn bench_fastfield_bitpack_get(b: &mut Bencher) {
let data: Vec<_> = get_data();
bench_get::<BitpackedCodec>(b, &data);
}
#[bench]
fn bench_fastfield_bitpack_get_dynamic(b: &mut Bencher) {
let data: Vec<_> = get_data();
bench_get_dynamic::<BitpackedCodec>(b, &data);
}
#[bench]
fn bench_fastfield_linearinterpol_get(b: &mut Bencher) {
let data: Vec<_> = get_data();
bench_get::<LinearCodec>(b, &data);
}
#[bench]
fn bench_fastfield_linearinterpol_get_dynamic(b: &mut Bencher) {
let data: Vec<_> = get_data();
bench_get_dynamic::<LinearCodec>(b, &data);
}
#[bench]
fn bench_fastfield_multilinearinterpol_get(b: &mut Bencher) {
let data: Vec<_> = get_data();
bench_get::<BlockwiseLinearCodec>(b, &data);
}
#[bench]
fn bench_fastfield_multilinearinterpol_get_dynamic(b: &mut Bencher) {
let data: Vec<_> = get_data();
bench_get_dynamic::<BlockwiseLinearCodec>(b, &data);
}

3
columnar/src/column_values/mod.rs
View File

@@ -242,6 +242,3 @@ impl<T: Copy + PartialOrd + Debug + 'static> ColumnValues<T> for Arc<dyn ColumnV
.get_row_ids_for_value_range(range, doc_id_range, positions)
}
}
#[cfg(all(test, feature = "unstable"))]
mod bench;

16
columnar/src/column_values/u128_based/compact_space/build_compact_space.rs
View File

@@ -185,10 +185,10 @@ impl CompactSpaceBuilder {
let mut covered_space = Vec::with_capacity(self.blanks.len());
// beginning of the blanks
if let Some(first_blank_start) = self.blanks.first().map(RangeInclusive::start) {
if *first_blank_start != 0 {
covered_space.push(0..=first_blank_start - 1);
}
if let Some(first_blank_start) = self.blanks.first().map(RangeInclusive::start)
&& *first_blank_start != 0
{
covered_space.push(0..=first_blank_start - 1);
}
// Between the blanks
@@ -202,10 +202,10 @@ impl CompactSpaceBuilder {
covered_space.extend(between_blanks);
// end of the blanks
if let Some(last_blank_end) = self.blanks.last().map(RangeInclusive::end) {
if *last_blank_end != u128::MAX {
covered_space.push(last_blank_end + 1..=u128::MAX);
}
if let Some(last_blank_end) = self.blanks.last().map(RangeInclusive::end)
&& *last_blank_end != u128::MAX
{
covered_space.push(last_blank_end + 1..=u128::MAX);
}
if covered_space.is_empty() {

2
columnar/src/column_values/u64_based/bitpacked.rs
View File

@@ -105,7 +105,7 @@ impl ColumnCodecEstimator for BitpackedCodecEstimator {
fn estimate(&self, stats: &ColumnStats) -> Option<u64> {
let num_bits_per_value = num_bits(stats);
Some(stats.num_bytes() + (stats.num_rows as u64 * (num_bits_per_value as u64) + 7) / 8)
Some(stats.num_bytes() + (stats.num_rows as u64 * (num_bits_per_value as u64)).div_ceil(8))
}
fn serialize(

2
columnar/src/column_values/u64_based/linear.rs
View File

@@ -117,7 +117,7 @@ impl ColumnCodecEstimator for LinearCodecEstimator {
Some(
stats.num_bytes()
+ linear_params.num_bytes()
+ (num_bits as u64 * stats.num_rows as u64 + 7) / 8,
+ (num_bits as u64 * stats.num_rows as u64).div_ceil(8),
)
}

2
columnar/src/columnar/merge/mod.rs
View File

@@ -367,7 +367,7 @@ fn is_empty_after_merge(
ColumnIndex::Empty { .. } => true,
ColumnIndex::Full => alive_bitset.len() == 0,
ColumnIndex::Optional(optional_index) => {
for doc in optional_index.iter_docs() {
for doc in optional_index.iter_non_null_docs() {
if alive_bitset.contains(doc) {
return false;
}

2
columnar/src/columnar/writer/column_operation.rs
View File

@@ -244,7 +244,7 @@ impl SymbolValue for UnorderedId {
fn compute_num_bytes_for_u64(val: u64) -> usize {
let msb = (64u32 - val.leading_zeros()) as usize;
(msb + 7) / 8
msb.div_ceil(8)
}
fn encode_zig_zag(n: i64) -> u64 {

5
columnar/src/lib.rs
View File

@@ -17,15 +17,10 @@
//! column.
//! - [column_values]: Stores the values of a column in a dense format.
#![cfg_attr(all(feature = "unstable", test), feature(test))]
#[cfg(test)]
#[macro_use]
extern crate more_asserts;
#[cfg(all(test, feature = "unstable"))]
extern crate test;
use std::fmt::Display;
use std::io;

76
columnar/src/value.rs
View File

@@ -1,3 +1,5 @@
use std::str::FromStr;
use common::DateTime;
use crate::InvalidData;
@@ -9,6 +11,23 @@ pub enum NumericalValue {
F64(f64),
}
impl FromStr for NumericalValue {
type Err = ();
fn from_str(s: &str) -> Result<Self, ()> {
if let Ok(val_i64) = s.parse::<i64>() {
return Ok(val_i64.into());
}
if let Ok(val_u64) = s.parse::<u64>() {
return Ok(val_u64.into());
}
if let Ok(val_f64) = s.parse::<f64>() {
return Ok(NumericalValue::from(val_f64).normalize());
}
Err(())
}
}
impl NumericalValue {
pub fn numerical_type(&self) -> NumericalType {
match self {
@@ -26,7 +45,7 @@ impl NumericalValue {
if val <= i64::MAX as u64 {
NumericalValue::I64(val as i64)
} else {
NumericalValue::F64(val as f64)
NumericalValue::U64(val)
}
}
NumericalValue::I64(val) => NumericalValue::I64(val),
@@ -141,6 +160,7 @@ impl Coerce for DateTime {
#[cfg(test)]
mod tests {
use super::NumericalType;
use crate::NumericalValue;
#[test]
fn test_numerical_type_code() {
@@ -153,4 +173,58 @@ mod tests {
}
assert_eq!(num_numerical_type, 3);
}
#[test]
fn test_parse_numerical() {
assert_eq!(
"123".parse::<NumericalValue>().unwrap(),
NumericalValue::I64(123)
);
assert_eq!(
"18446744073709551615".parse::<NumericalValue>().unwrap(),
NumericalValue::U64(18446744073709551615u64)
);
assert_eq!(
"1.0".parse::<NumericalValue>().unwrap(),
NumericalValue::I64(1i64)
);
assert_eq!(
"1.1".parse::<NumericalValue>().unwrap(),
NumericalValue::F64(1.1f64)
);
assert_eq!(
"-1.0".parse::<NumericalValue>().unwrap(),
NumericalValue::I64(-1i64)
);
}
#[test]
fn test_normalize_numerical() {
assert_eq!(
NumericalValue::from(1u64).normalize(),
NumericalValue::I64(1i64),
);
let limit_val = i64::MAX as u64 + 1u64;
assert_eq!(
NumericalValue::from(limit_val).normalize(),
NumericalValue::U64(limit_val),
);
assert_eq!(
NumericalValue::from(-1i64).normalize(),
NumericalValue::I64(-1i64),
);
assert_eq!(
NumericalValue::from(-2.0f64).normalize(),
NumericalValue::I64(-2i64),
);
assert_eq!(
NumericalValue::from(-2.1f64).normalize(),
NumericalValue::F64(-2.1f64),
);
let large_float = 2.0f64.powf(70.0f64);
assert_eq!(
NumericalValue::from(large_float).normalize(),
NumericalValue::F64(large_float),
);
}
}

2
common/Cargo.toml
View File

@@ -1,6 +1,6 @@
[package]
name = "tantivy-common"
version = "0.9.0"
version = "0.10.0"
authors = ["Paul Masurel <paul@quickwit.io>", "Pascal Seitz <pascal@quickwit.io>"]
license = "MIT"
edition = "2024"

2
common/src/bitset.rs
View File

@@ -183,7 +183,7 @@ pub struct BitSet {
}
fn num_buckets(max_val: u32) -> u32 {
(max_val + 63u32) / 64u32
max_val.div_ceil(64u32)
}
impl BitSet {

BIN
doc/assets/images/paradedb.png
View File

Binary file not shown.
Side by Side Swipe Overlay

Before

Width:  |  Height:  |  Size: 30 KiB

After

Width:  |  Height:  |  Size: 7.4 KiB

BIN
doc/assets/images/searchbenchmark.png
View File

Binary file not shown.
Side by Side

Before

Width:  |  Height:  |  Size: 653 KiB

2
examples/basic_search.rs
View File

@@ -51,7 +51,7 @@ fn main() -> tantivy::Result<()> {
// Our second field is body.
// We want full-text search for it, but we do not
// need to be able to be able to retrieve it
// need to be able to retrieve it
// for our application.
//
// We can make our index lighter by omitting the `STORED` flag.

212
examples/filter_aggregation.rs Normal file
View File

@@ -0,0 +1,212 @@
// # Filter Aggregation Example
//
// This example demonstrates filter aggregations - creating buckets of documents
// matching specific queries, with nested aggregations computed on each bucket.
//
// Filter aggregations are useful for computing metrics on different subsets of
// your data in a single query, like "average price overall + average price for
// electronics + count of in-stock items".
use serde_json::json;
use tantivy::aggregation::agg_req::Aggregations;
use tantivy::aggregation::AggregationCollector;
use tantivy::query::AllQuery;
use tantivy::schema::{Schema, FAST, INDEXED, TEXT};
use tantivy::{doc, Index};
fn main() -> tantivy::Result<()> {
// Create a simple product schema
let mut schema_builder = Schema::builder();
schema_builder.add_text_field("category", TEXT | FAST);
schema_builder.add_text_field("brand", TEXT | FAST);
schema_builder.add_u64_field("price", FAST);
schema_builder.add_f64_field("rating", FAST);
schema_builder.add_bool_field("in_stock", FAST | INDEXED);
let schema = schema_builder.build();
// Create index and add sample products
let index = Index::create_in_ram(schema.clone());
let mut writer = index.writer(50_000_000)?;
writer.add_document(doc!(
schema.get_field("category")? => "electronics",
schema.get_field("brand")? => "apple",
schema.get_field("price")? => 999u64,
schema.get_field("rating")? => 4.5f64,
schema.get_field("in_stock")? => true
))?;
writer.add_document(doc!(
schema.get_field("category")? => "electronics",
schema.get_field("brand")? => "samsung",
schema.get_field("price")? => 799u64,
schema.get_field("rating")? => 4.2f64,
schema.get_field("in_stock")? => true
))?;
writer.add_document(doc!(
schema.get_field("category")? => "clothing",
schema.get_field("brand")? => "nike",
schema.get_field("price")? => 120u64,
schema.get_field("rating")? => 4.1f64,
schema.get_field("in_stock")? => false
))?;
writer.add_document(doc!(
schema.get_field("category")? => "books",
schema.get_field("brand")? => "penguin",
schema.get_field("price")? => 25u64,
schema.get_field("rating")? => 4.8f64,
schema.get_field("in_stock")? => true
))?;
writer.commit()?;
let reader = index.reader()?;
let searcher = reader.searcher();
// Example 1: Basic filter with metric aggregation
println!("=== Example 1: Electronics average price ===");
let agg_req = json!({
"electronics": {
"filter": "category:electronics",
"aggs": {
"avg_price": { "avg": { "field": "price" } }
}
}
});
let agg: Aggregations = serde_json::from_value(agg_req)?;
let collector = AggregationCollector::from_aggs(agg, Default::default());
let result = searcher.search(&AllQuery, &collector)?;
let expected = json!({
"electronics": {
"doc_count": 2,
"avg_price": { "value": 899.0 }
}
});
assert_eq!(serde_json::to_value(&result)?, expected);
println!("{}\n", serde_json::to_string_pretty(&result)?);
// Example 2: Multiple independent filters
println!("=== Example 2: Multiple filters in one query ===");
let agg_req = json!({
"electronics": {
"filter": "category:electronics",
"aggs": { "avg_price": { "avg": { "field": "price" } } }
},
"in_stock": {
"filter": "in_stock:true",
"aggs": { "count": { "value_count": { "field": "brand" } } }
},
"high_rated": {
"filter": "rating:[4.5 TO *]",
"aggs": { "count": { "value_count": { "field": "brand" } } }
}
});
let agg: Aggregations = serde_json::from_value(agg_req)?;
let collector = AggregationCollector::from_aggs(agg, Default::default());
let result = searcher.search(&AllQuery, &collector)?;
let expected = json!({
"electronics": {
"doc_count": 2,
"avg_price": { "value": 899.0 }
},
"in_stock": {
"doc_count": 3,
"count": { "value": 3.0 }
},
"high_rated": {
"doc_count": 2,
"count": { "value": 2.0 }
}
});
assert_eq!(serde_json::to_value(&result)?, expected);
println!("{}\n", serde_json::to_string_pretty(&result)?);
// Example 3: Nested filters - progressive refinement
println!("=== Example 3: Nested filters ===");
let agg_req = json!({
"in_stock": {
"filter": "in_stock:true",
"aggs": {
"electronics": {
"filter": "category:electronics",
"aggs": {
"expensive": {
"filter": "price:[800 TO *]",
"aggs": {
"avg_rating": { "avg": { "field": "rating" } }
}
}
}
}
}
}
});
let agg: Aggregations = serde_json::from_value(agg_req)?;
let collector = AggregationCollector::from_aggs(agg, Default::default());
let result = searcher.search(&AllQuery, &collector)?;
let expected = json!({
"in_stock": {
"doc_count": 3, // apple, samsung, penguin
"electronics": {
"doc_count": 2, // apple, samsung
"expensive": {
"doc_count": 1, // only apple (999)
"avg_rating": { "value": 4.5 }
}
}
}
});
assert_eq!(serde_json::to_value(&result)?, expected);
println!("{}\n", serde_json::to_string_pretty(&result)?);
// Example 4: Filter with sub-aggregation (terms)
println!("=== Example 4: Filter with terms sub-aggregation ===");
let agg_req = json!({
"electronics": {
"filter": "category:electronics",
"aggs": {
"by_brand": {
"terms": { "field": "brand" },
"aggs": {
"avg_price": { "avg": { "field": "price" } }
}
}
}
}
});
let agg: Aggregations = serde_json::from_value(agg_req)?;
let collector = AggregationCollector::from_aggs(agg, Default::default());
let result = searcher.search(&AllQuery, &collector)?;
let expected = json!({
"electronics": {
"doc_count": 2,
"by_brand": {
"buckets": [
{
"key": "samsung",
"doc_count": 1,
"avg_price": { "value": 799.0 }
},
{
"key": "apple",
"doc_count": 1,
"avg_price": { "value": 999.0 }
}
],
"sum_other_doc_count": 0,
"doc_count_error_upper_bound": 0
}
}
});
assert_eq!(serde_json::to_value(&result)?, expected);
println!("{}", serde_json::to_string_pretty(&result)?);
Ok(())
}

1
examples/fuzzy_search.rs
View File

@@ -85,7 +85,6 @@ fn main() -> tantivy::Result<()> {
index_writer.add_document(doc!(
title => "The Diary of a Young Girl",
))?;
index_writer.commit()?;
// ### Committing
//

4
query-grammar/Cargo.toml
View File

@@ -1,6 +1,6 @@
[package]
name = "tantivy-query-grammar"
version = "0.24.0"
version = "0.25.0"
authors = ["Paul Masurel <paul.masurel@gmail.com>"]
license = "MIT"
categories = ["database-implementations", "data-structures"]
@@ -15,3 +15,5 @@ edition = "2024"
nom = "7"
serde = { version = "1.0.219", features = ["derive"] }
serde_json = "1.0.140"
ordered-float = "5.0.0"
fnv = "1.0.7"

16
query-grammar/src/infallible.rs
View File

@@ -117,6 +117,22 @@ where F: nom::Parser<I, (O, ErrorList), Infallible> {
}
}
pub(crate) fn terminated_infallible<I, O1, O2, F, G>(
mut first: F,
mut second: G,
) -> impl FnMut(I) -> JResult<I, O1>
where
F: nom::Parser<I, (O1, ErrorList), Infallible>,
G: nom::Parser<I, (O2, ErrorList), Infallible>,
{
move |input: I| {
let (input, (o1, mut err)) = first.parse(input)?;
let (input, (_, mut err2)) = second.parse(input)?;
err.append(&mut err2);
Ok((input, (o1, err)))
}
}
pub(crate) fn delimited_infallible<I, O1, O2, O3, F, G, H>(
mut first: F,
mut second: G,

12
query-grammar/src/lib.rs
View File

@@ -31,7 +31,17 @@ pub fn parse_query_lenient(query: &str) -> (UserInputAst, Vec<LenientError>) {
#[cfg(test)]
mod tests {
use crate::{parse_query, parse_query_lenient};
use crate::{UserInputAst, parse_query, parse_query_lenient};
#[test]
fn test_deduplication() {
let ast: UserInputAst = parse_query("a a").unwrap();
let json = serde_json::to_string(&ast).unwrap();
assert_eq!(
json,
r#"{"type":"bool","clauses":[[null,{"type":"literal","field_name":null,"phrase":"a","delimiter":"none","slop":0,"prefix":false}]]}"#
);
}
#[test]
fn test_parse_query_serialization() {

173
query-grammar/src/query_grammar.rs
View File

@@ -1,6 +1,7 @@
use std::borrow::Cow;
use std::iter::once;
use fnv::FnvHashSet;
use nom::IResult;
use nom::branch::alt;
use nom::bytes::complete::tag;
@@ -36,7 +37,7 @@ fn field_name(inp: &str) -> IResult<&str, String> {
alt((first_char, escape_sequence())),
many0(alt((simple_char, escape_sequence(), char('\\')))),
)),
char(':'),
tuple((multispace0, char(':'), multispace0)),
),
|(first_char, next)| once(first_char).chain(next).collect(),
)(inp)
@@ -68,7 +69,7 @@ fn interpret_escape(source: &str) -> String {
/// Consume a word outside of any context.
// TODO should support escape sequences
fn word(inp: &str) -> IResult<&str, Cow<str>> {
fn word(inp: &str) -> IResult<&str, Cow<'_, str>> {
map_res(
recognize(tuple((
alt((
@@ -305,15 +306,14 @@ fn term_group_infallible(inp: &str) -> JResult<&str, UserInputAst> {
let (inp, (field_name, _, _, _)) =
tuple((field_name, multispace0, char('('), multispace0))(inp).expect("precondition failed");
let res = delimited_infallible(
delimited_infallible(
nothing,
map(ast_infallible, |(mut ast, errors)| {
ast.set_default_field(field_name.to_string());
(ast, errors)
}),
opt_i_err(char(')'), "expected ')'"),
)(inp);
res
)(inp)
}
fn exists(inp: &str) -> IResult<&str, UserInputLeaf> {
@@ -367,7 +367,10 @@ fn literal(inp: &str) -> IResult<&str, UserInputAst> {
// something (a field name) got parsed before
alt((
map(
tuple((opt(field_name), alt((range, set, exists, term_or_phrase)))),
tuple((
opt(field_name),
alt((range, set, exists, regex, term_or_phrase)),
)),
|(field_name, leaf): (Option<String>, UserInputLeaf)| leaf.set_field(field_name).into(),
),
term_group,
@@ -389,6 +392,10 @@ fn literal_no_group_infallible(inp: &str) -> JResult<&str, Option<UserInputAst>>
value((), peek(one_of("{[><"))),
map(range_infallible, |(range, errs)| (Some(range), errs)),
),
(
value((), peek(one_of("/"))),
map(regex_infallible, |(regex, errs)| (Some(regex), errs)),
),
),
delimited_infallible(space0_infallible, term_or_phrase_infallible, nothing),
),
@@ -689,6 +696,61 @@ fn set_infallible(mut inp: &str) -> JResult<&str, UserInputLeaf> {
}
}
fn regex(inp: &str) -> IResult<&str, UserInputLeaf> {
map(
terminated(
delimited(
char('/'),
many1(alt((preceded(char('\\'), char('/')), none_of("/")))),
char('/'),
),
peek(alt((multispace1, eof))),
),
|elements| UserInputLeaf::Regex {
field: None,
pattern: elements.into_iter().collect::<String>(),
},
)(inp)
}
fn regex_infallible(inp: &str) -> JResult<&str, UserInputLeaf> {
match terminated_infallible(
delimited_infallible(
opt_i_err(char('/'), "missing delimiter /"),
opt_i(many1(alt((preceded(char('\\'), char('/')), none_of("/"))))),
opt_i_err(char('/'), "missing delimiter /"),
),
opt_i_err(
peek(alt((multispace1, eof))),
"expected whitespace or end of input",
),
)(inp)
{
Ok((rest, (elements_part, errors))) => {
let pattern = match elements_part {
Some(elements_part) => elements_part.into_iter().collect(),
None => String::new(),
};
let res = UserInputLeaf::Regex {
field: None,
pattern,
};
Ok((rest, (res, errors)))
}
Err(e) => {
let errs = vec![LenientErrorInternal {
pos: inp.len(),
message: e.to_string(),
}];
let res = UserInputLeaf::Regex {
field: None,
pattern: String::new(),
};
Ok((inp, (res, errs)))
}
}
}
fn negate(expr: UserInputAst) -> UserInputAst {
expr.unary(Occur::MustNot)
}
@@ -753,7 +815,7 @@ fn boosted_leaf(inp: &str) -> IResult<&str, UserInputAst> {
tuple((leaf, fallible(boost))),
|(leaf, boost_opt)| match boost_opt {
Some(boost) if (boost - 1.0).abs() > f64::EPSILON => {
UserInputAst::Boost(Box::new(leaf), boost)
UserInputAst::Boost(Box::new(leaf), boost.into())
}
_ => leaf,
},
@@ -765,7 +827,7 @@ fn boosted_leaf_infallible(inp: &str) -> JResult<&str, Option<UserInputAst>> {
tuple_infallible((leaf_infallible, boost)),
|((leaf, boost_opt), error)| match boost_opt {
Some(boost) if (boost - 1.0).abs() > f64::EPSILON => (
leaf.map(|leaf| UserInputAst::Boost(Box::new(leaf), boost)),
leaf.map(|leaf| UserInputAst::Boost(Box::new(leaf), boost.into())),
error,
),
_ => (leaf, error),
@@ -1016,12 +1078,25 @@ pub fn parse_to_ast_lenient(query_str: &str) -> (UserInputAst, Vec<LenientError>
(rewrite_ast(res), errors)
}
/// Removes unnecessary children clauses in AST
///
/// Motivated by [issue #1433](https://github.com/quickwit-oss/tantivy/issues/1433)
fn rewrite_ast(mut input: UserInputAst) -> UserInputAst {
if let UserInputAst::Clause(terms) = &mut input {
for term in terms {
if let UserInputAst::Clause(sub_clauses) = &mut input {
// call rewrite_ast recursively on children clauses if applicable
let mut new_clauses = Vec::with_capacity(sub_clauses.len());
for (occur, clause) in sub_clauses.drain(..) {
let rewritten_clause = rewrite_ast(clause);
new_clauses.push((occur, rewritten_clause));
}
*sub_clauses = new_clauses;
// remove duplicate child clauses
// e.g. (+a +b) OR (+c +d) OR (+a +b) => (+a +b) OR (+c +d)
let mut seen = FnvHashSet::default();
sub_clauses.retain(|term| seen.insert(term.clone()));
// Removes unnecessary children clauses in AST
//
// Motivated by [issue #1433](https://github.com/quickwit-oss/tantivy/issues/1433)
for term in sub_clauses {
rewrite_ast_clause(term);
}
}
@@ -1283,6 +1358,10 @@ mod test {
super::field_name("~my~field:a"),
Ok(("a", "~my~field".to_string()))
);
assert_eq!(
super::field_name(".my.field.name : a"),
Ok(("a", ".my.field.name".to_string()))
);
for special_char in SPECIAL_CHARS.iter() {
let query = &format!("\\{special_char}my\\{special_char}field:a");
assert_eq!(
@@ -1689,4 +1768,72 @@ mod test {
fn test_invalid_field() {
test_is_parse_err(r#"!bc:def"#, "!bc:def");
}
#[test]
fn test_regex_parser() {
let r = parse_to_ast(r#"a:/joh?n(ath[oa]n)/"#);
assert!(r.is_ok(), "Failed to parse custom query: {r:?}");
let (_, input) = r.unwrap();
match input {
UserInputAst::Leaf(leaf) => match leaf.as_ref() {
UserInputLeaf::Regex { field, pattern } => {
assert_eq!(field, &Some("a".to_string()));
assert_eq!(pattern, "joh?n(ath[oa]n)");
}
_ => panic!("Expected a regex leaf, got {leaf:?}"),
},
_ => panic!("Expected a leaf"),
}
let r = parse_to_ast(r#"a:/\\/cgi-bin\\/luci.*/"#);
assert!(r.is_ok(), "Failed to parse custom query: {r:?}");
let (_, input) = r.unwrap();
match input {
UserInputAst::Leaf(leaf) => match leaf.as_ref() {
UserInputLeaf::Regex { field, pattern } => {
assert_eq!(field, &Some("a".to_string()));
assert_eq!(pattern, "\\/cgi-bin\\/luci.*");
}
_ => panic!("Expected a regex leaf, got {leaf:?}"),
},
_ => panic!("Expected a leaf"),
}
}
#[test]
fn test_regex_parser_lenient() {
let literal = |query| literal_infallible(query).unwrap().1;
let (res, errs) = literal(r#"a:/joh?n(ath[oa]n)/"#);
let expected = UserInputLeaf::Regex {
field: Some("a".to_string()),
pattern: "joh?n(ath[oa]n)".to_string(),
}
.into();
assert_eq!(res.unwrap(), expected);
assert!(errs.is_empty(), "Expected no errors, got: {errs:?}");
let (res, errs) = literal("title:/joh?n(ath[oa]n)");
let expected = UserInputLeaf::Regex {
field: Some("title".to_string()),
pattern: "joh?n(ath[oa]n)".to_string(),
}
.into();
assert_eq!(res.unwrap(), expected);
assert_eq!(errs.len(), 1, "Expected 1 error, got: {errs:?}");
assert_eq!(
errs[0].message, "missing delimiter /",
"Unexpected error message",
);
}
#[test]
fn test_space_before_value() {
test_parse_query_to_ast_helper("field : a", r#""field":a"#);
test_parse_query_to_ast_helper("field: a", r#""field":a"#);
test_parse_query_to_ast_helper("field :a", r#""field":a"#);
test_parse_query_to_ast_helper(
"field : 'happy tax payer' AND other_field : 1",
r#"(+"field":'happy tax payer' +"other_field":1)"#,
);
}
}

36
query-grammar/src/user_input_ast.rs
View File

@@ -5,7 +5,7 @@ use serde::Serialize;
use crate::Occur;
#[derive(PartialEq, Clone, Serialize)]
#[derive(PartialEq, Eq, Hash, Clone, Serialize)]
#[serde(tag = "type")]
#[serde(rename_all = "snake_case")]
pub enum UserInputLeaf {
@@ -23,6 +23,10 @@ pub enum UserInputLeaf {
Exists {
field: String,
},
Regex {
field: Option<String>,
pattern: String,
},
}
impl UserInputLeaf {
@@ -46,6 +50,7 @@ impl UserInputLeaf {
UserInputLeaf::Exists { field: _ } => UserInputLeaf::Exists {
field: field.expect("Exist query without a field isn't allowed"),
},
UserInputLeaf::Regex { field: _, pattern } => UserInputLeaf::Regex { field, pattern },
}
}
@@ -103,11 +108,19 @@ impl Debug for UserInputLeaf {
UserInputLeaf::Exists { field } => {
write!(formatter, "$exists(\"{field}\")")
}
UserInputLeaf::Regex { field, pattern } => {
if let Some(field) = field {
// TODO properly escape field (in case of \")
write!(formatter, "\"{field}\":")?;
}
// TODO properly escape pattern (in case of \")
write!(formatter, "/{pattern}/")
}
}
}
}
#[derive(Copy, Clone, Eq, PartialEq, Debug, Serialize)]
#[derive(Copy, Clone, Eq, PartialEq, Hash, Debug, Serialize)]
#[serde(rename_all = "snake_case")]
pub enum Delimiter {
SingleQuotes,
@@ -115,7 +128,7 @@ pub enum Delimiter {
None,
}
#[derive(PartialEq, Clone, Serialize)]
#[derive(PartialEq, Eq, Hash, Clone, Serialize)]
#[serde(rename_all = "snake_case")]
pub struct UserInputLiteral {
pub field_name: Option<String>,
@@ -154,7 +167,7 @@ impl fmt::Debug for UserInputLiteral {
}
}
#[derive(PartialEq, Debug, Clone, Serialize)]
#[derive(PartialEq, Eq, Hash, Debug, Clone, Serialize)]
#[serde(tag = "type", content = "value")]
#[serde(rename_all = "snake_case")]
pub enum UserInputBound {
@@ -191,11 +204,11 @@ impl UserInputBound {
}
}
#[derive(PartialEq, Clone, Serialize)]
#[derive(PartialEq, Eq, Hash, Clone, Serialize)]
#[serde(into = "UserInputAstSerde")]
pub enum UserInputAst {
Clause(Vec<(Option<Occur>, UserInputAst)>),
Boost(Box<UserInputAst>, f64),
Boost(Box<UserInputAst>, ordered_float::OrderedFloat<f64>),
Leaf(Box<UserInputLeaf>),
}
@@ -217,9 +230,10 @@ impl From<UserInputAst> for UserInputAstSerde {
fn from(ast: UserInputAst) -> Self {
match ast {
UserInputAst::Clause(clause) => UserInputAstSerde::Bool { clauses: clause },
UserInputAst::Boost(underlying, boost) => {
UserInputAstSerde::Boost { underlying, boost }
}
UserInputAst::Boost(underlying, boost) => UserInputAstSerde::Boost {
underlying,
boost: boost.into_inner(),
},
UserInputAst::Leaf(leaf) => UserInputAstSerde::Leaf(leaf),
}
}
@@ -378,7 +392,7 @@ mod tests {
#[test]
fn test_boost_serialization() {
let inner_ast = UserInputAst::Leaf(Box::new(UserInputLeaf::All));
let boost_ast = UserInputAst::Boost(Box::new(inner_ast), 2.5);
let boost_ast = UserInputAst::Boost(Box::new(inner_ast), 2.5.into());
let json = serde_json::to_string(&boost_ast).unwrap();
assert_eq!(
json,
@@ -405,7 +419,7 @@ mod tests {
}))),
),
])),
2.5,
2.5.into(),
);
let json = serde_json::to_string(&boost_ast).unwrap();
assert_eq!(

7
src/aggregation/README.md
View File

@@ -20,17 +20,16 @@ Contains all metric aggregations, like average aggregation. Metric aggregations
#### agg_req
agg_req contains the users aggregation request. Deserialization from json is compatible with elasticsearch aggregation requests.
#### agg_req_with_accessor
agg_req_with_accessor contains the users aggregation request enriched with fast field accessors etc, which are
#### agg_data
agg_data contains the users aggregation request enriched with fast field accessors etc, which are
used during collection.
#### segment_agg_result
segment_agg_result contains the aggregation result tree, which is used for collection of a segment.
The tree from agg_req_with_accessor is passed during collection.
agg_data is passed during collection.
#### intermediate_agg_result
intermediate_agg_result contains the aggregation tree for merging with other trees.
#### agg_result
agg_result contains the final aggregation tree.

104
src/aggregation/accessor_helpers.rs Normal file
View File

@@ -0,0 +1,104 @@
//! This will enhance the request tree with access to the fastfield and metadata.
use std::io;
use columnar::{Column, ColumnType};
use crate::aggregation::{f64_to_fastfield_u64, Key};
use crate::index::SegmentReader;
/// Get the missing value as internal u64 representation
///
/// For terms we use u64::MAX as sentinel value
/// For numerical data we convert the value into the representation
/// we would get from the fast field, when we open it as u64_lenient_for_type.
///
/// That way we can use it the same way as if it would come from the fastfield.
pub(crate) fn get_missing_val_as_u64_lenient(
column_type: ColumnType,
missing: &Key,
field_name: &str,
) -> crate::Result<Option<u64>> {
let missing_val = match missing {
Key::Str(_) if column_type == ColumnType::Str => Some(u64::MAX),
// Allow fallback to number on text fields
Key::F64(_) if column_type == ColumnType::Str => Some(u64::MAX),
Key::U64(_) if column_type == ColumnType::Str => Some(u64::MAX),
Key::I64(_) if column_type == ColumnType::Str => Some(u64::MAX),
Key::F64(val) if column_type.numerical_type().is_some() => {
f64_to_fastfield_u64(*val, &column_type)
}
// NOTE: We may loose precision of the passed missing value by casting i64 and u64 to f64.
Key::I64(val) if column_type.numerical_type().is_some() => {
f64_to_fastfield_u64(*val as f64, &column_type)
}
Key::U64(val) if column_type.numerical_type().is_some() => {
f64_to_fastfield_u64(*val as f64, &column_type)
}
_ => {
return Err(crate::TantivyError::InvalidArgument(format!(
"Missing value {missing:?} for field {field_name} is not supported for column \
type {column_type:?}"
)));
}
};
Ok(missing_val)
}
pub(crate) fn get_numeric_or_date_column_types() -> &'static [ColumnType] {
&[
ColumnType::F64,
ColumnType::U64,
ColumnType::I64,
ColumnType::DateTime,
]
}
/// Get fast field reader or empty as default.
pub(crate) fn get_ff_reader(
reader: &SegmentReader,
field_name: &str,
allowed_column_types: Option<&[ColumnType]>,
) -> crate::Result<(columnar::Column<u64>, ColumnType)> {
let ff_fields = reader.fast_fields();
let ff_field_with_type = ff_fields
.u64_lenient_for_type(allowed_column_types, field_name)?
.unwrap_or_else(|| {
(
Column::build_empty_column(reader.num_docs()),
ColumnType::U64,
)
});
Ok(ff_field_with_type)
}
pub(crate) fn get_dynamic_columns(
reader: &SegmentReader,
field_name: &str,
) -> crate::Result<Vec<columnar::DynamicColumn>> {
let ff_fields = reader.fast_fields().dynamic_column_handles(field_name)?;
let cols = ff_fields
.iter()
.map(|h| h.open())
.collect::<io::Result<_>>()?;
assert!(!ff_fields.is_empty(), "field {field_name} not found");
Ok(cols)
}
/// Get all fast field reader or empty as default.
///
/// Is guaranteed to return at least one column.
pub(crate) fn get_all_ff_reader_or_empty(
reader: &SegmentReader,
field_name: &str,
allowed_column_types: Option<&[ColumnType]>,
fallback_type: ColumnType,
) -> crate::Result<Vec<(columnar::Column<u64>, ColumnType)>> {
let ff_fields = reader.fast_fields();
let mut ff_field_with_type =
ff_fields.u64_lenient_for_type_all(allowed_column_types, field_name)?;
if ff_field_with_type.is_empty() {
ff_field_with_type.push((Column::build_empty_column(reader.num_docs()), fallback_type));
}
Ok(ff_field_with_type)
}

1083
src/aggregation/agg_data.rs Normal file
View File

File diff suppressed because it is too large Load Diff

2
src/aggregation/agg_limits.rs
View File

@@ -70,7 +70,7 @@ impl AggregationLimitsGuard {
/// *memory_limit*
/// memory_limit is defined in bytes.
/// Aggregation fails when the estimated memory consumption of the aggregation is higher than
/// memory_limit.
/// memory_limit.
/// memory_limit will default to `DEFAULT_MEMORY_LIMIT` (500MB)
///
/// *bucket_limit*

19
src/aggregation/agg_req.rs
View File

@@ -26,12 +26,14 @@
//! let _agg_req: Aggregations = serde_json::from_str(elasticsearch_compatible_json_req).unwrap();
//! ```
use std::collections::{HashMap, HashSet};
use std::collections::HashSet;
use rustc_hash::FxHashMap;
use serde::{Deserialize, Serialize};
use super::bucket::{
DateHistogramAggregationReq, HistogramAggregation, RangeAggregation, TermsAggregation,
DateHistogramAggregationReq, FilterAggregation, HistogramAggregation, RangeAggregation,
TermsAggregation,
};
use super::metric::{
AverageAggregation, CardinalityAggregationReq, CountAggregation, ExtendedStatsAggregation,
@@ -43,7 +45,7 @@ use super::metric::{
/// defined names. It is also used in buckets aggregations to define sub-aggregations.
///
/// The key is the user defined name of the aggregation.
pub type Aggregations = HashMap<String, Aggregation>;
pub type Aggregations = FxHashMap<String, Aggregation>;
/// Aggregation request.
///
@@ -129,6 +131,9 @@ pub enum AggregationVariants {
/// Put data into buckets of terms.
#[serde(rename = "terms")]
Terms(TermsAggregation),
/// Filter documents into a single bucket.
#[serde(rename = "filter")]
Filter(FilterAggregation),
// Metric aggregation types
/// Computes the average of the extracted values.
@@ -174,6 +179,7 @@ impl AggregationVariants {
AggregationVariants::Range(range) => vec![range.field.as_str()],
AggregationVariants::Histogram(histogram) => vec![histogram.field.as_str()],
AggregationVariants::DateHistogram(histogram) => vec![histogram.field.as_str()],
AggregationVariants::Filter(filter) => filter.get_fast_field_names(),
AggregationVariants::Average(avg) => vec![avg.field_name()],
AggregationVariants::Count(count) => vec![count.field_name()],
AggregationVariants::Max(max) => vec![max.field_name()],
@@ -208,13 +214,6 @@ impl AggregationVariants {
_ => None,
}
}
pub(crate) fn as_top_hits(&self) -> Option<&TopHitsAggregationReq> {
match &self {
AggregationVariants::TopHits(top_hits) => Some(top_hits),
_ => None,
}
}
pub(crate) fn as_percentile(&self) -> Option<&PercentilesAggregationReq> {
match &self {
AggregationVariants::Percentiles(percentile_req) => Some(percentile_req),

471
src/aggregation/agg_req_with_accessor.rs
View File

@@ -1,471 +0,0 @@
//! This will enhance the request tree with access to the fastfield and metadata.
use std::collections::HashMap;
use std::io;
use columnar::{Column, ColumnBlockAccessor, ColumnType, DynamicColumn, StrColumn};
use super::agg_req::{Aggregation, AggregationVariants, Aggregations};
use super::bucket::{
DateHistogramAggregationReq, HistogramAggregation, RangeAggregation, TermsAggregation,
};
use super::metric::{
AverageAggregation, CardinalityAggregationReq, CountAggregation, ExtendedStatsAggregation,
MaxAggregation, MinAggregation, StatsAggregation, SumAggregation,
};
use super::segment_agg_result::AggregationLimitsGuard;
use super::VecWithNames;
use crate::aggregation::{f64_to_fastfield_u64, Key};
use crate::index::SegmentReader;
use crate::SegmentOrdinal;
#[derive(Default)]
pub(crate) struct AggregationsWithAccessor {
pub aggs: VecWithNames<AggregationWithAccessor>,
}
impl AggregationsWithAccessor {
fn from_data(aggs: VecWithNames<AggregationWithAccessor>) -> Self {
Self { aggs }
}
pub fn is_empty(&self) -> bool {
self.aggs.is_empty()
}
}
pub struct AggregationWithAccessor {
pub(crate) segment_ordinal: SegmentOrdinal,
/// In general there can be buckets without fast field access, e.g. buckets that are created
/// based on search terms. That is not that case currently, but eventually this needs to be
/// Option or moved.
pub(crate) accessor: Column<u64>,
/// Load insert u64 for missing use case
pub(crate) missing_value_for_accessor: Option<u64>,
pub(crate) str_dict_column: Option<StrColumn>,
pub(crate) field_type: ColumnType,
pub(crate) sub_aggregation: AggregationsWithAccessor,
pub(crate) limits: AggregationLimitsGuard,
pub(crate) column_block_accessor: ColumnBlockAccessor<u64>,
/// Used for missing term aggregation, which checks all columns for existence.
/// And also for `top_hits` aggregation, which may sort on multiple fields.
/// By convention the missing aggregation is chosen, when this property is set
/// (instead bein set in `agg`).
/// If this needs to used by other aggregations, we need to refactor this.
// NOTE: we can make all other aggregations use this instead of the `accessor` and `field_type`
// (making them obsolete) But will it have a performance impact?
pub(crate) accessors: Vec<(Column<u64>, ColumnType)>,
/// Map field names to all associated column accessors.
/// This field is used for `docvalue_fields`, which is currently only supported for `top_hits`.
pub(crate) value_accessors: HashMap<String, Vec<DynamicColumn>>,
pub(crate) agg: Aggregation,
}
impl AggregationWithAccessor {
/// May return multiple accessors if the aggregation is e.g. on mixed field types.
fn try_from_agg(
agg: &Aggregation,
sub_aggregation: &Aggregations,
reader: &SegmentReader,
segment_ordinal: SegmentOrdinal,
limits: AggregationLimitsGuard,
) -> crate::Result<Vec<AggregationWithAccessor>> {
let mut agg = agg.clone();
let add_agg_with_accessor = |agg: &Aggregation,
accessor: Column<u64>,
column_type: ColumnType,
aggs: &mut Vec<AggregationWithAccessor>|
-> crate::Result<()> {
let res = AggregationWithAccessor {
segment_ordinal,
accessor,
accessors: Default::default(),
value_accessors: Default::default(),
field_type: column_type,
sub_aggregation: get_aggs_with_segment_accessor_and_validate(
sub_aggregation,
reader,
segment_ordinal,
&limits,
)?,
agg: agg.clone(),
limits: limits.clone(),
missing_value_for_accessor: None,
str_dict_column: None,
column_block_accessor: Default::default(),
};
aggs.push(res);
Ok(())
};
let add_agg_with_accessors = |agg: &Aggregation,
accessors: Vec<(Column<u64>, ColumnType)>,
aggs: &mut Vec<AggregationWithAccessor>,
value_accessors: HashMap<String, Vec<DynamicColumn>>|
-> crate::Result<()> {
let (accessor, field_type) = accessors.first().expect("at least one accessor");
let limits = limits.clone();
let res = AggregationWithAccessor {
segment_ordinal,
// TODO: We should do away with the `accessor` field altogether
accessor: accessor.clone(),
value_accessors,
field_type: *field_type,
accessors,
sub_aggregation: get_aggs_with_segment_accessor_and_validate(
sub_aggregation,
reader,
segment_ordinal,
&limits,
)?,
agg: agg.clone(),
limits,
missing_value_for_accessor: None,
str_dict_column: None,
column_block_accessor: Default::default(),
};
aggs.push(res);
Ok(())
};
let mut res: Vec<AggregationWithAccessor> = Vec::new();
use AggregationVariants::*;
match agg.agg {
Range(RangeAggregation {
field: ref field_name,
..
}) => {
let (accessor, column_type) =
get_ff_reader(reader, field_name, Some(get_numeric_or_date_column_types()))?;
add_agg_with_accessor(&agg, accessor, column_type, &mut res)?;
}
Histogram(HistogramAggregation {
field: ref field_name,
..
}) => {
let (accessor, column_type) =
get_ff_reader(reader, field_name, Some(get_numeric_or_date_column_types()))?;
add_agg_with_accessor(&agg, accessor, column_type, &mut res)?;
}
DateHistogram(DateHistogramAggregationReq {
field: ref field_name,
..
}) => {
let (accessor, column_type) =
// Only DateTime is supported for DateHistogram
get_ff_reader(reader, field_name, Some(&[ColumnType::DateTime]))?;
add_agg_with_accessor(&agg, accessor, column_type, &mut res)?;
}
Terms(TermsAggregation {
field: ref field_name,
ref missing,
..
})
| Cardinality(CardinalityAggregationReq {
field: ref field_name,
ref missing,
..
}) => {
let str_dict_column = reader.fast_fields().str(field_name)?;
let allowed_column_types = [
ColumnType::I64,
ColumnType::U64,
ColumnType::F64,
ColumnType::Str,
ColumnType::DateTime,
ColumnType::Bool,
ColumnType::IpAddr,
// ColumnType::Bytes Unsupported
];
// In case the column is empty we want the shim column to match the missing type
let fallback_type = missing
.as_ref()
.map(|missing| match missing {
Key::Str(_) => ColumnType::Str,
Key::F64(_) => ColumnType::F64,
Key::I64(_) => ColumnType::I64,
Key::U64(_) => ColumnType::U64,
})
.unwrap_or(ColumnType::U64);
let column_and_types = get_all_ff_reader_or_empty(
reader,
field_name,
Some(&allowed_column_types),
fallback_type,
)?;
let missing_and_more_than_one_col = column_and_types.len() > 1 && missing.is_some();
let text_on_non_text_col = column_and_types.len() == 1
&& column_and_types[0].1.numerical_type().is_some()
&& missing
.as_ref()
.map(|m| matches!(m, Key::Str(_)))
.unwrap_or(false);
// Actually we could convert the text to a number and have the fast path, if it is
// provided in Rfc3339 format. But this use case is probably common
// enough to justify the effort.
let text_on_date_col = column_and_types.len() == 1
&& column_and_types[0].1 == ColumnType::DateTime
&& missing
.as_ref()
.map(|m| matches!(m, Key::Str(_)))
.unwrap_or(false);
let use_special_missing_agg =
missing_and_more_than_one_col || text_on_non_text_col || text_on_date_col;
if use_special_missing_agg {
let column_and_types =
get_all_ff_reader_or_empty(reader, field_name, None, fallback_type)?;
let accessors = column_and_types
.iter()
.map(|c_t| (c_t.0.clone(), c_t.1))
.collect();
add_agg_with_accessors(&agg, accessors, &mut res, Default::default())?;
}
for (accessor, column_type) in column_and_types {
let missing_value_term_agg = if use_special_missing_agg {
None
} else {
missing.clone()
};
let missing_value_for_accessor =
if let Some(missing) = missing_value_term_agg.as_ref() {
get_missing_val_as_u64_lenient(
column_type,
missing,
agg.agg.get_fast_field_names()[0],
)?
} else {
None
};
let limits = limits.clone();
let agg = AggregationWithAccessor {
segment_ordinal,
missing_value_for_accessor,
accessor,
accessors: Default::default(),
value_accessors: Default::default(),
field_type: column_type,
sub_aggregation: get_aggs_with_segment_accessor_and_validate(
sub_aggregation,
reader,
segment_ordinal,
&limits,
)?,
agg: agg.clone(),
str_dict_column: str_dict_column.clone(),
limits,
column_block_accessor: Default::default(),
};
res.push(agg);
}
}
Average(AverageAggregation {
field: ref field_name,
..
})
| Max(MaxAggregation {
field: ref field_name,
..
})
| Min(MinAggregation {
field: ref field_name,
..
})
| Stats(StatsAggregation {
field: ref field_name,
..
})
| ExtendedStats(ExtendedStatsAggregation {
field: ref field_name,
..
})
| Sum(SumAggregation {
field: ref field_name,
..
}) => {
let (accessor, column_type) =
get_ff_reader(reader, field_name, Some(get_numeric_or_date_column_types()))?;
add_agg_with_accessor(&agg, accessor, column_type, &mut res)?;
}
Count(CountAggregation {
field: ref field_name,
..
}) => {
let allowed_column_types = [
ColumnType::I64,
ColumnType::U64,
ColumnType::F64,
ColumnType::Str,
ColumnType::DateTime,
ColumnType::Bool,
ColumnType::IpAddr,
// ColumnType::Bytes Unsupported
];
let (accessor, column_type) =
get_ff_reader(reader, field_name, Some(&allowed_column_types))?;
add_agg_with_accessor(&agg, accessor, column_type, &mut res)?;
}
Percentiles(ref percentiles) => {
let (accessor, column_type) = get_ff_reader(
reader,
percentiles.field_name(),
Some(get_numeric_or_date_column_types()),
)?;
add_agg_with_accessor(&agg, accessor, column_type, &mut res)?;
}
TopHits(ref mut top_hits) => {
top_hits.validate_and_resolve_field_names(reader.fast_fields().columnar())?;
let accessors: Vec<(Column<u64>, ColumnType)> = top_hits
.field_names()
.iter()
.map(|field| {
get_ff_reader(reader, field, Some(get_numeric_or_date_column_types()))
})
.collect::<crate::Result<_>>()?;
let value_accessors = top_hits
.value_field_names()
.iter()
.map(|field_name| {
Ok((
field_name.to_string(),
get_dynamic_columns(reader, field_name)?,
))
})
.collect::<crate::Result<_>>()?;
add_agg_with_accessors(&agg, accessors, &mut res, value_accessors)?;
}
};
Ok(res)
}
}
/// Get the missing value as internal u64 representation
///
/// For terms we use u64::MAX as sentinel value
/// For numerical data we convert the value into the representation
/// we would get from the fast field, when we open it as u64_lenient_for_type.
///
/// That way we can use it the same way as if it would come from the fastfield.
fn get_missing_val_as_u64_lenient(
column_type: ColumnType,
missing: &Key,
field_name: &str,
) -> crate::Result<Option<u64>> {
let missing_val = match missing {
Key::Str(_) if column_type == ColumnType::Str => Some(u64::MAX),
// Allow fallback to number on text fields
Key::F64(_) if column_type == ColumnType::Str => Some(u64::MAX),
Key::U64(_) if column_type == ColumnType::Str => Some(u64::MAX),
Key::I64(_) if column_type == ColumnType::Str => Some(u64::MAX),
Key::F64(val) if column_type.numerical_type().is_some() => {
f64_to_fastfield_u64(*val, &column_type)
}
// NOTE: We may loose precision of the passed missing value by casting i64 and u64 to f64.
Key::I64(val) if column_type.numerical_type().is_some() => {
f64_to_fastfield_u64(*val as f64, &column_type)
}
Key::U64(val) if column_type.numerical_type().is_some() => {
f64_to_fastfield_u64(*val as f64, &column_type)
}
_ => {
return Err(crate::TantivyError::InvalidArgument(format!(
"Missing value {missing:?} for field {field_name} is not supported for column \
type {column_type:?}"
)));
}
};
Ok(missing_val)
}
fn get_numeric_or_date_column_types() -> &'static [ColumnType] {
&[
ColumnType::F64,
ColumnType::U64,
ColumnType::I64,
ColumnType::DateTime,
]
}
pub(crate) fn get_aggs_with_segment_accessor_and_validate(
aggs: &Aggregations,
reader: &SegmentReader,
segment_ordinal: SegmentOrdinal,
limits: &AggregationLimitsGuard,
) -> crate::Result<AggregationsWithAccessor> {
let mut aggss = Vec::new();
for (key, agg) in aggs.iter() {
let aggs = AggregationWithAccessor::try_from_agg(
agg,
agg.sub_aggregation(),
reader,
segment_ordinal,
limits.clone(),
)?;
for agg in aggs {
aggss.push((key.to_string(), agg));
}
}
Ok(AggregationsWithAccessor::from_data(
VecWithNames::from_entries(aggss),
))
}
/// Get fast field reader or empty as default.
fn get_ff_reader(
reader: &SegmentReader,
field_name: &str,
allowed_column_types: Option<&[ColumnType]>,
) -> crate::Result<(columnar::Column<u64>, ColumnType)> {
let ff_fields = reader.fast_fields();
let ff_field_with_type = ff_fields
.u64_lenient_for_type(allowed_column_types, field_name)?
.unwrap_or_else(|| {
(
Column::build_empty_column(reader.num_docs()),
ColumnType::U64,
)
});
Ok(ff_field_with_type)
}
fn get_dynamic_columns(
reader: &SegmentReader,
field_name: &str,
) -> crate::Result<Vec<columnar::DynamicColumn>> {
let ff_fields = reader.fast_fields().dynamic_column_handles(field_name)?;
let cols = ff_fields
.iter()
.map(|h| h.open())
.collect::<io::Result<_>>()?;
assert!(!ff_fields.is_empty(), "field {field_name} not found");
Ok(cols)
}
/// Get all fast field reader or empty as default.
///
/// Is guaranteed to return at least one column.
fn get_all_ff_reader_or_empty(
reader: &SegmentReader,
field_name: &str,
allowed_column_types: Option<&[ColumnType]>,
fallback_type: ColumnType,
) -> crate::Result<Vec<(columnar::Column<u64>, ColumnType)>> {
let ff_fields = reader.fast_fields();
let mut ff_field_with_type =
ff_fields.u64_lenient_for_type_all(allowed_column_types, field_name)?;
if ff_field_with_type.is_empty() {
ff_field_with_type.push((Column::build_empty_column(reader.num_docs()), fallback_type));
}
Ok(ff_field_with_type)
}

29
src/aggregation/agg_result.rs
View File

@@ -156,6 +156,8 @@ pub enum BucketResult {
/// The upper bound error for the doc count of each term.
doc_count_error_upper_bound: Option<u64>,
},
/// This is the filter result - a single bucket with sub-aggregations
Filter(FilterBucketResult),
}
impl BucketResult {
@@ -172,6 +174,11 @@ impl BucketResult {
sum_other_doc_count: _,
doc_count_error_upper_bound: _,
} => buckets.iter().map(|bucket| bucket.get_bucket_count()).sum(),
BucketResult::Filter(filter_result) => {
// Filter doesn't add to bucket count - it's not a user-facing bucket
// Only count sub-aggregation buckets
filter_result.sub_aggregations.get_bucket_count()
}
}
}
}
@@ -308,3 +315,25 @@ impl RangeBucketEntry {
1 + self.sub_aggregation.get_bucket_count()
}
}
/// This is the filter bucket result, which contains the document count and sub-aggregations.
///
/// # JSON Format
/// ```json
/// {
/// "electronics_only": {
/// "doc_count": 2,
/// "avg_price": {
/// "value": 150.0
/// }
/// }
/// }
/// ```
#[derive(Clone, Debug, PartialEq, Serialize, Deserialize)]
pub struct FilterBucketResult {
/// Number of documents in the filter bucket
pub doc_count: u64,
/// Sub-aggregation results
#[serde(flatten)]
pub sub_aggregations: AggregationResults,
}

7
src/aggregation/agg_tests.rs
View File

@@ -5,7 +5,6 @@ use crate::aggregation::agg_result::AggregationResults;
use crate::aggregation::buf_collector::DOC_BLOCK_SIZE;
use crate::aggregation::collector::AggregationCollector;
use crate::aggregation::intermediate_agg_result::IntermediateAggregationResults;
use crate::aggregation::segment_agg_result::AggregationLimitsGuard;
use crate::aggregation::tests::{get_test_index_2_segments, get_test_index_from_values_and_terms};
use crate::aggregation::DistributedAggregationCollector;
use crate::query::{AllQuery, TermQuery};
@@ -128,10 +127,8 @@ fn test_aggregation_flushing(
.unwrap();
let agg_res: AggregationResults = if use_distributed_collector {
let collector = DistributedAggregationCollector::from_aggs(
agg_req.clone(),
AggregationLimitsGuard::default(),
);
let collector =
DistributedAggregationCollector::from_aggs(agg_req.clone(), Default::default());
let searcher = reader.searcher();
let intermediate_agg_result = searcher.search(&AllQuery, &collector).unwrap();

1755
src/aggregation/bucket/filter.rs Normal file
View File

File diff suppressed because it is too large Load Diff

157
src/aggregation/bucket/histogram/histogram.rs
View File

@@ -1,25 +1,54 @@
use std::cmp::Ordering;
use columnar::{Column, ColumnBlockAccessor, ColumnType};
use rustc_hash::FxHashMap;
use serde::{Deserialize, Serialize};
use tantivy_bitpacker::minmax;
use crate::aggregation::agg_data::{
build_segment_agg_collectors, AggRefNode, AggregationsSegmentCtx,
};
use crate::aggregation::agg_limits::MemoryConsumption;
use crate::aggregation::agg_req::Aggregations;
use crate::aggregation::agg_req_with_accessor::{
AggregationWithAccessor, AggregationsWithAccessor,
};
use crate::aggregation::agg_result::BucketEntry;
use crate::aggregation::intermediate_agg_result::{
IntermediateAggregationResult, IntermediateAggregationResults, IntermediateBucketResult,
IntermediateHistogramBucketEntry,
};
use crate::aggregation::segment_agg_result::{
build_segment_agg_collector, SegmentAggregationCollector,
};
use crate::aggregation::segment_agg_result::SegmentAggregationCollector;
use crate::aggregation::*;
use crate::TantivyError;
/// Contains all information required by the SegmentHistogramCollector to perform the
/// histogram or date_histogram aggregation on a segment.
pub struct HistogramAggReqData {
/// The column accessor to access the fast field values.
pub accessor: Column<u64>,
/// The field type of the fast field.
pub field_type: ColumnType,
/// The column block accessor to access the fast field values.
pub column_block_accessor: ColumnBlockAccessor<u64>,
/// The name of the aggregation.
pub name: String,
/// The sub aggregation blueprint, used to create sub aggregations for each bucket.
/// Will be filled during initialization of the collector.
pub sub_aggregation_blueprint: Option<Box<dyn SegmentAggregationCollector>>,
/// The histogram aggregation request.
pub req: HistogramAggregation,
/// True if this is a date_histogram aggregation.
pub is_date_histogram: bool,
/// The bounds to limit the buckets to.
pub bounds: HistogramBounds,
/// The offset used to calculate the bucket position.
pub offset: f64,
}
impl HistogramAggReqData {
/// Estimate the memory consumption of this struct in bytes.
pub fn get_memory_consumption(&self) -> usize {
std::mem::size_of::<Self>()
}
}
/// Histogram is a bucket aggregation, where buckets are created dynamically for given `interval`.
/// Each document value is rounded down to its bucket.
///
@@ -234,12 +263,12 @@ impl SegmentHistogramBucketEntry {
pub(crate) fn into_intermediate_bucket_entry(
self,
sub_aggregation: Option<Box<dyn SegmentAggregationCollector>>,
agg_with_accessor: &AggregationsWithAccessor,
agg_data: &AggregationsSegmentCtx,
) -> crate::Result<IntermediateHistogramBucketEntry> {
let mut sub_aggregation_res = IntermediateAggregationResults::default();
if let Some(sub_aggregation) = sub_aggregation {
sub_aggregation
.add_intermediate_aggregation_result(agg_with_accessor, &mut sub_aggregation_res)?;
.add_intermediate_aggregation_result(agg_data, &mut sub_aggregation_res)?;
}
Ok(IntermediateHistogramBucketEntry {
key: self.key,
@@ -256,24 +285,20 @@ pub struct SegmentHistogramCollector {
/// The buckets containing the aggregation data.
buckets: FxHashMap<i64, SegmentHistogramBucketEntry>,
sub_aggregations: FxHashMap<i64, Box<dyn SegmentAggregationCollector>>,
sub_aggregation_blueprint: Option<Box<dyn SegmentAggregationCollector>>,
column_type: ColumnType,
interval: f64,
offset: f64,
bounds: HistogramBounds,
accessor_idx: usize,
}
impl SegmentAggregationCollector for SegmentHistogramCollector {
fn add_intermediate_aggregation_result(
self: Box<Self>,
agg_with_accessor: &AggregationsWithAccessor,
agg_data: &AggregationsSegmentCtx,
results: &mut IntermediateAggregationResults,
) -> crate::Result<()> {
let name = agg_with_accessor.aggs.keys[self.accessor_idx].to_string();
let agg_with_accessor = &agg_with_accessor.aggs.values[self.accessor_idx];
let bucket = self.into_intermediate_bucket_result(agg_with_accessor)?;
let name = agg_data
.get_histogram_req_data(self.accessor_idx)
.name
.clone();
let bucket = self.into_intermediate_bucket_result(agg_data)?;
results.push(name, IntermediateAggregationResult::Bucket(bucket))?;
Ok(())
@@ -283,56 +308,52 @@ impl SegmentAggregationCollector for SegmentHistogramCollector {
fn collect(
&mut self,
doc: crate::DocId,
agg_with_accessor: &mut AggregationsWithAccessor,
agg_data: &mut AggregationsSegmentCtx,
) -> crate::Result<()> {
self.collect_block(&[doc], agg_with_accessor)
self.collect_block(&[doc], agg_data)
}
#[inline]
fn collect_block(
&mut self,
docs: &[crate::DocId],
agg_with_accessor: &mut AggregationsWithAccessor,
agg_data: &mut AggregationsSegmentCtx,
) -> crate::Result<()> {
let bucket_agg_accessor = &mut agg_with_accessor.aggs.values[self.accessor_idx];
let mut req = agg_data.take_histogram_req_data(self.accessor_idx);
let mem_pre = self.get_memory_consumption();
let bounds = self.bounds;
let interval = self.interval;
let offset = self.offset;
let get_bucket_pos = |val| (get_bucket_pos_f64(val, interval, offset) as i64);
let bounds = req.bounds;
let interval = req.req.interval;
let offset = req.offset;
let get_bucket_pos = |val| get_bucket_pos_f64(val, interval, offset) as i64;
bucket_agg_accessor
req.column_block_accessor.fetch_block(docs, &req.accessor);
for (doc, val) in req
.column_block_accessor
.fetch_block(docs, &bucket_agg_accessor.accessor);
for (doc, val) in bucket_agg_accessor
.column_block_accessor
.iter_docid_vals(docs, &bucket_agg_accessor.accessor)
.iter_docid_vals(docs, &req.accessor)
{
let val = self.f64_from_fastfield_u64(val);
let val = f64_from_fastfield_u64(val, &req.field_type);
let bucket_pos = get_bucket_pos(val);
if bounds.contains(val) {
let bucket = self.buckets.entry(bucket_pos).or_insert_with(|| {
let key = get_bucket_key_from_pos(bucket_pos as f64, interval, offset);
SegmentHistogramBucketEntry { key, doc_count: 0 }
});
bucket.doc_count += 1;
if let Some(sub_aggregation_blueprint) = self.sub_aggregation_blueprint.as_mut() {
if let Some(sub_aggregation_blueprint) = req.sub_aggregation_blueprint.as_ref() {
self.sub_aggregations
.entry(bucket_pos)
.or_insert_with(|| sub_aggregation_blueprint.clone())
.collect(doc, &mut bucket_agg_accessor.sub_aggregation)?;
.collect(doc, agg_data)?;
}
}
}
agg_data.put_back_histogram_req_data(self.accessor_idx, req);
let mem_delta = self.get_memory_consumption() - mem_pre;
if mem_delta > 0 {
bucket_agg_accessor
agg_data
.context
.limits
.add_memory_consumed(mem_delta as u64)?;
}
@@ -340,12 +361,9 @@ impl SegmentAggregationCollector for SegmentHistogramCollector {
Ok(())
}
fn flush(&mut self, agg_with_accessor: &mut AggregationsWithAccessor) -> crate::Result<()> {
let sub_aggregation_accessor =
&mut agg_with_accessor.aggs.values[self.accessor_idx].sub_aggregation;
fn flush(&mut self, agg_data: &mut AggregationsSegmentCtx) -> crate::Result<()> {
for sub_aggregation in self.sub_aggregations.values_mut() {
sub_aggregation.flush(sub_aggregation_accessor)?;
sub_aggregation.flush(agg_data)?;
}
Ok(())
@@ -362,65 +380,58 @@ impl SegmentHistogramCollector {
/// Converts the collector result into a intermediate bucket result.
pub fn into_intermediate_bucket_result(
self,
agg_with_accessor: &AggregationWithAccessor,
agg_data: &AggregationsSegmentCtx,
) -> crate::Result<IntermediateBucketResult> {
let mut buckets = Vec::with_capacity(self.buckets.len());
for (bucket_pos, bucket) in self.buckets {
let bucket_res = bucket.into_intermediate_bucket_entry(
self.sub_aggregations.get(&bucket_pos).cloned(),
&agg_with_accessor.sub_aggregation,
agg_data,
);
buckets.push(bucket_res?);
}
buckets.sort_unstable_by(|b1, b2| b1.key.total_cmp(&b2.key));
let is_date_agg = agg_data
.get_histogram_req_data(self.accessor_idx)
.field_type
== ColumnType::DateTime;
Ok(IntermediateBucketResult::Histogram {
buckets,
is_date_agg: self.column_type == ColumnType::DateTime,
is_date_agg,
})
}
pub(crate) fn from_req_and_validate(
mut req: HistogramAggregation,
sub_aggregation: &mut AggregationsWithAccessor,
field_type: ColumnType,
accessor_idx: usize,
agg_data: &mut AggregationsSegmentCtx,
node: &AggRefNode,
) -> crate::Result<Self> {
req.validate()?;
if field_type == ColumnType::DateTime {
req.normalize_date_time();
}
let sub_aggregation_blueprint = if sub_aggregation.is_empty() {
None
let blueprint = if !node.children.is_empty() {
Some(build_segment_agg_collectors(agg_data, &node.children)?)
} else {
let sub_aggregation = build_segment_agg_collector(sub_aggregation)?;
Some(sub_aggregation)
None
};
let bounds = req.hard_bounds.unwrap_or(HistogramBounds {
let req_data = agg_data.get_histogram_req_data_mut(node.idx_in_req_data);
req_data.req.validate()?;
if req_data.field_type == ColumnType::DateTime && !req_data.is_date_histogram {
req_data.req.normalize_date_time();
}
req_data.bounds = req_data.req.hard_bounds.unwrap_or(HistogramBounds {
min: f64::MIN,
max: f64::MAX,
});
req_data.offset = req_data.req.offset.unwrap_or(0.0);
req_data.sub_aggregation_blueprint = blueprint;
Ok(Self {
buckets: Default::default(),
column_type: field_type,
interval: req.interval,
offset: req.offset.unwrap_or(0.0),
bounds,
sub_aggregations: Default::default(),
sub_aggregation_blueprint,
accessor_idx,
accessor_idx: node.idx_in_req_data,
})
}
#[inline]
fn f64_from_fastfield_u64(&self, val: u64) -> f64 {
f64_from_fastfield_u64(val, &self.column_type)
}
}
#[inline]

2
src/aggregation/bucket/mod.rs
View File

@@ -22,6 +22,7 @@
//! - [Range](RangeAggregation)
//! - [Terms](TermsAggregation)
mod filter;
mod histogram;
mod range;
mod term_agg;
@@ -30,6 +31,7 @@ mod term_missing_agg;
use std::collections::HashMap;
use std::fmt;
pub use filter::*;
pub use histogram::*;
pub use range::*;
use serde::{de, Deserialize, Deserializer, Serialize, Serializer};

153
src/aggregation/bucket/range.rs
View File

@@ -1,20 +1,43 @@
use std::fmt::Debug;
use std::ops::Range;
use columnar::{Column, ColumnBlockAccessor, ColumnType};
use rustc_hash::FxHashMap;
use serde::{Deserialize, Serialize};
use crate::aggregation::agg_req_with_accessor::AggregationsWithAccessor;
use crate::aggregation::agg_data::{
build_segment_agg_collectors, AggRefNode, AggregationsSegmentCtx,
};
use crate::aggregation::intermediate_agg_result::{
IntermediateAggregationResult, IntermediateAggregationResults, IntermediateBucketResult,
IntermediateRangeBucketEntry, IntermediateRangeBucketResult,
};
use crate::aggregation::segment_agg_result::{
build_segment_agg_collector, SegmentAggregationCollector,
};
use crate::aggregation::segment_agg_result::SegmentAggregationCollector;
use crate::aggregation::*;
use crate::TantivyError;
/// Contains all information required by the SegmentRangeCollector to perform the
/// range aggregation on a segment.
pub struct RangeAggReqData {
/// The column accessor to access the fast field values.
pub accessor: Column<u64>,
/// The type of the fast field.
pub field_type: ColumnType,
/// The column block accessor to access the fast field values.
pub column_block_accessor: ColumnBlockAccessor<u64>,
/// The range aggregation request.
pub req: RangeAggregation,
/// The name of the aggregation.
pub name: String,
}
impl RangeAggReqData {
/// Estimate the memory consumption of this struct in bytes.
pub fn get_memory_consumption(&self) -> usize {
std::mem::size_of::<Self>()
}
}
/// Provide user-defined buckets to aggregate on.
///
/// Two special buckets will automatically be created to cover the whole range of values.
@@ -161,12 +184,12 @@ impl Debug for SegmentRangeBucketEntry {
impl SegmentRangeBucketEntry {
pub(crate) fn into_intermediate_bucket_entry(
self,
agg_with_accessor: &AggregationsWithAccessor,
agg_data: &AggregationsSegmentCtx,
) -> crate::Result<IntermediateRangeBucketEntry> {
let mut sub_aggregation_res = IntermediateAggregationResults::default();
if let Some(sub_aggregation) = self.sub_aggregation {
sub_aggregation
.add_intermediate_aggregation_result(agg_with_accessor, &mut sub_aggregation_res)?
.add_intermediate_aggregation_result(agg_data, &mut sub_aggregation_res)?
} else {
Default::default()
};
@@ -184,12 +207,14 @@ impl SegmentRangeBucketEntry {
impl SegmentAggregationCollector for SegmentRangeCollector {
fn add_intermediate_aggregation_result(
self: Box<Self>,
agg_with_accessor: &AggregationsWithAccessor,
agg_data: &AggregationsSegmentCtx,
results: &mut IntermediateAggregationResults,
) -> crate::Result<()> {
let field_type = self.column_type;
let name = agg_with_accessor.aggs.keys[self.accessor_idx].to_string();
let sub_agg = &agg_with_accessor.aggs.values[self.accessor_idx].sub_aggregation;
let name = agg_data
.get_range_req_data(self.accessor_idx)
.name
.to_string();
let buckets: FxHashMap<SerializedKey, IntermediateRangeBucketEntry> = self
.buckets
@@ -199,7 +224,7 @@ impl SegmentAggregationCollector for SegmentRangeCollector {
range_to_string(&range_bucket.range, &field_type)?,
range_bucket
.bucket
.into_intermediate_bucket_entry(sub_agg)?,
.into_intermediate_bucket_entry(agg_data)?,
))
})
.collect::<crate::Result<_>>()?;
@@ -218,66 +243,70 @@ impl SegmentAggregationCollector for SegmentRangeCollector {
fn collect(
&mut self,
doc: crate::DocId,
agg_with_accessor: &mut AggregationsWithAccessor,
agg_data: &mut AggregationsSegmentCtx,
) -> crate::Result<()> {
self.collect_block(&[doc], agg_with_accessor)
self.collect_block(&[doc], agg_data)
}
#[inline]
fn collect_block(
&mut self,
docs: &[crate::DocId],
agg_with_accessor: &mut AggregationsWithAccessor,
agg_data: &mut AggregationsSegmentCtx,
) -> crate::Result<()> {
let bucket_agg_accessor = &mut agg_with_accessor.aggs.values[self.accessor_idx];
// Take request data to avoid borrow conflicts during sub-aggregation
let mut req = agg_data.take_range_req_data(self.accessor_idx);
bucket_agg_accessor
.column_block_accessor
.fetch_block(docs, &bucket_agg_accessor.accessor);
req.column_block_accessor.fetch_block(docs, &req.accessor);
for (doc, val) in bucket_agg_accessor
for (doc, val) in req
.column_block_accessor
.iter_docid_vals(docs, &bucket_agg_accessor.accessor)
.iter_docid_vals(docs, &req.accessor)
{
let bucket_pos = self.get_bucket_pos(val);
let bucket = &mut self.buckets[bucket_pos];
bucket.bucket.doc_count += 1;
if let Some(sub_aggregation) = &mut bucket.bucket.sub_aggregation {
sub_aggregation.collect(doc, &mut bucket_agg_accessor.sub_aggregation)?;
if let Some(sub_agg) = bucket.bucket.sub_aggregation.as_mut() {
sub_agg.collect(doc, agg_data)?;
}
}
agg_data.put_back_range_req_data(self.accessor_idx, req);
Ok(())
}
fn flush(&mut self, agg_with_accessor: &mut AggregationsWithAccessor) -> crate::Result<()> {
let sub_aggregation_accessor =
&mut agg_with_accessor.aggs.values[self.accessor_idx].sub_aggregation;
fn flush(&mut self, agg_data: &mut AggregationsSegmentCtx) -> crate::Result<()> {
for bucket in self.buckets.iter_mut() {
if let Some(sub_agg) = bucket.bucket.sub_aggregation.as_mut() {
sub_agg.flush(sub_aggregation_accessor)?;
sub_agg.flush(agg_data)?;
}
}
Ok(())
}
}
impl SegmentRangeCollector {
pub(crate) fn from_req_and_validate(
req: &RangeAggregation,
sub_aggregation: &mut AggregationsWithAccessor,
limits: &mut AggregationLimitsGuard,
field_type: ColumnType,
accessor_idx: usize,
req_data: &mut AggregationsSegmentCtx,
node: &AggRefNode,
) -> crate::Result<Self> {
let accessor_idx = node.idx_in_req_data;
let (field_type, ranges) = {
let req_view = req_data.get_range_req_data(node.idx_in_req_data);
(req_view.field_type, req_view.req.ranges.clone())
};
// The range input on the request is f64.
// We need to convert to u64 ranges, because we read the values as u64.
// The mapping from the conversion is monotonic so ordering is preserved.
let buckets: Vec<_> = extend_validate_ranges(&req.ranges, &field_type)?
let sub_agg_prototype = if !node.children.is_empty() {
Some(build_segment_agg_collectors(req_data, &node.children)?)
} else {
None
};
let buckets: Vec<_> = extend_validate_ranges(&ranges, &field_type)?
.iter()
.map(|range| {
let key = range
@@ -295,11 +324,7 @@ impl SegmentRangeCollector {
} else {
Some(f64_from_fastfield_u64(range.range.start, &field_type))
};
let sub_aggregation = if sub_aggregation.is_empty() {
None
} else {
Some(build_segment_agg_collector(sub_aggregation)?)
};
let sub_aggregation = sub_agg_prototype.clone();
Ok(SegmentRangeAndBucketEntry {
range: range.range.clone(),
@@ -314,7 +339,7 @@ impl SegmentRangeCollector {
})
.collect::<crate::Result<_>>()?;
limits.add_memory_consumed(
req_data.context.limits.add_memory_consumed(
buckets.len() as u64 * std::mem::size_of::<SegmentRangeAndBucketEntry>() as u64,
)?;
@@ -467,15 +492,45 @@ mod tests {
ranges,
..Default::default()
};
// Build buckets directly as in from_req_and_validate without AggregationsData
let buckets: Vec<_> = extend_validate_ranges(&req.ranges, &field_type)
.expect("unexpected error in extend_validate_ranges")
.iter()
.map(|range| {
let key = range
.key
.clone()
.map(|key| Ok(Key::Str(key)))
.unwrap_or_else(|| range_to_key(&range.range, &field_type))
.expect("unexpected error in range_to_key");
let to = if range.range.end == u64::MAX {
None
} else {
Some(f64_from_fastfield_u64(range.range.end, &field_type))
};
let from = if range.range.start == u64::MIN {
None
} else {
Some(f64_from_fastfield_u64(range.range.start, &field_type))
};
SegmentRangeAndBucketEntry {
range: range.range.clone(),
bucket: SegmentRangeBucketEntry {
doc_count: 0,
sub_aggregation: None,
key,
from,
to,
},
}
})
.collect();
SegmentRangeCollector::from_req_and_validate(
&req,
&mut Default::default(),
&mut AggregationLimitsGuard::default(),
field_type,
0,
)
.expect("unexpected error")
SegmentRangeCollector {
buckets,
column_type: field_type,
accessor_idx: 0,
}
}
#[test]

196
src/aggregation/bucket/term_agg/default_impl.rs Normal file
View File

@@ -0,0 +1,196 @@
use std::fmt::Debug;
use columnar::ColumnType;
use rustc_hash::FxHashMap;
use super::OrderTarget;
use crate::aggregation::agg_data::{
build_segment_agg_collectors, AggRefNode, AggregationsSegmentCtx,
};
use crate::aggregation::agg_limits::MemoryConsumption;
use crate::aggregation::bucket::get_agg_name_and_property;
use crate::aggregation::intermediate_agg_result::{
IntermediateAggregationResult, IntermediateAggregationResults,
};
use crate::aggregation::segment_agg_result::SegmentAggregationCollector;
use crate::TantivyError;
#[derive(Clone, Debug, Default)]
/// Container to store term_ids/or u64 values and their buckets.
struct TermBuckets {
pub(crate) entries: FxHashMap<u64, u32>,
pub(crate) sub_aggs: FxHashMap<u64, Box<dyn SegmentAggregationCollector>>,
}
impl TermBuckets {
fn get_memory_consumption(&self) -> usize {
let sub_aggs_mem = self.sub_aggs.memory_consumption();
let buckets_mem = self.entries.memory_consumption();
sub_aggs_mem + buckets_mem
}
fn force_flush(&mut self, agg_data: &mut AggregationsSegmentCtx) -> crate::Result<()> {
for sub_aggregations in &mut self.sub_aggs.values_mut() {
sub_aggregations.as_mut().flush(agg_data)?;
}
Ok(())
}
}
/// The collector puts values from the fast field into the correct buckets and does a conversion to
/// the correct datatype.
#[derive(Clone, Debug)]
pub struct SegmentTermCollector {
/// The buckets containing the aggregation data.
term_buckets: TermBuckets,
accessor_idx: usize,
}
impl SegmentAggregationCollector for SegmentTermCollector {
fn add_intermediate_aggregation_result(
self: Box<Self>,
agg_data: &AggregationsSegmentCtx,
results: &mut IntermediateAggregationResults,
) -> crate::Result<()> {
let name = agg_data.get_term_req_data(self.accessor_idx).name.clone();
let entries: Vec<(u64, u32)> = self.term_buckets.entries.into_iter().collect();
let bucket = super::into_intermediate_bucket_result(
self.accessor_idx,
entries,
self.term_buckets.sub_aggs,
agg_data,
)?;
results.push(name, IntermediateAggregationResult::Bucket(bucket))?;
Ok(())
}
#[inline]
fn collect(
&mut self,
doc: crate::DocId,
agg_data: &mut AggregationsSegmentCtx,
) -> crate::Result<()> {
self.collect_block(&[doc], agg_data)
}
#[inline]
fn collect_block(
&mut self,
docs: &[crate::DocId],
agg_data: &mut AggregationsSegmentCtx,
) -> crate::Result<()> {
let mut req_data = agg_data.take_term_req_data(self.accessor_idx);
let mem_pre = self.get_memory_consumption();
if let Some(missing) = req_data.missing_value_for_accessor {
req_data.column_block_accessor.fetch_block_with_missing(
docs,
&req_data.accessor,
missing,
);
} else {
req_data
.column_block_accessor
.fetch_block(docs, &req_data.accessor);
}
for term_id in req_data.column_block_accessor.iter_vals() {
if let Some(allowed_bs) = req_data.allowed_term_ids.as_ref() {
if !allowed_bs.contains(term_id as u32) {
continue;
}
}
let entry = self.term_buckets.entries.entry(term_id).or_default();
*entry += 1;
}
// has subagg
if let Some(blueprint) = req_data.sub_aggregation_blueprint.as_ref() {
for (doc, term_id) in req_data
.column_block_accessor
.iter_docid_vals(docs, &req_data.accessor)
{
if let Some(allowed_bs) = req_data.allowed_term_ids.as_ref() {
if !allowed_bs.contains(term_id as u32) {
continue;
}
}
let sub_aggregations = self
.term_buckets
.sub_aggs
.entry(term_id)
.or_insert_with(|| blueprint.clone());
sub_aggregations.collect(doc, agg_data)?;
}
}
let mem_delta = self.get_memory_consumption() - mem_pre;
if mem_delta > 0 {
agg_data
.context
.limits
.add_memory_consumed(mem_delta as u64)?;
}
agg_data.put_back_term_req_data(self.accessor_idx, req_data);
Ok(())
}
fn flush(&mut self, agg_data: &mut AggregationsSegmentCtx) -> crate::Result<()> {
self.term_buckets.force_flush(agg_data)?;
Ok(())
}
}
impl SegmentTermCollector {
pub fn from_req_and_validate(
req_data: &mut AggregationsSegmentCtx,
node: &AggRefNode,
) -> crate::Result<Self> {
let terms_req_data = req_data.get_term_req_data(node.idx_in_req_data);
let column_type = terms_req_data.column_type;
let accessor_idx = node.idx_in_req_data;
if column_type == ColumnType::Bytes {
return Err(TantivyError::InvalidArgument(format!(
"terms aggregation is not supported for column type {column_type:?}"
)));
}
let term_buckets = TermBuckets::default();
// Validate sub aggregation exists
if let OrderTarget::SubAggregation(sub_agg_name) = &terms_req_data.req.order.target {
let (agg_name, _agg_property) = get_agg_name_and_property(sub_agg_name);
node.get_sub_agg(agg_name, &req_data.per_request)
.ok_or_else(|| {
TantivyError::InvalidArgument(format!(
"could not find aggregation with name {agg_name} in metric \
sub_aggregations"
))
})?;
}
let has_sub_aggregations = !node.children.is_empty();
let blueprint = if has_sub_aggregations {
let sub_aggregation = build_segment_agg_collectors(req_data, &node.children)?;
Some(sub_aggregation)
} else {
None
};
let terms_req_data = req_data.get_term_req_data_mut(node.idx_in_req_data);
terms_req_data.sub_aggregation_blueprint = blueprint;
Ok(SegmentTermCollector {
term_buckets,
accessor_idx,
})
}
fn get_memory_consumption(&self) -> usize {
let self_mem = std::mem::size_of::<Self>();
let term_buckets_mem = self.term_buckets.get_memory_consumption();
self_mem + term_buckets_mem
}
}

228
src/aggregation/bucket/term_agg/low_cardinality_impl.rs Normal file
View File

@@ -0,0 +1,228 @@
use std::vec;
use rustc_hash::FxHashMap;
use crate::aggregation::agg_data::{
build_segment_agg_collectors, AggRefNode, AggregationsSegmentCtx,
};
use crate::aggregation::bucket::{get_agg_name_and_property, OrderTarget};
use crate::aggregation::intermediate_agg_result::{
IntermediateAggregationResult, IntermediateAggregationResults,
};
use crate::aggregation::segment_agg_result::SegmentAggregationCollector;
use crate::{DocId, TantivyError};
const MAX_BATCH_SIZE: usize = 1_024;
#[derive(Debug, Clone)]
struct LowCardTermBuckets {
entries: Box<[u32]>,
sub_aggs: Vec<Box<dyn SegmentAggregationCollector>>,
doc_buffers: Box<[Vec<DocId>]>,
}
impl LowCardTermBuckets {
pub fn with_num_buckets(
num_buckets: usize,
sub_aggs_blueprint_opt: Option<&Box<dyn SegmentAggregationCollector>>,
) -> Self {
let sub_aggs = sub_aggs_blueprint_opt
.as_ref()
.map(|blueprint| {
std::iter::repeat_with(|| blueprint.clone_box())
.take(num_buckets)
.collect::<Vec<_>>()
})
.unwrap_or_default();
Self {
entries: vec![0; num_buckets].into_boxed_slice(),
sub_aggs,
doc_buffers: std::iter::repeat_with(|| Vec::with_capacity(MAX_BATCH_SIZE))
.take(num_buckets)
.collect::<Vec<_>>()
.into_boxed_slice(),
}
}
fn get_memory_consumption(&self) -> usize {
std::mem::size_of::<Self>()
+ self.entries.len() * std::mem::size_of::<u32>()
+ self.doc_buffers.len()
* (std::mem::size_of::<Vec<DocId>>()
+ std::mem::size_of::<DocId>() * MAX_BATCH_SIZE)
}
}
#[derive(Debug, Clone)]
pub struct LowCardSegmentTermCollector {
term_buckets: LowCardTermBuckets,
accessor_idx: usize,
}
impl LowCardSegmentTermCollector {
pub fn from_req_and_validate(
req_data: &mut AggregationsSegmentCtx,
node: &AggRefNode,
) -> crate::Result<Self> {
let terms_req_data = req_data.get_term_req_data(node.idx_in_req_data);
let accessor_idx = node.idx_in_req_data;
let cardinality = terms_req_data
.accessor
.max_value()
.max(terms_req_data.missing_value_for_accessor.unwrap_or(0))
+ 1;
assert!(cardinality <= super::LOW_CARDINALITY_THRESHOLD);
// Validate sub aggregation exists
if let OrderTarget::SubAggregation(sub_agg_name) = &terms_req_data.req.order.target {
let (agg_name, _agg_property) = get_agg_name_and_property(sub_agg_name);
node.get_sub_agg(agg_name, &req_data.per_request)
.ok_or_else(|| {
TantivyError::InvalidArgument(format!(
"could not find aggregation with name {agg_name} in metric \
sub_aggregations"
))
})?;
}
let has_sub_aggregations = !node.children.is_empty();
let blueprint = if has_sub_aggregations {
let sub_aggregation = build_segment_agg_collectors(req_data, &node.children)?;
Some(sub_aggregation)
} else {
None
};
let terms_req_data = req_data.get_term_req_data_mut(node.idx_in_req_data);
let term_buckets =
LowCardTermBuckets::with_num_buckets(cardinality as usize, blueprint.as_ref());
terms_req_data.sub_aggregation_blueprint = blueprint;
Ok(LowCardSegmentTermCollector {
term_buckets,
accessor_idx,
})
}
fn get_memory_consumption(&self) -> usize {
let self_mem = std::mem::size_of::<Self>();
let term_buckets_mem = self.term_buckets.get_memory_consumption();
self_mem + term_buckets_mem
}
}
impl SegmentAggregationCollector for LowCardSegmentTermCollector {
fn add_intermediate_aggregation_result(
self: Box<Self>,
agg_data: &AggregationsSegmentCtx,
results: &mut IntermediateAggregationResults,
) -> crate::Result<()> {
let name = agg_data.get_term_req_data(self.accessor_idx).name.clone();
let sub_aggs: FxHashMap<u64, Box<dyn SegmentAggregationCollector>> = self
.term_buckets
.sub_aggs
.into_iter()
.enumerate()
.filter(|(bucket_id, _sub_agg)| self.term_buckets.entries[*bucket_id] > 0)
.map(|(bucket_id, sub_agg)| (bucket_id as u64, sub_agg))
.collect();
let entries: Vec<(u64, u32)> = self
.term_buckets
.entries
.iter()
.enumerate()
.filter(|(_, count)| **count > 0)
.map(|(bucket_id, count)| (bucket_id as u64, *count))
.collect();
let bucket =
super::into_intermediate_bucket_result(self.accessor_idx, entries, sub_aggs, agg_data)?;
results.push(name, IntermediateAggregationResult::Bucket(bucket))?;
Ok(())
}
fn collect_block(
&mut self,
docs: &[crate::DocId],
agg_data: &mut AggregationsSegmentCtx,
) -> crate::Result<()> {
if docs.len() > MAX_BATCH_SIZE {
for batch in docs.chunks(MAX_BATCH_SIZE) {
self.collect_block(batch, agg_data)?;
}
}
let mut req_data = agg_data.take_term_req_data(self.accessor_idx);
let mem_pre = self.get_memory_consumption();
if let Some(missing) = req_data.missing_value_for_accessor {
req_data.column_block_accessor.fetch_block_with_missing(
docs,
&req_data.accessor,
missing,
);
} else {
req_data
.column_block_accessor
.fetch_block(docs, &req_data.accessor);
}
// has subagg
if req_data.sub_aggregation_blueprint.is_some() {
for (doc, term_id) in req_data
.column_block_accessor
.iter_docid_vals(docs, &req_data.accessor)
{
if let Some(allowed_bs) = req_data.allowed_term_ids.as_ref() {
if !allowed_bs.contains(term_id as u32) {
continue;
}
}
self.term_buckets.doc_buffers[term_id as usize].push(doc);
}
for (bucket_id, docs) in self.term_buckets.doc_buffers.iter_mut().enumerate() {
self.term_buckets.entries[bucket_id] += docs.len() as u32;
self.term_buckets.sub_aggs[bucket_id].collect_block(&docs[..], agg_data)?;
docs.clear();
}
} else {
for term_id in req_data.column_block_accessor.iter_vals() {
if let Some(allowed_bs) = req_data.allowed_term_ids.as_ref() {
if !allowed_bs.contains(term_id as u32) {
continue;
}
}
self.term_buckets.entries[term_id as usize] += 1;
}
}
let mem_delta = self.get_memory_consumption() - mem_pre;
if mem_delta > 0 {
agg_data
.context
.limits
.add_memory_consumed(mem_delta as u64)?;
}
agg_data.put_back_term_req_data(self.accessor_idx, req_data);
Ok(())
}
fn collect(
&mut self,
doc: crate::DocId,
agg_data: &mut AggregationsSegmentCtx,
) -> crate::Result<()> {
self.collect_block(&[doc], agg_data)
}
fn flush(&mut self, agg_data: &mut AggregationsSegmentCtx) -> crate::Result<()> {
for sub_aggregations in &mut self.term_buckets.sub_aggs.iter_mut() {
sub_aggregations.as_mut().flush(agg_data)?;
}
Ok(())
}
}

1000
src/aggregation/bucket/term_agg.rs → src/aggregation/bucket/term_agg/mod.rs
View File

File diff suppressed because it is too large Load Diff

76
src/aggregation/bucket/term_missing_agg.rs
View File

@@ -1,13 +1,39 @@
use columnar::{Column, ColumnType};
use rustc_hash::FxHashMap;
use crate::aggregation::agg_req_with_accessor::AggregationsWithAccessor;
use crate::aggregation::agg_data::{
build_segment_agg_collectors, AggRefNode, AggregationsSegmentCtx,
};
use crate::aggregation::bucket::term_agg::TermsAggregation;
use crate::aggregation::intermediate_agg_result::{
IntermediateAggregationResult, IntermediateAggregationResults, IntermediateBucketResult,
IntermediateKey, IntermediateTermBucketEntry, IntermediateTermBucketResult,
};
use crate::aggregation::segment_agg_result::{
build_segment_agg_collector, SegmentAggregationCollector,
};
use crate::aggregation::segment_agg_result::SegmentAggregationCollector;
/// Special aggregation to handle missing values for term aggregations.
/// This missing aggregation will check multiple columns for existence.
///
/// This is needed when:
/// - The field is multi-valued and we therefore have multiple columns
/// - The field is not text and missing is provided as string (we cannot use the numeric missing
/// value optimization)
#[derive(Default)]
pub struct MissingTermAggReqData {
/// The accessors to check for existence of a value.
pub accessors: Vec<(Column<u64>, ColumnType)>,
/// The name of the aggregation.
pub name: String,
/// The original terms aggregation request.
pub req: TermsAggregation,
}
impl MissingTermAggReqData {
/// Estimate the memory consumption of this struct in bytes.
pub fn get_memory_consumption(&self) -> usize {
std::mem::size_of::<Self>()
}
}
/// The specialized missing term aggregation.
#[derive(Default, Debug, Clone)]
@@ -18,12 +44,13 @@ pub struct TermMissingAgg {
}
impl TermMissingAgg {
pub(crate) fn new(
accessor_idx: usize,
sub_aggregations: &mut AggregationsWithAccessor,
req_data: &mut AggregationsSegmentCtx,
node: &AggRefNode,
) -> crate::Result<Self> {
let has_sub_aggregations = !sub_aggregations.is_empty();
let has_sub_aggregations = !node.children.is_empty();
let accessor_idx = node.idx_in_req_data;
let sub_agg = if has_sub_aggregations {
let sub_aggregation = build_segment_agg_collector(sub_aggregations)?;
let sub_aggregation = build_segment_agg_collectors(req_data, &node.children)?;
Some(sub_aggregation)
} else {
None
@@ -40,16 +67,11 @@ impl TermMissingAgg {
impl SegmentAggregationCollector for TermMissingAgg {
fn add_intermediate_aggregation_result(
self: Box<Self>,
agg_with_accessor: &AggregationsWithAccessor,
agg_data: &AggregationsSegmentCtx,
results: &mut IntermediateAggregationResults,
) -> crate::Result<()> {
let name = agg_with_accessor.aggs.keys[self.accessor_idx].to_string();
let agg_with_accessor = &agg_with_accessor.aggs.values[self.accessor_idx];
let term_agg = agg_with_accessor
.agg
.agg
.as_term()
.expect("TermMissingAgg collector must be term agg req");
let req_data = agg_data.get_missing_term_req_data(self.accessor_idx);
let term_agg = &req_data.req;
let missing = term_agg
.missing
.as_ref()
@@ -64,10 +86,7 @@ impl SegmentAggregationCollector for TermMissingAgg {
};
if let Some(sub_agg) = self.sub_agg {
let mut res = IntermediateAggregationResults::default();
sub_agg.add_intermediate_aggregation_result(
&agg_with_accessor.sub_aggregation,
&mut res,
)?;
sub_agg.add_intermediate_aggregation_result(agg_data, &mut res)?;
missing_entry.sub_aggregation = res;
}
entries.insert(missing.into(), missing_entry);
@@ -80,7 +99,10 @@ impl SegmentAggregationCollector for TermMissingAgg {
},
};
results.push(name, IntermediateAggregationResult::Bucket(bucket))?;
results.push(
req_data.name.to_string(),
IntermediateAggregationResult::Bucket(bucket),
)?;
Ok(())
}
@@ -88,17 +110,17 @@ impl SegmentAggregationCollector for TermMissingAgg {
fn collect(
&mut self,
doc: crate::DocId,
agg_with_accessor: &mut AggregationsWithAccessor,
agg_data: &mut AggregationsSegmentCtx,
) -> crate::Result<()> {
let agg = &mut agg_with_accessor.aggs.values[self.accessor_idx];
let has_value = agg
let req_data = agg_data.get_missing_term_req_data(self.accessor_idx);
let has_value = req_data
.accessors
.iter()
.any(|(acc, _)| acc.index.has_value(doc));
if !has_value {
self.missing_count += 1;
if let Some(sub_agg) = self.sub_agg.as_mut() {
sub_agg.collect(doc, &mut agg.sub_aggregation)?;
sub_agg.collect(doc, agg_data)?;
}
}
Ok(())
@@ -107,10 +129,10 @@ impl SegmentAggregationCollector for TermMissingAgg {
fn collect_block(
&mut self,
docs: &[crate::DocId],
agg_with_accessor: &mut AggregationsWithAccessor,
agg_data: &mut AggregationsSegmentCtx,
) -> crate::Result<()> {
for doc in docs {
self.collect(*doc, agg_with_accessor)?;
self.collect(*doc, agg_data)?;
}
Ok(())
}

20
src/aggregation/buf_collector.rs
View File

@@ -1,6 +1,6 @@
use super::agg_req_with_accessor::AggregationsWithAccessor;
use super::intermediate_agg_result::IntermediateAggregationResults;
use super::segment_agg_result::SegmentAggregationCollector;
use crate::aggregation::agg_data::AggregationsSegmentCtx;
use crate::DocId;
pub(crate) const DOC_BLOCK_SIZE: usize = 64;
@@ -37,23 +37,23 @@ impl SegmentAggregationCollector for BufAggregationCollector {
#[inline]
fn add_intermediate_aggregation_result(
self: Box<Self>,
agg_with_accessor: &AggregationsWithAccessor,
agg_data: &AggregationsSegmentCtx,
results: &mut IntermediateAggregationResults,
) -> crate::Result<()> {
Box::new(self.collector).add_intermediate_aggregation_result(agg_with_accessor, results)
Box::new(self.collector).add_intermediate_aggregation_result(agg_data, results)
}
#[inline]
fn collect(
&mut self,
doc: crate::DocId,
agg_with_accessor: &mut AggregationsWithAccessor,
agg_data: &mut AggregationsSegmentCtx,
) -> crate::Result<()> {
self.staged_docs[self.num_staged_docs] = doc;
self.num_staged_docs += 1;
if self.num_staged_docs == self.staged_docs.len() {
self.collector
.collect_block(&self.staged_docs[..self.num_staged_docs], agg_with_accessor)?;
.collect_block(&self.staged_docs[..self.num_staged_docs], agg_data)?;
self.num_staged_docs = 0;
}
Ok(())
@@ -63,20 +63,20 @@ impl SegmentAggregationCollector for BufAggregationCollector {
fn collect_block(
&mut self,
docs: &[crate::DocId],
agg_with_accessor: &mut AggregationsWithAccessor,
agg_data: &mut AggregationsSegmentCtx,
) -> crate::Result<()> {
self.collector.collect_block(docs, agg_with_accessor)?;
self.collector.collect_block(docs, agg_data)?;
Ok(())
}
#[inline]
fn flush(&mut self, agg_with_accessor: &mut AggregationsWithAccessor) -> crate::Result<()> {
fn flush(&mut self, agg_data: &mut AggregationsSegmentCtx) -> crate::Result<()> {
self.collector
.collect_block(&self.staged_docs[..self.num_staged_docs], agg_with_accessor)?;
.collect_block(&self.staged_docs[..self.num_staged_docs], agg_data)?;
self.num_staged_docs = 0;
self.collector.flush(agg_with_accessor)?;
self.collector.flush(agg_data)?;
Ok(())
}

39
src/aggregation/collector.rs
View File

@@ -1,12 +1,12 @@
use super::agg_req::Aggregations;
use super::agg_req_with_accessor::AggregationsWithAccessor;
use super::agg_result::AggregationResults;
use super::buf_collector::BufAggregationCollector;
use super::intermediate_agg_result::IntermediateAggregationResults;
use super::segment_agg_result::{
build_segment_agg_collector, AggregationLimitsGuard, SegmentAggregationCollector,
use super::segment_agg_result::SegmentAggregationCollector;
use super::AggContextParams;
use crate::aggregation::agg_data::{
build_aggregations_data_from_req, build_segment_agg_collectors_root, AggregationsSegmentCtx,
};
use crate::aggregation::agg_req_with_accessor::get_aggs_with_segment_accessor_and_validate;
use crate::collector::{Collector, SegmentCollector};
use crate::index::SegmentReader;
use crate::{DocId, SegmentOrdinal, TantivyError};
@@ -22,7 +22,7 @@ pub const DEFAULT_MEMORY_LIMIT: u64 = 500_000_000;
/// The collector collects all aggregations by the underlying aggregation request.
pub struct AggregationCollector {
agg: Aggregations,
limits: AggregationLimitsGuard,
context: AggContextParams,
}
impl AggregationCollector {
@@ -30,8 +30,8 @@ impl AggregationCollector {
///
/// Aggregation fails when the limits in `AggregationLimits` is exceeded. (memory limit and
/// bucket limit)
pub fn from_aggs(agg: Aggregations, limits: AggregationLimitsGuard) -> Self {
Self { agg, limits }
pub fn from_aggs(agg: Aggregations, context: AggContextParams) -> Self {
Self { agg, context }
}
}
@@ -45,7 +45,7 @@ impl AggregationCollector {
/// into the final `AggregationResults` via the `into_final_result()` method.
pub struct DistributedAggregationCollector {
agg: Aggregations,
limits: AggregationLimitsGuard,
context: AggContextParams,
}
impl DistributedAggregationCollector {
@@ -53,8 +53,8 @@ impl DistributedAggregationCollector {
///
/// Aggregation fails when the limits in `AggregationLimits` is exceeded. (memory limit and
/// bucket limit)
pub fn from_aggs(agg: Aggregations, limits: AggregationLimitsGuard) -> Self {
Self { agg, limits }
pub fn from_aggs(agg: Aggregations, context: AggContextParams) -> Self {
Self { agg, context }
}
}
@@ -72,7 +72,7 @@ impl Collector for DistributedAggregationCollector {
&self.agg,
reader,
segment_local_id,
&self.limits,
&self.context,
)
}
@@ -102,7 +102,7 @@ impl Collector for AggregationCollector {
&self.agg,
reader,
segment_local_id,
&self.limits,
&self.context,
)
}
@@ -115,7 +115,7 @@ impl Collector for AggregationCollector {
segment_fruits: Vec<<Self::Child as SegmentCollector>::Fruit>,
) -> crate::Result<Self::Fruit> {
let res = merge_fruits(segment_fruits)?;
res.into_final_result(self.agg.clone(), self.limits.clone())
res.into_final_result(self.agg.clone(), self.context.limits.clone())
}
}
@@ -135,7 +135,7 @@ fn merge_fruits(
/// `AggregationSegmentCollector` does the aggregation collection on a segment.
pub struct AggregationSegmentCollector {
aggs_with_accessor: AggregationsWithAccessor,
aggs_with_accessor: AggregationsSegmentCtx,
agg_collector: BufAggregationCollector,
error: Option<TantivyError>,
}
@@ -147,14 +147,15 @@ impl AggregationSegmentCollector {
agg: &Aggregations,
reader: &SegmentReader,
segment_ordinal: SegmentOrdinal,
limits: &AggregationLimitsGuard,
context: &AggContextParams,
) -> crate::Result<Self> {
let mut aggs_with_accessor =
get_aggs_with_segment_accessor_and_validate(agg, reader, segment_ordinal, limits)?;
let mut agg_data =
build_aggregations_data_from_req(agg, reader, segment_ordinal, context.clone())?;
let result =
BufAggregationCollector::new(build_segment_agg_collector(&mut aggs_with_accessor)?);
BufAggregationCollector::new(build_segment_agg_collectors_root(&mut agg_data)?);
Ok(AggregationSegmentCollector {
aggs_with_accessor,
aggs_with_accessor: agg_data,
agg_collector: result,
error: None,
})

61
src/aggregation/intermediate_agg_result.rs
View File

@@ -24,7 +24,9 @@ use super::metric::{
};
use super::segment_agg_result::AggregationLimitsGuard;
use super::{format_date, AggregationError, Key, SerializedKey};
use crate::aggregation::agg_result::{AggregationResults, BucketEntries, BucketEntry};
use crate::aggregation::agg_result::{
AggregationResults, BucketEntries, BucketEntry, FilterBucketResult,
};
use crate::aggregation::bucket::TermsAggregationInternal;
use crate::aggregation::metric::CardinalityCollector;
use crate::TantivyError;
@@ -179,12 +181,17 @@ impl IntermediateAggregationResults {
}
/// Merge another intermediate aggregation result into this result.
///
/// The order of the values need to be the same on both results. This is ensured when the same
/// (key values) are present on the underlying `VecWithNames` struct.
pub fn merge_fruits(&mut self, other: IntermediateAggregationResults) -> crate::Result<()> {
for (left, right) in self.aggs_res.values_mut().zip(other.aggs_res.into_values()) {
left.merge_fruits(right)?;
pub fn merge_fruits(&mut self, mut other: IntermediateAggregationResults) -> crate::Result<()> {
for (key, left) in self.aggs_res.iter_mut() {
if let Some(key) = other.aggs_res.remove(key) {
left.merge_fruits(key)?;
}
}
// Move remainder of other aggs_res into self.
// Note: Currently we don't expect this to happen, as we create empty intermediate results
// via [IntermediateAggregationResults::empty_from_req].
for (key, value) in other.aggs_res {
self.aggs_res.insert(key, value);
}
Ok(())
}
@@ -241,11 +248,16 @@ pub(crate) fn empty_from_req(req: &Aggregation) -> IntermediateAggregationResult
Cardinality(_) => IntermediateAggregationResult::Metric(
IntermediateMetricResult::Cardinality(CardinalityCollector::default()),
),
Filter(_) => IntermediateAggregationResult::Bucket(IntermediateBucketResult::Filter {
doc_count: 0,
sub_aggregations: IntermediateAggregationResults::default(),
}),
}
}
/// An aggregation is either a bucket or a metric.
#[derive(Clone, Debug, PartialEq, Serialize, Deserialize)]
#[allow(clippy::large_enum_variant)]
pub enum IntermediateAggregationResult {
/// Bucket variant
Bucket(IntermediateBucketResult),
@@ -426,6 +438,13 @@ pub enum IntermediateBucketResult {
/// The term buckets
buckets: IntermediateTermBucketResult,
},
/// Filter aggregation - a single bucket with sub-aggregations
Filter {
/// Document count in the filter bucket
doc_count: u64,
/// Sub-aggregation results
sub_aggregations: IntermediateAggregationResults,
},
}
impl IntermediateBucketResult {
@@ -509,6 +528,18 @@ impl IntermediateBucketResult {
req.sub_aggregation(),
limits,
),
IntermediateBucketResult::Filter {
doc_count,
sub_aggregations,
} => {
// Convert sub-aggregation results to final format
let final_sub_aggregations = sub_aggregations
.into_final_result(req.sub_aggregation().clone(), limits.clone())?;
Ok(BucketResult::Filter(FilterBucketResult {
doc_count,
sub_aggregations: final_sub_aggregations,
}))
}
}
}
@@ -562,6 +593,19 @@ impl IntermediateBucketResult {
*buckets_left = buckets?;
}
(
IntermediateBucketResult::Filter {
doc_count: doc_count_left,
sub_aggregations: sub_aggs_left,
},
IntermediateBucketResult::Filter {
doc_count: doc_count_right,
sub_aggregations: sub_aggs_right,
},
) => {
*doc_count_left += doc_count_right;
sub_aggs_left.merge_fruits(sub_aggs_right)?;
}
(IntermediateBucketResult::Range(_), _) => {
panic!("try merge on different types")
}
@@ -571,6 +615,9 @@ impl IntermediateBucketResult {
(IntermediateBucketResult::Terms { .. }, _) => {
panic!("try merge on different types")
}
(IntermediateBucketResult::Filter { .. }, _) => {
panic!("try merge on different types")
}
}
Ok(())
}

91
src/aggregation/metric/cardinality.rs
View File

@@ -2,15 +2,13 @@ use std::collections::hash_map::DefaultHasher;
use std::hash::{BuildHasher, Hasher};
use columnar::column_values::CompactSpaceU64Accessor;
use columnar::Dictionary;
use columnar::{Column, ColumnBlockAccessor, ColumnType, Dictionary, StrColumn};
use common::f64_to_u64;
use hyperloglogplus::{HyperLogLog, HyperLogLogPlus};
use rustc_hash::FxHashSet;
use serde::{Deserialize, Serialize};
use crate::aggregation::agg_req_with_accessor::{
AggregationWithAccessor, AggregationsWithAccessor,
};
use crate::aggregation::agg_data::AggregationsSegmentCtx;
use crate::aggregation::intermediate_agg_result::{
IntermediateAggregationResult, IntermediateAggregationResults, IntermediateMetricResult,
};
@@ -97,6 +95,32 @@ pub struct CardinalityAggregationReq {
pub missing: Option<Key>,
}
/// Contains all information required by the SegmentCardinalityCollector to perform the
/// cardinality aggregation on a segment.
pub struct CardinalityAggReqData {
/// The column accessor to access the fast field values.
pub accessor: Column<u64>,
/// The column_type of the field.
pub column_type: ColumnType,
/// The string dictionary column if the field is of type string.
pub str_dict_column: Option<StrColumn>,
/// The missing value normalized to the internal u64 representation of the field type.
pub missing_value_for_accessor: Option<u64>,
/// The column block accessor to access the fast field values.
pub(crate) column_block_accessor: ColumnBlockAccessor<u64>,
/// The name of the aggregation.
pub name: String,
/// The aggregation request.
pub req: CardinalityAggregationReq,
}
impl CardinalityAggReqData {
/// Estimate the memory consumption of this struct in bytes.
pub fn get_memory_consumption(&self) -> usize {
std::mem::size_of::<Self>()
}
}
impl CardinalityAggregationReq {
/// Creates a new [`CardinalityAggregationReq`] instance from a field name.
pub fn from_field_name(field_name: String) -> Self {
@@ -115,47 +139,44 @@ impl CardinalityAggregationReq {
pub(crate) struct SegmentCardinalityCollector {
cardinality: CardinalityCollector,
entries: FxHashSet<u64>,
column_type: ColumnType,
accessor_idx: usize,
missing: Option<Key>,
}
impl SegmentCardinalityCollector {
pub fn from_req(column_type: ColumnType, accessor_idx: usize, missing: &Option<Key>) -> Self {
pub fn from_req(column_type: ColumnType, accessor_idx: usize) -> Self {
Self {
cardinality: CardinalityCollector::new(column_type as u8),
entries: Default::default(),
column_type,
accessor_idx,
missing: missing.clone(),
}
}
fn fetch_block_with_field(
&mut self,
docs: &[crate::DocId],
agg_accessor: &mut AggregationWithAccessor,
agg_data: &mut CardinalityAggReqData,
) {
if let Some(missing) = agg_accessor.missing_value_for_accessor {
agg_accessor.column_block_accessor.fetch_block_with_missing(
if let Some(missing) = agg_data.missing_value_for_accessor {
agg_data.column_block_accessor.fetch_block_with_missing(
docs,
&agg_accessor.accessor,
&agg_data.accessor,
missing,
);
} else {
agg_accessor
agg_data
.column_block_accessor
.fetch_block(docs, &agg_accessor.accessor);
.fetch_block(docs, &agg_data.accessor);
}
}
fn into_intermediate_metric_result(
mut self,
agg_with_accessor: &AggregationWithAccessor,
agg_data: &AggregationsSegmentCtx,
) -> crate::Result<IntermediateMetricResult> {
if self.column_type == ColumnType::Str {
let req_data = &agg_data.get_cardinality_req_data(self.accessor_idx);
if req_data.column_type == ColumnType::Str {
let fallback_dict = Dictionary::empty();
let dict = agg_with_accessor
let dict = req_data
.str_dict_column
.as_ref()
.map(|el| el.dictionary())
@@ -180,10 +201,10 @@ impl SegmentCardinalityCollector {
})?;
if has_missing {
// Replace missing with the actual value provided
let missing_key = self
.missing
.as_ref()
.expect("Found sentinel value u64::MAX for term_ord but `missing` is not set");
let missing_key =
req_data.req.missing.as_ref().expect(
"Found sentinel value u64::MAX for term_ord but `missing` is not set",
);
match missing_key {
Key::Str(missing) => {
self.cardinality.sketch.insert_any(&missing);
@@ -209,13 +230,13 @@ impl SegmentCardinalityCollector {
impl SegmentAggregationCollector for SegmentCardinalityCollector {
fn add_intermediate_aggregation_result(
self: Box<Self>,
agg_with_accessor: &AggregationsWithAccessor,
agg_data: &AggregationsSegmentCtx,
results: &mut IntermediateAggregationResults,
) -> crate::Result<()> {
let name = agg_with_accessor.aggs.keys[self.accessor_idx].to_string();
let agg_with_accessor = &agg_with_accessor.aggs.values[self.accessor_idx];
let req_data = &agg_data.get_cardinality_req_data(self.accessor_idx);
let name = req_data.name.to_string();
let intermediate_result = self.into_intermediate_metric_result(agg_with_accessor)?;
let intermediate_result = self.into_intermediate_metric_result(agg_data)?;
results.push(
name,
IntermediateAggregationResult::Metric(intermediate_result),
@@ -227,26 +248,26 @@ impl SegmentAggregationCollector for SegmentCardinalityCollector {
fn collect(
&mut self,
doc: crate::DocId,
agg_with_accessor: &mut AggregationsWithAccessor,
agg_data: &mut AggregationsSegmentCtx,
) -> crate::Result<()> {
self.collect_block(&[doc], agg_with_accessor)
self.collect_block(&[doc], agg_data)
}
fn collect_block(
&mut self,
docs: &[crate::DocId],
agg_with_accessor: &mut AggregationsWithAccessor,
agg_data: &mut AggregationsSegmentCtx,
) -> crate::Result<()> {
let bucket_agg_accessor = &mut agg_with_accessor.aggs.values[self.accessor_idx];
self.fetch_block_with_field(docs, bucket_agg_accessor);
let req_data = agg_data.get_cardinality_req_data_mut(self.accessor_idx);
self.fetch_block_with_field(docs, req_data);
let col_block_accessor = &bucket_agg_accessor.column_block_accessor;
if self.column_type == ColumnType::Str {
let col_block_accessor = &req_data.column_block_accessor;
if req_data.column_type == ColumnType::Str {
for term_ord in col_block_accessor.iter_vals() {
self.entries.insert(term_ord);
}
} else if self.column_type == ColumnType::IpAddr {
let compact_space_accessor = bucket_agg_accessor
} else if req_data.column_type == ColumnType::IpAddr {
let compact_space_accessor = req_data
.accessor
.values
.clone()

35
src/aggregation/metric/extended_stats.rs
View File

@@ -4,12 +4,11 @@ use std::mem;
use serde::{Deserialize, Serialize};
use super::*;
use crate::aggregation::agg_req_with_accessor::{
AggregationWithAccessor, AggregationsWithAccessor,
};
use crate::aggregation::agg_data::AggregationsSegmentCtx;
use crate::aggregation::intermediate_agg_result::{
IntermediateAggregationResult, IntermediateAggregationResults, IntermediateMetricResult,
};
use crate::aggregation::metric::MetricAggReqData;
use crate::aggregation::segment_agg_result::SegmentAggregationCollector;
use crate::aggregation::*;
use crate::{DocId, TantivyError};
@@ -348,20 +347,20 @@ impl SegmentExtendedStatsCollector {
pub(crate) fn collect_block_with_field(
&mut self,
docs: &[DocId],
agg_accessor: &mut AggregationWithAccessor,
req_data: &mut MetricAggReqData,
) {
if let Some(missing) = self.missing.as_ref() {
agg_accessor.column_block_accessor.fetch_block_with_missing(
req_data.column_block_accessor.fetch_block_with_missing(
docs,
&agg_accessor.accessor,
&req_data.accessor,
*missing,
);
} else {
agg_accessor
req_data
.column_block_accessor
.fetch_block(docs, &agg_accessor.accessor);
.fetch_block(docs, &req_data.accessor);
}
for val in agg_accessor.column_block_accessor.iter_vals() {
for val in req_data.column_block_accessor.iter_vals() {
let val1 = f64_from_fastfield_u64(val, &self.field_type);
self.extended_stats.collect(val1);
}
@@ -372,10 +371,10 @@ impl SegmentAggregationCollector for SegmentExtendedStatsCollector {
#[inline]
fn add_intermediate_aggregation_result(
self: Box<Self>,
agg_with_accessor: &AggregationsWithAccessor,
agg_data: &AggregationsSegmentCtx,
results: &mut IntermediateAggregationResults,
) -> crate::Result<()> {
let name = agg_with_accessor.aggs.keys[self.accessor_idx].to_string();
let name = agg_data.get_metric_req_data(self.accessor_idx).name.clone();
results.push(
name,
IntermediateAggregationResult::Metric(IntermediateMetricResult::ExtendedStats(
@@ -390,12 +389,12 @@ impl SegmentAggregationCollector for SegmentExtendedStatsCollector {
fn collect(
&mut self,
doc: crate::DocId,
agg_with_accessor: &mut AggregationsWithAccessor,
agg_data: &mut AggregationsSegmentCtx,
) -> crate::Result<()> {
let field = &agg_with_accessor.aggs.values[self.accessor_idx].accessor;
let req_data = agg_data.get_metric_req_data(self.accessor_idx);
if let Some(missing) = self.missing {
let mut has_val = false;
for val in field.values_for_doc(doc) {
for val in req_data.accessor.values_for_doc(doc) {
let val1 = f64_from_fastfield_u64(val, &self.field_type);
self.extended_stats.collect(val1);
has_val = true;
@@ -405,7 +404,7 @@ impl SegmentAggregationCollector for SegmentExtendedStatsCollector {
.collect(f64_from_fastfield_u64(missing, &self.field_type));
}
} else {
for val in field.values_for_doc(doc) {
for val in req_data.accessor.values_for_doc(doc) {
let val1 = f64_from_fastfield_u64(val, &self.field_type);
self.extended_stats.collect(val1);
}
@@ -418,10 +417,10 @@ impl SegmentAggregationCollector for SegmentExtendedStatsCollector {
fn collect_block(
&mut self,
docs: &[crate::DocId],
agg_with_accessor: &mut AggregationsWithAccessor,
agg_data: &mut AggregationsSegmentCtx,
) -> crate::Result<()> {
let field = &mut agg_with_accessor.aggs.values[self.accessor_idx];
self.collect_block_with_field(docs, field);
let req_data = agg_data.get_metric_req_data_mut(self.accessor_idx);
self.collect_block_with_field(docs, req_data);
Ok(())
}
}

30
src/aggregation/metric/mod.rs
View File

@@ -31,6 +31,7 @@ use std::collections::HashMap;
pub use average::*;
pub use cardinality::*;
use columnar::{Column, ColumnBlockAccessor, ColumnType};
pub use count::*;
pub use extended_stats::*;
pub use max::*;
@@ -44,6 +45,35 @@ pub use top_hits::*;
use crate::schema::OwnedValue;
/// Contains all information required by metric aggregations like avg, min, max, sum, stats,
/// extended_stats, count, percentiles.
#[repr(C)]
pub struct MetricAggReqData {
/// True if the field is of number or date type.
pub is_number_or_date_type: bool,
/// The type of the field.
pub field_type: ColumnType,
/// The missing value normalized to the internal u64 representation of the field type.
pub missing_u64: Option<u64>,
/// The column block accessor to access the fast field values.
pub column_block_accessor: ColumnBlockAccessor<u64>,
/// The column accessor to access the fast field values.
pub accessor: Column<u64>,
/// Used when converting to intermediate result
pub collecting_for: StatsType,
/// The missing value
pub missing: Option<f64>,
/// The name of the aggregation.
pub name: String,
}
impl MetricAggReqData {
/// Estimate the memory consumption of this struct in bytes.
pub fn get_memory_consumption(&self) -> usize {
std::mem::size_of::<Self>()
}
}
/// Single-metric aggregations use this common result structure.
///
/// Main reason to wrap it in value is to match elasticsearch output structure.

65
src/aggregation/metric/percentiles.rs
View File

@@ -3,12 +3,11 @@ use std::fmt::Debug;
use serde::{Deserialize, Serialize};
use super::*;
use crate::aggregation::agg_req_with_accessor::{
AggregationWithAccessor, AggregationsWithAccessor,
};
use crate::aggregation::agg_data::AggregationsSegmentCtx;
use crate::aggregation::intermediate_agg_result::{
IntermediateAggregationResult, IntermediateAggregationResults, IntermediateMetricResult,
};
use crate::aggregation::metric::MetricAggReqData;
use crate::aggregation::segment_agg_result::SegmentAggregationCollector;
use crate::aggregation::*;
use crate::{DocId, TantivyError};
@@ -112,7 +111,8 @@ impl PercentilesAggregationReq {
&self.field
}
fn validate(&self) -> crate::Result<()> {
/// Validates the request parameters.
pub fn validate(&self) -> crate::Result<()> {
if let Some(percents) = self.percents.as_ref() {
let all_in_range = percents
.iter()
@@ -133,10 +133,8 @@ impl PercentilesAggregationReq {
#[derive(Clone, Debug, PartialEq)]
pub(crate) struct SegmentPercentilesCollector {
field_type: ColumnType,
pub(crate) percentiles: PercentilesCollector,
pub(crate) accessor_idx: usize,
missing: Option<u64>,
}
#[derive(Clone, Serialize, Deserialize)]
@@ -231,43 +229,32 @@ impl PercentilesCollector {
}
impl SegmentPercentilesCollector {
pub fn from_req_and_validate(
req: &PercentilesAggregationReq,
field_type: ColumnType,
accessor_idx: usize,
) -> crate::Result<Self> {
req.validate()?;
let missing = req
.missing
.and_then(|val| f64_to_fastfield_u64(val, &field_type));
pub fn from_req_and_validate(accessor_idx: usize) -> crate::Result<Self> {
Ok(Self {
field_type,
percentiles: PercentilesCollector::new(),
accessor_idx,
missing,
})
}
#[inline]
pub(crate) fn collect_block_with_field(
&mut self,
docs: &[DocId],
agg_accessor: &mut AggregationWithAccessor,
req_data: &mut MetricAggReqData,
) {
if let Some(missing) = self.missing.as_ref() {
agg_accessor.column_block_accessor.fetch_block_with_missing(
if let Some(missing) = req_data.missing_u64.as_ref() {
req_data.column_block_accessor.fetch_block_with_missing(
docs,
&agg_accessor.accessor,
&req_data.accessor,
*missing,
);
} else {
agg_accessor
req_data
.column_block_accessor
.fetch_block(docs, &agg_accessor.accessor);
.fetch_block(docs, &req_data.accessor);
}
for val in agg_accessor.column_block_accessor.iter_vals() {
let val1 = f64_from_fastfield_u64(val, &self.field_type);
for val in req_data.column_block_accessor.iter_vals() {
let val1 = f64_from_fastfield_u64(val, &req_data.field_type);
self.percentiles.collect(val1);
}
}
@@ -277,10 +264,10 @@ impl SegmentAggregationCollector for SegmentPercentilesCollector {
#[inline]
fn add_intermediate_aggregation_result(
self: Box<Self>,
agg_with_accessor: &AggregationsWithAccessor,
agg_data: &AggregationsSegmentCtx,
results: &mut IntermediateAggregationResults,
) -> crate::Result<()> {
let name = agg_with_accessor.aggs.keys[self.accessor_idx].to_string();
let name = agg_data.get_metric_req_data(self.accessor_idx).name.clone();
let intermediate_metric_result = IntermediateMetricResult::Percentiles(self.percentiles);
results.push(
@@ -295,24 +282,24 @@ impl SegmentAggregationCollector for SegmentPercentilesCollector {
fn collect(
&mut self,
doc: crate::DocId,
agg_with_accessor: &mut AggregationsWithAccessor,
agg_data: &mut AggregationsSegmentCtx,
) -> crate::Result<()> {
let field = &agg_with_accessor.aggs.values[self.accessor_idx].accessor;
let req_data = agg_data.get_metric_req_data(self.accessor_idx);
if let Some(missing) = self.missing {
if let Some(missing) = req_data.missing_u64 {
let mut has_val = false;
for val in field.values_for_doc(doc) {
let val1 = f64_from_fastfield_u64(val, &self.field_type);
for val in req_data.accessor.values_for_doc(doc) {
let val1 = f64_from_fastfield_u64(val, &req_data.field_type);
self.percentiles.collect(val1);
has_val = true;
}
if !has_val {
self.percentiles
.collect(f64_from_fastfield_u64(missing, &self.field_type));
.collect(f64_from_fastfield_u64(missing, &req_data.field_type));
}
} else {
for val in field.values_for_doc(doc) {
let val1 = f64_from_fastfield_u64(val, &self.field_type);
for val in req_data.accessor.values_for_doc(doc) {
let val1 = f64_from_fastfield_u64(val, &req_data.field_type);
self.percentiles.collect(val1);
}
}
@@ -324,10 +311,10 @@ impl SegmentAggregationCollector for SegmentPercentilesCollector {
fn collect_block(
&mut self,
docs: &[crate::DocId],
agg_with_accessor: &mut AggregationsWithAccessor,
agg_data: &mut AggregationsSegmentCtx,
) -> crate::Result<()> {
let field = &mut agg_with_accessor.aggs.values[self.accessor_idx];
self.collect_block_with_field(docs, field);
let req_data = agg_data.get_metric_req_data_mut(self.accessor_idx);
self.collect_block_with_field(docs, req_data);
Ok(())
}
}

117
src/aggregation/metric/stats.rs
View File

@@ -3,12 +3,11 @@ use std::fmt::Debug;
use serde::{Deserialize, Serialize};
use super::*;
use crate::aggregation::agg_req_with_accessor::{
AggregationWithAccessor, AggregationsWithAccessor,
};
use crate::aggregation::agg_data::AggregationsSegmentCtx;
use crate::aggregation::intermediate_agg_result::{
IntermediateAggregationResult, IntermediateAggregationResults, IntermediateMetricResult,
};
use crate::aggregation::metric::MetricAggReqData;
use crate::aggregation::segment_agg_result::SegmentAggregationCollector;
use crate::aggregation::*;
use crate::{DocId, TantivyError};
@@ -166,74 +165,65 @@ impl IntermediateStats {
}
}
#[derive(Clone, Debug, PartialEq)]
pub(crate) enum SegmentStatsType {
/// The type of stats aggregation to perform.
/// Note that not all stats types are supported in the stats aggregation.
#[derive(Clone, Copy, Debug)]
pub enum StatsType {
/// The average of the values.
Average,
/// The count of the values.
Count,
/// The maximum value.
Max,
/// The minimum value.
Min,
/// The stats (count, sum, min, max, avg) of the values.
Stats,
/// The extended stats (count, sum, min, max, avg, sum_of_squares, variance, std_deviation,
ExtendedStats(Option<f64>), // sigma
/// The sum of the values.
Sum,
/// The percentiles of the values.
Percentiles,
}
#[derive(Clone, Debug, PartialEq)]
#[derive(Clone, Debug)]
pub(crate) struct SegmentStatsCollector {
missing: Option<u64>,
field_type: ColumnType,
pub(crate) collecting_for: SegmentStatsType,
pub(crate) stats: IntermediateStats,
pub(crate) accessor_idx: usize,
val_cache: Vec<u64>,
}
impl SegmentStatsCollector {
pub fn from_req(
field_type: ColumnType,
collecting_for: SegmentStatsType,
accessor_idx: usize,
missing: Option<f64>,
) -> Self {
let missing = missing.and_then(|val| f64_to_fastfield_u64(val, &field_type));
pub fn from_req(accessor_idx: usize) -> Self {
Self {
field_type,
collecting_for,
stats: IntermediateStats::default(),
accessor_idx,
missing,
val_cache: Default::default(),
}
}
#[inline]
pub(crate) fn collect_block_with_field(
&mut self,
docs: &[DocId],
agg_accessor: &mut AggregationWithAccessor,
req_data: &mut MetricAggReqData,
) {
if let Some(missing) = self.missing.as_ref() {
agg_accessor.column_block_accessor.fetch_block_with_missing(
if let Some(missing) = req_data.missing_u64.as_ref() {
req_data.column_block_accessor.fetch_block_with_missing(
docs,
&agg_accessor.accessor,
&req_data.accessor,
*missing,
);
} else {
agg_accessor
req_data
.column_block_accessor
.fetch_block(docs, &agg_accessor.accessor);
.fetch_block(docs, &req_data.accessor);
}
if [
ColumnType::I64,
ColumnType::U64,
ColumnType::F64,
ColumnType::DateTime,
]
.contains(&self.field_type)
{
for val in agg_accessor.column_block_accessor.iter_vals() {
let val1 = f64_from_fastfield_u64(val, &self.field_type);
if req_data.is_number_or_date_type {
for val in req_data.column_block_accessor.iter_vals() {
let val1 = f64_from_fastfield_u64(val, &req_data.field_type);
self.stats.collect(val1);
}
} else {
for _val in agg_accessor.column_block_accessor.iter_vals() {
for _val in req_data.column_block_accessor.iter_vals() {
// we ignore the value and simply record that we got something
self.stats.collect(0.0);
}
@@ -245,27 +235,28 @@ impl SegmentAggregationCollector for SegmentStatsCollector {
#[inline]
fn add_intermediate_aggregation_result(
self: Box<Self>,
agg_with_accessor: &AggregationsWithAccessor,
agg_data: &AggregationsSegmentCtx,
results: &mut IntermediateAggregationResults,
) -> crate::Result<()> {
let name = agg_with_accessor.aggs.keys[self.accessor_idx].to_string();
let req = agg_data.get_metric_req_data(self.accessor_idx);
let name = req.name.clone();
let intermediate_metric_result = match self.collecting_for {
SegmentStatsType::Average => {
let intermediate_metric_result = match req.collecting_for {
StatsType::Average => {
IntermediateMetricResult::Average(IntermediateAverage::from_collector(*self))
}
SegmentStatsType::Count => {
StatsType::Count => {
IntermediateMetricResult::Count(IntermediateCount::from_collector(*self))
}
SegmentStatsType::Max => {
IntermediateMetricResult::Max(IntermediateMax::from_collector(*self))
}
SegmentStatsType::Min => {
IntermediateMetricResult::Min(IntermediateMin::from_collector(*self))
}
SegmentStatsType::Stats => IntermediateMetricResult::Stats(self.stats),
SegmentStatsType::Sum => {
IntermediateMetricResult::Sum(IntermediateSum::from_collector(*self))
StatsType::Max => IntermediateMetricResult::Max(IntermediateMax::from_collector(*self)),
StatsType::Min => IntermediateMetricResult::Min(IntermediateMin::from_collector(*self)),
StatsType::Stats => IntermediateMetricResult::Stats(self.stats),
StatsType::Sum => IntermediateMetricResult::Sum(IntermediateSum::from_collector(*self)),
_ => {
return Err(TantivyError::InvalidArgument(format!(
"Unsupported stats type for stats aggregation: {:?}",
req.collecting_for
)))
}
};
@@ -281,23 +272,23 @@ impl SegmentAggregationCollector for SegmentStatsCollector {
fn collect(
&mut self,
doc: crate::DocId,
agg_with_accessor: &mut AggregationsWithAccessor,
agg_data: &mut AggregationsSegmentCtx,
) -> crate::Result<()> {
let field = &agg_with_accessor.aggs.values[self.accessor_idx].accessor;
if let Some(missing) = self.missing {
let req_data = agg_data.get_metric_req_data(self.accessor_idx);
if let Some(missing) = req_data.missing_u64 {
let mut has_val = false;
for val in field.values_for_doc(doc) {
let val1 = f64_from_fastfield_u64(val, &self.field_type);
for val in req_data.accessor.values_for_doc(doc) {
let val1 = f64_from_fastfield_u64(val, &req_data.field_type);
self.stats.collect(val1);
has_val = true;
}
if !has_val {
self.stats
.collect(f64_from_fastfield_u64(missing, &self.field_type));
.collect(f64_from_fastfield_u64(missing, &req_data.field_type));
}
} else {
for val in field.values_for_doc(doc) {
let val1 = f64_from_fastfield_u64(val, &self.field_type);
for val in req_data.accessor.values_for_doc(doc) {
let val1 = f64_from_fastfield_u64(val, &req_data.field_type);
self.stats.collect(val1);
}
}
@@ -309,10 +300,10 @@ impl SegmentAggregationCollector for SegmentStatsCollector {
fn collect_block(
&mut self,
docs: &[crate::DocId],
agg_with_accessor: &mut AggregationsWithAccessor,
agg_data: &mut AggregationsSegmentCtx,
) -> crate::Result<()> {
let field = &mut agg_with_accessor.aggs.values[self.accessor_idx];
self.collect_block_with_field(docs, field);
let req_data = agg_data.get_metric_req_data_mut(self.accessor_idx);
self.collect_block_with_field(docs, req_data);
Ok(())
}
}

63
src/aggregation/metric/top_hits.rs
View File

@@ -9,6 +9,7 @@ use serde::ser::SerializeMap;
use serde::{Deserialize, Deserializer, Serialize, Serializer};
use super::{TopHitsMetricResult, TopHitsVecEntry};
use crate::aggregation::agg_data::AggregationsSegmentCtx;
use crate::aggregation::bucket::Order;
use crate::aggregation::intermediate_agg_result::{
IntermediateAggregationResult, IntermediateMetricResult,
@@ -18,6 +19,30 @@ use crate::aggregation::AggregationError;
use crate::collector::TopNComputer;
use crate::schema::OwnedValue;
use crate::{DocAddress, DocId, SegmentOrdinal};
// duplicate import removed; already imported above
/// Contains all information required by the TopHitsSegmentCollector to perform the
/// top_hits aggregation on a segment.
#[derive(Default)]
pub struct TopHitsAggReqData {
/// The accessors to access the fast field values.
pub accessors: Vec<(Column<u64>, ColumnType)>,
/// The accessors to access the fast field values for retrieving document fields.
pub value_accessors: HashMap<String, Vec<DynamicColumn>>,
/// The ordinal of the segment this request data is for.
pub segment_ordinal: SegmentOrdinal,
/// The name of the aggregation.
pub name: String,
/// The top_hits aggregation request.
pub req: TopHitsAggregationReq,
}
impl TopHitsAggReqData {
/// Estimate the memory consumption of this struct in bytes.
pub fn get_memory_consumption(&self) -> usize {
std::mem::size_of::<Self>()
}
}
/// # Top Hits
///
@@ -229,6 +254,7 @@ impl TopHitsAggregationReq {
self.sort
.iter()
.map(|KeyOrder { field, .. }| field.as_str())
.chain(self.doc_value_fields.iter().map(|s| s.as_str()))
.collect()
}
@@ -565,23 +591,18 @@ impl TopHitsSegmentCollector {
impl SegmentAggregationCollector for TopHitsSegmentCollector {
fn add_intermediate_aggregation_result(
self: Box<Self>,
agg_with_accessor: &crate::aggregation::agg_req_with_accessor::AggregationsWithAccessor,
agg_data: &AggregationsSegmentCtx,
results: &mut crate::aggregation::intermediate_agg_result::IntermediateAggregationResults,
) -> crate::Result<()> {
let name = agg_with_accessor.aggs.keys[self.accessor_idx].to_string();
let req_data = agg_data.get_top_hits_req_data(self.accessor_idx);
let value_accessors = &agg_with_accessor.aggs.values[self.accessor_idx].value_accessors;
let tophits_req = &agg_with_accessor.aggs.values[self.accessor_idx]
.agg
.agg
.as_top_hits()
.expect("aggregation request must be of type top hits");
let value_accessors = &req_data.value_accessors;
let intermediate_result = IntermediateMetricResult::TopHits(
self.into_top_hits_collector(value_accessors, tophits_req),
self.into_top_hits_collector(value_accessors, &req_data.req),
);
results.push(
name,
req_data.name.to_string(),
IntermediateAggregationResult::Metric(intermediate_result),
)
}
@@ -590,32 +611,22 @@ impl SegmentAggregationCollector for TopHitsSegmentCollector {
fn collect(
&mut self,
doc_id: crate::DocId,
agg_with_accessor: &mut crate::aggregation::agg_req_with_accessor::AggregationsWithAccessor,
agg_data: &mut AggregationsSegmentCtx,
) -> crate::Result<()> {
let tophits_req = &agg_with_accessor.aggs.values[self.accessor_idx]
.agg
.agg
.as_top_hits()
.expect("aggregation request must be of type top hits");
let accessors = &agg_with_accessor.aggs.values[self.accessor_idx].accessors;
self.collect_with(doc_id, tophits_req, accessors)?;
let req_data = agg_data.get_top_hits_req_data(self.accessor_idx);
self.collect_with(doc_id, &req_data.req, &req_data.accessors)?;
Ok(())
}
fn collect_block(
&mut self,
docs: &[crate::DocId],
agg_with_accessor: &mut crate::aggregation::agg_req_with_accessor::AggregationsWithAccessor,
agg_data: &mut AggregationsSegmentCtx,
) -> crate::Result<()> {
let tophits_req = &agg_with_accessor.aggs.values[self.accessor_idx]
.agg
.agg
.as_top_hits()
.expect("aggregation request must be of type top hits");
let accessors = &agg_with_accessor.aggs.values[self.accessor_idx].accessors;
let req_data = agg_data.get_top_hits_req_data(self.accessor_idx);
// TODO: Consider getting fields with the column block accessor.
for doc in docs {
self.collect_with(*doc, tophits_req, accessors)?;
self.collect_with(*doc, &req_data.req, &req_data.accessors)?;
}
Ok(())
}

105
src/aggregation/mod.rs
View File

@@ -127,9 +127,10 @@
//! [`AggregationResults`](agg_result::AggregationResults) via the
//! [`into_final_result`](intermediate_agg_result::IntermediateAggregationResults::into_final_result) method.
mod accessor_helpers;
mod agg_data;
mod agg_limits;
pub mod agg_req;
mod agg_req_with_accessor;
pub mod agg_result;
pub mod bucket;
mod buf_collector;
@@ -140,7 +141,6 @@ pub mod intermediate_agg_result;
pub mod metric;
mod segment_agg_result;
use std::collections::HashMap;
use std::fmt::Display;
#[cfg(test)]
@@ -160,6 +160,28 @@ use itertools::Itertools;
use serde::de::{self, Visitor};
use serde::{Deserialize, Deserializer, Serialize};
use crate::tokenizer::TokenizerManager;
/// Context parameters for aggregation execution
///
/// This struct holds shared resources needed during aggregation execution:
/// - `limits`: Memory and bucket limits for the aggregation
/// - `tokenizers`: TokenizerManager for parsing query strings in filter aggregations
#[derive(Clone, Default)]
pub struct AggContextParams {
/// Aggregation limits (memory and bucket count)
pub limits: AggregationLimitsGuard,
/// Tokenizer manager for query string parsing
pub tokenizers: TokenizerManager,
}
impl AggContextParams {
/// Create new aggregation context parameters
pub fn new(limits: AggregationLimitsGuard, tokenizers: TokenizerManager) -> Self {
Self { limits, tokenizers }
}
}
fn parse_str_into_f64<E: de::Error>(value: &str) -> Result<f64, E> {
let parsed = value
.parse::<f64>()
@@ -257,80 +279,6 @@ where D: Deserializer<'de> {
deserializer.deserialize_any(StringOrFloatVisitor)
}
/// Represents an associative array `(key => values)` in a very efficient manner.
#[derive(PartialEq, Serialize, Deserialize)]
pub(crate) struct VecWithNames<T> {
pub(crate) values: Vec<T>,
keys: Vec<String>,
}
impl<T: Clone> Clone for VecWithNames<T> {
fn clone(&self) -> Self {
Self {
values: self.values.clone(),
keys: self.keys.clone(),
}
}
}
impl<T> Default for VecWithNames<T> {
fn default() -> Self {
Self {
values: Default::default(),
keys: Default::default(),
}
}
}
impl<T: std::fmt::Debug> std::fmt::Debug for VecWithNames<T> {
fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
f.debug_map().entries(self.iter()).finish()
}
}
impl<T> From<HashMap<String, T>> for VecWithNames<T> {
fn from(map: HashMap<String, T>) -> Self {
VecWithNames::from_entries(map.into_iter().collect_vec())
}
}
impl<T> VecWithNames<T> {
fn from_entries(mut entries: Vec<(String, T)>) -> Self {
// Sort to ensure order of elements match across multiple instances
entries.sort_by(|left, right| left.0.cmp(&right.0));
let mut data = Vec::with_capacity(entries.len());
let mut data_names = Vec::with_capacity(entries.len());
for entry in entries {
data_names.push(entry.0);
data.push(entry.1);
}
VecWithNames {
values: data,
keys: data_names,
}
}
fn iter(&self) -> impl Iterator<Item = (&str, &T)> + '_ {
self.keys().zip(self.values.iter())
}
fn keys(&self) -> impl Iterator<Item = &str> + '_ {
self.keys.iter().map(|key| key.as_str())
}
fn values_mut(&mut self) -> impl Iterator<Item = &mut T> + '_ {
self.values.iter_mut()
}
fn is_empty(&self) -> bool {
self.keys.is_empty()
}
fn len(&self) -> usize {
self.keys.len()
}
fn get(&self, name: &str) -> Option<&T> {
self.keys()
.position(|key| key == name)
.map(|pos| &self.values[pos])
}
}
/// The serialized key is used in a `HashMap`.
pub type SerializedKey = String;
@@ -464,7 +412,10 @@ mod tests {
query: Option<(&str, &str)>,
limits: AggregationLimitsGuard,
) -> crate::Result<Value> {
let collector = AggregationCollector::from_aggs(agg_req, limits);
let collector = AggregationCollector::from_aggs(
agg_req,
AggContextParams::new(limits, index.tokenizers().clone()),
);
let reader = index.reader()?;
let searcher = reader.searcher();

175
src/aggregation/segment_agg_result.rs
View File

@@ -6,48 +6,41 @@
use std::fmt::Debug;
pub(crate) use super::agg_limits::AggregationLimitsGuard;
use super::agg_req::AggregationVariants;
use super::agg_req_with_accessor::{AggregationWithAccessor, AggregationsWithAccessor};
use super::bucket::{SegmentHistogramCollector, SegmentRangeCollector, SegmentTermCollector};
use super::intermediate_agg_result::IntermediateAggregationResults;
use super::metric::{
AverageAggregation, CountAggregation, ExtendedStatsAggregation, MaxAggregation, MinAggregation,
SegmentPercentilesCollector, SegmentStatsCollector, SegmentStatsType, StatsAggregation,
SumAggregation,
};
use crate::aggregation::bucket::TermMissingAgg;
use crate::aggregation::metric::{
CardinalityAggregationReq, SegmentCardinalityCollector, SegmentExtendedStatsCollector,
TopHitsSegmentCollector,
};
use crate::aggregation::agg_data::AggregationsSegmentCtx;
pub(crate) trait SegmentAggregationCollector: CollectorClone + Debug {
/// A SegmentAggregationCollector is used to collect aggregation results.
pub trait SegmentAggregationCollector: CollectorClone + Debug {
fn add_intermediate_aggregation_result(
self: Box<Self>,
agg_with_accessor: &AggregationsWithAccessor,
agg_data: &AggregationsSegmentCtx,
results: &mut IntermediateAggregationResults,
) -> crate::Result<()>;
#[inline]
fn collect(
&mut self,
doc: crate::DocId,
agg_with_accessor: &mut AggregationsWithAccessor,
) -> crate::Result<()>;
agg_data: &mut AggregationsSegmentCtx,
) -> crate::Result<()> {
self.collect_block(&[doc], agg_data)
}
fn collect_block(
&mut self,
docs: &[crate::DocId],
agg_with_accessor: &mut AggregationsWithAccessor,
agg_data: &mut AggregationsSegmentCtx,
) -> crate::Result<()>;
/// Finalize method. Some Aggregator collect blocks of docs before calling `collect_block`.
/// This method ensures those staged docs will be collected.
fn flush(&mut self, _agg_with_accessor: &mut AggregationsWithAccessor) -> crate::Result<()> {
fn flush(&mut self, _agg_data: &mut AggregationsSegmentCtx) -> crate::Result<()> {
Ok(())
}
}
pub(crate) trait CollectorClone {
/// A helper trait to enable cloning of Box<dyn SegmentAggregationCollector>
pub trait CollectorClone {
fn clone_box(&self) -> Box<dyn SegmentAggregationCollector>;
}
@@ -65,119 +58,6 @@ impl Clone for Box<dyn SegmentAggregationCollector> {
}
}
pub(crate) fn build_segment_agg_collector(
req: &mut AggregationsWithAccessor,
) -> crate::Result<Box<dyn SegmentAggregationCollector>> {
// Single collector special case
if req.aggs.len() == 1 {
let req = &mut req.aggs.values[0];
let accessor_idx = 0;
return build_single_agg_segment_collector(req, accessor_idx);
}
let agg = GenericSegmentAggregationResultsCollector::from_req_and_validate(req)?;
Ok(Box::new(agg))
}
pub(crate) fn build_single_agg_segment_collector(
req: &mut AggregationWithAccessor,
accessor_idx: usize,
) -> crate::Result<Box<dyn SegmentAggregationCollector>> {
use AggregationVariants::*;
match &req.agg.agg {
Terms(terms_req) => {
if req.accessors.is_empty() {
Ok(Box::new(SegmentTermCollector::from_req_and_validate(
terms_req,
&mut req.sub_aggregation,
req.field_type,
accessor_idx,
)?))
} else {
Ok(Box::new(TermMissingAgg::new(
accessor_idx,
&mut req.sub_aggregation,
)?))
}
}
Range(range_req) => Ok(Box::new(SegmentRangeCollector::from_req_and_validate(
range_req,
&mut req.sub_aggregation,
&mut req.limits,
req.field_type,
accessor_idx,
)?)),
Histogram(histogram) => Ok(Box::new(SegmentHistogramCollector::from_req_and_validate(
histogram.clone(),
&mut req.sub_aggregation,
req.field_type,
accessor_idx,
)?)),
DateHistogram(histogram) => Ok(Box::new(SegmentHistogramCollector::from_req_and_validate(
histogram.to_histogram_req()?,
&mut req.sub_aggregation,
req.field_type,
accessor_idx,
)?)),
Average(AverageAggregation { missing, .. }) => {
Ok(Box::new(SegmentStatsCollector::from_req(
req.field_type,
SegmentStatsType::Average,
accessor_idx,
*missing,
)))
}
Count(CountAggregation { missing, .. }) => Ok(Box::new(SegmentStatsCollector::from_req(
req.field_type,
SegmentStatsType::Count,
accessor_idx,
*missing,
))),
Max(MaxAggregation { missing, .. }) => Ok(Box::new(SegmentStatsCollector::from_req(
req.field_type,
SegmentStatsType::Max,
accessor_idx,
*missing,
))),
Min(MinAggregation { missing, .. }) => Ok(Box::new(SegmentStatsCollector::from_req(
req.field_type,
SegmentStatsType::Min,
accessor_idx,
*missing,
))),
Stats(StatsAggregation { missing, .. }) => Ok(Box::new(SegmentStatsCollector::from_req(
req.field_type,
SegmentStatsType::Stats,
accessor_idx,
*missing,
))),
ExtendedStats(ExtendedStatsAggregation { missing, sigma, .. }) => Ok(Box::new(
SegmentExtendedStatsCollector::from_req(req.field_type, *sigma, accessor_idx, *missing),
)),
Sum(SumAggregation { missing, .. }) => Ok(Box::new(SegmentStatsCollector::from_req(
req.field_type,
SegmentStatsType::Sum,
accessor_idx,
*missing,
))),
Percentiles(percentiles_req) => Ok(Box::new(
SegmentPercentilesCollector::from_req_and_validate(
percentiles_req,
req.field_type,
accessor_idx,
)?,
)),
TopHits(top_hits_req) => Ok(Box::new(TopHitsSegmentCollector::from_req(
top_hits_req,
accessor_idx,
req.segment_ordinal,
))),
Cardinality(CardinalityAggregationReq { missing, .. }) => Ok(Box::new(
SegmentCardinalityCollector::from_req(req.field_type, accessor_idx, missing),
)),
}
}
#[derive(Clone, Default)]
/// The GenericSegmentAggregationResultsCollector is the generic version of the collector, which
/// can handle arbitrary complexity of sub-aggregations. Ideally we never have to pick this one
@@ -197,11 +77,11 @@ impl Debug for GenericSegmentAggregationResultsCollector {
impl SegmentAggregationCollector for GenericSegmentAggregationResultsCollector {
fn add_intermediate_aggregation_result(
self: Box<Self>,
agg_with_accessor: &AggregationsWithAccessor,
agg_data: &AggregationsSegmentCtx,
results: &mut IntermediateAggregationResults,
) -> crate::Result<()> {
for agg in self.aggs {
agg.add_intermediate_aggregation_result(agg_with_accessor, results)?;
agg.add_intermediate_aggregation_result(agg_data, results)?;
}
Ok(())
@@ -210,9 +90,9 @@ impl SegmentAggregationCollector for GenericSegmentAggregationResultsCollector {
fn collect(
&mut self,
doc: crate::DocId,
agg_with_accessor: &mut AggregationsWithAccessor,
agg_data: &mut AggregationsSegmentCtx,
) -> crate::Result<()> {
self.collect_block(&[doc], agg_with_accessor)?;
self.collect_block(&[doc], agg_data)?;
Ok(())
}
@@ -220,32 +100,19 @@ impl SegmentAggregationCollector for GenericSegmentAggregationResultsCollector {
fn collect_block(
&mut self,
docs: &[crate::DocId],
agg_with_accessor: &mut AggregationsWithAccessor,
agg_data: &mut AggregationsSegmentCtx,
) -> crate::Result<()> {
for collector in &mut self.aggs {
collector.collect_block(docs, agg_with_accessor)?;
collector.collect_block(docs, agg_data)?;
}
Ok(())
}
fn flush(&mut self, agg_with_accessor: &mut AggregationsWithAccessor) -> crate::Result<()> {
fn flush(&mut self, agg_data: &mut AggregationsSegmentCtx) -> crate::Result<()> {
for collector in &mut self.aggs {
collector.flush(agg_with_accessor)?;
collector.flush(agg_data)?;
}
Ok(())
}
}
impl GenericSegmentAggregationResultsCollector {
pub(crate) fn from_req_and_validate(req: &mut AggregationsWithAccessor) -> crate::Result<Self> {
let aggs = req
.aggs
.values_mut()
.enumerate()
.map(|(accessor_idx, req)| build_single_agg_segment_collector(req, accessor_idx))
.collect::<crate::Result<Vec<Box<dyn SegmentAggregationCollector>>>>()?;
Ok(GenericSegmentAggregationResultsCollector { aggs })
}
}

5
src/collector/facet_collector.rs
View File

@@ -484,7 +484,6 @@ impl FacetCounts {
#[cfg(test)]
mod tests {
use std::collections::BTreeSet;
use std::iter;
use columnar::Dictionary;
use rand::distributions::Uniform;
@@ -739,7 +738,7 @@ mod tests {
.flat_map(|(c, count)| {
let facet = Facet::from(&format!("/facet/{c}"));
let doc = doc!(facet_field => facet);
iter::repeat(doc).take(count)
std::iter::repeat_n(doc, count)
})
.map(|mut doc| {
doc.add_facet(
@@ -787,7 +786,7 @@ mod tests {
.flat_map(|(c, count)| {
let facet = Facet::from(&format!("/facet/{c}"));
let doc = doc!(facet_field => facet);
iter::repeat(doc).take(count)
std::iter::repeat_n(doc, count)
})
.collect();

358
src/collector/top_score_collector.rs
View File

@@ -970,7 +970,7 @@ impl<Score, D, const R: bool> From<TopNComputerDeser<Score, D, R>> for TopNCompu
}
}
impl<Score, D, const R: bool> TopNComputer<Score, D, R>
impl<Score, D, const REVERSE_ORDER: bool> TopNComputer<Score, D, REVERSE_ORDER>
where
Score: PartialOrd + Clone,
D: Ord,
@@ -991,7 +991,10 @@ where
#[inline]
pub fn push(&mut self, feature: Score, doc: D) {
if let Some(last_median) = self.threshold.clone() {
if feature < last_median {
if !REVERSE_ORDER && feature > last_median {
return;
}
if REVERSE_ORDER && feature < last_median {
return;
}
}
@@ -1026,7 +1029,7 @@ where
}
/// Returns the top n elements in sorted order.
pub fn into_sorted_vec(mut self) -> Vec<ComparableDoc<Score, D, R>> {
pub fn into_sorted_vec(mut self) -> Vec<ComparableDoc<Score, D, REVERSE_ORDER>> {
if self.buffer.len() > self.top_n {
self.truncate_top_n();
}
@@ -1037,7 +1040,7 @@ where
/// Returns the top n elements in stored order.
/// Useful if you do not need the elements in sorted order,
/// for example when merging the results of multiple segments.
pub fn into_vec(mut self) -> Vec<ComparableDoc<Score, D, R>> {
pub fn into_vec(mut self) -> Vec<ComparableDoc<Score, D, REVERSE_ORDER>> {
if self.buffer.len() > self.top_n {
self.truncate_top_n();
}
@@ -1047,9 +1050,11 @@ where
#[cfg(test)]
mod tests {
use proptest::prelude::*;
use super::{TopDocs, TopNComputer};
use crate::collector::top_collector::ComparableDoc;
use crate::collector::Collector;
use crate::collector::{Collector, DocSetCollector};
use crate::query::{AllQuery, Query, QueryParser};
use crate::schema::{Field, Schema, FAST, STORED, TEXT};
use crate::time::format_description::well_known::Rfc3339;
@@ -1144,6 +1149,44 @@ mod tests {
}
}
proptest! {
#[test]
fn test_topn_computer_asc_prop(
limit in 0..10_usize,
docs in proptest::collection::vec((0..100_u64, 0..100_u64), 0..100_usize),
) {
let mut computer: TopNComputer<_, _, false> = TopNComputer::new(limit);
for (feature, doc) in &docs {
computer.push(*feature, *doc);
}
let mut comparable_docs = docs.into_iter().map(|(feature, doc)| ComparableDoc { feature, doc }).collect::<Vec<_>>();
comparable_docs.sort();
comparable_docs.truncate(limit);
prop_assert_eq!(
computer.into_sorted_vec(),
comparable_docs,
);
}
#[test]
fn test_topn_computer_desc_prop(
limit in 0..10_usize,
docs in proptest::collection::vec((0..100_u64, 0..100_u64), 0..100_usize),
) {
let mut computer: TopNComputer<_, _, true> = TopNComputer::new(limit);
for (feature, doc) in &docs {
computer.push(*feature, *doc);
}
let mut comparable_docs = docs.into_iter().map(|(feature, doc)| ComparableDoc { feature, doc }).collect::<Vec<_>>();
comparable_docs.sort();
comparable_docs.truncate(limit);
prop_assert_eq!(
computer.into_sorted_vec(),
comparable_docs,
);
}
}
#[test]
fn test_top_collector_not_at_capacity_without_offset() -> crate::Result<()> {
let index = make_index()?;
@@ -1250,6 +1293,220 @@ mod tests {
assert_eq!(page_0, &page_2[..page_0.len()]);
}
proptest! {
#![proptest_config(ProptestConfig::with_cases(20))]
/// Build multiple segments with equal-scoring docs and verify stable ordering
/// across pages when increasing limit or offset.
#[test]
fn proptest_stable_ordering_across_segments_with_pagination(
docs_per_segment in proptest::collection::vec(1usize..50, 2..5)
) {
use crate::indexer::NoMergePolicy;
// Build an index with multiple segments; all docs will have the same score using AllQuery.
let mut schema_builder = Schema::builder();
let text = schema_builder.add_text_field("text", TEXT);
let schema = schema_builder.build();
let index = Index::create_in_ram(schema);
let mut writer = index.writer_for_tests().unwrap();
writer.set_merge_policy(Box::new(NoMergePolicy));
for num_docs in &docs_per_segment {
for _ in 0..*num_docs {
writer.add_document(doc!(text => "x")).unwrap();
}
writer.commit().unwrap();
}
let reader = index.reader().unwrap();
let searcher = reader.searcher();
let total_docs: usize = docs_per_segment.iter().sum();
// Full result set, first assert all scores are identical.
let full_with_scores: Vec<(Score, DocAddress)> = searcher
.search(&AllQuery, &TopDocs::with_limit(total_docs))
.unwrap();
// Sanity: at least one document was returned.
prop_assert!(!full_with_scores.is_empty());
let first_score = full_with_scores[0].0;
prop_assert!(full_with_scores.iter().all(|(score, _)| *score == first_score));
// Keep only the addresses for the remaining checks.
let full: Vec<DocAddress> = full_with_scores
.into_iter()
.map(|(_score, addr)| addr)
.collect();
// Sanity: we actually created multiple segments and have documents.
prop_assert!(docs_per_segment.len() >= 2);
prop_assert!(total_docs >= 2);
// 1) Increasing limit should preserve prefix ordering.
for k in 1..=total_docs {
let page: Vec<DocAddress> = searcher
.search(&AllQuery, &TopDocs::with_limit(k))
.unwrap()
.into_iter()
.map(|(_score, addr)| addr)
.collect();
prop_assert_eq!(page, full[..k].to_vec());
}
// 2) Offset + limit pages should always match the corresponding slice.
// For each offset, check three representative page sizes:
// - first page (size 1)
// - a middle page (roughly half of remaining)
// - the last page (size = remaining)
for offset in 0..total_docs {
let remaining = total_docs - offset;
let assert_page_eq = |limit: usize| -> proptest::test_runner::TestCaseResult {
let page: Vec<DocAddress> = searcher
.search(&AllQuery, &TopDocs::with_limit(limit).and_offset(offset))
.unwrap()
.into_iter()
.map(|(_score, addr)| addr)
.collect();
prop_assert_eq!(page, full[offset..offset + limit].to_vec());
Ok(())
};
// Smallest page.
assert_page_eq(1)?;
// A middle-sized page (dedupes to 1 if remaining == 1).
assert_page_eq((remaining / 2).max(1))?;
// Largest page for this offset.
assert_page_eq(remaining)?;
}
// 3) Concatenating fixed-size pages by offset reproduces the full order.
for page_size in 1..=total_docs.min(5) {
let mut concat: Vec<DocAddress> = Vec::new();
let mut offset = 0;
while offset < total_docs {
let size = page_size.min(total_docs - offset);
let page: Vec<DocAddress> = searcher
.search(&AllQuery, &TopDocs::with_limit(size).and_offset(offset))
.unwrap()
.into_iter()
.map(|(_score, addr)| addr)
.collect();
concat.extend(page);
offset += size;
}
// Avoid moving `full` across loop iterations.
prop_assert_eq!(concat, full.clone());
}
}
}
proptest! {
#![proptest_config(ProptestConfig::with_cases(20))]
/// Build multiple segments with same-scoring term matches and verify stable ordering
/// across pages for a real scoring query (TermQuery with identical TF and fieldnorm).
#[test]
fn proptest_stable_ordering_across_segments_with_term_query_and_pagination(
docs_per_segment in proptest::collection::vec(1usize..50, 2..5)
) {
use crate::indexer::NoMergePolicy;
use crate::schema::IndexRecordOption;
use crate::query::TermQuery;
use crate::Term;
// Build an index with multiple segments; each doc has exactly one token "x",
// ensuring equal BM25 scores across all matching docs (same TF=1 and fieldnorm=1).
let mut schema_builder = Schema::builder();
let text = schema_builder.add_text_field("text", TEXT);
let schema = schema_builder.build();
let index = Index::create_in_ram(schema);
let mut writer = index.writer_for_tests().unwrap();
writer.set_merge_policy(Box::new(NoMergePolicy));
for num_docs in &docs_per_segment {
for _ in 0..*num_docs {
writer.add_document(doc!(text => "x")).unwrap();
}
writer.commit().unwrap();
}
let reader = index.reader().unwrap();
let searcher = reader.searcher();
let total_docs: usize = docs_per_segment.iter().sum();
let term = Term::from_field_text(text, "x");
let tq = TermQuery::new(term, IndexRecordOption::WithFreqs);
// Full result set, first assert all scores are identical across docs.
let full_with_scores: Vec<(Score, DocAddress)> = searcher
.search(&tq, &TopDocs::with_limit(total_docs))
.unwrap();
// Sanity: at least one document was returned.
prop_assert!(!full_with_scores.is_empty());
let first_score = full_with_scores[0].0;
prop_assert!(full_with_scores.iter().all(|(score, _)| *score == first_score));
// Keep only the addresses for the remaining checks.
let full: Vec<DocAddress> = full_with_scores
.into_iter()
.map(|(_score, addr)| addr)
.collect();
// Sanity: we actually created multiple segments and have documents.
prop_assert!(docs_per_segment.len() >= 2);
prop_assert!(total_docs >= 2);
// 1) Increasing limit should preserve prefix ordering.
for k in 1..=total_docs {
let page: Vec<DocAddress> = searcher
.search(&tq, &TopDocs::with_limit(k))
.unwrap()
.into_iter()
.map(|(_score, addr)| addr)
.collect();
prop_assert_eq!(page, full[..k].to_vec());
}
// 2) Offset + limit pages should always match the corresponding slice.
// Check three representative page sizes for each offset: 1, ~half, and remaining.
for offset in 0..total_docs {
let remaining = total_docs - offset;
let assert_page_eq = |limit: usize| -> proptest::test_runner::TestCaseResult {
let page: Vec<DocAddress> = searcher
.search(&tq, &TopDocs::with_limit(limit).and_offset(offset))
.unwrap()
.into_iter()
.map(|(_score, addr)| addr)
.collect();
prop_assert_eq!(page, full[offset..offset + limit].to_vec());
Ok(())
};
assert_page_eq(1)?;
assert_page_eq((remaining / 2).max(1))?;
assert_page_eq(remaining)?;
}
// 3) Concatenating fixed-size pages by offset reproduces the full order.
for page_size in 1..=total_docs.min(5) {
let mut concat: Vec<DocAddress> = Vec::new();
let mut offset = 0;
while offset < total_docs {
let size = page_size.min(total_docs - offset);
let page: Vec<DocAddress> = searcher
.search(&tq, &TopDocs::with_limit(size).and_offset(offset))
.unwrap()
.into_iter()
.map(|(_score, addr)| addr)
.collect();
concat.extend(page);
offset += size;
}
prop_assert_eq!(concat, full.clone());
}
}
}
#[test]
#[should_panic]
fn test_top_0() {
@@ -1486,6 +1743,72 @@ mod tests {
Ok(())
}
proptest! {
#[test]
fn test_top_field_collect_string_prop(
order in prop_oneof!(Just(Order::Desc), Just(Order::Asc)),
limit in 1..256_usize,
offset in 0..256_usize,
segments_terms in
proptest::collection::vec(
proptest::collection::vec(0..32_u8, 1..32_usize),
0..8_usize,
)
) {
let mut schema_builder = Schema::builder();
let city = schema_builder.add_text_field("city", TEXT | FAST);
let schema = schema_builder.build();
let index = Index::create_in_ram(schema);
let mut index_writer = index.writer_for_tests()?;
// A Vec<Vec<u8>>, where the outer Vec represents segments, and the inner Vec
// represents terms.
for segment_terms in segments_terms.into_iter() {
for term in segment_terms.into_iter() {
let term = format!("{term:0>3}");
index_writer.add_document(doc!(
city => term,
))?;
}
index_writer.commit()?;
}
let searcher = index.reader()?.searcher();
let top_n_results = searcher.search(&AllQuery, &TopDocs::with_limit(limit)
.and_offset(offset)
.order_by_string_fast_field("city", order.clone()))?;
let all_results = searcher.search(&AllQuery, &DocSetCollector)?.into_iter().map(|doc_address| {
// Get the term for this address.
// NOTE: We can't determine the SegmentIds that will be generated for Segments
// ahead of time, so we can't pre-compute the expected `DocAddress`es.
let column = searcher.segment_readers()[doc_address.segment_ord as usize].fast_fields().str("city").unwrap().unwrap();
let term_ord = column.term_ords(doc_address.doc_id).next().unwrap();
let mut city = Vec::new();
column.dictionary().ord_to_term(term_ord, &mut city).unwrap();
(String::try_from(city).unwrap(), doc_address)
});
// Using the TopDocs collector should always be equivalent to sorting, skipping the
// offset, and then taking the limit.
let sorted_docs: Vec<_> = if order.is_desc() {
let mut comparable_docs: Vec<ComparableDoc<_, _, true>> =
all_results.into_iter().map(|(feature, doc)| ComparableDoc { feature, doc}).collect();
comparable_docs.sort();
comparable_docs.into_iter().map(|cd| (cd.feature, cd.doc)).collect()
} else {
let mut comparable_docs: Vec<ComparableDoc<_, _, false>> =
all_results.into_iter().map(|(feature, doc)| ComparableDoc { feature, doc}).collect();
comparable_docs.sort();
comparable_docs.into_iter().map(|cd| (cd.feature, cd.doc)).collect()
};
let expected_docs = sorted_docs.into_iter().skip(offset).take(limit).collect::<Vec<_>>();
prop_assert_eq!(
expected_docs,
top_n_results
);
}
}
#[test]
#[should_panic]
fn test_field_does_not_exist() {
@@ -1645,4 +1968,29 @@ mod tests {
);
Ok(())
}
#[test]
fn test_topn_computer_asc() {
let mut computer: TopNComputer<u32, u32, false> = TopNComputer::new(2);
computer.push(1u32, 1u32);
computer.push(2u32, 2u32);
computer.push(3u32, 3u32);
computer.push(2u32, 4u32);
computer.push(4u32, 5u32);
computer.push(1u32, 6u32);
assert_eq!(
computer.into_sorted_vec(),
&[
ComparableDoc {
feature: 1u32,
doc: 1u32,
},
ComparableDoc {
feature: 1u32,
doc: 6u32,
}
]
);
}
}

2
src/compat_tests.rs
View File

@@ -30,7 +30,7 @@ fn create_format() {
}
fn path_for_version(version: &str) -> String {
format!("./tests/compat_tests_data/index_v{}/", version)
format!("./tests/compat_tests_data/index_v{version}/")
}
/// feature flag quickwit uses a different dictionary type

3
src/core/executor.rs
View File

@@ -65,8 +65,7 @@ impl Executor {
if let Err(err) = fruit_sender_ref.send((idx, fruit)) {
error!(
"Failed to send search task. It probably means all search \
threads have panicked. {:?}",
err
threads have panicked. {err:?}"
);
}
});

92
src/core/json_utils.rs
View File

@@ -1,3 +1,4 @@
use columnar::NumericalValue;
use common::json_path_writer::{JSON_END_OF_PATH, JSON_PATH_SEGMENT_SEP};
use common::{replace_in_place, JsonPathWriter};
use rustc_hash::FxHashMap;
@@ -152,7 +153,7 @@ pub(crate) fn index_json_value<'a, V: Value<'a>>(
if let Ok(i64_val) = val.try_into() {
term_buffer.append_type_and_fast_value::<i64>(i64_val);
} else {
term_buffer.append_type_and_fast_value(val);
term_buffer.append_type_and_fast_value::<u64>(val);
}
postings_writer.subscribe(doc, 0u32, term_buffer, ctx);
}
@@ -166,12 +167,30 @@ pub(crate) fn index_json_value<'a, V: Value<'a>>(
postings_writer.subscribe(doc, 0u32, term_buffer, ctx);
}
ReferenceValueLeaf::F64(val) => {
if !val.is_finite() {
return;
};
set_path_id(
term_buffer,
ctx.path_to_unordered_id
.get_or_allocate_unordered_id(json_path_writer.as_str()),
);
term_buffer.append_type_and_fast_value(val);
// Normalize here is important.
// In the inverted index, we coerce all numerical values to their canonical
// representation.
//
// (We do the same thing on the query side)
match NumericalValue::F64(val).normalize() {
NumericalValue::I64(val_i64) => {
term_buffer.append_type_and_fast_value::<i64>(val_i64);
}
NumericalValue::U64(val_u64) => {
term_buffer.append_type_and_fast_value::<u64>(val_u64);
}
NumericalValue::F64(val_f64) => {
term_buffer.append_type_and_fast_value::<f64>(val_f64);
}
}
postings_writer.subscribe(doc, 0u32, term_buffer, ctx);
}
ReferenceValueLeaf::Bool(val) => {
@@ -241,8 +260,8 @@ pub(crate) fn index_json_value<'a, V: Value<'a>>(
///
/// The term must be json + JSON path.
pub fn convert_to_fast_value_and_append_to_json_term(
mut term: Term,
phrase: &str,
term: &Term,
text: &str,
truncate_date_for_search: bool,
) -> Option<Term> {
assert_eq!(
@@ -254,31 +273,50 @@ pub fn convert_to_fast_value_and_append_to_json_term(
0,
"JSON value bytes should be empty"
);
if let Ok(dt) = OffsetDateTime::parse(phrase, &Rfc3339) {
let mut dt = DateTime::from_utc(dt.to_offset(UtcOffset::UTC));
if truncate_date_for_search {
dt = dt.truncate(DATE_TIME_PRECISION_INDEXED);
try_convert_to_datetime_and_append_to_json_term(term, text, truncate_date_for_search)
.or_else(|| try_convert_to_number_and_append_to_json_term(term, text))
.or_else(|| try_convert_to_bool_and_append_to_json_term_typed(term, text))
}
fn try_convert_to_datetime_and_append_to_json_term(
term: &Term,
text: &str,
truncate_date_for_search: bool,
) -> Option<Term> {
let dt = OffsetDateTime::parse(text, &Rfc3339).ok()?;
let mut dt = DateTime::from_utc(dt.to_offset(UtcOffset::UTC));
if truncate_date_for_search {
dt = dt.truncate(DATE_TIME_PRECISION_INDEXED);
}
let mut term_clone = term.clone();
term_clone.append_type_and_fast_value(dt);
Some(term_clone)
}
fn try_convert_to_number_and_append_to_json_term(term: &Term, text: &str) -> Option<Term> {
let numerical_value: NumericalValue = str::parse::<NumericalValue>(text).ok()?;
let mut term_clone = term.clone();
// Parse is actually returning normalized values already today, but let's not
// not rely on that hidden contract.
match numerical_value.normalize() {
NumericalValue::I64(i64_value) => {
term_clone.append_type_and_fast_value::<i64>(i64_value);
}
NumericalValue::U64(u64_value) => {
term_clone.append_type_and_fast_value::<u64>(u64_value);
}
NumericalValue::F64(f64_value) => {
term_clone.append_type_and_fast_value::<f64>(f64_value);
}
term.append_type_and_fast_value(dt);
return Some(term);
}
if let Ok(i64_val) = str::parse::<i64>(phrase) {
term.append_type_and_fast_value(i64_val);
return Some(term);
}
if let Ok(u64_val) = str::parse::<u64>(phrase) {
term.append_type_and_fast_value(u64_val);
return Some(term);
}
if let Ok(f64_val) = str::parse::<f64>(phrase) {
term.append_type_and_fast_value(f64_val);
return Some(term);
}
if let Ok(bool_val) = str::parse::<bool>(phrase) {
term.append_type_and_fast_value(bool_val);
return Some(term);
}
None
Some(term_clone)
}
fn try_convert_to_bool_and_append_to_json_term_typed(term: &Term, text: &str) -> Option<Term> {
let val = str::parse::<bool>(text).ok()?;
let mut term_clone = term.clone();
term_clone.append_type_and_fast_value(val);
Some(term_clone)
}
/// Splits a json path supplied to the query parser in such a way that

2
src/directory/directory.rs
View File

@@ -56,7 +56,7 @@ impl<T: Send + Sync + 'static> From<Box<T>> for DirectoryLock {
impl Drop for DirectoryLockGuard {
fn drop(&mut self) {
if let Err(e) = self.directory.delete(&self.path) {
error!("Failed to remove the lock file. {:?}", e);
error!("Failed to remove the lock file. {e:?}");
}
}
}

4
src/directory/file_watcher.rs
View File

@@ -51,7 +51,7 @@ impl FileWatcher {
.map(|current_checksum| current_checksum != checksum)
.unwrap_or(true);
if metafile_has_changed {
info!("Meta file {:?} was modified", path);
info!("Meta file {path:?} was modified");
current_checksum_opt = Some(checksum);
// We actually ignore callbacks failing here.
// We just wait for the end of their execution.
@@ -75,7 +75,7 @@ impl FileWatcher {
let reader = match fs::File::open(path) {
Ok(f) => io::BufReader::new(f),
Err(e) => {
warn!("Failed to open meta file {:?}: {:?}", path, e);
warn!("Failed to open meta file {path:?}: {e:?}");
return Err(e);
}
};

4
src/directory/managed_directory.rs
View File

@@ -157,7 +157,7 @@ impl ManagedDirectory {
for file_to_delete in files_to_delete {
match self.delete(&file_to_delete) {
Ok(_) => {
info!("Deleted {:?}", file_to_delete);
info!("Deleted {file_to_delete:?}");
deleted_files.push(file_to_delete);
}
Err(file_error) => {
@@ -170,7 +170,7 @@ impl ManagedDirectory {
if !cfg!(target_os = "windows") {
// On windows, delete is expected to fail if the file
// is mmapped.
error!("Failed to delete {:?}", file_to_delete);
error!("Failed to delete {file_to_delete:?}");
}
}
}

14
src/directory/mmap_directory.rs
View File

@@ -7,7 +7,7 @@ use std::path::{Path, PathBuf};
use std::sync::{Arc, RwLock, Weak};
use common::StableDeref;
use fs4::FileExt;
use fs4::fs_std::FileExt;
#[cfg(all(feature = "mmap", unix))]
pub use memmap2::Advice;
use memmap2::Mmap;
@@ -29,7 +29,7 @@ pub type WeakArcBytes = Weak<dyn Deref<Target = [u8]> + Send + Sync + 'static>;
/// Create a default io error given a string.
pub(crate) fn make_io_err(msg: String) -> io::Error {
io::Error::new(io::ErrorKind::Other, msg)
io::Error::other(msg)
}
/// Returns `None` iff the file exists, can be read, but is empty (and hence
@@ -369,7 +369,7 @@ pub(crate) fn atomic_write(path: &Path, content: &[u8]) -> io::Result<()> {
impl Directory for MmapDirectory {
fn get_file_handle(&self, path: &Path) -> Result<Arc<dyn FileHandle>, OpenReadError> {
debug!("Open Read {:?}", path);
debug!("Open Read {path:?}");
let full_path = self.resolve_path(path);
let mut mmap_cache = self.inner.mmap_cache.write().map_err(|_| {
@@ -414,7 +414,7 @@ impl Directory for MmapDirectory {
}
fn open_write(&self, path: &Path) -> Result<WritePtr, OpenWriteError> {
debug!("Open Write {:?}", path);
debug!("Open Write {path:?}");
let full_path = self.resolve_path(path);
let open_res = OpenOptions::new()
@@ -467,7 +467,7 @@ impl Directory for MmapDirectory {
}
fn atomic_write(&self, path: &Path, content: &[u8]) -> io::Result<()> {
debug!("Atomic Write {:?}", path);
debug!("Atomic Write {path:?}");
let full_path = self.resolve_path(path);
atomic_write(&full_path, content)?;
Ok(())
@@ -484,8 +484,8 @@ impl Directory for MmapDirectory {
.map_err(LockError::wrap_io_error)?;
if lock.is_blocking {
file.lock_exclusive().map_err(LockError::wrap_io_error)?;
} else {
file.try_lock_exclusive().map_err(|_| LockError::LockBusy)?
} else if !file.try_lock_exclusive().map_err(|_| LockError::LockBusy)? {
return Err(LockError::LockBusy);
}
// dropping the file handle will release the lock.
Ok(DirectoryLock::from(Box::new(ReleaseLockFile {

2
src/directory/ram_directory.rs
View File

@@ -191,7 +191,7 @@ impl Directory for RamDirectory {
.fs
.read()
.map_err(|e| OpenReadError::IoError {
io_error: Arc::new(io::Error::new(io::ErrorKind::Other, e.to_string())),
io_error: Arc::new(io::Error::other(e.to_string())),
filepath: path.to_path_buf(),
})?
.exists(path))

5
src/directory/watch_event_router.rs
View File

@@ -90,10 +90,7 @@ impl WatchCallbackList {
let _ = sender.send(Ok(()));
});
if let Err(err) = spawn_res {
error!(
"Failed to spawn thread to call watch callbacks. Cause: {:?}",
err
);
error!("Failed to spawn thread to call watch callbacks. Cause: {err:?}");
}
result
}

20
src/docset.rs
View File

@@ -87,6 +87,17 @@ pub trait DocSet: Send {
/// length of the docset.
fn size_hint(&self) -> u32;
/// Returns a best-effort hint of the cost to consume the entire docset.
///
/// Consuming means calling advance until [`TERMINATED`] is returned.
/// The cost should be relative to the cost of driving a Term query,
/// which would be the number of documents in the DocSet.
///
/// By default this returns `size_hint()`.
fn cost(&self) -> u64 {
self.size_hint() as u64
}
/// Returns the number documents matching.
/// Calling this method consumes the `DocSet`.
fn count(&mut self, alive_bitset: &AliveBitSet) -> u32 {
@@ -134,6 +145,10 @@ impl DocSet for &mut dyn DocSet {
(**self).size_hint()
}
fn cost(&self) -> u64 {
(**self).cost()
}
fn count(&mut self, alive_bitset: &AliveBitSet) -> u32 {
(**self).count(alive_bitset)
}
@@ -169,6 +184,11 @@ impl<TDocSet: DocSet + ?Sized> DocSet for Box<TDocSet> {
unboxed.size_hint()
}
fn cost(&self) -> u64 {
let unboxed: &TDocSet = self.borrow();
unboxed.cost()
}
fn count(&mut self, alive_bitset: &AliveBitSet) -> u32 {
let unboxed: &mut TDocSet = self.borrow_mut();
unboxed.count(alive_bitset)

4
src/index/index.rs
View File

@@ -216,7 +216,7 @@ impl IndexBuilder {
/// Opens or creates a new index in the provided directory
pub fn open_or_create<T: Into<Box<dyn Directory>>>(self, dir: T) -> crate::Result<Index> {
let dir = dir.into();
let dir: Box<dyn Directory> = dir.into();
if !Index::exists(&*dir)? {
return self.create(dir);
}
@@ -494,7 +494,7 @@ impl Index {
.into_iter()
.map(|segment| SegmentReader::open(&segment)?.fields_metadata())
.collect::<Result<_, _>>()?;
Ok(merge_field_meta_data(fields_metadata, &self.schema()))
Ok(merge_field_meta_data(fields_metadata))
}
/// Creates a new segment_meta (Advanced user only).

84
src/index/inverted_index_reader.rs
View File

@@ -1,8 +1,7 @@
use std::io;
use common::json_path_writer::JSON_END_OF_PATH;
use common::BinarySerializable;
use fnv::FnvHashSet;
use common::{BinarySerializable, ByteCount};
#[cfg(feature = "quickwit")]
use futures_util::{FutureExt, StreamExt, TryStreamExt};
#[cfg(feature = "quickwit")]
@@ -36,6 +35,33 @@ pub struct InvertedIndexReader {
total_num_tokens: u64,
}
/// Object that records the amount of space used by a field in an inverted index.
pub(crate) struct InvertedIndexFieldSpace {
pub field_name: String,
pub field_type: Type,
pub postings_size: ByteCount,
pub positions_size: ByteCount,
pub num_terms: u64,
}
/// Returns None if the term is not a valid JSON path.
fn extract_field_name_and_field_type_from_json_path(term: &[u8]) -> Option<(String, Type)> {
let index = term.iter().position(|&byte| byte == JSON_END_OF_PATH)?;
let field_type_code = term.get(index + 1).copied()?;
let field_type = Type::from_code(field_type_code)?;
// Let's flush the current field.
let field_name = String::from_utf8_lossy(&term[..index]).to_string();
Some((field_name, field_type))
}
impl InvertedIndexFieldSpace {
fn record(&mut self, term_info: &TermInfo) {
self.postings_size += ByteCount::from(term_info.posting_num_bytes() as u64);
self.positions_size += ByteCount::from(term_info.positions_num_bytes() as u64);
self.num_terms += 1;
}
}
impl InvertedIndexReader {
pub(crate) fn new(
termdict: TermDictionary,
@@ -81,20 +107,56 @@ impl InvertedIndexReader {
///
/// Notice: This requires a full scan and therefore **very expensive**.
/// TODO: Move to sstable to use the index.
pub fn list_encoded_fields(&self) -> io::Result<Vec<(String, Type)>> {
pub(crate) fn list_encoded_json_fields(&self) -> io::Result<Vec<InvertedIndexFieldSpace>> {
let mut stream = self.termdict.stream()?;
let mut fields = Vec::new();
let mut fields_set = FnvHashSet::default();
while let Some((term, _term_info)) = stream.next() {
if let Some(index) = term.iter().position(|&byte| byte == JSON_END_OF_PATH) {
if !fields_set.contains(&term[..index + 2]) {
fields_set.insert(term[..index + 2].to_vec());
let typ = Type::from_code(term[index + 1]).unwrap();
fields.push((String::from_utf8_lossy(&term[..index]).to_string(), typ));
let mut fields: Vec<InvertedIndexFieldSpace> = Vec::new();
let mut current_field_opt: Option<InvertedIndexFieldSpace> = None;
// Current field bytes, including the JSON_END_OF_PATH.
let mut current_field_bytes: Vec<u8> = Vec::new();
while let Some((term, term_info)) = stream.next() {
if let Some(current_field) = &mut current_field_opt {
if term.starts_with(&current_field_bytes) {
// We are still in the same field.
current_field.record(term_info);
continue;
}
}
// This is a new field!
// Let's flush the current field.
fields.extend(current_field_opt.take());
current_field_bytes.clear();
// And create a new one.
let Some((field_name, field_type)) =
extract_field_name_and_field_type_from_json_path(term)
else {
error!(
"invalid term bytes encountered {term:?}. this only happens if the term \
dictionary is corrupted. please report"
);
continue;
};
let mut field_space = InvertedIndexFieldSpace {
field_name,
field_type,
postings_size: ByteCount::default(),
positions_size: ByteCount::default(),
num_terms: 0u64,
};
field_space.record(term_info);
// We include the json type and the json end of path to make sure the prefix check
// is meaningful.
current_field_bytes.extend_from_slice(&term[..field_space.field_name.len() + 2]);
current_field_opt = Some(field_space);
}
// We need to flush the last field as well.
fields.extend(current_field_opt.take());
Ok(fields)
}

309
src/index/segment_reader.rs
View File

@@ -1,8 +1,8 @@
use std::collections::HashMap;
use std::ops::BitOrAssign;
use std::sync::{Arc, RwLock};
use std::{fmt, io};
use common::{ByteCount, HasLen};
use fnv::FnvHashMap;
use itertools::Itertools;
@@ -304,12 +304,16 @@ impl SegmentReader {
for (field, field_entry) in self.schema().fields() {
let field_name = field_entry.name().to_string();
let is_indexed = field_entry.is_indexed();
if is_indexed {
let is_json = field_entry.field_type().value_type() == Type::Json;
if is_json {
let term_dictionary_json_field_num_bytes: u64 = self
.termdict_composite
.open_read(field)
.map(|file_slice| file_slice.len() as u64)
.unwrap_or(0u64);
let inv_index = self.inverted_index(field)?;
let encoded_fields_in_index = inv_index.list_encoded_fields()?;
let encoded_fields_in_index = inv_index.list_encoded_json_fields()?;
let mut build_path = |field_name: &str, mut json_path: String| {
// In this case we need to map the potential fast field to the field name
// accepted by the query parser.
@@ -328,30 +332,65 @@ impl SegmentReader {
format!("{field_name}.{json_path}")
}
};
indexed_fields.extend(
encoded_fields_in_index
.into_iter()
.map(|(name, typ)| (build_path(&field_name, name), typ))
.map(|(field_name, typ)| FieldMetadata {
indexed: true,
stored: false,
field_name,
fast: false,
typ,
}),
);
let total_num_terms = encoded_fields_in_index
.iter()
.map(|field_space| field_space.num_terms)
.sum();
indexed_fields.extend(encoded_fields_in_index.into_iter().map(|field_space| {
let field_name = build_path(&field_name, field_space.field_name);
// It is complex to attribute the exact amount of bytes required by specific
// field in the json field. Instead, as a proxy, we
// attribute the total amount of bytes for the entire json field,
// proportionally to the number of terms in each
// fields.
let term_dictionary_size = (term_dictionary_json_field_num_bytes
* field_space.num_terms)
.checked_div(total_num_terms)
.unwrap_or(0);
FieldMetadata {
postings_size: Some(field_space.postings_size),
positions_size: Some(field_space.positions_size),
term_dictionary_size: Some(ByteCount::from(term_dictionary_size)),
fast_size: None,
// The stored flag will be set at the end of this function!
stored: field_entry.is_stored(),
field_name,
typ: field_space.field_type,
}
}));
} else {
let postings_size: ByteCount = self
.postings_composite
.open_read(field)
.map(|posting_fileslice| posting_fileslice.len())
.unwrap_or(0)
.into();
let positions_size: ByteCount = self
.positions_composite
.open_read(field)
.map(|positions_fileslice| positions_fileslice.len())
.unwrap_or(0)
.into();
let term_dictionary_size: ByteCount = self
.termdict_composite
.open_read(field)
.map(|term_dictionary_fileslice| term_dictionary_fileslice.len())
.unwrap_or(0)
.into();
indexed_fields.push(FieldMetadata {
indexed: true,
stored: false,
field_name: field_name.to_string(),
fast: false,
typ: field_entry.field_type().value_type(),
// The stored flag will be set at the end of this function!
stored: field_entry.is_stored(),
fast_size: None,
term_dictionary_size: Some(term_dictionary_size),
postings_size: Some(postings_size),
positions_size: Some(positions_size),
});
}
}
}
let mut fast_fields: Vec<FieldMetadata> = self
let fast_fields: Vec<FieldMetadata> = self
.fast_fields()
.columnar()
.iter_columns()?
@@ -363,23 +402,21 @@ impl SegmentReader {
.get(&field_name)
.unwrap_or(&field_name)
.to_string();
let stored = is_field_stored(&field_name, &self.schema);
FieldMetadata {
indexed: false,
stored: false,
field_name,
fast: true,
typ: Type::from(handle.column_type()),
stored,
fast_size: Some(handle.num_bytes()),
term_dictionary_size: None,
postings_size: None,
positions_size: None,
}
})
.collect();
// Since the type is encoded differently in the fast field and in the inverted index,
// the order of the fields is not guaranteed to be the same. Therefore, we sort the fields.
// If we are sure that the order is the same, we can remove this sort.
indexed_fields.sort_unstable();
fast_fields.sort_unstable();
let merged = merge_field_meta_data(vec![indexed_fields, fast_fields], &self.schema);
Ok(merged)
let merged_field_metadatas: Vec<FieldMetadata> =
merge_field_meta_data(vec![indexed_fields, fast_fields]);
Ok(merged_field_metadatas)
}
/// Returns the segment id
@@ -443,20 +480,47 @@ pub struct FieldMetadata {
// Notice: Don't reorder the declaration of 1.field_name 2.typ, as it is used for ordering by
// field_name then typ.
pub typ: Type,
/// Is the field indexed for search
pub indexed: bool,
/// Is the field stored in the doc store
pub stored: bool,
/// Is the field stored in the columnar storage
pub fast: bool,
/// Size occupied in the columnar storage (None if not fast)
pub fast_size: Option<ByteCount>,
/// term_dictionary
pub term_dictionary_size: Option<ByteCount>,
/// Size occupied in the index postings storage (None if not indexed)
pub postings_size: Option<ByteCount>,
/// Size occupied in the index postings storage (None if positions are not recorded)
pub positions_size: Option<ByteCount>,
}
impl BitOrAssign for FieldMetadata {
fn bitor_assign(&mut self, rhs: Self) {
assert!(self.field_name == rhs.field_name);
assert!(self.typ == rhs.typ);
self.indexed |= rhs.indexed;
fn merge_options(left: Option<ByteCount>, right: Option<ByteCount>) -> Option<ByteCount> {
match (left, right) {
(Some(l), Some(r)) => Some(l + r),
(None, right) => right,
(left, None) => left,
}
}
impl FieldMetadata {
/// Returns true if and only if the field is indexed.
pub fn is_indexed(&self) -> bool {
self.postings_size.is_some()
}
/// Returns true if and only if the field is a fast field (i.e.: recorded in columnar format).
pub fn is_fast(&self) -> bool {
self.fast_size.is_some()
}
/// Merges two field metadata.
pub fn merge(&mut self, rhs: Self) {
assert_eq!(self.field_name, rhs.field_name);
assert_eq!(self.typ, rhs.typ);
self.stored |= rhs.stored;
self.fast |= rhs.fast;
self.fast_size = merge_options(self.fast_size, rhs.fast_size);
self.term_dictionary_size =
merge_options(self.term_dictionary_size, rhs.term_dictionary_size);
self.postings_size = merge_options(self.postings_size, rhs.postings_size);
self.positions_size = merge_options(self.positions_size, rhs.positions_size);
}
}
@@ -469,23 +533,29 @@ fn is_field_stored(field_name: &str, schema: &Schema) -> bool {
}
/// Helper to merge the field metadata from multiple segments.
pub fn merge_field_meta_data(
field_metadatas: Vec<Vec<FieldMetadata>>,
schema: &Schema,
) -> Vec<FieldMetadata> {
pub fn merge_field_meta_data(mut field_metadatas: Vec<Vec<FieldMetadata>>) -> Vec<FieldMetadata> {
// READ BEFORE REMOVING THIS!
//
// Because we replace field sep by `.`, fields are not always sorted.
// Also, to enforce such an implicit contract, we would have to add
// assert here.
//
// Sorting is linear time on pre-sorted data, so we are simply better off sorting data here.
for field_metadatas in &mut field_metadatas {
field_metadatas.sort_unstable();
}
let mut merged_field_metadata = Vec::new();
for (_key, mut group) in &field_metadatas
.into_iter()
.kmerge_by(|left, right| left < right)
.kmerge()
// TODO: Remove allocation
.chunk_by(|el| (el.field_name.to_string(), el.typ))
{
let mut merged: FieldMetadata = group.next().unwrap();
for el in group {
merged |= el;
merged.merge(el);
}
// Currently is_field_stored is maybe too slow for the high cardinality case
merged.stored = is_field_stored(&merged.field_name, schema);
merged_field_metadata.push(merged);
}
merged_field_metadata
@@ -507,7 +577,7 @@ fn intersect_alive_bitset(
}
impl fmt::Debug for SegmentReader {
fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
fn fmt(&self, f: &mut fmt::Formatter) -> fmt::Result {
write!(f, "SegmentReader({:?})", self.segment_id)
}
}
@@ -516,122 +586,168 @@ impl fmt::Debug for SegmentReader {
mod test {
use super::*;
use crate::index::Index;
use crate::schema::{SchemaBuilder, Term, STORED, TEXT};
use crate::schema::{Term, STORED, TEXT};
use crate::IndexWriter;
#[track_caller]
fn assert_merge(fields_metadatas: &[Vec<FieldMetadata>], expected: &[FieldMetadata]) {
use itertools::Itertools;
let num_els = fields_metadatas.len();
for permutation in fields_metadatas.iter().cloned().permutations(num_els) {
let res = merge_field_meta_data(permutation);
assert_eq!(&res, &expected);
}
}
#[test]
fn test_merge_field_meta_data_same() {
let schema = SchemaBuilder::new().build();
fn test_merge_field_meta_data_same_field() {
let field_metadata1 = FieldMetadata {
field_name: "a".to_string(),
typ: crate::schema::Type::Str,
indexed: true,
stored: false,
fast: true,
term_dictionary_size: Some(ByteCount::from(100u64)),
postings_size: Some(ByteCount::from(1_000u64)),
positions_size: Some(ByteCount::from(2_000u64)),
fast_size: Some(ByteCount::from(1_000u64).into()),
};
let field_metadata2 = FieldMetadata {
field_name: "a".to_string(),
typ: crate::schema::Type::Str,
indexed: true,
stored: false,
fast: true,
term_dictionary_size: Some(ByteCount::from(80u64)),
postings_size: Some(ByteCount::from(1_500u64)),
positions_size: Some(ByteCount::from(2_500u64)),
fast_size: Some(ByteCount::from(3_000u64).into()),
};
let res = merge_field_meta_data(
vec![vec![field_metadata1.clone()], vec![field_metadata2]],
&schema,
let expected = FieldMetadata {
field_name: "a".to_string(),
typ: crate::schema::Type::Str,
stored: false,
term_dictionary_size: Some(ByteCount::from(180u64)),
postings_size: Some(ByteCount::from(2_500u64)),
positions_size: Some(ByteCount::from(4_500u64)),
fast_size: Some(ByteCount::from(4_000u64).into()),
};
assert_merge(
&[vec![field_metadata1.clone()], vec![field_metadata2]],
&[expected],
);
assert_eq!(res, vec![field_metadata1]);
}
#[track_caller]
#[test]
fn test_merge_field_meta_data_different() {
let schema = SchemaBuilder::new().build();
let field_metadata1 = FieldMetadata {
field_name: "a".to_string(),
typ: crate::schema::Type::Str,
indexed: false,
stored: false,
fast: true,
fast_size: Some(1_000u64.into()),
term_dictionary_size: Some(100u64.into()),
postings_size: Some(2_000u64.into()),
positions_size: Some(4_000u64.into()),
};
let field_metadata2 = FieldMetadata {
field_name: "b".to_string(),
typ: crate::schema::Type::Str,
indexed: false,
stored: false,
fast: true,
fast_size: Some(1_002u64.into()),
term_dictionary_size: None,
postings_size: None,
positions_size: None,
};
let field_metadata3 = FieldMetadata {
field_name: "a".to_string(),
typ: crate::schema::Type::Str,
indexed: true,
term_dictionary_size: Some(101u64.into()),
postings_size: Some(2_001u64.into()),
positions_size: Some(4_001u64.into()),
stored: false,
fast: false,
fast_size: None,
};
let res = merge_field_meta_data(
vec![
let expected = vec![
FieldMetadata {
field_name: "a".to_string(),
typ: crate::schema::Type::Str,
stored: false,
term_dictionary_size: Some(201u64.into()),
postings_size: Some(4_001u64.into()),
positions_size: Some(8_001u64.into()),
fast_size: Some(1_000u64.into()),
},
FieldMetadata {
field_name: "b".to_string(),
typ: crate::schema::Type::Str,
stored: false,
term_dictionary_size: None,
postings_size: None,
positions_size: None,
fast_size: Some(1_002u64.into()),
},
];
assert_merge(
&[
vec![field_metadata1.clone(), field_metadata2.clone()],
vec![field_metadata3],
],
&schema,
&expected,
);
let field_metadata_expected1 = FieldMetadata {
field_name: "a".to_string(),
typ: crate::schema::Type::Str,
indexed: true,
stored: false,
fast: true,
};
assert_eq!(res, vec![field_metadata_expected1, field_metadata2.clone()]);
}
#[test]
fn test_merge_field_meta_data_merge() {
use pretty_assertions::assert_eq;
let get_meta_data = |name: &str, typ: Type| FieldMetadata {
field_name: name.to_string(),
typ,
indexed: false,
term_dictionary_size: None,
postings_size: None,
positions_size: None,
stored: false,
fast: true,
fast_size: Some(1u64.into()),
};
let schema = SchemaBuilder::new().build();
let mut metas = vec![get_meta_data("d", Type::Str), get_meta_data("e", Type::U64)];
metas.sort();
let res = merge_field_meta_data(vec![vec![get_meta_data("e", Type::Str)], metas], &schema);
assert_eq!(
res,
vec![
let metas = vec![get_meta_data("d", Type::Str), get_meta_data("e", Type::U64)];
assert_merge(
&[vec![get_meta_data("e", Type::Str)], metas],
&[
get_meta_data("d", Type::Str),
get_meta_data("e", Type::Str),
get_meta_data("e", Type::U64),
]
],
);
}
#[test]
fn test_merge_field_meta_data_bitxor() {
let field_metadata1 = FieldMetadata {
field_name: "a".to_string(),
typ: crate::schema::Type::Str,
indexed: false,
term_dictionary_size: None,
postings_size: None,
positions_size: None,
stored: false,
fast: true,
fast_size: Some(10u64.into()),
};
let field_metadata2 = FieldMetadata {
field_name: "a".to_string(),
typ: crate::schema::Type::Str,
indexed: true,
term_dictionary_size: Some(10u64.into()),
postings_size: Some(11u64.into()),
positions_size: Some(12u64.into()),
stored: false,
fast: false,
fast_size: None,
};
let field_metadata_expected = FieldMetadata {
field_name: "a".to_string(),
typ: crate::schema::Type::Str,
indexed: true,
term_dictionary_size: Some(10u64.into()),
postings_size: Some(11u64.into()),
positions_size: Some(12u64.into()),
stored: false,
fast: true,
fast_size: Some(10u64.into()),
};
let mut res1 = field_metadata1.clone();
res1 |= field_metadata2.clone();
res1.merge(field_metadata2.clone());
let mut res2 = field_metadata2.clone();
res2 |= field_metadata1;
res2.merge(field_metadata1);
assert_eq!(res1, field_metadata_expected);
assert_eq!(res2, field_metadata_expected);
}
@@ -662,6 +778,7 @@ mod test {
assert_eq!(4, searcher.segment_reader(0).max_doc());
Ok(())
}
#[test]
fn test_alive_docs_iterator() -> crate::Result<()> {
let mut schema_builder = Schema::builder();

6
src/indexer/index_writer.rs
View File

@@ -370,7 +370,7 @@ impl<D: Document> IndexWriter<D> {
.map_err(|_| error_in_index_worker_thread("Failed to join merging thread."));
if let Err(ref e) = result {
error!("Some merging thread failed {:?}", e);
error!("Some merging thread failed {e:?}");
}
result
@@ -615,7 +615,7 @@ impl<D: Document> IndexWriter<D> {
/// It is also possible to add a payload to the `commit`
/// using this API.
/// See [`PreparedCommit::set_payload()`].
pub fn prepare_commit(&mut self) -> crate::Result<PreparedCommit<D>> {
pub fn prepare_commit(&mut self) -> crate::Result<PreparedCommit<'_, D>> {
// Here, because we join all of the worker threads,
// all of the segment update for this commit have been
// sent.
@@ -644,7 +644,7 @@ impl<D: Document> IndexWriter<D> {
let commit_opstamp = self.stamper.stamp();
let prepared_commit = PreparedCommit::new(self, commit_opstamp);
info!("Prepared commit {}", commit_opstamp);
info!("Prepared commit {commit_opstamp}");
Ok(prepared_commit)
}

209
src/indexer/mod.rs
View File

@@ -61,6 +61,8 @@ type AddBatchReceiver<D> = channel::Receiver<AddBatch<D>>;
#[cfg(test)]
mod tests_mmap {
use common::ByteCount;
use crate::aggregation::agg_req::Aggregations;
use crate::aggregation::agg_result::AggregationResults;
use crate::aggregation::AggregationCollector;
@@ -280,11 +282,14 @@ mod tests_mmap {
field_name_out
};
let mut fields = reader.searcher().segment_readers()[0]
let mut fields: Vec<(String, Type)> = reader.searcher().segment_readers()[0]
.inverted_index(field)
.unwrap()
.list_encoded_fields()
.unwrap();
.list_encoded_json_fields()
.unwrap()
.into_iter()
.map(|field_space| (field_space.field_name, field_space.field_type))
.collect();
assert_eq!(fields.len(), 8);
fields.sort();
let mut expected_fields = vec![
@@ -385,7 +390,12 @@ mod tests_mmap {
let reader = &searcher.segment_readers()[0];
let inverted_index = reader.inverted_index(json_field).unwrap();
assert_eq!(
inverted_index.list_encoded_fields().unwrap(),
inverted_index
.list_encoded_json_fields()
.unwrap()
.into_iter()
.map(|field_space| (field_space.field_name, field_space.field_type))
.collect::<Vec<_>>(),
[
("k8s.container.name".to_string(), Type::Str),
("sub\u{1}a".to_string(), Type::I64),
@@ -402,19 +412,41 @@ mod tests_mmap {
fn test_json_fields_metadata_expanded_dots_one_segment() {
test_json_fields_metadata(true, true);
}
#[test]
fn test_json_fields_metadata_expanded_dots_multi_segment() {
test_json_fields_metadata(true, false);
}
#[test]
fn test_json_fields_metadata_no_expanded_dots_one_segment() {
test_json_fields_metadata(false, true);
}
#[test]
fn test_json_fields_metadata_no_expanded_dots_multi_segment() {
test_json_fields_metadata(false, false);
}
#[track_caller]
fn assert_size_eq(lhs: Option<ByteCount>, rhs: Option<ByteCount>) {
let ignore_actual_values = |size_opt: Option<ByteCount>| size_opt.map(|val| val > 0);
assert_eq!(ignore_actual_values(lhs), ignore_actual_values(rhs));
}
#[track_caller]
fn assert_field_metadata_eq_but_ignore_field_size(
expected: &FieldMetadata,
actual: &FieldMetadata,
) {
assert_eq!(&expected.field_name, &actual.field_name);
assert_eq!(&expected.typ, &actual.typ);
assert_eq!(&expected.stored, &actual.stored);
assert_size_eq(expected.postings_size, actual.postings_size);
assert_size_eq(expected.positions_size, actual.positions_size);
assert_size_eq(expected.fast_size, actual.fast_size);
}
fn test_json_fields_metadata(expanded_dots: bool, one_segment: bool) {
use pretty_assertions::assert_eq;
let mut schema_builder = Schema::builder();
@@ -453,81 +485,101 @@ mod tests_mmap {
assert_eq!(searcher.num_docs(), 3);
let fields_metadata = index.fields_metadata().unwrap();
assert_eq!(
fields_metadata,
[
FieldMetadata {
field_name: "empty".to_string(),
indexed: true,
stored: true,
fast: true,
typ: Type::U64
let expected_fields = &[
FieldMetadata {
field_name: "empty".to_string(),
stored: true,
typ: Type::U64,
term_dictionary_size: Some(0u64.into()),
fast_size: Some(1u64.into()),
postings_size: Some(0u64.into()),
positions_size: Some(0u64.into()),
},
FieldMetadata {
field_name: if expanded_dots {
"json.shadow.k8s.container.name".to_string()
} else {
"json.shadow.k8s\\.container\\.name".to_string()
},
FieldMetadata {
field_name: if expanded_dots {
"json.shadow.k8s.container.name".to_string()
} else {
"json.shadow.k8s\\.container\\.name".to_string()
},
indexed: true,
stored: true,
fast: true,
typ: Type::Str
},
FieldMetadata {
field_name: "json.shadow.sub.a".to_string(),
indexed: true,
stored: true,
fast: true,
typ: Type::I64
},
FieldMetadata {
field_name: "json.shadow.sub.b".to_string(),
indexed: true,
stored: true,
fast: true,
typ: Type::I64
},
FieldMetadata {
field_name: "json.shadow.suber.a".to_string(),
indexed: true,
stored: true,
fast: true,
typ: Type::I64
},
FieldMetadata {
field_name: "json.shadow.suber.a".to_string(),
indexed: true,
stored: true,
fast: true,
typ: Type::Str
},
FieldMetadata {
field_name: "json.shadow.suber.b".to_string(),
indexed: true,
stored: true,
fast: true,
typ: Type::I64
},
FieldMetadata {
field_name: "json.shadow.val".to_string(),
indexed: true,
stored: true,
fast: true,
typ: Type::Str
},
FieldMetadata {
field_name: "numbers".to_string(),
indexed: false,
stored: false,
fast: true,
typ: Type::U64
}
]
);
stored: true,
typ: Type::Str,
term_dictionary_size: Some(1u64.into()),
fast_size: Some(1u64.into()),
postings_size: Some(1u64.into()),
positions_size: Some(1u64.into()),
},
FieldMetadata {
field_name: "json.shadow.sub.a".to_string(),
typ: Type::I64,
stored: true,
fast_size: Some(1u64.into()),
term_dictionary_size: Some(1u64.into()),
postings_size: Some(1u64.into()),
positions_size: Some(1u64.into()),
},
FieldMetadata {
field_name: "json.shadow.sub.b".to_string(),
typ: Type::I64,
stored: true,
fast_size: Some(1u64.into()),
term_dictionary_size: Some(1u64.into()),
postings_size: Some(1u64.into()),
positions_size: Some(1u64.into()),
},
FieldMetadata {
field_name: "json.shadow.suber.a".to_string(),
stored: true,
typ: Type::I64,
fast_size: Some(1u64.into()),
term_dictionary_size: Some(1u64.into()),
postings_size: Some(1u64.into()),
positions_size: Some(1u64.into()),
},
FieldMetadata {
field_name: "json.shadow.suber.a".to_string(),
typ: Type::Str,
stored: true,
fast_size: Some(1u64.into()),
term_dictionary_size: Some(1u64.into()),
postings_size: Some(1u64.into()),
positions_size: Some(1u64.into()),
},
FieldMetadata {
field_name: "json.shadow.suber.b".to_string(),
typ: Type::I64,
stored: true,
fast_size: Some(1u64.into()),
term_dictionary_size: Some(1u64.into()),
postings_size: Some(1u64.into()),
positions_size: Some(1u64.into()),
},
FieldMetadata {
field_name: "json.shadow.val".to_string(),
typ: Type::Str,
stored: true,
fast_size: Some(1u64.into()),
term_dictionary_size: Some(1u64.into()),
postings_size: Some(1u64.into()),
positions_size: Some(1u64.into()),
},
FieldMetadata {
field_name: "numbers".to_string(),
stored: false,
typ: Type::U64,
fast_size: Some(1u64.into()),
term_dictionary_size: None,
postings_size: None,
positions_size: None,
},
];
assert_eq!(fields_metadata.len(), expected_fields.len());
for (expected, value) in expected_fields.iter().zip(fields_metadata.iter()) {
assert_field_metadata_eq_but_ignore_field_size(expected, value);
}
let query_parser = QueryParser::for_index(&index, vec![]);
// Test if returned field name can be queried
for indexed_field in fields_metadata.iter().filter(|meta| meta.indexed) {
for indexed_field in fields_metadata.iter().filter(|meta| meta.is_indexed()) {
let val = if indexed_field.typ == Type::Str {
"a"
} else {
@@ -543,7 +595,10 @@ mod tests_mmap {
}
}
// Test if returned field name can be used for aggregation
for fast_field in fields_metadata.iter().filter(|meta| meta.fast) {
for fast_field in fields_metadata
.iter()
.filter(|field_metadata| field_metadata.is_fast())
{
let agg_req_str = json!(
{
"termagg": {

3
src/indexer/segment_updater.rs
View File

@@ -501,8 +501,7 @@ impl SegmentUpdater {
Ok(segment_entries) => segment_entries,
Err(err) => {
warn!(
"Starting the merge failed for the following reason. This is not fatal. {}",
err
"Starting the merge failed for the following reason. This is not fatal. {err}"
);
return err.into();
}

4
src/indexer/segment_writer.rs
View File

@@ -873,8 +873,8 @@ mod tests {
fn assert_type(reader: &SegmentReader, field: &str, typ: ColumnType) {
let cols = reader.fast_fields().dynamic_column_handles(field).unwrap();
assert_eq!(cols.len(), 1, "{}", field);
assert_eq!(cols[0].column_type(), typ, "{}", field);
assert_eq!(cols.len(), 1, "{field}");
assert_eq!(cols[0].column_type(), typ, "{field}");
}
assert_type(segment_reader, "json.toto", ColumnType::Str);
assert_type(segment_reader, "json.float", ColumnType::F64);

53
src/lib.rs
View File

@@ -55,7 +55,7 @@
//! // between indexing threads.
//! let mut index_writer: IndexWriter = index.writer(100_000_000)?;
//!
//! // Let's index one documents!
//! // Let's index a document!
//! index_writer.add_document(doc!(
//! title => "The Old Man and the Sea",
//! body => "He was an old man who fished alone in a skiff in \
@@ -165,7 +165,7 @@ mod macros;
mod future_result;
// Re-exports
pub use common::DateTime;
pub use common::{ByteCount, DateTime};
pub use {columnar, query_grammar, time};
pub use crate::error::TantivyError;
@@ -370,6 +370,8 @@ macro_rules! fail_point {
/// Common test utilities.
#[cfg(test)]
pub mod tests {
use std::collections::BTreeMap;
use common::{BinarySerializable, FixedSize};
use query_grammar::{UserInputAst, UserInputLeaf, UserInputLiteral};
use rand::distributions::{Bernoulli, Uniform};
@@ -382,7 +384,7 @@ pub mod tests {
use crate::index::SegmentReader;
use crate::merge_policy::NoMergePolicy;
use crate::postings::Postings;
use crate::query::BooleanQuery;
use crate::query::{BooleanQuery, QueryParser};
use crate::schema::*;
use crate::{DateTime, DocAddress, Index, IndexWriter, ReloadPolicy};
@@ -1223,4 +1225,49 @@ pub mod tests {
);
assert_eq!(dt_from_ts_nanos.to_hms_micro(), offset_dt.to_hms_micro());
}
#[test]
fn test_json_number_ambiguity() {
let mut schema_builder = Schema::builder();
let json_field = schema_builder.add_json_field("number", crate::schema::TEXT);
let schema = schema_builder.build();
let index = Index::create_in_ram(schema);
let mut index_writer = index.writer_for_tests().unwrap();
{
let mut doc = TantivyDocument::new();
let mut obj = BTreeMap::default();
obj.insert("key".to_string(), OwnedValue::I64(1i64));
doc.add_object(json_field, obj);
index_writer.add_document(doc).unwrap();
}
{
let mut doc = TantivyDocument::new();
let mut obj = BTreeMap::default();
obj.insert("key".to_string(), OwnedValue::U64(1u64));
doc.add_object(json_field, obj);
index_writer.add_document(doc).unwrap();
}
{
let mut doc = TantivyDocument::new();
let mut obj = BTreeMap::default();
obj.insert("key".to_string(), OwnedValue::F64(1.0f64));
doc.add_object(json_field, obj);
index_writer.add_document(doc).unwrap();
}
index_writer.commit().unwrap();
let searcher = index.reader().unwrap().searcher();
assert_eq!(searcher.num_docs(), 3);
{
let parser = QueryParser::for_index(&index, vec![]);
let query = parser.parse_query("number.key:1").unwrap();
let count = searcher.search(&query, &crate::collector::Count).unwrap();
assert_eq!(count, 3);
}
{
let parser = QueryParser::for_index(&index, vec![]);
let query = parser.parse_query("number.key:1.0").unwrap();
let count = searcher.search(&query, &crate::collector::Count).unwrap();
assert_eq!(count, 3);
}
}
}

4
src/positions/mod.rs
View File

@@ -41,8 +41,6 @@ const COMPRESSION_BLOCK_SIZE: usize = BitPacker4x::BLOCK_LEN;
#[cfg(test)]
pub(crate) mod tests {
use std::iter;
use proptest::prelude::*;
use proptest::sample::select;
@@ -206,7 +204,7 @@ pub(crate) mod tests {
#[test]
fn test_position() -> crate::Result<()> {
const CONST_VAL: u32 = 9u32;
let positions_delta: Vec<u32> = iter::repeat(CONST_VAL).take(2_000_000).collect();
let positions_delta: Vec<u32> = std::iter::repeat_n(CONST_VAL, 2_000_000).collect();
let positions_data = create_positions_data(&positions_delta[..])?;
assert_eq!(positions_data.len(), 1_015_627);
let mut position_reader = PositionReader::open(positions_data)?;

35
src/postings/block_segment_postings.rs
View File

@@ -227,19 +227,6 @@ impl BlockSegmentPostings {
self.doc_decoder.output_array()
}
/// Returns a full block, regardless of whether the block is complete or incomplete (
/// as it happens for the last block of the posting list).
///
/// In the latter case, the block is guaranteed to be padded with the sentinel value:
/// `TERMINATED`. The array is also guaranteed to be aligned on 16 bytes = 128 bits.
///
/// This method is useful to run SSE2 linear search.
#[inline]
pub(crate) fn full_block(&self) -> &[DocId; COMPRESSION_BLOCK_SIZE] {
debug_assert!(self.block_is_loaded());
self.doc_decoder.full_output()
}
/// Return the document at index `idx` of the block.
#[inline]
pub fn doc(&self, idx: usize) -> u32 {
@@ -275,22 +262,36 @@ impl BlockSegmentPostings {
///
/// If all docs are smaller than target, the block loaded may be empty,
/// or be the last an incomplete VInt block.
pub fn seek(&mut self, target_doc: DocId) {
self.shallow_seek(target_doc);
pub fn seek(&mut self, target_doc: DocId) -> usize {
// Move to the block that might contain our document.
self.seek_block(target_doc);
self.load_block();
// At this point we are on the block that might contain our document.
let doc = self.doc_decoder.seek_within_block(target_doc);
// The last block is not full and padded with TERMINATED,
// so we are guaranteed to have at least one value (real or padding)
// that is >= target_doc.
debug_assert!(doc < COMPRESSION_BLOCK_SIZE);
// `doc` is now the first element >= `target_doc`.
// If all docs are smaller than target, the current block is incomplete and padded
// with TERMINATED. After the search, the cursor points to the first TERMINATED.
doc
}
pub(crate) fn position_offset(&self) -> u64 {
self.skip_reader.position_offset()
}
/// Dangerous API! This calls seek on the skip list,
/// Dangerous API! This calls seeks the next block on the skip list,
/// but does not `.load_block()` afterwards.
///
/// `.load_block()` needs to be called manually afterwards.
/// If all docs are smaller than target, the block loaded may be empty,
/// or be the last an incomplete VInt block.
pub(crate) fn shallow_seek(&mut self, target_doc: DocId) {
pub(crate) fn seek_block(&mut self, target_doc: DocId) {
if self.skip_reader.seek(target_doc) {
self.block_max_score_cache = None;
self.block_loaded = false;

6
src/postings/compression/mod.rs
View File

@@ -151,9 +151,11 @@ impl BlockDecoder {
&self.output[..self.output_len]
}
/// Return in-block index of first value >= `target`.
/// Uses the padded buffer to enable branchless search.
#[inline]
pub(crate) fn full_output(&self) -> &[u32; COMPRESSION_BLOCK_SIZE] {
&self.output
pub(crate) fn seek_within_block(&self, target: u32) -> usize {
crate::postings::branchless_binary_search(&self.output, target)
}
#[inline]

15
src/postings/mod.rs
View File

@@ -667,12 +667,15 @@ mod bench {
.read_postings(&TERM_D, IndexRecordOption::Basic)
.unwrap()
.unwrap();
let mut intersection = Intersection::new(vec![
segment_postings_a,
segment_postings_b,
segment_postings_c,
segment_postings_d,
]);
let mut intersection = Intersection::new(
vec![
segment_postings_a,
segment_postings_b,
segment_postings_c,
segment_postings_d,
],
reader.searcher().num_docs() as u32,
);
while intersection.advance() != TERMINATED {}
});
}

25
src/postings/segment_postings.rs
View File

@@ -4,7 +4,7 @@ use crate::docset::DocSet;
use crate::fastfield::AliveBitSet;
use crate::positions::PositionReader;
use crate::postings::compression::COMPRESSION_BLOCK_SIZE;
use crate::postings::{branchless_binary_search, BlockSegmentPostings, Postings};
use crate::postings::{BlockSegmentPostings, Postings};
use crate::{DocId, TERMINATED};
/// `SegmentPostings` represents the inverted list or postings associated with
@@ -175,26 +175,11 @@ impl DocSet for SegmentPostings {
return self.doc();
}
self.block_cursor.seek(target);
// At this point we are on the block, that might contain our document.
let output = self.block_cursor.full_block();
self.cur = branchless_binary_search(output, target);
// The last block is not full and padded with the value TERMINATED,
// so that we are guaranteed to have at least doc in the block (a real one or the padding)
// that is greater or equal to the target.
debug_assert!(self.cur < COMPRESSION_BLOCK_SIZE);
// `doc` is now the first element >= `target`
// If all docs are smaller than target the current block should be incomplemented and padded
// with the value `TERMINATED`.
//
// After the search, the cursor should point to the first value of TERMINATED.
let doc = output[self.cur];
// Delegate block-local search to BlockSegmentPostings::seek, which returns
// the in-block index of the first doc >= target.
self.cur = self.block_cursor.seek(target);
let doc = self.doc();
debug_assert!(doc >= target);
debug_assert_eq!(doc, self.doc());
doc
}

4
src/postings/serializer.rs
View File

@@ -75,7 +75,7 @@ impl InvertedIndexSerializer {
field: Field,
total_num_tokens: u64,
fieldnorm_reader: Option<FieldNormReader>,
) -> io::Result<FieldSerializer> {
) -> io::Result<FieldSerializer<'_>> {
let field_entry: &FieldEntry = self.schema.get_field_entry(field);
let term_dictionary_write = self.terms_write.for_field(field);
let postings_write = self.postings_write.for_field(field);
@@ -126,7 +126,7 @@ impl<'a> FieldSerializer<'a> {
let term_dictionary_builder = TermDictionaryBuilder::create(term_dictionary_write)?;
let average_fieldnorm = fieldnorm_reader
.as_ref()
.map(|ff_reader| (total_num_tokens as Score / ff_reader.num_docs() as Score))
.map(|ff_reader| total_num_tokens as Score / ff_reader.num_docs() as Score)
.unwrap_or(0.0);
let postings_serializer = PostingsSerializer::new(
postings_write,

8
src/query/bm25.rs
View File

@@ -1,5 +1,3 @@
use serde::{Deserialize, Serialize};
use crate::fieldnorm::FieldNormReader;
use crate::query::Explanation;
use crate::schema::Field;
@@ -68,12 +66,6 @@ fn compute_tf_cache(average_fieldnorm: Score) -> [Score; 256] {
cache
}
#[derive(Clone, PartialEq, Debug, Serialize, Deserialize)]
pub struct Bm25Params {
pub idf: Score,
pub avg_fieldnorm: Score,
}
/// A struct used for computing BM25 scores.
#[derive(Clone)]
pub struct Bm25Weight {

20
src/query/boolean_query/block_wand.rs
View File

@@ -167,7 +167,7 @@ pub fn block_wand(
let block_max_score_upperbound: Score = scorers[..pivot_len]
.iter_mut()
.map(|scorer| {
scorer.shallow_seek(pivot_doc);
scorer.seek_block(pivot_doc);
scorer.block_max_score()
})
.sum();
@@ -234,7 +234,7 @@ pub fn block_wand_single_scorer(
return;
}
doc = last_doc_in_block + 1;
scorer.shallow_seek(doc);
scorer.seek_block(doc);
}
// Seek will effectively load that block.
doc = scorer.seek(doc);
@@ -256,7 +256,7 @@ pub fn block_wand_single_scorer(
}
}
doc += 1;
scorer.shallow_seek(doc);
scorer.seek_block(doc);
}
}
@@ -302,7 +302,6 @@ fn is_sorted<I: Iterator<Item = DocId>>(mut it: I) -> bool {
mod tests {
use std::cmp::Ordering;
use std::collections::BinaryHeap;
use std::iter;
use proptest::prelude::*;
@@ -368,10 +367,14 @@ mod tests {
checkpoints
}
fn compute_checkpoints_manual(term_scorers: Vec<TermScorer>, n: usize) -> Vec<(DocId, Score)> {
fn compute_checkpoints_manual(
term_scorers: Vec<TermScorer>,
n: usize,
max_doc: u32,
) -> Vec<(DocId, Score)> {
let mut heap: BinaryHeap<Float> = BinaryHeap::with_capacity(n);
let mut checkpoints: Vec<(DocId, Score)> = Vec::new();
let mut scorer = BufferedUnionScorer::build(term_scorers, SumCombiner::default);
let mut scorer = BufferedUnionScorer::build(term_scorers, SumCombiner::default, max_doc);
let mut limit = Score::MIN;
loop {
@@ -436,7 +439,7 @@ mod tests {
let fieldnorms_expanded = fieldnorms
.iter()
.cloned()
.flat_map(|fieldnorm| iter::repeat(fieldnorm).take(REPEAT))
.flat_map(|fieldnorm| std::iter::repeat_n(fieldnorm, REPEAT))
.collect::<Vec<u32>>();
let postings_lists_expanded: Vec<Vec<(DocId, u32)>> = posting_lists
@@ -479,7 +482,8 @@ mod tests {
for top_k in 1..4 {
let checkpoints_for_each_pruning =
compute_checkpoints_for_each_pruning(term_scorers.clone(), top_k);
let checkpoints_manual = compute_checkpoints_manual(term_scorers.clone(), top_k);
let checkpoints_manual =
compute_checkpoints_manual(term_scorers.clone(), top_k, 100_000);
assert_eq!(checkpoints_for_each_pruning.len(), checkpoints_manual.len());
for (&(left_doc, left_score), &(right_doc, right_score)) in checkpoints_for_each_pruning
.iter()

2
src/query/boolean_query/boolean_query.rs
View File

@@ -180,7 +180,7 @@ impl BooleanQuery {
pub fn new(subqueries: Vec<(Occur, Box<dyn Query>)>) -> BooleanQuery {
// If the bool query includes at least one should clause
// and no Must or MustNot clauses, the default value is 1. Otherwise, the default value is
// 0. Keep pace with Elasticsearch.
// 0. Keep compatible with Elasticsearch.
let mut minimum_required = 0;
for (occur, _) in &subqueries {
match occur {

Some files were not shown because too many files have changed in this diff Show More