rust/tantivy
1
0
Fork 0
mirror of https://github.com/quickwit-oss/tantivy.git synced 2026-01-06 09:12:55 +00:00
Code Issues Packages Projects Releases Wiki Activity

Compare commits

base: rust:unit-test-aggregation-sum-other
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:stuhood.lazy-scorers-blocks
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

132 Commits
unit-test- ... stuhood.la

Author SHA1 Message Date
Stu Hood
147214b0eb Implement GreaterThanOrEqual and LessThanOrEqual to handle boundary cases in Chain. 2025-12-29 15:38:28 -07:00
Stu Hood
865a12f4bb Simpler implementation of first_vals_in_value_range. 2025-12-29 14:51:21 -07:00
Stu Hood
00110312c9 Fix compound filters, and remove redundant implementation in Chain implementation 2025-12-29 14:51:17 -07:00
Stu Hood
b2e980b450 Property test for Comparator/ValueRange consistency, and fixes. 2025-12-27 21:03:02 -07:00
Stu Hood
1a701b86bd Remove allow-dead-code annotation. 2025-12-27 17:56:55 -07:00
Stu Hood
ee4538d6c2 test_order_by_u64_prop 2025-12-27 16:04:24 -07:00
Stu Hood
25f1e9aa9f Move ComparableDoc to a reusable location, allowing for pushing directly from ColumnValues into a TopNComputer buffer in some cases. 2025-12-27 14:01:02 -07:00
Stu Hood
6b03b28bac Use a Buffer generic scratch buffer parameter on TopNComputer to allow for internal iteration in SegmentSortKeyComputer. 2025-12-27 13:03:56 -07:00
Stu Hood
7a5241cb83 Update comments. 2025-12-26 17:52:40 -07:00
Stu Hood
0f5e0f6f87 TODO: Audit. 2025-12-26 16:12:15 -07:00
Stu Hood
a654115d9a Squash with Optional. WIP: Still needs work: we are allocating. 2025-12-26 15:18:17 -07:00
Stu Hood
1a17515ead Convert test_order_by_compound_filtering_with_none to a proptest. 2025-12-26 14:25:01 -07:00
Stu Hood
0f1b0ce527 Optimize Optional indexes. TODO: Audit. 2025-12-26 13:01:30 -07:00
Stu Hood
0c920dfc61 Add a ValueRange filter to SegmentSortKeyComputer::segment_sort_keys. 2025-12-26 12:27:11 -07:00
Stu Hood
996fc936f6 Add null handling to first_vals_in_value_range. 2025-12-26 11:12:53 -07:00
Stu Hood
5ff38e1605 WIP: Add ValueRange cases for Comparators. 2025-12-26 11:02:19 -07:00
Stu Hood
e8a4adeedd Replace Column::first_vals with Column::first_vals_in_value_range. 2025-12-25 15:39:18 -07:00
Stu Hood
efc9e585a9 WIP: Add ValueRange::All 2025-12-25 15:16:26 -07:00
Stu Hood
f4252fc184 WIP: Add ValueRange. 2025-12-25 14:53:15 -07:00
Stu Hood
53c067d1f3 Restore laziness in ChainSegmentSortKeyComputer. 2025-12-24 10:39:26 -07:00
Stu Hood
259c1ed965 Isolate accept_sort_key_lazy to ChainSegmentSortKeyComputer. 2025-12-23 17:37:33 -07:00
Stu Hood
1afc432df8 Use an internal buffer in the SegmentSortKeyComputer. 2025-12-23 17:23:10 -07:00
Stu Hood
b8acd3ac94 WIP: Add and use segment_sort_keys to remove dynamic dispatch to the column. 2025-12-23 16:44:50 -07:00
Stu Hood
b5321d2125 Implement laziness for collect_block. 2025-12-23 15:48:36 -07:00
Stu Hood
ad3e2363fe WIP: Add failing test. 2025-12-23 15:48:34 -07:00
Stu Hood
9ec5750c25 Implement collect_block for lazy scorers. 2025-12-23 15:46:41 -07:00
Stu Hood
03f09a2b5b chore: Add support for natural-order-with-none-highest in TopDocs::order_by (#90) 
Add `ComparatorEnum::NaturalNoneHigher`, which matches Postgres's `DESC NULLS FIRST` behavior in `TopDocs::order_by`.

Expands comments on `Comparator` implementations to ensure that behavior for `None` is explicit.

Upstream as https://github.com/quickwit-oss/tantivy/pull/2780
 2025-12-23 09:15:31 -08:00
Stu Hood
9ffe4af096 Fix TopN performance regression. 
https://github.com/quickwit-oss/tantivy/pull/2777
 2025-12-17 10:43:29 -07:00
Stu Hood
c56ddcb6d7 Add an erased SortKeyComputer to sort on types which are not known until runtime. 
https://github.com/quickwit-oss/tantivy/pull/2770
 2025-12-17 10:43:29 -07:00
Ming
5b8fff154b fix: overflow in vint buffer (#88) 2025-12-17 10:43:29 -07:00
Mohammad Dashti
ff6ee3a5db fix: post-rebase fixes 
- Add missing size_hint module declaration
- Remove test-only export serialize_and_load_u64_based_column_values
- fixed quickwit CI issues
 2025-12-10 10:17:28 -08:00
Moe
eda9aa437f fix: boolean query incorrectly dropping documents when AllScorer is present (#84) 
Co-authored-by: Stu Hood <stuhood@gmail.com>
 2025-12-10 10:17:28 -08:00
Piotr Olszak
538da08eb5 Add polish stemmer (#82) 
This commit adds support for Polish language stemming.
The previously used rust-stemmers crate is abandoned and unmaintained, which blocked the addition of new languages. This change addresses a user request for Polish stemming to improve BM25 recall in their use case. The tantivy-stemmers crate is a modern, maintained alternative that also opens the door for supporting many other languages in the future.
- Added the tantivy-stemmers crate as a dependency to the workspace, alongside the existing rust-stemmers dependency (for backward compatibility)
- Introduced an internal enum that can hold an algorithm from either rust-stemmers or tantivy-stemmers
- Added Polish to the main Language enum, mapped to the new tantivy-stemmers implementation
- Updated the token stream to handle both types of stemmers internally
- Added the POLISH variant to the stopwords list
- Existing tests pass
- Added test_pl_tokenizer to verify that the Polish stemmer works correctly
 2025-12-10 10:17:28 -08:00
Moe
7bd5cc5417 fix: fixed integer overflow in ExpUnrolledLinkedList for large datasets (#80) 2025-12-10 10:17:28 -08:00
Moe
5d46137556 feat: Added multiple snippet support (#76) 
Adds `SnippetGenerator::snippets` to render multiple snippets in either score or position order.

Additionally: renames the existing `limit` and `offset` arguments to disambiguate between "match" positions (which are concatenated into fragments), and "snippet" positions.

Co-authored-by: Stu Hood <stuhood@gmail.com>
 2025-12-10 10:17:28 -08:00
Stu Hood
92c784f697 perf: Optimize TermSet for very large sets of terms. (#75) 
* Removes allocation in a bunch of places
* Removes sorting of terms if we're going to use the fast field execution method
* Adds back the (accidentally dropped) cardinality threshold
* Removes `bool` support -- using the posting lists is always more efficient for a `bool`, since there are at most two of them
* More eagerly constructs the term `HashSet` so that it happens once, rather than once per segment
 2025-12-10 10:17:28 -08:00
Stu Hood
b3541d10e1 chore: Use smaller merge buffers. (#74) 
## What

Reduce the per-segment buffer sizes from 4MB to 512KB.

## Why

#71 moved from buffers which covered the entire file to maximum 4MB buffers. But for merges with very large segment counts, we need to be using more conservative buffer sizes. 512KB will still eliminate most posting list reads: posting lists larger than 512KB will skip the buffer.
 2025-12-10 10:17:28 -08:00
Stu Hood
7183ac6cbc fix: Use smaller buffers during merging (#71) 
`MergeOptimizedInvertedIndexReader` was added in #32 in order to avoid making small reads to our underlying `FileHandle`. It did so by reading the entire content of the posting lists and positions at open time.

As that PR says:
> A likely downside to this approach is that now pg_search will be, indirectly, holding onto a lot of heap-allocated memory that was read from its block storage. Perhaps in the (near) future we can further optimize the new `MergeOptimizedInvertedIndexReader` such that it pages in blocks of a few megabytes at a time, on demand, rather than the whole file.

This PR makes that change. But it additionally removes code that was later added in #47 to borrow individual entries rather than creating `OwnedBytes` for them. I believe that this code was added due to a misunderstanding:

`OwnedBytes` is a total misnomer: the bytes are not "owned": they are immutably borrowed and reference counted. An `OwnedBytes` object can be created for any type which derefs to a slice of bytes, and can be cheaply cloned and sliced. So there is no need to actually borrow _or_ copy the buffer under the `OwnedBytes`. Removing the code that was doing so allows us to safely recreate our buffer without worrying about the lifetimes of buffers that we've handed out.
 2025-12-10 10:17:28 -08:00
Stu Hood
e0476d2eb2 fix: Add support for bool to the fast field TermSet implementation (#70) 
Missed in #69.

The `TermSet` fast fields implementation cribbed from `RangeQuery`'s fast fields implementation: ... which also has this bug. Will fix upstream.
 2025-12-10 10:17:28 -08:00
Stu Hood
9fe0899934 perf: Implement a TermSet variant which uses fast fields (#69) 
The `TermSet` `Query` currently produces one `Scorer`/`DocSet` per matched term by scanning the term dictionary and then consuming posting lists. For very large sets of terms and a fast field, it is faster to scan the fast field column while intersecting with a `HashSet` of (encoded) term values.

Following the pattern set by the two execution modes of `RangeQuery`, this PR introduces a variant of `TermSet` which uses fast fields, and then uses it when there are more than 1024 input terms (an arbitrary threshold!).

Performance is significantly improved for large `TermSet`s of primitives.
 2025-12-10 10:17:28 -08:00
Stu Hood
aaa5abb7d6 chore: Expose a method to create a segment with a particular id (#68) 
In support of https://github.com/paradedb/paradedb/pull/3203
 2025-12-10 10:17:28 -08:00
Ming
f8b8fd0321 feat: SnippetGenerator accepts limit/offset (#66) 2025-12-10 10:17:27 -08:00
Eric Ridge
cd878a5c90 fix: support MemoryArena allocations up to 4GB (#62) 
A MemoryArena should support allocations up to 4GB and https://github.com/paradedb/tantivy/pull/60 broke this by not accounting for the "max page id" when pages are now 50% the size of what the originally were.

This cleans up the code so things stay in sync if we change NUM_BITS_PAGE_ADDR again and adds a unit test
 2025-12-10 10:17:27 -08:00
Eric Ridge
30c237e895 perf: various optimizations around arenas (#60) 
- Use a bitset to track used buckets in the `SharedArenaHashmap`, allowing for more efficient iteration
- Create a global pool for both `MemoryArena` and `IndexingContext`
- Reduce the MemoryArea page size by half (it's now 512KB instead of 1MB)
- Centralize thread pool instances in `SegmentUpdater` so we can elide making them if all nthread sizes are zero
 2025-12-10 10:17:27 -08:00
Eric Ridge
b6cd39872b fix: Allow zero indexing & merging threads (#59) 
This removes a check against `IndexWriterOptions` which disallowed zero indexing worker threads (`num_worker_threads`).
 2025-12-10 10:17:27 -08:00
Stu Hood
c96d801c68 perf: Lazily load in BitpackedCodec (#56) 
We would like to be able to lazily load `BitpackedCodec` columns (similar to what 020bdffd61 did for `BlockwiseLinearCodec`), because in the context of `pg_search`, immediately constructing `OwnedBytes` means copying the entire content of the column into memory.

To do so, we expose some (slightly overlapped) block boundaries from `BitUnpacker`, and then lazily load each block when it is requested. Only the `get_val` function uses the cache: `get_row_ids_for_value_range` does not (yet), because it would be necessary to partition the row ids by block, and most of the time consumers using it are already loading reasonably large ranges anyway.

See https://github.com/paradedb/paradedb/pull/2894 for usage. There are a few 2x speedups in the benchmark suite, as well as a 1.8x speedup on a representative customer query. Unfortunately there are also some 13-19% slowdowns on aggregates: it looks like that is because aggregates use `get_vals`, for which the default implementation is to just call `get_val` in a loop.
 2025-12-10 10:17:27 -08:00
Stu Hood
7a13e0294d Avoid copying into OwnedBytes when opening a fast field column Dictionary. (#55) 
When a fast fields string/bytes `Dictionary` is opened, we currently read the entire dictionary from `FileSlice` -> `OwnedBytes`... and then immediately wrap it back into a `FileSlice`.

Switching to `Dictionary::open` preserves the `FileSlice`, such that only the portions of the `Dictionary` which are actually accessed are read from disk/buffers.
 2025-12-10 10:17:27 -08:00
Eric Ridge
20d00701ee perf: lazily open positions file (#54) 
Not all queries use positions and it's okay if we (from the perspective of `pg_search`, anyways) defer opening/loading them until they're first needed.

This is probably completely wrong for a mmap-based Directory, but we (again, `pg_search`) decided long ago that we don't care about that use case.

This saves a lot of disk I/O when an index has lots of segments and the query doesn't need positions.

As a drive by, make sure a random Vec has enough space before pushing items to it.  This showed up in the profiler, believe it or not.
 2025-12-10 10:17:27 -08:00
Eric Ridge
526afc6111 chore: internal API visibility adjustments (#53) 2025-12-10 10:17:27 -08:00
Ming
f9e4a8413b make the directory BufWriter capacity configurable (#52) 2025-12-10 10:17:27 -08:00
Ming
58124bb164 changes to make merging work (#48) 2025-12-10 10:17:27 -08:00
Eric Ridge
176f7e852a perf: remove general overhead during segment merging (#47) 2025-12-10 10:17:27 -08:00
Ming
cfa5f94114 chore: Make some delete-related functions public (#46) 2025-12-10 10:17:26 -08:00
Ming
5e449e7dda feat: SnippetGenerator can handle JSON fields (#42) 2025-12-10 10:17:26 -08:00
Stu Hood
1617459b01 Expose some methods which are necessary to create a streaming version of sorted_ords_to_term_cb. (#43) 
See https://github.com/paradedb/paradedb/pull/2612.

We might eventually want that function upstreamed, but there are more changes planned to it for https://github.com/paradedb/paradedb/issues/2619, so doing the expedient thing now.
 2025-12-10 10:17:26 -08:00
Ming
0e1a7e213e chore: allow merge_foreground to ignore the store (#40) 2025-12-10 10:17:26 -08:00
Ming
b0660ba196 chore: make some structs pub (#39) 2025-12-10 10:17:26 -08:00
Eric Ridge
936d6af471 feat: ability to directly merge segments in the foregound (#36) 
This adds new public-facing (and internal) APIs for being able to merge a list of segments in the foreground, without using any threads.  It's largely a cut-n-paste of the existing background merge code.

For pg_search, this is beneficial because it allows us to merge directly using our `MVCCDirectory` rather than going through the `ChannelDirectory`, which has quite a bit of overhead.
 2025-12-10 10:17:26 -08:00
Eric Ridge
2560de3a01 feat: IndexWriter::wait_merging_threads() return Err on merge failure (#34) 2025-12-10 10:17:26 -08:00
Eric Ridge
75a8384c2b feat: remove Directory::reconsider_merge_policy() and add other niceties to Directory API (#33) 
This removes `Directory::reconsider_merge_policy()`.  After reconsidering this, it's better to make this decision ahead of time.

Also adds a `Directory::log(message: &str)` function along with passing a `Directory` reference to `MergePolicy::compute_merge_candidates()`.

It also hits some `#[derive(Debug)]` and `#[derive(Serialize)]` on a couple of structs that can benefit.
 2025-12-10 10:17:26 -08:00
Eric Ridge
5b6da9123c feat: introduce a MergeOptimizedInvertedIndexReader (#32) 
This is probably a bit of a misnomer as it's really a "PgSearchOptimizedInvertedIndexReaderForMerge".

What we've done here is copied `InvertedIndexReader` and internally adjusted it to hold onto the complete `OwnedBytes` of the index's postings and positions.  One or two other small touch points were required to make other internal APIs compatabile with this but they don't otherwise change functionality or I/O patterns.

`MergeOptimizedInvertedIndexReader` does change I/O patterns, however, in that the merge process now does two (potentially) very large reads when it obtains the new "merge optimized inverted index reader" for each segment.  This changes access patterns such that all the reads happen up-front rather than term-by-term as the merge process is solving.

A likely downside to this approach is that now pg_search will be, indirectly, holding onto a lot of heap-allocated memory that was read from its block storage.  Perhaps in the (near) future we can further optimize the new `MergeOptimizedInvertedIndexReader` such that it pages in blocks of a few megabytes at a time, on demand, rather than the whole file.

---

Some unit tests were also updated to resolve compilation problems by PR https://github.com/paradedb/tantivy/pull/31 that for some reason didn't show in CI.  #weird
 2025-12-10 10:17:26 -08:00
Eric Ridge
8b7db36c99 feat: Add Directory::wants_cancel() function (#31) 
This adds a function named `wants_cancel() -> bool` to the `Directory` trait.  It allows a Directory implementation to indicate that it would like Tantivy to cancel an operation.

Right now, querying this function only happens during key points of index merging, but _could_ be used in other places.  Technically, segment merging is the only "black box" in tantivy that users don't otherwise have the direct ability to control.

The default implementaiton of `wants_cancel()` returns false, so there's no fear of default tantivy spuriously cancelling a merge.

The cancels happen "cleanly" such that if `wants_cancel()` returns true an `Err(TantivyError::Cancelled)` is returned from the calling function at that point, and the error result will be propogated up the stack.  No panics are raised.
 2025-12-10 10:17:26 -08:00
Eric Ridge
eabe589814 feat: ability to assign a panic handler to a Directory (#30) 
Tantivy creates thread pools for some of its background work, specifically committing and merging.

It's possible if one of the thread workers panics that rayon will simply abort the process.  This is terrible from pg_search as that takes down the entire Postgres cluster.

These changes allow a Directory to assign a panic handler that gets called in such cases.  Which allows pg_search to gracefully rollback the current transaction, while presenting the panic message to the user.
 2025-12-10 10:17:26 -08:00
Eric Ridge
65d3574dfd feat: make garbage collection opt-out (#28) 2025-12-10 10:17:26 -08:00
Ming
26d623c411 Change default index precision to microseconds (#27) 2025-12-10 10:17:25 -08:00
Eric Ridge
0552dddeb9 feat: delete docs by (SegmentId, DocId) (#26) 
This teaches tantivy how to "directly" delete a document in a segment.
    
Our use case from pg_search is that we already know the segment_id and doc_id so it's waaaaay more efficient for us to delete docs through our `ambulkdelete()` routine.

It avoids doing a search, and all the stuff around that, for each of our "ctid" terms that we want to delete.
 2025-12-10 10:17:25 -08:00
Eric Ridge
1b88bb61f9 feat: Add ability to construct a SegmentId from raw bytes (#24) 
This allows a `SegmentId` to be constructed from a `[u8; 16]` byte array.  

It also adds a `impl Default for SegementId`, which defaults to all nulls
 2025-12-10 10:17:25 -08:00
Eric Ridge
16da31cf06 perf: make the footer fixed width (#23) 
Prior to this commit, the Footer tantivy serialized at the end of every file included a json blob that could be an arbitrary size.

This changes the Footer to be exactly 24 bytes (6 u32s), making sure to keep the `crc` value.  The other change we make here is to not actually read/validate the footer bytes when opening a file.

From pg_search's perspective, this is quite unnecessary Postgres buffer cache I/O and increases index/segment opening overhead, which is something pg_search does often for each query.

Two tests are ignored here as they test physical index files stored here in the repo that this change completely breaks.
 2025-12-10 10:17:25 -08:00
Eric Ridge
658b9b22e0 perf: remove some fast fields loading overhead (#22) 
This removes up some overhead the profiler exposed.  In the case I was testing, fast fields no longer shows up in the profile at all.

I also renamed `BlockWithLength` to `BlockWithData`
 2025-12-10 10:17:25 -08:00
Eric Ridge
95661fba30 perf: teach SegmentReader to lazily open/read its various SegmentComponents (#20) 
This overhauls `SegmentReader` to put its various components behind `OnceLock`s such that they can be opened and read on their first use, as oppoed when a SegmentReader is constructed -- which is once for every segment when an Index is opened.

This has a negative impact on some of Tantivy's expectations in that an existing SegementReader can still read from physical files that were deleted by a merge.  This isn't true now that the segment's physical files aren't opened until needed.  As such, I've `#[ignore]`'d six tests that expose this problem.

From our (pg_search's) side of things, we don't really have physical files and don't need to rely on the filesystem/kernel to allow reading unlinked files that are still open.

Overall, this cuts down a signficiant number of disk reads during pg_search's query planning.  With my test data it goes from 808 individual reads totalling 999,799 bytes, to 18 reads totalling 814,514 bytes.

This reduces the time it takes to plan a simple query from about 1.4ms to 0.436ms -- roughly a 3.2x improvement.
 2025-12-10 10:17:25 -08:00
Philippe Noël
ddd169b77c chore: Don't do codecov (#21) 2025-12-10 10:17:25 -08:00
Eric Ridge
bb4c4b8522 perf: push FileSlices down through most of fast fields (#19) 
This PR modifies internal API signatures and implementation details so that `FileSlice`s are passed down into the innards of (at least) the `BlockwiseLinearCodec`.  This allows tantivy to defer dereferencing large slices of bytes when reading numeric fast fields, and instead dereference only the slice of bytes it needs for any given compressed Block.

The motivation here is for external `Directory` implementations where it's not exactly efficient to dereference large slices of bytes.
 2025-12-10 10:17:25 -08:00
Neil Hansen
ffa558e3a9 fix: tests in ci (#18) 2025-12-10 10:17:25 -08:00
Neil Hansen
a35e3dcb5a suppress warnings after rebase 2025-12-10 10:17:25 -08:00
Neil Hansen
1e3998fbad implement fuzzy scoring in sstable 2025-12-10 10:17:25 -08:00
Neil Hansen
f3df079d6b chore: point tantivy-fst to paradedb fork to fix regex 2025-12-10 10:17:24 -08:00
Ming Ying
f7c0335857 comments 2025-12-10 10:17:24 -08:00
Ming Ying
2584325e0d add reconsider_merge_policy to directory 2025-12-10 10:17:24 -08:00
Eric B. Ridge
1f2c2d0c8a fix compilation warnings on rust v1.83 2025-12-10 10:17:24 -08:00
Eric Ridge
91db6909d1 Add a payload: &mut (dyn Any + '_) argument to Directory::save_meta() (#17) 2025-12-10 10:17:24 -08:00
Ming Ying
7639b47615 small changes to make MVCC work with delete 2025-12-10 10:17:24 -08:00
Ming Ying
8b55f0f355 Make DeleteMeta pub 2025-12-10 10:17:24 -08:00
Ming Ying
8d29f19110 make save_metas provide previous metas 2025-12-10 10:17:24 -08:00
Ming Ying
d742d3277a undo changes to segment_updater.rs 2025-12-10 10:17:24 -08:00
Eric B. Ridge
3afe3714a2 no pgrx, please 2025-12-10 10:17:24 -08:00
Ming Ying
67ea8e53a8 quickwit compiles 2025-12-10 10:17:24 -08:00
Ming Ying
3adc85c017 Directory trait can read/write meta/managed 2025-12-10 10:17:24 -08:00
Ming
6bb3a22c98 expose AddOperation and with_max_doc (#7) 2025-12-10 10:17:23 -08:00
Ming
5503cfb8ef Fix managed paths (#5) 2025-12-10 10:17:23 -08:00
Alexander Alexandrov
ea0e88ae4b feat: implement TokenFilter for Option<F> (#4) 2025-12-10 10:17:23 -08:00
Neil Hansen
dee2dd3f21 Use Levenshtein distance to score documents in fuzzy term queries 2025-12-10 10:17:19 -08:00
Philippe Noël
794ff1ffc9 chore: Make Language hashable (#79) (#2763) 
Co-authored-by: Ming <ming.ying.nyc@gmail.com>
 2025-12-10 15:38:43 +01:00
PSeitz-dd
c6912ce89a Handle JSON fields and columnar in space_usage (#2761) 
return field names in space_usage instead of `Field`
more detailed info for columns
 2025-12-10 20:33:33 +08:00
PSeitz
618e3bd11b Term and IndexingTerm cleanup (#2750) 
* refactor term

* add deprecated functions

---------

Co-authored-by: Pascal Seitz <pascal.seitz@datadoghq.com>
 2025-12-05 09:48:40 +08:00
PSeitz
b2f99c6217 add term->histogram benchmark (#2758) 
* add term->histogram benchmark

* add more term aggs

---------

Co-authored-by: Pascal Seitz <pascal.seitz@datadoghq.com>
 2025-12-04 02:29:37 +01:00
PSeitz
76de5bab6f fix unsafe warnings (#2757) 2025-12-03 20:15:21 +08:00
rustmailer
b7eb31162b docs: add usage example to README (#2743) 2025-12-02 21:56:57 +01:00
Paul Masurel
63c66005db Lazy scorers (#2726) 
* Refactoring of the score tweaker into `SortKeyComputer`s to unlock two features.

- Allow lazy evaluation of score. As soon as we identified that a doc won't
reach the topK threshold, we can stop the evaluation.
- Allow for a different segment level score, segment level score and their conversion.

This PR breaks public API, but fixing code is straightforward.

* Bumping tantivy version

---------

Co-authored-by: Paul Masurel <paul.masurel@datadoghq.com>
 2025-12-01 15:38:57 +01:00
Paul Masurel
7d513a44c5 Added some benchmark for top K by a fast field (#2754) 
Also removed query parsing from the bench code.

Co-authored-by: Paul Masurel <paul.masurel@datadoghq.com>
 2025-12-01 14:58:29 +01:00
Stu Hood
ca87fcd454 Implement collect_block for Collectors which wrap other Collectors (#2727) 
* Implement `collect_block` for tuple Collectors, and for MultiCollector.

* Two more.
 2025-12-01 12:26:29 +01:00
Ang
08a92675dc Fix typos again (#2753) 
Found via `codespell -S benches,stopwords.rs -L
womens,parth,abd,childs,ond,ser,ue,mot,hel,atleast,pris,claus,allo`
 2025-12-01 12:15:41 +01:00
Raphaël Cohen
f7f4b354d6 fix: Handle phrase prefixed with star (#2751) 
Signed-off-by: Darkheir <raphael.cohen@sekoia.io>
 2025-12-01 11:43:25 +01:00
Paul Masurel
25d44fcec8 Revert "remove unused columnar api (#2742)" (#2748) 
* Revert "remove unused columnar api (#2742)"

This reverts commit 8725594d47.

* Clippy comment + removing fill_vals

---------

Co-authored-by: Paul Masurel <paul.masurel@datadoghq.com>
 2025-11-26 17:44:02 +01:00
PSeitz-dd
842fe9295f split Term in Term and IndexingTerm (#2744) 
* split Term in Term and IndexingTerm

* add append_json_path to JsonTermSerializer
 2025-11-26 16:48:59 +01:00
Paul Masurel
f88b7200b2 Optimization when posting list are saturated. (#2745) 
* Optimization when posting list are saturated.

If a posting list doc freq is the segment reader's
max_doc, and if scoring does not matter, we can replace it
by a AllScorer.

In turn, in a boolean query, we can dismiss  all scorers and
empty scorers, to accelerate the request.

* Added range query optimization

* CR comment

* CR comments

* CR comment

---------

Co-authored-by: Paul Masurel <paul.masurel@datadoghq.com>
 2025-11-26 15:50:57 +01:00
PSeitz-dd
8725594d47 remove unused columnar api (#2742) 2025-11-21 18:07:25 +01:00
PSeitz
43a784671a clippy (#2741) 
Co-authored-by: Pascal Seitz <pascal.seitz@datadoghq.com>
 2025-11-21 18:07:03 +01:00
Paul Masurel
c363bbd23d Optimize term aggregation with low cardinality + some refactoring (#2740) 
This introduce an optimization of top level term aggregation on field with a low cardinality.

We then use a Vec as the underlying map.
In addition, we buffer subaggregations.

---------

Co-authored-by: Pascal Seitz <pascal.seitz@datadoghq.com>
Co-authored-by: Paul Masurel <paul@quickwit.io>
 2025-11-21 14:46:29 +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
253 changed files with 21948 additions and 4716 deletions
Expand all files Collapse all files

29
.github/workflows/coverage.yml vendored
View File

@@ -1,29 +0,0 @@
name: Coverage
on:
push:
branches: [main]
# Ensures that we cancel running jobs for the same PR / same workflow.
concurrency:
group: ${{ github.workflow }}-${{ github.event.pull_request.number || github.ref }}
cancel-in-progress: true
jobs:
coverage:
runs-on: ubuntu-latest
steps:
- uses: actions/checkout@v4
- name: Install Rust
run: rustup toolchain install nightly-2024-07-01 --profile minimal --component llvm-tools-preview
- uses: Swatinem/rust-cache@v2
- uses: taiki-e/install-action@cargo-llvm-cov
- name: Generate code coverage
run: cargo +nightly-2024-07-01 llvm-cov --all-features --workspace --doctests --lcov --output-path lcov.info
- name: Upload coverage to Codecov
uses: codecov/codecov-action@v3
continue-on-error: true
with:
token: ${{ secrets.CODECOV_TOKEN }} # not required for public repos
files: lcov.info
fail_ci_if_error: true

4
.github/workflows/test.yml vendored
View File

@@ -76,7 +76,9 @@ jobs:
profile: minimal
override: true
- uses: taiki-e/install-action@nextest
- uses: taiki-e/install-action@v2
with:
tool: 'nextest'
- uses: Swatinem/rust-cache@v2
- name: Run tests

5
.gitignore vendored
View File

@@ -6,7 +6,6 @@ target
target/debug
.vscode
target/release
Cargo.lock
benchmark
.DS_Store
*.bk
@@ -15,3 +14,7 @@ trace.dat
cargo-timing*
control
variable
# for `sample record -p`
profile.json
profile.json.gz

22
CHANGELOG.md
View File

@@ -14,6 +14,18 @@ Tantivy 0.25
- 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.
@@ -66,7 +78,7 @@ This will slightly increase space and access time. [#2439](https://github.com/qu
- **Store DateTime as nanoseconds in doc store** DateTime in the doc store was truncated to microseconds previously. This removes this truncation, while still keeping backwards compatibility. [#2486](https://github.com/quickwit-oss/tantivy/pull/2486)(@PSeitz)
- **Performace/Memory**
- **Performance/Memory**
- lift clauses in LogicalAst for optimized ast during execution [#2449](https://github.com/quickwit-oss/tantivy/pull/2449)(@PSeitz)
- Use Vec instead of BTreeMap to back OwnedValue object [#2364](https://github.com/quickwit-oss/tantivy/pull/2364)(@fulmicoton)
- Replace TantivyDocument with CompactDoc. CompactDoc is much smaller and provides similar performance. [#2402](https://github.com/quickwit-oss/tantivy/pull/2402)(@PSeitz)
@@ -96,6 +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
================================

2361
Cargo.lock generated Normal file
View File

File diff suppressed because it is too large Load Diff

41
Cargo.toml
View File

@@ -1,6 +1,6 @@
[package]
name = "tantivy"
version = "0.25.0"
version = "0.26.0"
authors = ["Paul Masurel <paul.masurel@gmail.com>"]
license = "MIT"
categories = ["database-implementations", "data-structures"]
@@ -21,11 +21,11 @@ byteorder = "1.4.3"
crc32fast = "1.3.2"
once_cell = "1.10.0"
regex = { version = "1.5.5", default-features = false, features = [
"std",
"unicode",
"std",
"unicode",
] }
aho-corasick = "1.0"
tantivy-fst = "0.5"
tantivy-fst = { git = "https://github.com/paradedb/fst.git" }
memmap2 = { version = "0.9.0", optional = true }
lz4_flex = { version = "0.11", default-features = false, optional = true }
zstd = { version = "0.13", optional = true, default-features = false }
@@ -38,9 +38,10 @@ levenshtein_automata = "0.2.1"
uuid = { version = "1.0.0", features = ["v4", "serde"] }
crossbeam-channel = "0.5.4"
rust-stemmers = "1.2.0"
tantivy-stemmers = { version = "0.4.0", default-features = false, features = ["polish_yarovoy"] }
downcast-rs = "2.0.1"
bitpacking = { version = "0.9.2", default-features = false, features = [
"bitpacker4x",
"bitpacker4x",
] }
census = "0.4.2"
rustc-hash = "2.0.0"
@@ -48,6 +49,10 @@ thiserror = "2.0.1"
htmlescape = "0.3.1"
fail = { version = "0.5.0", optional = true }
time = { version = "0.3.35", features = ["serde-well-known"] }
# TODO: We have integer wrappers with PartialOrd, and a misfeature of
# `deranged` causes inference to fail in a bunch of cases. See
# https://github.com/jhpratt/deranged/issues/18#issuecomment-2746844093
deranged = "=0.4.0"
smallvec = "1.8.0"
rayon = "1.5.2"
lru = "0.12.0"
@@ -69,6 +74,8 @@ 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"
parking_lot = "0.12.4"
typetag = "0.2.21"
[target.'cfg(windows)'.dependencies]
winapi = "0.3.9"
@@ -134,14 +141,14 @@ compare_hash_only = ["stacker/compare_hash_only"]
[workspace]
members = [
"query-grammar",
"bitpacker",
"common",
"ownedbytes",
"stacker",
"sstable",
"tokenizer-api",
"columnar",
"query-grammar",
"bitpacker",
"common",
"ownedbytes",
"stacker",
"sstable",
"tokenizer-api",
"columnar",
]
# Following the "fail" crate best practises, we isolate
@@ -167,3 +174,11 @@ harness = false
[[bench]]
name = "agg_bench"
harness = false
[[bench]]
name = "exists_json"
harness = false
[[bench]]
name = "and_or_queries"
harness = false

3
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
@@ -125,6 +123,7 @@ You can also find other bindings on [GitHub](https://github.com/search?q=tantivy
- [seshat](https://github.com/matrix-org/seshat/): A matrix message database/indexer
- [tantiny](https://github.com/baygeldin/tantiny): Tiny full-text search for Ruby
- [lnx](https://github.com/lnx-search/lnx): adaptable, typo tolerant search engine with a REST API
- [Bichon](https://github.com/rustmailer/bichon): A lightweight, high-performance Rust email archiver with WebUI
- and [more](https://github.com/search?q=tantivy)!
### On average, how much faster is Tantivy compared to Lucene?

2
TODO.txt
View File

@@ -10,7 +10,7 @@ rename FastFieldReaders::open to load
remove fast field reader
find a way to unify the two DateTime.
readd type check in the filter wrapper
re-add type check in the filter wrapper
add unit test on columnar list columns.

180
benches/agg_bench.rs
View File

@@ -1,5 +1,6 @@
use binggan::plugins::PeakMemAllocPlugin;
use binggan::{black_box, InputGroup, PeakMemAlloc, INSTRUMENTED_SYSTEM};
use rand::distributions::WeightedIndex;
use rand::prelude::SliceRandom;
use rand::rngs::StdRng;
use rand::{Rng, SeedableRng};
@@ -54,11 +55,19 @@ fn bench_agg(mut group: InputGroup<Index>) {
register!(group, extendedstats_f64);
register!(group, percentiles_f64);
register!(group, terms_few);
register!(group, terms_all_unique);
register!(group, terms_many);
register!(group, terms_many_top_1000);
register!(group, terms_many_order_by_term);
register!(group, terms_many_with_top_hits);
register!(group, terms_all_unique_with_avg_sub_agg);
register!(group, terms_many_with_avg_sub_agg);
register!(group, terms_few_with_avg_sub_agg);
register!(group, terms_status_with_avg_sub_agg);
register!(group, terms_status);
register!(group, terms_few_with_histogram);
register!(group, terms_status_with_histogram);
register!(group, terms_many_json_mixed_type_with_avg_sub_agg);
register!(group, cardinality_agg);
@@ -71,8 +80,15 @@ 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();
}
@@ -123,12 +139,12 @@ fn extendedstats_f64(index: &Index) {
}
fn percentiles_f64(index: &Index) {
let agg_req = json!({
"mypercentiles": {
"percentiles": {
"field": "score_f64",
"percents": [ 95, 99, 99.9 ]
"mypercentiles": {
"percentiles": {
"field": "score_f64",
"percents": [ 95, 99, 99.9 ]
}
}
}
});
execute_agg(index, agg_req);
}
@@ -165,6 +181,19 @@ fn terms_few(index: &Index) {
});
execute_agg(index, agg_req);
}
fn terms_status(index: &Index) {
let agg_req = json!({
"my_texts": { "terms": { "field": "text_few_terms_status" } },
});
execute_agg(index, agg_req);
}
fn terms_all_unique(index: &Index) {
let agg_req = json!({
"my_texts": { "terms": { "field": "text_all_unique_terms" } },
});
execute_agg(index, agg_req);
}
fn terms_many(index: &Index) {
let agg_req = json!({
"my_texts": { "terms": { "field": "text_many_terms" } },
@@ -213,6 +242,63 @@ fn terms_many_with_avg_sub_agg(index: &Index) {
});
execute_agg(index, agg_req);
}
fn terms_all_unique_with_avg_sub_agg(index: &Index) {
let agg_req = json!({
"my_texts": {
"terms": { "field": "text_all_unique_terms" },
"aggs": {
"average_f64": { "avg": { "field": "score_f64" } }
}
},
});
execute_agg(index, agg_req);
}
fn terms_few_with_histogram(index: &Index) {
let agg_req = json!({
"my_texts": {
"terms": { "field": "text_few_terms" },
"aggs": {
"histo": {"histogram": { "field": "score_f64", "interval": 10 }}
}
}
});
execute_agg(index, agg_req);
}
fn terms_status_with_histogram(index: &Index) {
let agg_req = json!({
"my_texts": {
"terms": { "field": "text_few_terms_status" },
"aggs": {
"histo": {"histogram": { "field": "score_f64", "interval": 10 }}
}
}
});
execute_agg(index, agg_req);
}
fn terms_few_with_avg_sub_agg(index: &Index) {
let agg_req = json!({
"my_texts": {
"terms": { "field": "text_few_terms" },
"aggs": {
"average_f64": { "avg": { "field": "score_f64" } }
}
},
});
execute_agg(index, agg_req);
}
fn terms_status_with_avg_sub_agg(index: &Index) {
let agg_req = json!({
"my_texts": {
"terms": { "field": "text_few_terms_status" },
"aggs": {
"average_f64": { "avg": { "field": "score_f64" } }
}
},
});
execute_agg(index, agg_req);
}
fn terms_many_json_mixed_type_with_avg_sub_agg(index: &Index) {
let agg_req = json!({
"my_texts": {
@@ -339,6 +425,17 @@ 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": {
@@ -386,14 +483,21 @@ fn get_test_index_bench(cardinality: Cardinality) -> tantivy::Result<Index> {
.set_stored();
let text_field = schema_builder.add_text_field("text", text_fieldtype);
let json_field = schema_builder.add_json_field("json", FAST);
let text_field_all_unique_terms =
schema_builder.add_text_field("text_all_unique_terms", STRING | FAST);
let text_field_many_terms = schema_builder.add_text_field("text_many_terms", STRING | FAST);
let text_field_many_terms = schema_builder.add_text_field("text_many_terms", STRING | FAST);
let text_field_few_terms = schema_builder.add_text_field("text_few_terms", STRING | FAST);
let text_field_few_terms_status =
schema_builder.add_text_field("text_few_terms_status", STRING | FAST);
let score_fieldtype = tantivy::schema::NumericOptions::default().set_fast();
let score_field = schema_builder.add_u64_field("score", score_fieldtype.clone());
let score_field_f64 = schema_builder.add_f64_field("score_f64", score_fieldtype.clone());
let score_field_i64 = schema_builder.add_i64_field("score_i64", score_fieldtype);
let index = Index::create_from_tempdir(schema_builder.build())?;
let few_terms_data = ["INFO", "ERROR", "WARN", "DEBUG"];
// Approximate production log proportions: INFO dominant, WARN and DEBUG occasional, ERROR rare.
let log_level_distribution = WeightedIndex::new([80u32, 3, 12, 5]).unwrap();
let lg_norm = rand_distr::LogNormal::new(2.996f64, 0.979f64).unwrap();
@@ -409,15 +513,21 @@ fn get_test_index_bench(cardinality: Cardinality) -> tantivy::Result<Index> {
index_writer.add_document(doc!())?;
}
if cardinality == Cardinality::Multivalued {
let log_level_sample_a = few_terms_data[log_level_distribution.sample(&mut rng)];
let log_level_sample_b = few_terms_data[log_level_distribution.sample(&mut rng)];
index_writer.add_document(doc!(
json_field => json!({"mixed_type": 10.0}),
json_field => json!({"mixed_type": 10.0}),
text_field => "cool",
text_field => "cool",
text_field_all_unique_terms => "cool",
text_field_all_unique_terms => "coolo",
text_field_many_terms => "cool",
text_field_many_terms => "cool",
text_field_few_terms => "cool",
text_field_few_terms => "cool",
text_field_few_terms_status => log_level_sample_a,
text_field_few_terms_status => log_level_sample_b,
score_field => 1u64,
score_field => 1u64,
score_field_f64 => lg_norm.sample(&mut rng),
@@ -442,8 +552,10 @@ fn get_test_index_bench(cardinality: Cardinality) -> tantivy::Result<Index> {
index_writer.add_document(doc!(
text_field => "cool",
json_field => json,
text_field_all_unique_terms => format!("unique_term_{}", rng.gen::<u64>()),
text_field_many_terms => many_terms_data.choose(&mut rng).unwrap().to_string(),
text_field_few_terms => few_terms_data.choose(&mut rng).unwrap().to_string(),
text_field_few_terms_status => few_terms_data[log_level_distribution.sample(&mut rng)],
score_field => val as u64,
score_field_f64 => lg_norm.sample(&mut rng),
score_field_i64 => val as i64,
@@ -460,3 +572,61 @@ 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);
}

218
benches/and_or_queries.rs Normal file
View File

@@ -0,0 +1,218 @@
// 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, BenchGroup, BenchRunner};
use rand::prelude::*;
use rand::rngs::StdRng;
use rand::SeedableRng;
use tantivy::collector::sort_key::SortByStaticFastValue;
use tantivy::collector::{Collector, Count, TopDocs};
use tantivy::query::{Query, QueryParser};
use tantivy::schema::{Schema, FAST, TEXT};
use tantivy::{doc, Index, Order, ReloadPolicy, Searcher};
#[derive(Clone)]
struct BenchIndex {
#[allow(dead_code)]
index: Index,
searcher: Searcher,
query_parser: QueryParser,
}
/// 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 f_score = schema_builder.add_u64_field("score", FAST);
let f_score2 = schema_builder.add_u64_field("score2", FAST);
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_with_num_threads(1, 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 score = rng.gen_range(0u64..100u64);
let score2 = rng.gen_range(0u64..100_000u64);
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(" "),
f_score=>score,
f_score2=>score2,
))
.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 queries = &["a", "+a +b", "+a +b +c", "a OR b", "a OR b OR c"];
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);
for (view_name, bench_index) in [("single_field", single_view), ("multi_field", multi_view)]
{
// Single-field group: default field is body only
let mut group = runner.new_group();
group.set_name(format!("{} — {}", view_name, label));
for query_str in queries {
add_bench_task(&mut group, &bench_index, query_str, Count, "count");
add_bench_task(
&mut group,
&bench_index,
query_str,
TopDocs::with_limit(10).order_by_score(),
"top10",
);
add_bench_task(
&mut group,
&bench_index,
query_str,
TopDocs::with_limit(10).order_by_fast_field::<u64>("score", Order::Asc),
"top10_by_ff",
);
add_bench_task(
&mut group,
&bench_index,
query_str,
TopDocs::with_limit(10).order_by((
SortByStaticFastValue::<u64>::for_field("score"),
SortByStaticFastValue::<u64>::for_field("score2"),
)),
"top10_by_2ff",
);
}
group.run();
}
}
}
fn add_bench_task<C: Collector + 'static>(
bench_group: &mut BenchGroup,
bench_index: &BenchIndex,
query_str: &str,
collector: C,
collector_name: &str,
) {
let task_name = format!("{}_{}", query_str.replace(" ", "_"), collector_name);
let query = bench_index.query_parser.parse_query(query_str).unwrap();
let search_task = SearchTask {
searcher: bench_index.searcher.clone(),
collector,
query,
};
bench_group.register(task_name, move |_| black_box(search_task.run()));
}
struct SearchTask<C: Collector> {
searcher: Searcher,
collector: C,
query: Box<dyn Query>,
}
impl<C: Collector> SearchTask<C> {
#[inline(never)]
pub fn run(&self) -> usize {
self.searcher.search(&self.query, &self.collector).unwrap();
1
}
}

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
}

3
bitpacker/Cargo.toml
View File

@@ -11,9 +11,6 @@ keywords = []
documentation = "https://docs.rs/tantivy-bitpacker/latest/tantivy_bitpacker"
homepage = "https://github.com/quickwit-oss/tantivy"
# See more keys and their definitions at https://doc.rust-lang.org/cargo/reference/manifest.html
[dependencies]
bitpacking = { version = "0.9.2", default-features = false, features = ["bitpacker1x"] }

125
bitpacker/src/bitpacker.rs
View File

@@ -65,10 +65,16 @@ impl BitPacker {
#[derive(Clone, Debug, Default, Copy)]
pub struct BitUnpacker {
num_bits: usize,
num_bits: u32,
mask: u64,
}
pub type BlockNumber = usize;
// 16k
const BLOCK_SIZE_MIN_POW: u8 = 14;
const BLOCK_SIZE_MIN: usize = 2 << BLOCK_SIZE_MIN_POW;
impl BitUnpacker {
/// Creates a bit unpacker, that assumes the same bitwidth for all values.
///
@@ -82,8 +88,9 @@ impl BitUnpacker {
} else {
(1u64 << num_bits) - 1u64
};
BitUnpacker {
num_bits: usize::from(num_bits),
num_bits: u32::from(num_bits),
mask,
}
}
@@ -92,16 +99,69 @@ impl BitUnpacker {
self.num_bits as u8
}
/// Calculates a block number for the given `idx`.
#[inline]
pub fn block_num(&self, idx: u32) -> BlockNumber {
// Find the address in bits of the index.
let addr_in_bits = (idx * self.num_bits) as usize;
// Then round down to the nearest byte.
let addr_in_bytes = addr_in_bits >> 3;
// And compute the containing BlockNumber.
addr_in_bytes >> (BLOCK_SIZE_MIN_POW + 1)
}
/// Given a block number and dataset length, calculates a data Range for the block.
pub fn block(&self, block: BlockNumber, data_len: usize) -> Range<usize> {
let block_addr = block << (BLOCK_SIZE_MIN_POW + 1);
// We extend the end of the block by a constant factor, so that it overlaps the next
// block. That ensures that we never need to read on a block boundary.
block_addr..(std::cmp::min(block_addr + BLOCK_SIZE_MIN + 8, data_len))
}
/// Calculates the number of blocks for the given data_len.
///
/// Usually only called at startup to pre-allocate structures.
pub fn block_count(&self, data_len: usize) -> usize {
let block_count = data_len / (BLOCK_SIZE_MIN as usize);
if data_len % (BLOCK_SIZE_MIN as usize) == 0 {
block_count
} else {
block_count + 1
}
}
/// Returns a range within the data which covers the given id_range.
///
/// NOTE: This method is used for batch reads which bypass blocks to avoid dealing with block
/// boundaries.
#[inline]
pub fn block_oblivious_range(&self, id_range: Range<u32>, data_len: usize) -> Range<usize> {
let start_in_bits = id_range.start * self.num_bits;
let start = (start_in_bits >> 3) as usize;
let end_in_bits = id_range.end * self.num_bits;
let end = (end_in_bits >> 3) as usize;
// TODO: We fetch more than we need and then truncate.
start..(std::cmp::min(end + 8, data_len))
}
#[inline]
pub fn get(&self, idx: u32, data: &[u8]) -> u64 {
let addr_in_bits = idx as usize * self.num_bits;
let addr = addr_in_bits >> 3;
self.get_from_subset(idx, 0, data)
}
/// Get the value at the given idx, which must exist within the given subset of the data.
#[inline]
pub fn get_from_subset(&self, idx: u32, data_offset: usize, data: &[u8]) -> u64 {
let addr_in_bits = idx * self.num_bits;
let addr = (addr_in_bits >> 3) as usize - data_offset;
if addr + 8 > data.len() {
if self.num_bits == 0 {
return 0;
}
let bit_shift = addr_in_bits & 7;
return self.get_slow_path(addr, bit_shift as u32, data);
return self.get_slow_path(addr, bit_shift, data);
}
let bit_shift = addr_in_bits & 7;
let bytes: [u8; 8] = (&data[addr..addr + 8]).try_into().unwrap();
@@ -113,6 +173,7 @@ impl BitUnpacker {
#[inline(never)]
fn get_slow_path(&self, addr: usize, bit_shift: u32, data: &[u8]) -> u64 {
let mut bytes: [u8; 8] = [0u8; 8];
let available_bytes = data.len() - addr;
// This function is meant to only be called if we did not have 8 bytes to load.
debug_assert!(available_bytes < 8);
@@ -128,26 +189,25 @@ impl BitUnpacker {
// #Panics
//
// This methods panics if `num_bits` is > 32.
fn get_batch_u32s(&self, start_idx: u32, data: &[u8], output: &mut [u32]) {
fn get_batch_u32s(&self, start_idx: u32, data_offset: usize, data: &[u8], output: &mut [u32]) {
assert!(
self.bit_width() <= 32,
"Bitwidth must be <= 32 to use this method."
);
let end_idx: u32 = start_idx + output.len() as u32;
let end_idx = start_idx + output.len() as u32;
// We use `usize` here to avoid overflow issues.
let end_bit_read = (end_idx as usize) * self.num_bits;
let end_bit_read = end_idx * self.num_bits;
let end_byte_read = (end_bit_read + 7) / 8;
assert!(
end_byte_read <= data.len(),
end_byte_read as usize <= data_offset + data.len(),
"Requested index is out of bounds."
);
// Simple slow implementation of get_batch_u32s, to deal with our ramps.
let get_batch_ramp = |start_idx: u32, output: &mut [u32]| {
for (out, idx) in output.iter_mut().zip(start_idx..) {
*out = self.get(idx, data) as u32;
*out = self.get_from_subset(idx, data_offset, data) as u32;
}
};
@@ -160,24 +220,24 @@ impl BitUnpacker {
// We want the start of the fast track to start align with bytes.
// A sufficient condition is to start with an idx that is a multiple of 8,
// so highway start is the closest multiple of 8 that is >= start_idx.
let entrance_ramp_len: u32 = 8 - (start_idx % 8) % 8;
let entrance_ramp_len = 8 - (start_idx % 8) % 8;
let highway_start: u32 = start_idx + entrance_ramp_len;
if highway_start + (BitPacker1x::BLOCK_LEN as u32) > end_idx {
if highway_start + BitPacker1x::BLOCK_LEN as u32 > end_idx {
// We don't have enough values to have even a single block of highway.
// Let's just supply the values the simple way.
get_batch_ramp(start_idx, output);
return;
}
let num_blocks: usize = (end_idx - highway_start) as usize / BitPacker1x::BLOCK_LEN;
let num_blocks: u32 = (end_idx - highway_start) / BitPacker1x::BLOCK_LEN as u32;
// Entrance ramp
get_batch_ramp(start_idx, &mut output[..entrance_ramp_len as usize]);
// Highway
let mut offset = (highway_start as usize * self.num_bits) / 8;
let mut offset = ((highway_start * self.num_bits) as usize / 8) - data_offset;
let mut output_cursor = (highway_start - start_idx) as usize;
for _ in 0..num_blocks {
offset += BitPacker1x.decompress(
@@ -189,7 +249,7 @@ impl BitUnpacker {
}
// Exit ramp
let highway_end: u32 = highway_start + (num_blocks * BitPacker1x::BLOCK_LEN) as u32;
let highway_end = highway_start + num_blocks * BitPacker1x::BLOCK_LEN as u32;
get_batch_ramp(highway_end, &mut output[output_cursor..]);
}
@@ -199,16 +259,27 @@ impl BitUnpacker {
id_range: Range<u32>,
data: &[u8],
positions: &mut Vec<u32>,
) {
self.get_ids_for_value_range_from_subset(range, id_range, 0, data, positions)
}
pub fn get_ids_for_value_range_from_subset(
&self,
range: RangeInclusive<u64>,
id_range: Range<u32>,
data_offset: usize,
data: &[u8],
positions: &mut Vec<u32>,
) {
if self.bit_width() > 32 {
self.get_ids_for_value_range_slow(range, id_range, data, positions)
self.get_ids_for_value_range_slow(range, id_range, data_offset, data, positions)
} else {
if *range.start() > u32::MAX as u64 {
positions.clear();
return;
}
let range_u32 = (*range.start() as u32)..=(*range.end()).min(u32::MAX as u64) as u32;
self.get_ids_for_value_range_fast(range_u32, id_range, data, positions)
self.get_ids_for_value_range_fast(range_u32, id_range, data_offset, data, positions)
}
}
@@ -216,6 +287,7 @@ impl BitUnpacker {
&self,
range: RangeInclusive<u64>,
id_range: Range<u32>,
data_offset: usize,
data: &[u8],
positions: &mut Vec<u32>,
) {
@@ -223,7 +295,7 @@ impl BitUnpacker {
for i in id_range {
// If we cared we could make this branchless, but the slow implementation should rarely
// kick in.
let val = self.get(i, data);
let val = self.get_from_subset(i, data_offset, data);
if range.contains(&val) {
positions.push(i);
}
@@ -234,11 +306,12 @@ impl BitUnpacker {
&self,
value_range: RangeInclusive<u32>,
id_range: Range<u32>,
data_offset: usize,
data: &[u8],
positions: &mut Vec<u32>,
) {
positions.resize(id_range.len(), 0u32);
self.get_batch_u32s(id_range.start, data, positions);
self.get_batch_u32s(id_range.start, data_offset, data, positions);
crate::filter_vec::filter_vec_in_place(value_range, id_range.start, positions)
}
}
@@ -258,7 +331,7 @@ mod test {
bitpacker.write(val, num_bits, &mut data).unwrap();
}
bitpacker.close(&mut data).unwrap();
assert_eq!(data.len(), ((num_bits as usize) * len + 7) / 8);
assert_eq!(data.len(), ((num_bits as usize) * len).div_ceil(8));
let bitunpacker = BitUnpacker::new(num_bits);
(bitunpacker, vals, data)
}
@@ -304,7 +377,7 @@ mod test {
bitpacker.write(val, num_bits, &mut buffer).unwrap();
}
bitpacker.flush(&mut buffer).unwrap();
assert_eq!(buffer.len(), (vals.len() * num_bits as usize + 7) / 8);
assert_eq!(buffer.len(), (vals.len() * num_bits as usize).div_ceil(8));
let bitunpacker = BitUnpacker::new(num_bits);
let max_val = if num_bits == 64 {
u64::MAX
@@ -329,14 +402,14 @@ mod test {
fn test_get_batch_panics_over_32_bits() {
let bitunpacker = BitUnpacker::new(33);
let mut output: [u32; 1] = [0u32];
bitunpacker.get_batch_u32s(0, &[0, 0, 0, 0, 0, 0, 0, 0], &mut output[..]);
bitunpacker.get_batch_u32s(0, 0, &[0, 0, 0, 0, 0, 0, 0, 0], &mut output[..]);
}
#[test]
fn test_get_batch_limit() {
let bitunpacker = BitUnpacker::new(1);
let mut output: [u32; 3] = [0u32, 0u32, 0u32];
bitunpacker.get_batch_u32s(8 * 4 - 3, &[0u8, 0u8, 0u8, 0u8], &mut output[..]);
bitunpacker.get_batch_u32s(8 * 4 - 3, 0, &[0u8, 0u8, 0u8, 0u8], &mut output[..]);
}
#[test]
@@ -345,7 +418,7 @@ mod test {
let bitunpacker = BitUnpacker::new(1);
let mut output: [u32; 3] = [0u32, 0u32, 0u32];
// We are missing exactly one bit.
bitunpacker.get_batch_u32s(8 * 4 - 2, &[0u8, 0u8, 0u8, 0u8], &mut output[..]);
bitunpacker.get_batch_u32s(8 * 4 - 2, 0, &[0u8, 0u8, 0u8, 0u8], &mut output[..]);
}
proptest::proptest! {
@@ -368,7 +441,7 @@ mod test {
for len in [0, 1, 2, 32, 33, 34, 64] {
for start_idx in 0u32..32u32 {
output.resize(len, 0);
bitunpacker.get_batch_u32s(start_idx, &buffer, &mut output);
bitunpacker.get_batch_u32s(start_idx, 0, &buffer, &mut output);
for (i, output_byte) in output.iter().enumerate() {
let expected = (start_idx + i as u32) & mask;
assert_eq!(*output_byte, expected);

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())
}
}

27
bitpacker/src/filter_vec/avx2.rs
View File

@@ -19,7 +19,7 @@ fn u32_to_i32(val: u32) -> i32 {
#[inline]
unsafe fn u32_to_i32_avx2(vals_u32x8s: DataType) -> DataType {
const HIGHEST_BIT_MASK: DataType = from_u32x8([HIGHEST_BIT; NUM_LANES]);
op_xor(vals_u32x8s, HIGHEST_BIT_MASK)
unsafe { op_xor(vals_u32x8s, HIGHEST_BIT_MASK) }
}
pub fn filter_vec_in_place(range: RangeInclusive<u32>, offset: u32, output: &mut Vec<u32>) {
@@ -66,17 +66,19 @@ unsafe fn filter_vec_avx2_aux(
]);
const SHIFT: __m256i = from_u32x8([NUM_LANES as u32; NUM_LANES]);
for _ in 0..num_words {
let word = load_unaligned(input);
let word = u32_to_i32_avx2(word);
let keeper_bitset = compute_filter_bitset(word, range_simd.clone());
let added_len = keeper_bitset.count_ones();
let filtered_doc_ids = compact(ids, keeper_bitset);
store_unaligned(output_tail as *mut __m256i, filtered_doc_ids);
output_tail = output_tail.offset(added_len as isize);
ids = op_add(ids, SHIFT);
input = input.offset(1);
unsafe {
let word = load_unaligned(input);
let word = u32_to_i32_avx2(word);
let keeper_bitset = compute_filter_bitset(word, range_simd.clone());
let added_len = keeper_bitset.count_ones();
let filtered_doc_ids = compact(ids, keeper_bitset);
store_unaligned(output_tail as *mut __m256i, filtered_doc_ids);
output_tail = output_tail.offset(added_len as isize);
ids = op_add(ids, SHIFT);
input = input.offset(1);
}
}
output_tail.offset_from(output) as usize
unsafe { output_tail.offset_from(output) as usize }
}
#[inline]
@@ -92,8 +94,7 @@ unsafe fn compute_filter_bitset(val: __m256i, range: std::ops::RangeInclusive<__
let too_low = op_greater(*range.start(), val);
let too_high = op_greater(val, *range.end());
let inside = op_or(too_low, too_high);
255 - std::arch::x86_64::_mm256_movemask_ps(std::mem::transmute::<DataType, __m256>(inside))
as u8
255 - std::arch::x86_64::_mm256_movemask_ps(_mm256_castsi256_ps(inside)) as u8
}
union U8x32 {

2
columnar/Cargo.toml
View File

@@ -16,7 +16,7 @@ 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"
serde = { version = "1.0.152", features = ["derive"] }
downcast-rs = "2.0.1"
[dev-dependencies]

2
columnar/README.md
View File

@@ -73,7 +73,7 @@ The crate introduces the following concepts.
`Columnar` is an equivalent of a dataframe.
It maps `column_key` to `Column`.
A `Column<T>` asssociates a `RowId` (u32) to any
A `Column<T>` associates a `RowId` (u32) to any
number of values.
This is made possible by wrapping a `ColumnIndex` and a `ColumnValue` object.

14
columnar/benches/bench_access.rs
View File

@@ -1,6 +1,6 @@
use binggan::{InputGroup, black_box};
use common::*;
use tantivy_columnar::Column;
use tantivy_columnar::{Column, ValueRange};
pub mod common;
@@ -46,16 +46,16 @@ fn bench_group(mut runner: InputGroup<Column>) {
runner.register("access_first_vals", |column| {
let mut sum = 0;
const BLOCK_SIZE: usize = 32;
let mut docs = vec![0; BLOCK_SIZE];
let mut buffer = vec![None; BLOCK_SIZE];
let mut docs = Vec::with_capacity(BLOCK_SIZE);
let mut buffer = Vec::with_capacity(BLOCK_SIZE);
for i in (0..NUM_DOCS).step_by(BLOCK_SIZE) {
// fill docs
#[allow(clippy::needless_range_loop)]
docs.clear();
for idx in 0..BLOCK_SIZE {
docs[idx] = idx as u32 + i;
docs.push(idx as u32 + i);
}
column.first_vals(&docs, &mut buffer);
buffer.clear();
column.first_vals_in_value_range(&mut docs, &mut buffer, ValueRange::All);
for val in buffer.iter() {
let Some(val) = val else { continue };
sum += *val;

7
columnar/benches/bench_first_vals.rs
View File

@@ -89,13 +89,6 @@ fn main() {
black_box(sum);
});
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);
black_box(block[0]);
});
group.register("first_block_single_calls", |column| {
let mut block: Vec<Option<u64>> = vec![None; 64];
let fetch_docids = (0..64).collect::<Vec<_>>();

9
columnar/benches/bench_merge.rs
View File

@@ -40,7 +40,14 @@ fn main() {
let columnar_readers = columnar_readers.iter().collect::<Vec<_>>();
let merge_row_order = StackMergeOrder::stack(&columnar_readers[..]);
merge_columnar(&columnar_readers, &[], merge_row_order.into(), &mut out).unwrap();
merge_columnar(
&columnar_readers,
&[],
merge_row_order.into(),
&mut out,
|| false,
)
.unwrap();
Some(out.len() as u64)
},
);

219
columnar/src/column/mod.rs
View File

@@ -1,6 +1,7 @@
mod dictionary_encoded;
mod serialize;
use std::cell::RefCell;
use std::fmt::{self, Debug};
use std::io::Write;
use std::ops::{Range, RangeInclusive};
@@ -19,6 +20,11 @@ use crate::column_values::monotonic_mapping::StrictlyMonotonicMappingToInternal;
use crate::column_values::{ColumnValues, monotonic_map_column};
use crate::{Cardinality, DocId, EmptyColumnValues, MonotonicallyMappableToU64, RowId};
thread_local! {
static ROWS: RefCell<Vec<RowId>> = const { RefCell::new(Vec::new()) };
static DOCS: RefCell<Vec<DocId>> = const { RefCell::new(Vec::new()) };
}
#[derive(Clone)]
pub struct Column<T = u64> {
pub index: ColumnIndex,
@@ -89,31 +95,6 @@ impl<T: PartialOrd + Copy + Debug + Send + Sync + 'static> Column<T> {
self.values_for_doc(row_id).next()
}
/// Load the first value for each docid in the provided slice.
#[inline]
pub fn first_vals(&self, docids: &[DocId], output: &mut [Option<T>]) {
match &self.index {
ColumnIndex::Empty { .. } => {}
ColumnIndex::Full => self.values.get_vals_opt(docids, output),
ColumnIndex::Optional(optional_index) => {
for (i, docid) in docids.iter().enumerate() {
output[i] = optional_index
.rank_if_exists(*docid)
.map(|rowid| self.values.get_val(rowid));
}
}
ColumnIndex::Multivalued(multivalued_index) => {
for (i, docid) in docids.iter().enumerate() {
let range = multivalued_index.range(*docid);
let is_empty = range.start == range.end;
if !is_empty {
output[i] = Some(self.values.get_val(range.start));
}
}
}
}
}
/// Translates a block of docids to row_ids.
///
/// returns the row_ids and the matching docids on the same index
@@ -131,6 +112,8 @@ impl<T: PartialOrd + Copy + Debug + Send + Sync + 'static> Column<T> {
self.index.docids_to_rowids(doc_ids, doc_ids_out, row_ids)
}
/// Get an iterator over the values for the provided docid.
#[inline]
pub fn values_for_doc(&self, doc_id: DocId) -> impl Iterator<Item = T> + '_ {
self.index
.value_row_ids(doc_id)
@@ -141,7 +124,7 @@ impl<T: PartialOrd + Copy + Debug + Send + Sync + 'static> Column<T> {
#[inline]
pub fn get_docids_for_value_range(
&self,
value_range: RangeInclusive<T>,
value_range: ValueRange<T>,
selected_docid_range: Range<u32>,
doc_ids: &mut Vec<u32>,
) {
@@ -158,15 +141,6 @@ impl<T: PartialOrd + Copy + Debug + Send + Sync + 'static> Column<T> {
.select_batch_in_place(selected_docid_range.start, doc_ids);
}
/// Fills the output vector with the (possibly multiple values that are associated_with
/// `row_id`.
///
/// This method clears the `output` vector.
pub fn fill_vals(&self, row_id: RowId, output: &mut Vec<T>) {
output.clear();
output.extend(self.values_for_doc(row_id));
}
pub fn first_or_default_col(self, default_value: T) -> Arc<dyn ColumnValues<T>> {
Arc::new(FirstValueWithDefault {
column: self,
@@ -175,6 +149,181 @@ impl<T: PartialOrd + Copy + Debug + Send + Sync + 'static> Column<T> {
}
}
// Separate impl block for methods requiring `Default` for `T`.
impl<T: PartialOrd + Copy + Debug + Send + Sync + 'static + Default> Column<T> {
/// Load the first value for each docid in the provided slice.
///
/// The `docids` vector is mutated: documents that do not match the `value_range` are removed.
/// The `values` vector is populated with the values of the remaining documents.
#[inline]
pub fn first_vals_in_value_range(
&self,
input_docs: &[DocId],
output: &mut Vec<crate::ComparableDoc<Option<T>, DocId>>,
value_range: ValueRange<T>,
) {
match (&self.index, value_range) {
(ColumnIndex::Empty { .. }, value_range) => {
let nulls_match = match &value_range {
ValueRange::All => true,
ValueRange::Inclusive(_) => false,
ValueRange::GreaterThan(_, nulls_match) => *nulls_match,
ValueRange::GreaterThanOrEqual(_, nulls_match) => *nulls_match,
ValueRange::LessThan(_, nulls_match) => *nulls_match,
ValueRange::LessThanOrEqual(_, nulls_match) => *nulls_match,
};
if nulls_match {
for &doc in input_docs {
output.push(crate::ComparableDoc {
doc,
sort_key: None,
});
}
}
}
(ColumnIndex::Full, value_range) => {
self.values
.get_vals_in_value_range(input_docs, input_docs, output, value_range);
}
(ColumnIndex::Optional(optional_index), value_range) => {
let nulls_match = match &value_range {
ValueRange::All => true,
ValueRange::Inclusive(_) => false,
ValueRange::GreaterThan(_, nulls_match) => *nulls_match,
ValueRange::GreaterThanOrEqual(_, nulls_match) => *nulls_match,
ValueRange::LessThan(_, nulls_match) => *nulls_match,
ValueRange::LessThanOrEqual(_, nulls_match) => *nulls_match,
};
let fallback_needed = ROWS.with(|rows_cell| {
DOCS.with(|docs_cell| {
let mut rows = rows_cell.borrow_mut();
let mut docs = docs_cell.borrow_mut();
rows.clear();
docs.clear();
let mut has_nulls = false;
for &doc_id in input_docs {
if let Some(row_id) = optional_index.rank_if_exists(doc_id) {
rows.push(row_id);
docs.push(doc_id);
} else {
has_nulls = true;
if nulls_match {
break;
}
}
}
if !has_nulls || !nulls_match {
self.values.get_vals_in_value_range(
&rows,
&docs,
output,
value_range.clone(),
);
return false;
}
true
})
});
if fallback_needed {
for &doc_id in input_docs {
if let Some(row_id) = optional_index.rank_if_exists(doc_id) {
let val = self.values.get_val(row_id);
let value_matches = match &value_range {
ValueRange::All => true,
ValueRange::Inclusive(r) => r.contains(&val),
ValueRange::GreaterThan(t, _) => val > *t,
ValueRange::GreaterThanOrEqual(t, _) => val >= *t,
ValueRange::LessThan(t, _) => val < *t,
ValueRange::LessThanOrEqual(t, _) => val <= *t,
};
if value_matches {
output.push(crate::ComparableDoc {
doc: doc_id,
sort_key: Some(val),
});
}
} else if nulls_match {
output.push(crate::ComparableDoc {
doc: doc_id,
sort_key: None,
});
}
}
}
}
(ColumnIndex::Multivalued(multivalued_index), value_range) => {
let nulls_match = match &value_range {
ValueRange::All => true,
ValueRange::Inclusive(_) => false,
ValueRange::GreaterThan(_, nulls_match) => *nulls_match,
ValueRange::GreaterThanOrEqual(_, nulls_match) => *nulls_match,
ValueRange::LessThan(_, nulls_match) => *nulls_match,
ValueRange::LessThanOrEqual(_, nulls_match) => *nulls_match,
};
for i in 0..input_docs.len() {
let docid = input_docs[i];
let row_range = multivalued_index.range(docid);
let is_empty = row_range.start == row_range.end;
if !is_empty {
let val = self.values.get_val(row_range.start);
let matches = match &value_range {
ValueRange::All => true,
ValueRange::Inclusive(r) => r.contains(&val),
ValueRange::GreaterThan(t, _) => val > *t,
ValueRange::GreaterThanOrEqual(t, _) => val >= *t,
ValueRange::LessThan(t, _) => val < *t,
ValueRange::LessThanOrEqual(t, _) => val <= *t,
};
if matches {
output.push(crate::ComparableDoc {
doc: docid,
sort_key: Some(val),
});
}
} else if nulls_match {
output.push(crate::ComparableDoc {
doc: docid,
sort_key: None,
});
}
}
}
}
}
}
/// A range of values.
///
/// This type is intended to be used in batch APIs, where the cost of unpacking the enum
/// is outweighed by the time spent processing a batch.
///
/// Implementers should pattern match on the variants to use optimized loops for each case.
#[derive(Clone, Debug)]
pub enum ValueRange<T> {
/// A range that includes both start and end.
Inclusive(RangeInclusive<T>),
/// A range that matches all values.
All,
/// A range that matches all values greater than the threshold.
/// The boolean flag indicates if null values should be included.
GreaterThan(T, bool),
/// A range that matches all values greater than or equal to the threshold.
/// The boolean flag indicates if null values should be included.
GreaterThanOrEqual(T, bool),
/// A range that matches all values less than the threshold.
/// The boolean flag indicates if null values should be included.
LessThan(T, bool),
/// A range that matches all values less than or equal to the threshold.
/// The boolean flag indicates if null values should be included.
LessThanOrEqual(T, bool),
}
impl BinarySerializable for Cardinality {
fn serialize<W: Write + ?Sized>(&self, writer: &mut W) -> std::io::Result<()> {
self.to_code().serialize(writer)

39
columnar/src/column/serialize.rs
View File

@@ -2,7 +2,7 @@ use std::io;
use std::io::Write;
use std::sync::Arc;
use common::OwnedBytes;
use common::file_slice::FileSlice;
use sstable::Dictionary;
use crate::column::{BytesColumn, Column};
@@ -41,12 +41,13 @@ pub fn serialize_column_mappable_to_u64<T: MonotonicallyMappableToU64>(
}
pub fn open_column_u64<T: MonotonicallyMappableToU64>(
bytes: OwnedBytes,
file_slice: FileSlice,
format_version: Version,
) -> io::Result<Column<T>> {
let (body, column_index_num_bytes_payload) = bytes.rsplit(4);
let (body, column_index_num_bytes_payload) = file_slice.split_from_end(4);
let column_index_num_bytes = u32::from_le_bytes(
column_index_num_bytes_payload
.read_bytes()?
.as_slice()
.try_into()
.unwrap(),
@@ -61,12 +62,13 @@ pub fn open_column_u64<T: MonotonicallyMappableToU64>(
}
pub fn open_column_u128<T: MonotonicallyMappableToU128>(
bytes: OwnedBytes,
file_slice: FileSlice,
format_version: Version,
) -> io::Result<Column<T>> {
let (body, column_index_num_bytes_payload) = bytes.rsplit(4);
let (body, column_index_num_bytes_payload) = file_slice.split_from_end(4);
let column_index_num_bytes = u32::from_le_bytes(
column_index_num_bytes_payload
.read_bytes()?
.as_slice()
.try_into()
.unwrap(),
@@ -84,12 +86,13 @@ pub fn open_column_u128<T: MonotonicallyMappableToU128>(
///
/// See [`open_u128_as_compact_u64`] for more details.
pub fn open_column_u128_as_compact_u64(
bytes: OwnedBytes,
file_slice: FileSlice,
format_version: Version,
) -> io::Result<Column<u64>> {
let (body, column_index_num_bytes_payload) = bytes.rsplit(4);
let (body, column_index_num_bytes_payload) = file_slice.split_from_end(4);
let column_index_num_bytes = u32::from_le_bytes(
column_index_num_bytes_payload
.read_bytes()?
.as_slice()
.try_into()
.unwrap(),
@@ -103,11 +106,21 @@ pub fn open_column_u128_as_compact_u64(
})
}
pub fn open_column_bytes(data: OwnedBytes, format_version: Version) -> io::Result<BytesColumn> {
let (body, dictionary_len_bytes) = data.rsplit(4);
let dictionary_len = u32::from_le_bytes(dictionary_len_bytes.as_slice().try_into().unwrap());
pub fn open_column_bytes(
file_slice: FileSlice,
format_version: Version,
) -> io::Result<BytesColumn> {
let (body, dictionary_len_bytes) = file_slice.split_from_end(4);
let dictionary_len = u32::from_le_bytes(
dictionary_len_bytes
.read_bytes()?
.as_slice()
.try_into()
.unwrap(),
);
let (dictionary_bytes, column_bytes) = body.split(dictionary_len as usize);
let dictionary = Arc::new(Dictionary::from_bytes(dictionary_bytes)?);
let dictionary = Arc::new(Dictionary::open(dictionary_bytes)?);
let term_ord_column = crate::column::open_column_u64::<u64>(column_bytes, format_version)?;
Ok(BytesColumn {
dictionary,
@@ -115,7 +128,7 @@ pub fn open_column_bytes(data: OwnedBytes, format_version: Version) -> io::Resul
})
}
pub fn open_column_str(data: OwnedBytes, format_version: Version) -> io::Result<StrColumn> {
let bytes_column = open_column_bytes(data, format_version)?;
pub fn open_column_str(file_slice: FileSlice, format_version: Version) -> io::Result<StrColumn> {
let bytes_column = open_column_bytes(file_slice, format_version)?;
Ok(StrColumn::wrap(bytes_column))
}

6
columnar/src/column_index/merge/mod.rs
View File

@@ -95,7 +95,7 @@ pub fn merge_column_index<'a>(
#[cfg(test)]
mod tests {
use common::OwnedBytes;
use common::file_slice::FileSlice;
use crate::column_index::merge::detect_cardinality;
use crate::column_index::multivalued_index::{
@@ -178,7 +178,7 @@ mod tests {
let mut output = Vec::new();
serialize_multivalued_index(&start_index_iterable, &mut output).unwrap();
let multivalue =
open_multivalued_index(OwnedBytes::new(output), crate::Version::V2).unwrap();
open_multivalued_index(FileSlice::from(output), crate::Version::V2).unwrap();
let start_indexes: Vec<RowId> = multivalue.get_start_index_column().iter().collect();
assert_eq!(&start_indexes, &[0, 3, 5]);
}
@@ -216,7 +216,7 @@ mod tests {
let mut output = Vec::new();
serialize_multivalued_index(&start_index_iterable, &mut output).unwrap();
let multivalue =
open_multivalued_index(OwnedBytes::new(output), crate::Version::V2).unwrap();
open_multivalued_index(FileSlice::from(output), crate::Version::V2).unwrap();
let start_indexes: Vec<RowId> = multivalue.get_start_index_column().iter().collect();
assert_eq!(&start_indexes, &[0, 3, 5, 6]);
}

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(_) => {

52
columnar/src/column_index/multivalued_index.rs
View File

@@ -3,7 +3,8 @@ use std::io::Write;
use std::ops::Range;
use std::sync::Arc;
use common::{CountingWriter, OwnedBytes};
use common::CountingWriter;
use common::file_slice::FileSlice;
use super::optional_index::{open_optional_index, serialize_optional_index};
use super::{OptionalIndex, SerializableOptionalIndex, Set};
@@ -44,21 +45,26 @@ pub fn serialize_multivalued_index(
}
pub fn open_multivalued_index(
bytes: OwnedBytes,
file_slice: FileSlice,
format_version: Version,
) -> io::Result<MultiValueIndex> {
match format_version {
Version::V1 => {
let start_index_column: Arc<dyn ColumnValues<RowId>> =
load_u64_based_column_values(bytes)?;
load_u64_based_column_values(file_slice)?;
Ok(MultiValueIndex::MultiValueIndexV1(MultiValueIndexV1 {
start_index_column,
}))
}
Version::V2 => {
let (body_bytes, optional_index_len) = bytes.rsplit(4);
let optional_index_len =
u32::from_le_bytes(optional_index_len.as_slice().try_into().unwrap());
let (body_bytes, optional_index_len) = file_slice.split_from_end(4);
let optional_index_len = u32::from_le_bytes(
optional_index_len
.read_bytes()?
.as_slice()
.try_into()
.unwrap(),
);
let (optional_index_bytes, start_index_bytes) =
body_bytes.split(optional_index_len as usize);
let optional_index = open_optional_index(optional_index_bytes)?;
@@ -185,8 +191,8 @@ impl MultiValueIndex {
};
let mut buffer = Vec::new();
serialize_multivalued_index(&serializable_multivalued_index, &mut buffer).unwrap();
let bytes = OwnedBytes::new(buffer);
open_multivalued_index(bytes, Version::V2).unwrap()
let file_slice = FileSlice::from(buffer);
open_multivalued_index(file_slice, Version::V2).unwrap()
}
pub fn get_start_index_column(&self) -> &Arc<dyn crate::ColumnValues<RowId>> {
@@ -215,6 +221,32 @@ 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.
///
@@ -307,7 +339,7 @@ mod tests {
use std::ops::Range;
use super::MultiValueIndex;
use crate::{ColumnarReader, DynamicColumn};
use crate::{ColumnarReader, DynamicColumn, ValueRange};
fn index_to_pos_helper(
index: &MultiValueIndex,
@@ -387,7 +419,7 @@ mod tests {
assert_eq!(row_id_range, 0..4);
let check = |range, expected| {
let full_range = 0..=u64::MAX;
let full_range = ValueRange::All;
let mut docids = Vec::new();
column.get_docids_for_value_range(full_range, range, &mut docids);
assert_eq!(docids, expected);

63
columnar/src/column_index/optional_index/mod.rs
View File

@@ -1,9 +1,10 @@
use std::io::{self, Write};
use std::io;
use std::sync::Arc;
mod set;
mod set_block;
use common::file_slice::FileSlice;
use common::{BinarySerializable, OwnedBytes, VInt};
pub use set::{SelectCursor, Set, SetCodec};
use set_block::{
@@ -11,7 +12,7 @@ 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 +89,7 @@ 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())
}
}
@@ -268,8 +269,8 @@ impl OptionalIndex {
);
let mut buffer = Vec::new();
serialize_optional_index(&row_ids, num_rows, &mut buffer).unwrap();
let bytes = OwnedBytes::new(buffer);
open_optional_index(bytes).unwrap()
let file_slice = FileSlice::from(buffer);
open_optional_index(file_slice).unwrap()
}
pub fn num_docs(&self) -> RowId {
@@ -280,8 +281,9 @@ 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 +336,6 @@ 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 {
@@ -517,10 +487,17 @@ fn deserialize_optional_index_block_metadatas(
(block_metas.into_boxed_slice(), non_null_rows_before_block)
}
pub fn open_optional_index(bytes: OwnedBytes) -> io::Result<OptionalIndex> {
let (mut bytes, num_non_empty_blocks_bytes) = bytes.rsplit(2);
let num_non_empty_block_bytes =
u16::from_le_bytes(num_non_empty_blocks_bytes.as_slice().try_into().unwrap());
pub fn open_optional_index(file_slice: FileSlice) -> io::Result<OptionalIndex> {
let (bytes, num_non_empty_blocks_bytes) = file_slice.split_from_end(2);
let num_non_empty_block_bytes = u16::from_le_bytes(
num_non_empty_blocks_bytes
.read_bytes()?
.as_slice()
.try_into()
.unwrap(),
);
let mut bytes = bytes.read_bytes()?;
let num_docs = VInt::deserialize_u64(&mut bytes)? as u32;
let block_metas_num_bytes =
num_non_empty_block_bytes as usize * SERIALIZED_BLOCK_META_NUM_BYTES;

177
columnar/src/column_index/optional_index/tests.rs
View File

@@ -59,7 +59,7 @@ fn test_with_random_sets_simple() {
let vals = 10..ELEMENTS_PER_BLOCK * 2;
let mut out: Vec<u8> = Vec::new();
serialize_optional_index(&vals, 100, &mut out).unwrap();
let null_index = open_optional_index(OwnedBytes::new(out)).unwrap();
let null_index = open_optional_index(FileSlice::from(out)).unwrap();
let ranks: Vec<u32> = (65_472u32..65_473u32).collect();
let els: Vec<u32> = ranks.iter().copied().map(|rank| rank + 10).collect();
let mut select_cursor = null_index.select_cursor();
@@ -102,7 +102,7 @@ impl<'a> Iterable<RowId> for &'a [bool] {
fn test_null_index(data: &[bool]) {
let mut out: Vec<u8> = Vec::new();
serialize_optional_index(&data, data.len() as RowId, &mut out).unwrap();
let null_index = open_optional_index(OwnedBytes::new(out)).unwrap();
let null_index = open_optional_index(FileSlice::from(out)).unwrap();
let orig_idx_with_value: Vec<u32> = data
.iter()
.enumerate()
@@ -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]
@@ -219,3 +223,170 @@ 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(FileSlice::from(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);
}
}

16
columnar/src/column_index/serialize.rs
View File

@@ -1,7 +1,8 @@
use std::io;
use std::io::Write;
use common::{CountingWriter, OwnedBytes};
use common::file_slice::FileSlice;
use common::{CountingWriter, HasLen};
use super::OptionalIndex;
use super::multivalued_index::SerializableMultivalueIndex;
@@ -65,27 +66,28 @@ pub fn serialize_column_index(
/// Open a serialized column index.
pub fn open_column_index(
mut bytes: OwnedBytes,
file_slice: FileSlice,
format_version: Version,
) -> io::Result<ColumnIndex> {
if bytes.is_empty() {
if file_slice.len() == 0 {
return Err(io::Error::new(
io::ErrorKind::UnexpectedEof,
"Failed to deserialize column index. Empty buffer.",
));
}
let cardinality_code = bytes[0];
let (header, body) = file_slice.split(1);
let cardinality_code = header.read_bytes()?.as_slice()[0];
let cardinality = Cardinality::try_from_code(cardinality_code)?;
bytes.advance(1);
match cardinality {
Cardinality::Full => Ok(ColumnIndex::Full),
Cardinality::Optional => {
let optional_index = super::optional_index::open_optional_index(bytes)?;
let optional_index = super::optional_index::open_optional_index(body)?;
Ok(ColumnIndex::Optional(optional_index))
}
Cardinality::Multivalued => {
let multivalue_index =
super::multivalued_index::open_multivalued_index(bytes, format_version)?;
super::multivalued_index::open_multivalued_index(body, format_version)?;
Ok(ColumnIndex::Multivalued(multivalue_index))
}
}

382
columnar/src/column_values/mod.rs
View File

@@ -7,13 +7,15 @@
//! - Monotonically map values to u64/u128
use std::fmt::Debug;
use std::ops::{Range, RangeInclusive};
use std::ops::Range;
use std::sync::Arc;
use downcast_rs::DowncastSync;
pub use monotonic_mapping::{MonotonicallyMappableToU64, StrictlyMonotonicFn};
pub use monotonic_mapping_u128::MonotonicallyMappableToU128;
use crate::column::ValueRange;
mod merge;
pub(crate) mod monotonic_mapping;
pub(crate) mod monotonic_mapping_u128;
@@ -27,8 +29,7 @@ mod monotonic_column;
pub(crate) use merge::MergedColumnValues;
pub use stats::ColumnStats;
pub use u64_based::{
ALL_U64_CODEC_TYPES, CodecType, load_u64_based_column_values,
serialize_and_load_u64_based_column_values, serialize_u64_based_column_values,
ALL_U64_CODEC_TYPES, CodecType, load_u64_based_column_values, serialize_u64_based_column_values,
};
pub use u128_based::{
CompactSpaceU64Accessor, open_u128_as_compact_u64, open_u128_mapped,
@@ -109,6 +110,307 @@ pub trait ColumnValues<T: PartialOrd = u64>: Send + Sync + DowncastSync {
}
}
/// Load the values for the provided docids.
///
/// The values are filtered by the provided value range.
fn get_vals_in_value_range(
&self,
input_indexes: &[u32],
input_doc_ids: &[u32],
output: &mut Vec<crate::ComparableDoc<Option<T>, crate::DocId>>,
value_range: ValueRange<T>,
) {
let len = input_indexes.len();
let mut read_head = 0;
match value_range {
ValueRange::All => {
while read_head + 3 < len {
let idx0 = input_indexes[read_head];
let idx1 = input_indexes[read_head + 1];
let idx2 = input_indexes[read_head + 2];
let idx3 = input_indexes[read_head + 3];
let doc0 = input_doc_ids[read_head];
let doc1 = input_doc_ids[read_head + 1];
let doc2 = input_doc_ids[read_head + 2];
let doc3 = input_doc_ids[read_head + 3];
let val0 = self.get_val(idx0);
let val1 = self.get_val(idx1);
let val2 = self.get_val(idx2);
let val3 = self.get_val(idx3);
output.push(crate::ComparableDoc {
doc: doc0,
sort_key: Some(val0),
});
output.push(crate::ComparableDoc {
doc: doc1,
sort_key: Some(val1),
});
output.push(crate::ComparableDoc {
doc: doc2,
sort_key: Some(val2),
});
output.push(crate::ComparableDoc {
doc: doc3,
sort_key: Some(val3),
});
read_head += 4;
}
}
ValueRange::Inclusive(ref range) => {
while read_head + 3 < len {
let idx0 = input_indexes[read_head];
let idx1 = input_indexes[read_head + 1];
let idx2 = input_indexes[read_head + 2];
let idx3 = input_indexes[read_head + 3];
let doc0 = input_doc_ids[read_head];
let doc1 = input_doc_ids[read_head + 1];
let doc2 = input_doc_ids[read_head + 2];
let doc3 = input_doc_ids[read_head + 3];
let val0 = self.get_val(idx0);
let val1 = self.get_val(idx1);
let val2 = self.get_val(idx2);
let val3 = self.get_val(idx3);
if range.contains(&val0) {
output.push(crate::ComparableDoc {
doc: doc0,
sort_key: Some(val0),
});
}
if range.contains(&val1) {
output.push(crate::ComparableDoc {
doc: doc1,
sort_key: Some(val1),
});
}
if range.contains(&val2) {
output.push(crate::ComparableDoc {
doc: doc2,
sort_key: Some(val2),
});
}
if range.contains(&val3) {
output.push(crate::ComparableDoc {
doc: doc3,
sort_key: Some(val3),
});
}
read_head += 4;
}
}
ValueRange::GreaterThan(ref threshold, _) => {
while read_head + 3 < len {
let idx0 = input_indexes[read_head];
let idx1 = input_indexes[read_head + 1];
let idx2 = input_indexes[read_head + 2];
let idx3 = input_indexes[read_head + 3];
let doc0 = input_doc_ids[read_head];
let doc1 = input_doc_ids[read_head + 1];
let doc2 = input_doc_ids[read_head + 2];
let doc3 = input_doc_ids[read_head + 3];
let val0 = self.get_val(idx0);
let val1 = self.get_val(idx1);
let val2 = self.get_val(idx2);
let val3 = self.get_val(idx3);
if val0 > *threshold {
output.push(crate::ComparableDoc {
doc: doc0,
sort_key: Some(val0),
});
}
if val1 > *threshold {
output.push(crate::ComparableDoc {
doc: doc1,
sort_key: Some(val1),
});
}
if val2 > *threshold {
output.push(crate::ComparableDoc {
doc: doc2,
sort_key: Some(val2),
});
}
if val3 > *threshold {
output.push(crate::ComparableDoc {
doc: doc3,
sort_key: Some(val3),
});
}
read_head += 4;
}
}
ValueRange::GreaterThanOrEqual(ref threshold, _) => {
while read_head + 3 < len {
let idx0 = input_indexes[read_head];
let idx1 = input_indexes[read_head + 1];
let idx2 = input_indexes[read_head + 2];
let idx3 = input_indexes[read_head + 3];
let doc0 = input_doc_ids[read_head];
let doc1 = input_doc_ids[read_head + 1];
let doc2 = input_doc_ids[read_head + 2];
let doc3 = input_doc_ids[read_head + 3];
let val0 = self.get_val(idx0);
let val1 = self.get_val(idx1);
let val2 = self.get_val(idx2);
let val3 = self.get_val(idx3);
if val0 >= *threshold {
output.push(crate::ComparableDoc {
doc: doc0,
sort_key: Some(val0),
});
}
if val1 >= *threshold {
output.push(crate::ComparableDoc {
doc: doc1,
sort_key: Some(val1),
});
}
if val2 >= *threshold {
output.push(crate::ComparableDoc {
doc: doc2,
sort_key: Some(val2),
});
}
if val3 >= *threshold {
output.push(crate::ComparableDoc {
doc: doc3,
sort_key: Some(val3),
});
}
read_head += 4;
}
}
ValueRange::LessThan(ref threshold, _) => {
while read_head + 3 < len {
let idx0 = input_indexes[read_head];
let idx1 = input_indexes[read_head + 1];
let idx2 = input_indexes[read_head + 2];
let idx3 = input_indexes[read_head + 3];
let doc0 = input_doc_ids[read_head];
let doc1 = input_doc_ids[read_head + 1];
let doc2 = input_doc_ids[read_head + 2];
let doc3 = input_doc_ids[read_head + 3];
let val0 = self.get_val(idx0);
let val1 = self.get_val(idx1);
let val2 = self.get_val(idx2);
let val3 = self.get_val(idx3);
if val0 < *threshold {
output.push(crate::ComparableDoc {
doc: doc0,
sort_key: Some(val0),
});
}
if val1 < *threshold {
output.push(crate::ComparableDoc {
doc: doc1,
sort_key: Some(val1),
});
}
if val2 < *threshold {
output.push(crate::ComparableDoc {
doc: doc2,
sort_key: Some(val2),
});
}
if val3 < *threshold {
output.push(crate::ComparableDoc {
doc: doc3,
sort_key: Some(val3),
});
}
read_head += 4;
}
}
ValueRange::LessThanOrEqual(ref threshold, _) => {
while read_head + 3 < len {
let idx0 = input_indexes[read_head];
let idx1 = input_indexes[read_head + 1];
let idx2 = input_indexes[read_head + 2];
let idx3 = input_indexes[read_head + 3];
let doc0 = input_doc_ids[read_head];
let doc1 = input_doc_ids[read_head + 1];
let doc2 = input_doc_ids[read_head + 2];
let doc3 = input_doc_ids[read_head + 3];
let val0 = self.get_val(idx0);
let val1 = self.get_val(idx1);
let val2 = self.get_val(idx2);
let val3 = self.get_val(idx3);
if val0 <= *threshold {
output.push(crate::ComparableDoc {
doc: doc0,
sort_key: Some(val0),
});
}
if val1 <= *threshold {
output.push(crate::ComparableDoc {
doc: doc1,
sort_key: Some(val1),
});
}
if val2 <= *threshold {
output.push(crate::ComparableDoc {
doc: doc2,
sort_key: Some(val2),
});
}
if val3 <= *threshold {
output.push(crate::ComparableDoc {
doc: doc3,
sort_key: Some(val3),
});
}
read_head += 4;
}
}
}
// Process remaining elements (0 to 3)
while read_head < len {
let idx = input_indexes[read_head];
let doc = input_doc_ids[read_head];
let val = self.get_val(idx);
let matches = match value_range {
// 'value_range' is still moved here. This is the outer `value_range`
ValueRange::All => true,
ValueRange::Inclusive(ref r) => r.contains(&val),
ValueRange::GreaterThan(ref t, _) => val > *t,
ValueRange::GreaterThanOrEqual(ref t, _) => val >= *t,
ValueRange::LessThan(ref t, _) => val < *t,
ValueRange::LessThanOrEqual(ref t, _) => val <= *t,
};
if matches {
output.push(crate::ComparableDoc {
doc,
sort_key: Some(val),
});
}
read_head += 1;
}
}
/// Fills an output buffer with the fast field values
/// associated with the `DocId` going from
/// `start` to `start + output.len()`.
@@ -129,15 +431,54 @@ pub trait ColumnValues<T: PartialOrd = u64>: Send + Sync + DowncastSync {
/// Note that position == docid for single value fast fields
fn get_row_ids_for_value_range(
&self,
value_range: RangeInclusive<T>,
value_range: ValueRange<T>,
row_id_range: Range<RowId>,
row_id_hits: &mut Vec<RowId>,
) {
let row_id_range = row_id_range.start..row_id_range.end.min(self.num_vals());
for idx in row_id_range {
let val = self.get_val(idx);
if value_range.contains(&val) {
row_id_hits.push(idx);
match value_range {
ValueRange::Inclusive(range) => {
for idx in row_id_range {
let val = self.get_val(idx);
if range.contains(&val) {
row_id_hits.push(idx);
}
}
}
ValueRange::GreaterThan(threshold, _) => {
for idx in row_id_range {
let val = self.get_val(idx);
if val > threshold {
row_id_hits.push(idx);
}
}
}
ValueRange::GreaterThanOrEqual(threshold, _) => {
for idx in row_id_range {
let val = self.get_val(idx);
if val >= threshold {
row_id_hits.push(idx);
}
}
}
ValueRange::LessThan(threshold, _) => {
for idx in row_id_range {
let val = self.get_val(idx);
if val < threshold {
row_id_hits.push(idx);
}
}
}
ValueRange::LessThanOrEqual(threshold, _) => {
for idx in row_id_range {
let val = self.get_val(idx);
if val <= threshold {
row_id_hits.push(idx);
}
}
}
ValueRange::All => {
row_id_hits.extend(row_id_range);
}
}
}
@@ -193,6 +534,17 @@ impl<T: PartialOrd + Default> ColumnValues<T> for EmptyColumnValues {
fn num_vals(&self) -> u32 {
0
}
fn get_vals_in_value_range(
&self,
input_indexes: &[u32],
input_doc_ids: &[u32],
output: &mut Vec<crate::ComparableDoc<Option<T>, crate::DocId>>,
value_range: ValueRange<T>,
) {
let _ = (input_indexes, input_doc_ids, output, value_range);
panic!("Internal Error: Called get_vals_in_value_range of empty column.")
}
}
impl<T: Copy + PartialOrd + Debug + 'static> ColumnValues<T> for Arc<dyn ColumnValues<T>> {
@@ -206,6 +558,18 @@ impl<T: Copy + PartialOrd + Debug + 'static> ColumnValues<T> for Arc<dyn ColumnV
self.as_ref().get_vals_opt(indexes, output)
}
#[inline(always)]
fn get_vals_in_value_range(
&self,
input_indexes: &[u32],
input_doc_ids: &[u32],
output: &mut Vec<crate::ComparableDoc<Option<T>, crate::DocId>>,
value_range: ValueRange<T>,
) {
self.as_ref()
.get_vals_in_value_range(input_indexes, input_doc_ids, output, value_range)
}
#[inline(always)]
fn min_value(&self) -> T {
self.as_ref().min_value()
@@ -234,7 +598,7 @@ impl<T: Copy + PartialOrd + Debug + 'static> ColumnValues<T> for Arc<dyn ColumnV
#[inline(always)]
fn get_row_ids_for_value_range(
&self,
range: RangeInclusive<T>,
range: ValueRange<T>,
doc_id_range: Range<u32>,
positions: &mut Vec<u32>,
) {

53
columnar/src/column_values/monotonic_column.rs
View File

@@ -1,8 +1,9 @@
use std::fmt::Debug;
use std::marker::PhantomData;
use std::ops::{Range, RangeInclusive};
use std::ops::Range;
use crate::ColumnValues;
use crate::column::ValueRange;
use crate::column_values::monotonic_mapping::StrictlyMonotonicFn;
struct MonotonicMappingColumn<C, T, Input> {
@@ -80,16 +81,52 @@ where
fn get_row_ids_for_value_range(
&self,
range: RangeInclusive<Output>,
range: ValueRange<Output>,
doc_id_range: Range<u32>,
positions: &mut Vec<u32>,
) {
self.from_column.get_row_ids_for_value_range(
self.monotonic_mapping.inverse(range.start().clone())
..=self.monotonic_mapping.inverse(range.end().clone()),
doc_id_range,
positions,
)
match range {
ValueRange::Inclusive(range) => self.from_column.get_row_ids_for_value_range(
ValueRange::Inclusive(
self.monotonic_mapping.inverse(range.start().clone())
..=self.monotonic_mapping.inverse(range.end().clone()),
),
doc_id_range,
positions,
),
ValueRange::All => self.from_column.get_row_ids_for_value_range(
ValueRange::All,
doc_id_range,
positions,
),
ValueRange::GreaterThan(threshold, _) => self.from_column.get_row_ids_for_value_range(
ValueRange::GreaterThan(self.monotonic_mapping.inverse(threshold), false),
doc_id_range,
positions,
),
ValueRange::GreaterThanOrEqual(threshold, _) => {
self.from_column.get_row_ids_for_value_range(
ValueRange::GreaterThanOrEqual(
self.monotonic_mapping.inverse(threshold),
false,
),
doc_id_range,
positions,
)
}
ValueRange::LessThan(threshold, _) => self.from_column.get_row_ids_for_value_range(
ValueRange::LessThan(self.monotonic_mapping.inverse(threshold), false),
doc_id_range,
positions,
),
ValueRange::LessThanOrEqual(threshold, _) => {
self.from_column.get_row_ids_for_value_range(
ValueRange::LessThanOrEqual(self.monotonic_mapping.inverse(threshold), false),
doc_id_range,
positions,
)
}
}
}
// We voluntarily do not implement get_range as it yields a regression,

2
columnar/src/column_values/monotonic_mapping_u128.rs
View File

@@ -1,7 +1,7 @@
use std::fmt::Debug;
use std::net::Ipv6Addr;
/// Montonic maps a value to u128 value space
/// Monotonic maps a value to u128 value space
/// Monotonic mapping enables `PartialOrd` on u128 space without conversion to original space.
pub trait MonotonicallyMappableToU128: 'static + PartialOrd + Copy + Debug + Send + Sync {
/// Converts a value to u128.

52
columnar/src/column_values/stats.rs
View File

@@ -2,7 +2,8 @@ use std::io;
use std::io::Write;
use std::num::NonZeroU64;
use common::{BinarySerializable, VInt};
use common::file_slice::FileSlice;
use common::{BinarySerializable, HasLen, VInt};
use crate::RowId;
@@ -27,6 +28,55 @@ impl ColumnStats {
}
}
impl ColumnStats {
/// Deserialize from the tail of the given FileSlice, and return the stats and remaining prefix
/// FileSlice.
pub fn deserialize_from_tail(file_slice: FileSlice) -> io::Result<(Self, FileSlice)> {
// [`deserialize_with_size`] deserializes 4 variable-width encoded u64s, which
// could end up being, in the worst case, 9 bytes each. this is where the 36 comes from
let (stats, _) = file_slice.clone().split(36.min(file_slice.len())); // hope that's enough bytes
let mut stats = stats.read_bytes()?;
let (stats, stats_nbytes) = ColumnStats::deserialize_with_size(&mut stats)?;
let (_, remainder) = file_slice.split(stats_nbytes);
Ok((stats, remainder))
}
/// Same as [`BinarySeerializable::deserialize`] but also returns the number of bytes
/// consumed from the reader `R`
fn deserialize_with_size<R: io::Read>(reader: &mut R) -> io::Result<(Self, usize)> {
let mut nbytes = 0;
let (min_value, len) = VInt::deserialize_with_size(reader)?;
let min_value = min_value.0;
nbytes += len;
let (gcd, len) = VInt::deserialize_with_size(reader)?;
let gcd = gcd.0;
let gcd = NonZeroU64::new(gcd)
.ok_or_else(|| io::Error::new(io::ErrorKind::InvalidData, "GCD of 0 is forbidden"))?;
nbytes += len;
let (amplitude, len) = VInt::deserialize_with_size(reader)?;
let amplitude = amplitude.0 * gcd.get();
let max_value = min_value + amplitude;
nbytes += len;
let (num_rows, len) = VInt::deserialize_with_size(reader)?;
let num_rows = num_rows.0 as RowId;
nbytes += len;
Ok((
ColumnStats {
min_value,
max_value,
num_rows,
gcd,
},
nbytes,
))
}
}
impl BinarySerializable for ColumnStats {
fn serialize<W: Write + ?Sized>(&self, writer: &mut W) -> io::Result<()> {
VInt(self.min_value).serialize(writer)?;

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() {

101
columnar/src/column_values/u128_based/compact_space/mod.rs
View File

@@ -25,6 +25,7 @@ use common::{BinarySerializable, CountingWriter, OwnedBytes, VInt, VIntU128};
use tantivy_bitpacker::{BitPacker, BitUnpacker};
use crate::RowId;
use crate::column::ValueRange;
use crate::column_values::ColumnValues;
/// The cost per blank is quite hard actually, since blanks are delta encoded, the actual cost of
@@ -338,14 +339,48 @@ impl ColumnValues<u64> for CompactSpaceU64Accessor {
#[inline]
fn get_row_ids_for_value_range(
&self,
value_range: RangeInclusive<u64>,
value_range: ValueRange<u64>,
position_range: Range<u32>,
positions: &mut Vec<u32>,
) {
let value_range = self.0.compact_to_u128(*value_range.start() as u32)
..=self.0.compact_to_u128(*value_range.end() as u32);
self.0
.get_row_ids_for_value_range(value_range, position_range, positions)
match value_range {
ValueRange::Inclusive(value_range) => {
let value_range = ValueRange::Inclusive(
self.0.compact_to_u128(*value_range.start() as u32)
..=self.0.compact_to_u128(*value_range.end() as u32),
);
self.0
.get_row_ids_for_value_range(value_range, position_range, positions)
}
ValueRange::All => {
let position_range = position_range.start..position_range.end.min(self.num_vals());
positions.extend(position_range);
}
ValueRange::GreaterThan(threshold, _) => {
let value_range =
ValueRange::GreaterThan(self.0.compact_to_u128(threshold as u32), false);
self.0
.get_row_ids_for_value_range(value_range, position_range, positions)
}
ValueRange::GreaterThanOrEqual(threshold, _) => {
let value_range =
ValueRange::GreaterThanOrEqual(self.0.compact_to_u128(threshold as u32), false);
self.0
.get_row_ids_for_value_range(value_range, position_range, positions)
}
ValueRange::LessThan(threshold, _) => {
let value_range =
ValueRange::LessThan(self.0.compact_to_u128(threshold as u32), false);
self.0
.get_row_ids_for_value_range(value_range, position_range, positions)
}
ValueRange::LessThanOrEqual(threshold, _) => {
let value_range =
ValueRange::LessThanOrEqual(self.0.compact_to_u128(threshold as u32), false);
self.0
.get_row_ids_for_value_range(value_range, position_range, positions)
}
}
}
}
@@ -375,10 +410,47 @@ impl ColumnValues<u128> for CompactSpaceDecompressor {
#[inline]
fn get_row_ids_for_value_range(
&self,
value_range: RangeInclusive<u128>,
value_range: ValueRange<u128>,
position_range: Range<u32>,
positions: &mut Vec<u32>,
) {
let value_range = match value_range {
ValueRange::Inclusive(value_range) => value_range,
ValueRange::All => {
let position_range = position_range.start..position_range.end.min(self.num_vals());
positions.extend(position_range);
return;
}
ValueRange::GreaterThan(threshold, _) => {
let max = self.max_value();
if threshold >= max {
return;
}
(threshold + 1)..=max
}
ValueRange::GreaterThanOrEqual(threshold, _) => {
let max = self.max_value();
if threshold > max {
return;
}
threshold..=max
}
ValueRange::LessThan(threshold, _) => {
let min = self.min_value();
if threshold <= min {
return;
}
min..=(threshold - 1)
}
ValueRange::LessThanOrEqual(threshold, _) => {
let min = self.min_value();
if threshold < min {
return;
}
min..=threshold
}
};
if value_range.start() > value_range.end() {
return;
}
@@ -560,7 +632,7 @@ mod tests {
.collect::<Vec<_>>();
let mut positions = Vec::new();
decompressor.get_row_ids_for_value_range(
range,
ValueRange::Inclusive(range),
0..decompressor.num_vals(),
&mut positions,
);
@@ -604,7 +676,11 @@ mod tests {
let val = *val;
let pos = pos as u32;
let mut positions = Vec::new();
decomp.get_row_ids_for_value_range(val..=val, pos..pos + 1, &mut positions);
decomp.get_row_ids_for_value_range(
ValueRange::Inclusive(val..=val),
pos..pos + 1,
&mut positions,
);
assert_eq!(positions, vec![pos]);
}
@@ -746,7 +822,11 @@ mod tests {
doc_id_range: Range<u32>,
) -> Vec<u32> {
let mut positions = Vec::new();
column.get_row_ids_for_value_range(value_range, doc_id_range, &mut positions);
column.get_row_ids_for_value_range(
ValueRange::Inclusive(value_range),
doc_id_range,
&mut positions,
);
positions
}
@@ -769,7 +849,7 @@ mod tests {
];
let mut out = Vec::new();
serialize_column_values_u128(&&vals[..], &mut out).unwrap();
let decomp = open_u128_mapped(OwnedBytes::new(out)).unwrap();
let decomp = open_u128_mapped(FileSlice::from(out)).unwrap();
let complete_range = 0..vals.len() as u32;
assert_eq!(
@@ -823,6 +903,7 @@ mod tests {
let _data = test_aux_vals(vals);
}
use common::file_slice::FileSlice;
use proptest::prelude::*;
fn num_strategy() -> impl Strategy<Value = u128> {

9
columnar/src/column_values/u128_based/mod.rs
View File

@@ -5,7 +5,8 @@ use std::sync::Arc;
mod compact_space;
use common::{BinarySerializable, OwnedBytes, VInt};
use common::file_slice::FileSlice;
use common::{BinarySerializable, VInt};
pub use compact_space::{
CompactSpaceCompressor, CompactSpaceDecompressor, CompactSpaceU64Accessor,
};
@@ -101,8 +102,9 @@ impl U128FastFieldCodecType {
/// Returns the correct codec reader wrapped in the `Arc` for the data.
pub fn open_u128_mapped<T: MonotonicallyMappableToU128 + Debug>(
mut bytes: OwnedBytes,
file_slice: FileSlice,
) -> io::Result<Arc<dyn ColumnValues<T>>> {
let mut bytes = file_slice.read_bytes()?;
let header = U128Header::deserialize(&mut bytes)?;
assert_eq!(header.codec_type, U128FastFieldCodecType::CompactSpace);
let reader = CompactSpaceDecompressor::open(bytes)?;
@@ -120,7 +122,8 @@ pub fn open_u128_mapped<T: MonotonicallyMappableToU128 + Debug>(
/// # Notice
/// In case there are new codecs added, check for usages of `CompactSpaceDecompressorU64` and
/// also handle the new codecs.
pub fn open_u128_as_compact_u64(mut bytes: OwnedBytes) -> io::Result<Arc<dyn ColumnValues<u64>>> {
pub fn open_u128_as_compact_u64(file_slice: FileSlice) -> io::Result<Arc<dyn ColumnValues<u64>>> {
let mut bytes = file_slice.read_bytes()?;
let header = U128Header::deserialize(&mut bytes)?;
assert_eq!(header.codec_type, U128FastFieldCodecType::CompactSpace);
let reader = CompactSpaceU64Accessor::open(bytes)?;

384
columnar/src/column_values/u64_based/bitpacked.rs
View File

@@ -1,11 +1,14 @@
use std::io::{self, Write};
use std::num::NonZeroU64;
use std::ops::{Range, RangeInclusive};
use std::sync::{Arc, OnceLock};
use common::{BinarySerializable, OwnedBytes};
use common::file_slice::FileSlice;
use common::{BinarySerializable, HasLen, OwnedBytes};
use fastdivide::DividerU64;
use tantivy_bitpacker::{BitPacker, BitUnpacker, compute_num_bits};
use crate::column::ValueRange;
use crate::column_values::u64_based::{ColumnCodec, ColumnCodecEstimator, ColumnStats};
use crate::{ColumnValues, RowId};
@@ -13,9 +16,40 @@ use crate::{ColumnValues, RowId};
/// fast field is required.
#[derive(Clone)]
pub struct BitpackedReader {
data: OwnedBytes,
data: FileSlice,
bit_unpacker: BitUnpacker,
stats: ColumnStats,
blocks: Arc<[OnceLock<Block>]>,
}
impl BitpackedReader {
#[inline(always)]
fn unpack_val(&self, doc: u32) -> u64 {
let block_num = self.bit_unpacker.block_num(doc);
if block_num == 0 && self.blocks.len() == 0 {
return 0;
}
let block = self.blocks[block_num].get_or_init(|| {
let block_range = self.bit_unpacker.block(block_num, self.data.len());
let offset = block_range.start;
let data = self
.data
.slice(block_range)
.read_bytes()
.expect("Failed to read column values.");
Block { offset, data }
});
self.bit_unpacker
.get_from_subset(doc, block.offset, &block.data)
}
}
struct Block {
offset: usize,
data: OwnedBytes,
}
#[inline(always)]
@@ -57,8 +91,9 @@ fn transform_range_before_linear_transformation(
impl ColumnValues for BitpackedReader {
#[inline(always)]
fn get_val(&self, doc: u32) -> u64 {
self.stats.min_value + self.stats.gcd.get() * self.bit_unpacker.get(doc, &self.data)
self.stats.min_value + self.stats.gcd.get() * self.unpack_val(doc)
}
#[inline]
fn min_value(&self) -> u64 {
self.stats.min_value
@@ -72,24 +107,329 @@ impl ColumnValues for BitpackedReader {
self.stats.num_rows
}
fn get_vals_in_value_range(
&self,
input_indexes: &[u32],
input_doc_ids: &[u32],
output: &mut Vec<crate::ComparableDoc<Option<u64>, crate::DocId>>,
value_range: ValueRange<u64>,
) {
match value_range {
ValueRange::All => {
for (&idx, &doc) in input_indexes.iter().zip(input_doc_ids.iter()) {
output.push(crate::ComparableDoc {
doc,
sort_key: Some(self.get_val(idx)),
});
}
}
ValueRange::Inclusive(range) => {
if let Some(transformed_range) =
transform_range_before_linear_transformation(&self.stats, range)
{
for (&idx, &doc) in input_indexes.iter().zip(input_doc_ids.iter()) {
let raw_val = self.unpack_val(idx);
if transformed_range.contains(&raw_val) {
output.push(crate::ComparableDoc {
doc,
sort_key: Some(
self.stats.min_value + self.stats.gcd.get() * raw_val,
),
});
}
}
}
}
ValueRange::GreaterThan(threshold, _) => {
if threshold < self.stats.min_value {
for (&idx, &doc) in input_indexes.iter().zip(input_doc_ids.iter()) {
output.push(crate::ComparableDoc {
doc,
sort_key: Some(self.get_val(idx)),
});
}
} else if threshold >= self.stats.max_value {
// All filtered out
} else {
let raw_threshold = (threshold - self.stats.min_value) / self.stats.gcd.get();
for (&idx, &doc) in input_indexes.iter().zip(input_doc_ids.iter()) {
let raw_val = self.unpack_val(idx);
if raw_val > raw_threshold {
output.push(crate::ComparableDoc {
doc,
sort_key: Some(
self.stats.min_value + self.stats.gcd.get() * raw_val,
),
});
}
}
}
}
ValueRange::GreaterThanOrEqual(threshold, _) => {
if threshold <= self.stats.min_value {
for (&idx, &doc) in input_indexes.iter().zip(input_doc_ids.iter()) {
output.push(crate::ComparableDoc {
doc,
sort_key: Some(self.get_val(idx)),
});
}
} else if threshold > self.stats.max_value {
// All filtered out
} else {
let diff = threshold - self.stats.min_value;
let gcd = self.stats.gcd.get();
let raw_threshold = (diff + gcd - 1) / gcd;
for (&idx, &doc) in input_indexes.iter().zip(input_doc_ids.iter()) {
let raw_val = self.unpack_val(idx);
if raw_val >= raw_threshold {
output.push(crate::ComparableDoc {
doc,
sort_key: Some(
self.stats.min_value + self.stats.gcd.get() * raw_val,
),
});
}
}
}
}
ValueRange::LessThan(threshold, _) => {
if threshold > self.stats.max_value {
for (&idx, &doc) in input_indexes.iter().zip(input_doc_ids.iter()) {
output.push(crate::ComparableDoc {
doc,
sort_key: Some(self.get_val(idx)),
});
}
} else if threshold <= self.stats.min_value {
// All filtered out
} else {
let diff = threshold - self.stats.min_value;
let gcd = self.stats.gcd.get();
let raw_threshold = if diff % gcd == 0 {
diff / gcd
} else {
diff / gcd + 1
};
for (&idx, &doc) in input_indexes.iter().zip(input_doc_ids.iter()) {
let raw_val = self.unpack_val(idx);
if raw_val < raw_threshold {
output.push(crate::ComparableDoc {
doc,
sort_key: Some(
self.stats.min_value + self.stats.gcd.get() * raw_val,
),
});
}
}
}
}
ValueRange::LessThanOrEqual(threshold, _) => {
if threshold >= self.stats.max_value {
for (&idx, &doc) in input_indexes.iter().zip(input_doc_ids.iter()) {
output.push(crate::ComparableDoc {
doc,
sort_key: Some(self.get_val(idx)),
});
}
} else if threshold < self.stats.min_value {
// All filtered out
} else {
let diff = threshold - self.stats.min_value;
let gcd = self.stats.gcd.get();
let raw_threshold = diff / gcd;
for (&idx, &doc) in input_indexes.iter().zip(input_doc_ids.iter()) {
let raw_val = self.unpack_val(idx);
if raw_val <= raw_threshold {
output.push(crate::ComparableDoc {
doc,
sort_key: Some(
self.stats.min_value + self.stats.gcd.get() * raw_val,
),
});
}
}
}
}
}
}
fn get_row_ids_for_value_range(
&self,
range: RangeInclusive<u64>,
range: ValueRange<u64>,
doc_id_range: Range<u32>,
positions: &mut Vec<u32>,
) {
let Some(transformed_range) =
transform_range_before_linear_transformation(&self.stats, range)
else {
positions.clear();
return;
};
self.bit_unpacker.get_ids_for_value_range(
transformed_range,
doc_id_range,
&self.data,
positions,
);
match range {
ValueRange::All => {
positions.extend(doc_id_range);
return;
}
ValueRange::Inclusive(range) => {
let Some(transformed_range) =
transform_range_before_linear_transformation(&self.stats, range)
else {
positions.clear();
return;
};
// TODO: This does not use the `self.blocks` cache, because callers are usually
// already doing sequential, and fairly dense reads. Fix it to
// iterate over blocks if that assumption turns out to be incorrect!
let data_range = self
.bit_unpacker
.block_oblivious_range(doc_id_range.clone(), self.data.len());
let data_offset = data_range.start;
let data_subset = self
.data
.slice(data_range)
.read_bytes()
.expect("Failed to read column values.");
self.bit_unpacker.get_ids_for_value_range_from_subset(
transformed_range,
doc_id_range,
data_offset,
&data_subset,
positions,
);
}
ValueRange::GreaterThan(threshold, _) => {
if threshold < self.stats.min_value {
positions.extend(doc_id_range);
return;
}
if threshold >= self.stats.max_value {
return;
}
let raw_threshold = (threshold - self.stats.min_value) / self.stats.gcd.get();
// We want raw > raw_threshold.
// bit_unpacker.get_ids_for_value_range_from_subset takes a RangeInclusive.
// We can construct a RangeInclusive: (raw_threshold + 1) ..= u64::MAX
// But max raw value is known? (max_value - min_value) / gcd.
let max_raw = (self.stats.max_value - self.stats.min_value) / self.stats.gcd.get();
let transformed_range = (raw_threshold + 1)..=max_raw;
let data_range = self
.bit_unpacker
.block_oblivious_range(doc_id_range.clone(), self.data.len());
let data_offset = data_range.start;
let data_subset = self
.data
.slice(data_range)
.read_bytes()
.expect("Failed to read column values.");
self.bit_unpacker.get_ids_for_value_range_from_subset(
transformed_range,
doc_id_range,
data_offset,
&data_subset,
positions,
);
}
ValueRange::GreaterThanOrEqual(threshold, _) => {
if threshold <= self.stats.min_value {
positions.extend(doc_id_range);
return;
}
if threshold > self.stats.max_value {
return;
}
let diff = threshold - self.stats.min_value;
let gcd = self.stats.gcd.get();
let raw_threshold = (diff + gcd - 1) / gcd;
// We want raw >= raw_threshold.
let max_raw = (self.stats.max_value - self.stats.min_value) / self.stats.gcd.get();
let transformed_range = raw_threshold..=max_raw;
let data_range = self
.bit_unpacker
.block_oblivious_range(doc_id_range.clone(), self.data.len());
let data_offset = data_range.start;
let data_subset = self
.data
.slice(data_range)
.read_bytes()
.expect("Failed to read column values.");
self.bit_unpacker.get_ids_for_value_range_from_subset(
transformed_range,
doc_id_range,
data_offset,
&data_subset,
positions,
);
}
ValueRange::LessThan(threshold, _) => {
if threshold > self.stats.max_value {
positions.extend(doc_id_range);
return;
}
if threshold <= self.stats.min_value {
return;
}
let diff = threshold - self.stats.min_value;
let gcd = self.stats.gcd.get();
// We want raw < raw_threshold_limit
// raw <= raw_threshold_limit - 1
let raw_threshold_limit = if diff % gcd == 0 {
diff / gcd
} else {
diff / gcd + 1
};
if raw_threshold_limit == 0 {
return;
}
let transformed_range = 0..=(raw_threshold_limit - 1);
let data_range = self
.bit_unpacker
.block_oblivious_range(doc_id_range.clone(), self.data.len());
let data_offset = data_range.start;
let data_subset = self
.data
.slice(data_range)
.read_bytes()
.expect("Failed to read column values.");
self.bit_unpacker.get_ids_for_value_range_from_subset(
transformed_range,
doc_id_range,
data_offset,
&data_subset,
positions,
);
}
ValueRange::LessThanOrEqual(threshold, _) => {
if threshold >= self.stats.max_value {
positions.extend(doc_id_range);
return;
}
if threshold < self.stats.min_value {
return;
}
let diff = threshold - self.stats.min_value;
let gcd = self.stats.gcd.get();
// We want raw <= raw_threshold.
let raw_threshold = diff / gcd;
let transformed_range = 0..=raw_threshold;
let data_range = self
.bit_unpacker
.block_oblivious_range(doc_id_range.clone(), self.data.len());
let data_offset = data_range.start;
let data_subset = self
.data
.slice(data_range)
.read_bytes()
.expect("Failed to read column values.");
self.bit_unpacker.get_ids_for_value_range_from_subset(
transformed_range,
doc_id_range,
data_offset,
&data_subset,
positions,
);
}
}
}
}
@@ -105,7 +445,7 @@ 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(
@@ -133,14 +473,20 @@ impl ColumnCodec for BitpackedCodec {
type Estimator = BitpackedCodecEstimator;
/// Opens a fast field given a file.
fn load(mut data: OwnedBytes) -> io::Result<Self::ColumnValues> {
let stats = ColumnStats::deserialize(&mut data)?;
fn load(file_slice: FileSlice) -> io::Result<Self::ColumnValues> {
let (stats, data) = ColumnStats::deserialize_from_tail(file_slice)?;
let num_bits = num_bits(&stats);
let bit_unpacker = BitUnpacker::new(num_bits);
let block_count = bit_unpacker.block_count(data.len());
Ok(BitpackedReader {
data,
bit_unpacker,
stats,
blocks: (0..block_count)
.into_iter()
.map(|_| OnceLock::new())
.collect(),
})
}
}

69
columnar/src/column_values/u64_based/blockwise_linear.rs
View File

@@ -1,8 +1,10 @@
use std::io;
use std::io::Write;
use std::sync::Arc;
use std::{io, iter};
use std::ops::{Deref, DerefMut};
use std::sync::{Arc, OnceLock};
use common::{BinarySerializable, CountingWriter, DeserializeFrom, OwnedBytes};
use common::file_slice::FileSlice;
use common::{BinarySerializable, CountingWriter, DeserializeFrom, HasLen, OwnedBytes};
use fastdivide::DividerU64;
use tantivy_bitpacker::{BitPacker, BitUnpacker, compute_num_bits};
@@ -172,32 +174,63 @@ impl ColumnCodec<u64> for BlockwiseLinearCodec {
type Estimator = BlockwiseLinearEstimator;
fn load(mut bytes: OwnedBytes) -> io::Result<Self::ColumnValues> {
let stats = ColumnStats::deserialize(&mut bytes)?;
let footer_len: u32 = (&bytes[bytes.len() - 4..]).deserialize()?;
let footer_offset = bytes.len() - 4 - footer_len as usize;
let (data, mut footer) = bytes.split(footer_offset);
fn load(file_slice: FileSlice) -> io::Result<Self::ColumnValues> {
let (stats, body) = ColumnStats::deserialize_from_tail(file_slice)?;
let (_, footer) = body.clone().split_from_end(4);
let footer_len: u32 = footer.read_bytes()?.as_slice().deserialize()?;
let (data, footer) = body.split_from_end(footer_len as usize + 4);
let mut footer = footer.read_bytes()?;
let num_blocks = compute_num_blocks(stats.num_rows);
let mut blocks: Vec<Block> = iter::repeat_with(|| Block::deserialize(&mut footer))
.take(num_blocks as usize)
.collect::<io::Result<_>>()?;
let mut start_offset = 0;
for block in &mut blocks {
let mut blocks = Vec::with_capacity(num_blocks as usize);
for _ in 0..num_blocks {
let mut block = Block::deserialize(&mut footer)?;
let len = (block.bit_unpacker.bit_width() as usize) * BLOCK_SIZE as usize / 8;
block.data_start_offset = start_offset;
start_offset += (block.bit_unpacker.bit_width() as usize) * BLOCK_SIZE as usize / 8;
blocks.push(BlockWithData {
block,
file_slice: data.slice(start_offset..(start_offset + len).min(data.len())),
data: Default::default(),
});
start_offset += len;
}
Ok(BlockwiseLinearReader {
blocks: blocks.into_boxed_slice().into(),
data,
stats,
})
}
}
struct BlockWithData {
block: Block,
file_slice: FileSlice,
data: OnceLock<OwnedBytes>,
}
impl Deref for BlockWithData {
type Target = Block;
fn deref(&self) -> &Self::Target {
&self.block
}
}
impl DerefMut for BlockWithData {
fn deref_mut(&mut self) -> &mut Self::Target {
&mut self.block
}
}
#[derive(Clone)]
pub struct BlockwiseLinearReader {
blocks: Arc<[Block]>,
data: OwnedBytes,
blocks: Arc<[BlockWithData]>,
stats: ColumnStats,
}
@@ -208,7 +241,9 @@ impl ColumnValues for BlockwiseLinearReader {
let idx_within_block = idx % BLOCK_SIZE;
let block = &self.blocks[block_id];
let interpoled_val: u64 = block.line.eval(idx_within_block);
let block_bytes = &self.data[block.data_start_offset..];
let block_bytes = block
.data
.get_or_init(|| block.file_slice.read_bytes().unwrap());
let bitpacked_diff = block.bit_unpacker.get(idx_within_block, block_bytes);
// TODO optimize me! the line parameters could be tweaked to include the multiplication and
// remove the dependency.

4
columnar/src/column_values/u64_based/line.rs
View File

@@ -8,7 +8,7 @@ use crate::column_values::ColumnValues;
const MID_POINT: u64 = (1u64 << 32) - 1u64;
/// `Line` describes a line function `y: ax + b` using integer
/// arithmetics.
/// arithmetic.
///
/// The slope is in fact a decimal split into a 32 bit integer value,
/// and a 32-bit decimal value.
@@ -94,7 +94,7 @@ impl Line {
// `(i, ys[])`.
//
// The best intercept therefore has the form
// `y[i] - line.eval(i)` (using wrapping arithmetics).
// `y[i] - line.eval(i)` (using wrapping arithmetic).
// In other words, the best intercept is one of the `y - Line::eval(ys[i])`
// and our task is just to pick the one that minimizes our error.
//

6
columnar/src/column_values/u64_based/linear.rs
View File

@@ -1,5 +1,6 @@
use std::io;
use common::file_slice::FileSlice;
use common::{BinarySerializable, OwnedBytes};
use tantivy_bitpacker::{BitPacker, BitUnpacker, compute_num_bits};
@@ -117,7 +118,7 @@ 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),
)
}
@@ -190,7 +191,8 @@ impl ColumnCodec for LinearCodec {
type Estimator = LinearCodecEstimator;
fn load(mut data: OwnedBytes) -> io::Result<Self::ColumnValues> {
fn load(file_slice: FileSlice) -> io::Result<Self::ColumnValues> {
let mut data = file_slice.read_bytes()?;
let stats = ColumnStats::deserialize(&mut data)?;
let linear_params = LinearParams::deserialize(&mut data)?;
Ok(LinearReader {

38
columnar/src/column_values/u64_based/mod.rs
View File

@@ -8,7 +8,8 @@ use std::io;
use std::io::Write;
use std::sync::Arc;
use common::{BinarySerializable, OwnedBytes};
use common::BinarySerializable;
use common::file_slice::FileSlice;
use crate::column_values::monotonic_mapping::{
StrictlyMonotonicMappingInverter, StrictlyMonotonicMappingToInternal,
@@ -52,7 +53,7 @@ pub trait ColumnCodecEstimator<T = u64>: 'static {
) -> io::Result<()>;
}
/// A column codec describes a colunm serialization format.
/// A column codec describes a column serialization format.
pub trait ColumnCodec<T: PartialOrd = u64> {
/// Specialized `ColumnValues` type.
type ColumnValues: ColumnValues<T> + 'static;
@@ -60,7 +61,7 @@ pub trait ColumnCodec<T: PartialOrd = u64> {
type Estimator: ColumnCodecEstimator + Default;
/// Loads a column that has been serialized using this codec.
fn load(bytes: OwnedBytes) -> io::Result<Self::ColumnValues>;
fn load(file_slice: FileSlice) -> io::Result<Self::ColumnValues>;
/// Returns an estimator.
fn estimator() -> Self::Estimator {
@@ -111,20 +112,22 @@ impl CodecType {
fn load<T: MonotonicallyMappableToU64>(
&self,
bytes: OwnedBytes,
file_slice: FileSlice,
) -> io::Result<Arc<dyn ColumnValues<T>>> {
match self {
CodecType::Bitpacked => load_specific_codec::<BitpackedCodec, T>(bytes),
CodecType::Linear => load_specific_codec::<LinearCodec, T>(bytes),
CodecType::BlockwiseLinear => load_specific_codec::<BlockwiseLinearCodec, T>(bytes),
CodecType::Bitpacked => load_specific_codec::<BitpackedCodec, T>(file_slice),
CodecType::Linear => load_specific_codec::<LinearCodec, T>(file_slice),
CodecType::BlockwiseLinear => {
load_specific_codec::<BlockwiseLinearCodec, T>(file_slice)
}
}
}
}
fn load_specific_codec<C: ColumnCodec, T: MonotonicallyMappableToU64>(
bytes: OwnedBytes,
file_slice: FileSlice,
) -> io::Result<Arc<dyn ColumnValues<T>>> {
let reader = C::load(bytes)?;
let reader = C::load(file_slice)?;
let reader_typed = monotonic_map_column(
reader,
StrictlyMonotonicMappingInverter::from(StrictlyMonotonicMappingToInternal::<T>::new()),
@@ -189,25 +192,28 @@ pub fn serialize_u64_based_column_values<T: MonotonicallyMappableToU64>(
///
/// This method first identifies the codec off the first byte.
pub fn load_u64_based_column_values<T: MonotonicallyMappableToU64>(
mut bytes: OwnedBytes,
file_slice: FileSlice,
) -> io::Result<Arc<dyn ColumnValues<T>>> {
let codec_type: CodecType = bytes
.first()
.copied()
let (header, body) = file_slice.split(1);
let codec_type: CodecType = header
.read_bytes()?
.as_slice()
.get(0)
.cloned()
.and_then(CodecType::try_from_code)
.ok_or_else(|| io::Error::new(io::ErrorKind::InvalidData, "Failed to read codec type"))?;
bytes.advance(1);
codec_type.load(bytes)
codec_type.load(body)
}
/// Helper function to serialize a column (autodetect from all codecs) and then open it
#[cfg(test)]
pub fn serialize_and_load_u64_based_column_values<T: MonotonicallyMappableToU64>(
vals: &dyn Iterable,
codec_types: &[CodecType],
) -> Arc<dyn ColumnValues<T>> {
let mut buffer = Vec::new();
serialize_u64_based_column_values(vals, codec_types, &mut buffer).unwrap();
load_u64_based_column_values::<T>(OwnedBytes::new(buffer)).unwrap()
load_u64_based_column_values::<T>(FileSlice::from(buffer)).unwrap()
}
#[cfg(test)]

23
columnar/src/column_values/u64_based/tests.rs
View File

@@ -1,3 +1,4 @@
use common::HasLen;
use proptest::prelude::*;
use proptest::{prop_oneof, proptest};
use rand::Rng;
@@ -13,7 +14,7 @@ fn test_serialize_and_load_simple() {
)
.unwrap();
assert_eq!(buffer.len(), 7);
let col = load_u64_based_column_values::<u64>(OwnedBytes::new(buffer)).unwrap();
let col = load_u64_based_column_values::<u64>(FileSlice::from(buffer)).unwrap();
assert_eq!(col.num_vals(), 3);
assert_eq!(col.get_val(0), 1);
assert_eq!(col.get_val(1), 2);
@@ -30,7 +31,7 @@ fn test_empty_column_i64() {
continue;
}
num_acceptable_codecs += 1;
let col = load_u64_based_column_values::<i64>(OwnedBytes::new(buffer)).unwrap();
let col = load_u64_based_column_values::<i64>(FileSlice::from(buffer)).unwrap();
assert_eq!(col.num_vals(), 0);
assert_eq!(col.min_value(), i64::MIN);
assert_eq!(col.max_value(), i64::MIN);
@@ -48,7 +49,7 @@ fn test_empty_column_u64() {
continue;
}
num_acceptable_codecs += 1;
let col = load_u64_based_column_values::<u64>(OwnedBytes::new(buffer)).unwrap();
let col = load_u64_based_column_values::<u64>(FileSlice::from(buffer)).unwrap();
assert_eq!(col.num_vals(), 0);
assert_eq!(col.min_value(), u64::MIN);
assert_eq!(col.max_value(), u64::MIN);
@@ -66,7 +67,7 @@ fn test_empty_column_f64() {
continue;
}
num_acceptable_codecs += 1;
let col = load_u64_based_column_values::<f64>(OwnedBytes::new(buffer)).unwrap();
let col = load_u64_based_column_values::<f64>(FileSlice::from(buffer)).unwrap();
assert_eq!(col.num_vals(), 0);
// FIXME. f64::MIN would be better!
assert!(col.min_value().is_nan());
@@ -97,7 +98,7 @@ pub(crate) fn create_and_validate<TColumnCodec: ColumnCodec>(
let actual_compression = buffer.len() as u64;
let reader = TColumnCodec::load(OwnedBytes::new(buffer)).unwrap();
let reader = TColumnCodec::load(FileSlice::from(buffer)).unwrap();
assert_eq!(reader.num_vals(), vals.len() as u32);
let mut buffer = Vec::new();
for (doc, orig_val) in vals.iter().copied().enumerate() {
@@ -131,7 +132,7 @@ pub(crate) fn create_and_validate<TColumnCodec: ColumnCodec>(
.collect();
let mut positions = Vec::new();
reader.get_row_ids_for_value_range(
vals[test_rand_idx]..=vals[test_rand_idx],
crate::column::ValueRange::Inclusive(vals[test_rand_idx]..=vals[test_rand_idx]),
0..vals.len() as u32,
&mut positions,
);
@@ -326,7 +327,7 @@ fn test_fastfield_gcd_i64_with_codec(codec_type: CodecType, num_vals: usize) ->
&[codec_type],
&mut buffer,
)?;
let buffer = OwnedBytes::new(buffer);
let buffer = FileSlice::from(buffer);
let column = crate::column_values::load_u64_based_column_values::<i64>(buffer.clone())?;
assert_eq!(column.get_val(0), -4000i64);
assert_eq!(column.get_val(1), -3000i64);
@@ -343,7 +344,7 @@ fn test_fastfield_gcd_i64_with_codec(codec_type: CodecType, num_vals: usize) ->
&[codec_type],
&mut buffer_without_gcd,
)?;
let buffer_without_gcd = OwnedBytes::new(buffer_without_gcd);
let buffer_without_gcd = FileSlice::from(buffer_without_gcd);
assert!(buffer_without_gcd.len() > buffer.len());
Ok(())
@@ -369,7 +370,7 @@ fn test_fastfield_gcd_u64_with_codec(codec_type: CodecType, num_vals: usize) ->
&[codec_type],
&mut buffer,
)?;
let buffer = OwnedBytes::new(buffer);
let buffer = FileSlice::from(buffer);
let column = crate::column_values::load_u64_based_column_values::<u64>(buffer.clone())?;
assert_eq!(column.get_val(0), 1000u64);
assert_eq!(column.get_val(1), 2000u64);
@@ -386,7 +387,7 @@ fn test_fastfield_gcd_u64_with_codec(codec_type: CodecType, num_vals: usize) ->
&[codec_type],
&mut buffer_without_gcd,
)?;
let buffer_without_gcd = OwnedBytes::new(buffer_without_gcd);
let buffer_without_gcd = FileSlice::from(buffer_without_gcd);
assert!(buffer_without_gcd.len() > buffer.len());
Ok(())
}
@@ -405,7 +406,7 @@ fn test_fastfield_gcd_u64() -> io::Result<()> {
#[test]
pub fn test_fastfield2() {
let test_fastfield = crate::column_values::serialize_and_load_u64_based_column_values::<u64>(
let test_fastfield = serialize_and_load_u64_based_column_values::<u64>(
&&[100u64, 200u64, 300u64][..],
&ALL_U64_CODEC_TYPES,
);

7
columnar/src/columnar/merge/mod.rs
View File

@@ -4,6 +4,7 @@ mod term_merger;
use std::collections::{BTreeMap, HashSet};
use std::io;
use std::io::ErrorKind;
use std::net::Ipv6Addr;
use std::sync::Arc;
@@ -78,6 +79,7 @@ pub fn merge_columnar(
required_columns: &[(String, ColumnType)],
merge_row_order: MergeRowOrder,
output: &mut impl io::Write,
cancel: impl Fn() -> bool,
) -> io::Result<()> {
let mut serializer = ColumnarSerializer::new(output);
let num_docs_per_columnar = columnar_readers
@@ -87,6 +89,9 @@ pub fn merge_columnar(
let columns_to_merge = group_columns_for_merge(columnar_readers, required_columns)?;
for res in columns_to_merge {
if cancel() {
return Err(io::Error::new(ErrorKind::Interrupted, "Merge cancelled"));
}
let ((column_name, _column_type_category), grouped_columns) = res;
let grouped_columns = grouped_columns.open(&merge_row_order)?;
if grouped_columns.is_empty() {
@@ -367,7 +372,7 @@ 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;
}

8
columnar/src/columnar/merge/tests.rs
View File

@@ -205,6 +205,7 @@ fn test_merge_columnar_numbers() {
&[],
MergeRowOrder::Stack(stack_merge_order),
&mut buffer,
|| false,
)
.unwrap();
let columnar_reader = ColumnarReader::open(buffer).unwrap();
@@ -233,6 +234,7 @@ fn test_merge_columnar_texts() {
&[],
MergeRowOrder::Stack(stack_merge_order),
&mut buffer,
|| false,
)
.unwrap();
let columnar_reader = ColumnarReader::open(buffer).unwrap();
@@ -282,6 +284,7 @@ fn test_merge_columnar_byte() {
&[],
MergeRowOrder::Stack(stack_merge_order),
&mut buffer,
|| false,
)
.unwrap();
let columnar_reader = ColumnarReader::open(buffer).unwrap();
@@ -338,6 +341,7 @@ fn test_merge_columnar_byte_with_missing() {
&[],
MergeRowOrder::Stack(stack_merge_order),
&mut buffer,
|| false,
)
.unwrap();
let columnar_reader = ColumnarReader::open(buffer).unwrap();
@@ -390,6 +394,7 @@ fn test_merge_columnar_different_types() {
&[],
MergeRowOrder::Stack(stack_merge_order),
&mut buffer,
|| false,
)
.unwrap();
let columnar_reader = ColumnarReader::open(buffer).unwrap();
@@ -455,6 +460,7 @@ fn test_merge_columnar_different_empty_cardinality() {
&[],
MergeRowOrder::Stack(stack_merge_order),
&mut buffer,
|| false,
)
.unwrap();
let columnar_reader = ColumnarReader::open(buffer).unwrap();
@@ -565,6 +571,7 @@ proptest! {
&[],
MergeRowOrder::Stack(stack_merge_order),
&mut out,
|| false,
).unwrap();
let merged_reader = ColumnarReader::open(out).unwrap();
@@ -582,6 +589,7 @@ proptest! {
&[],
MergeRowOrder::Stack(stack_merge_order),
&mut out,
|| false,
).unwrap();
}

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 {

22
columnar/src/comparable_doc.rs Normal file
View File

@@ -0,0 +1,22 @@
use serde::{Deserialize, Serialize};
/// Contains a feature (field, score, etc.) of a document along with the document address.
///
/// Used only by TopNComputer, which implements the actual comparison via a `Comparator`.
#[derive(Clone, Default, Eq, PartialEq, Serialize, Deserialize)]
pub struct ComparableDoc<T, D> {
/// The feature of the document. In practice, this is
/// is a type which can be compared with a `Comparator<T>`.
pub sort_key: T,
/// The document address. In practice, this is either a `DocId` or `DocAddress`.
pub doc: D,
}
impl<T: std::fmt::Debug, D: std::fmt::Debug> std::fmt::Debug for ComparableDoc<T, D> {
fn fmt(&self, f: &mut std::fmt::Formatter) -> std::fmt::Result {
f.debug_struct("ComparableDoc")
.field("feature", &self.sort_key)
.field("doc", &self.doc)
.finish()
}
}

9
columnar/src/compat_tests.rs
View File

@@ -71,7 +71,14 @@ fn test_format(path: &str) {
let columnar_readers = vec![&reader, &reader2];
let merge_row_order = StackMergeOrder::stack(&columnar_readers[..]);
let mut out = Vec::new();
merge_columnar(&columnar_readers, &[], merge_row_order.into(), &mut out).unwrap();
merge_columnar(
&columnar_readers,
&[],
merge_row_order.into(),
&mut out,
|| false,
)
.unwrap();
let reader = ColumnarReader::open(out).unwrap();
check_columns(&reader);
}

118
columnar/src/dynamic_column.rs
View File

@@ -3,7 +3,8 @@ use std::sync::Arc;
use std::{fmt, io};
use common::file_slice::FileSlice;
use common::{ByteCount, DateTime, HasLen, OwnedBytes};
use common::{ByteCount, DateTime};
use serde::{Deserialize, Serialize};
use crate::column::{BytesColumn, Column, StrColumn};
use crate::column_values::{StrictlyMonotonicFn, monotonic_map_column};
@@ -238,8 +239,7 @@ pub struct DynamicColumnHandle {
impl DynamicColumnHandle {
// TODO rename load
pub fn open(&self) -> io::Result<DynamicColumn> {
let column_bytes: OwnedBytes = self.file_slice.read_bytes()?;
self.open_internal(column_bytes)
self.open_internal(self.file_slice.clone())
}
#[doc(hidden)]
@@ -258,16 +258,15 @@ impl DynamicColumnHandle {
/// If not, the fastfield reader will returns the u64-value associated with the original
/// FastValue.
pub fn open_u64_lenient(&self) -> io::Result<Option<Column<u64>>> {
let column_bytes = self.file_slice.read_bytes()?;
match self.column_type {
ColumnType::Str | ColumnType::Bytes => {
let column: BytesColumn =
crate::column::open_column_bytes(column_bytes, self.format_version)?;
crate::column::open_column_bytes(self.file_slice.clone(), self.format_version)?;
Ok(Some(column.term_ord_column))
}
ColumnType::IpAddr => {
let column = crate::column::open_column_u128_as_compact_u64(
column_bytes,
self.file_slice.clone(),
self.format_version,
)?;
Ok(Some(column))
@@ -277,50 +276,129 @@ impl DynamicColumnHandle {
| ColumnType::U64
| ColumnType::F64
| ColumnType::DateTime => {
let column =
crate::column::open_column_u64::<u64>(column_bytes, self.format_version)?;
let column = crate::column::open_column_u64::<u64>(
self.file_slice.clone(),
self.format_version,
)?;
Ok(Some(column))
}
}
}
fn open_internal(&self, column_bytes: OwnedBytes) -> io::Result<DynamicColumn> {
fn open_internal(&self, file_slice: FileSlice) -> io::Result<DynamicColumn> {
let dynamic_column: DynamicColumn = match self.column_type {
ColumnType::Bytes => {
crate::column::open_column_bytes(column_bytes, self.format_version)?.into()
crate::column::open_column_bytes(file_slice, self.format_version)?.into()
}
ColumnType::Str => {
crate::column::open_column_str(column_bytes, self.format_version)?.into()
crate::column::open_column_str(file_slice, self.format_version)?.into()
}
ColumnType::I64 => {
crate::column::open_column_u64::<i64>(column_bytes, self.format_version)?.into()
crate::column::open_column_u64::<i64>(file_slice, self.format_version)?.into()
}
ColumnType::U64 => {
crate::column::open_column_u64::<u64>(column_bytes, self.format_version)?.into()
crate::column::open_column_u64::<u64>(file_slice, self.format_version)?.into()
}
ColumnType::F64 => {
crate::column::open_column_u64::<f64>(column_bytes, self.format_version)?.into()
crate::column::open_column_u64::<f64>(file_slice, self.format_version)?.into()
}
ColumnType::Bool => {
crate::column::open_column_u64::<bool>(column_bytes, self.format_version)?.into()
crate::column::open_column_u64::<bool>(file_slice, self.format_version)?.into()
}
ColumnType::IpAddr => {
crate::column::open_column_u128::<Ipv6Addr>(column_bytes, self.format_version)?
.into()
crate::column::open_column_u128::<Ipv6Addr>(file_slice, self.format_version)?.into()
}
ColumnType::DateTime => {
crate::column::open_column_u64::<DateTime>(column_bytes, self.format_version)?
.into()
crate::column::open_column_u64::<DateTime>(file_slice, self.format_version)?.into()
}
};
Ok(dynamic_column)
}
pub fn num_bytes(&self) -> ByteCount {
self.file_slice.len().into()
self.file_slice.num_bytes()
}
/// Legacy helper returning the column space usage.
pub fn column_and_dictionary_num_bytes(&self) -> io::Result<ColumnSpaceUsage> {
self.space_usage()
}
/// Return the space usage of the column, optionally broken down by dictionary and column
/// values.
///
/// For dictionary encoded columns (strings and bytes), this splits the total footprint into
/// the dictionary and the remaining column data (including index and values).
/// For all other column types, the dictionary size is `None` and the column size
/// equals the total bytes.
pub fn space_usage(&self) -> io::Result<ColumnSpaceUsage> {
let total_num_bytes = self.num_bytes();
let dynamic_column = self.open()?;
let dictionary_num_bytes = match &dynamic_column {
DynamicColumn::Bytes(bytes_column) => bytes_column.dictionary().num_bytes(),
DynamicColumn::Str(str_column) => str_column.dictionary().num_bytes(),
_ => {
return Ok(ColumnSpaceUsage::new(self.num_bytes(), None));
}
};
assert!(dictionary_num_bytes <= total_num_bytes);
let column_num_bytes =
ByteCount::from(total_num_bytes.get_bytes() - dictionary_num_bytes.get_bytes());
Ok(ColumnSpaceUsage::new(
column_num_bytes,
Some(dictionary_num_bytes),
))
}
pub fn column_type(&self) -> ColumnType {
self.column_type
}
}
/// Represents space usage of a column.
///
/// `column_num_bytes` tracks the column payload (index, values and footer).
/// For dictionary encoded columns, `dictionary_num_bytes` captures the dictionary footprint.
/// [`ColumnSpaceUsage::total_num_bytes`] returns the sum of both parts.
#[derive(Clone, Debug, Serialize, Deserialize)]
pub struct ColumnSpaceUsage {
column_num_bytes: ByteCount,
dictionary_num_bytes: Option<ByteCount>,
}
impl ColumnSpaceUsage {
pub(crate) fn new(
column_num_bytes: ByteCount,
dictionary_num_bytes: Option<ByteCount>,
) -> Self {
ColumnSpaceUsage {
column_num_bytes,
dictionary_num_bytes,
}
}
pub fn column_num_bytes(&self) -> ByteCount {
self.column_num_bytes
}
pub fn dictionary_num_bytes(&self) -> Option<ByteCount> {
self.dictionary_num_bytes
}
pub fn total_num_bytes(&self) -> ByteCount {
self.column_num_bytes + self.dictionary_num_bytes.unwrap_or_default()
}
/// Merge two space usage values by summing their components.
pub fn merge(&self, other: &ColumnSpaceUsage) -> ColumnSpaceUsage {
let dictionary_num_bytes = match (self.dictionary_num_bytes, other.dictionary_num_bytes) {
(Some(lhs), Some(rhs)) => Some(lhs + rhs),
(Some(val), None) | (None, Some(val)) => Some(val),
(None, None) => None,
};
ColumnSpaceUsage {
column_num_bytes: self.column_num_bytes + other.column_num_bytes,
dictionary_num_bytes,
}
}
}

6
columnar/src/lib.rs
View File

@@ -29,6 +29,7 @@ mod column;
pub mod column_index;
pub mod column_values;
mod columnar;
mod comparable_doc;
mod dictionary;
mod dynamic_column;
mod iterable;
@@ -36,7 +37,7 @@ pub(crate) mod utils;
mod value;
pub use block_accessor::ColumnBlockAccessor;
pub use column::{BytesColumn, Column, StrColumn};
pub use column::{BytesColumn, Column, StrColumn, ValueRange};
pub use column_index::ColumnIndex;
pub use column_values::{
ColumnValues, EmptyColumnValues, MonotonicallyMappableToU64, MonotonicallyMappableToU128,
@@ -45,10 +46,11 @@ pub use columnar::{
CURRENT_VERSION, ColumnType, ColumnarReader, ColumnarWriter, HasAssociatedColumnType,
MergeRowOrder, ShuffleMergeOrder, StackMergeOrder, Version, merge_columnar,
};
pub use comparable_doc::ComparableDoc;
use sstable::VoidSSTable;
pub use value::{NumericalType, NumericalValue};
pub use self::dynamic_column::{DynamicColumn, DynamicColumnHandle};
pub use self::dynamic_column::{ColumnSpaceUsage, DynamicColumn, DynamicColumnHandle};
pub type RowId = u32;
pub type DocId = u32;

8
columnar/src/tests.rs
View File

@@ -641,7 +641,7 @@ proptest! {
let columnar_readers_arr: Vec<&ColumnarReader> = columnar_readers.iter().collect();
let mut output: Vec<u8> = Vec::new();
let stack_merge_order = StackMergeOrder::stack(&columnar_readers_arr[..]).into();
crate::merge_columnar(&columnar_readers_arr[..], &[], stack_merge_order, &mut output).unwrap();
crate::merge_columnar(&columnar_readers_arr[..], &[], stack_merge_order, &mut output, || false,).unwrap();
let merged_columnar = ColumnarReader::open(output).unwrap();
let concat_rows: Vec<Vec<(&'static str, ColumnValue)>> = columnar_docs.iter().flatten().cloned().collect();
let expected_merged_columnar = build_columnar(&concat_rows[..]);
@@ -665,6 +665,7 @@ fn test_columnar_merging_empty_columnar() {
&[],
crate::MergeRowOrder::Stack(stack_merge_order),
&mut output,
|| false,
)
.unwrap();
let merged_columnar = ColumnarReader::open(output).unwrap();
@@ -702,6 +703,7 @@ fn test_columnar_merging_number_columns() {
&[],
crate::MergeRowOrder::Stack(stack_merge_order),
&mut output,
|| false,
)
.unwrap();
let merged_columnar = ColumnarReader::open(output).unwrap();
@@ -775,6 +777,7 @@ fn test_columnar_merge_and_remap(
&[],
shuffle_merge_order.into(),
&mut output,
|| false,
)
.unwrap();
let merged_columnar = ColumnarReader::open(output).unwrap();
@@ -817,6 +820,7 @@ fn test_columnar_merge_empty() {
&[],
shuffle_merge_order.into(),
&mut output,
|| false,
)
.unwrap();
let merged_columnar = ColumnarReader::open(output).unwrap();
@@ -843,6 +847,7 @@ fn test_columnar_merge_single_str_column() {
&[],
shuffle_merge_order.into(),
&mut output,
|| false,
)
.unwrap();
let merged_columnar = ColumnarReader::open(output).unwrap();
@@ -875,6 +880,7 @@ fn test_delete_decrease_cardinality() {
&[],
shuffle_merge_order.into(),
&mut output,
|| false,
)
.unwrap();
let merged_columnar = ColumnarReader::open(output).unwrap();

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/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 {

106
common/src/buffered_file_slice.rs Normal file
View File

@@ -0,0 +1,106 @@
use std::cell::RefCell;
use std::cmp::min;
use std::io;
use std::ops::Range;
use super::file_slice::FileSlice;
use super::{HasLen, OwnedBytes};
const DEFAULT_BUFFER_MAX_SIZE: usize = 512 * 1024; // 512K
/// A buffered reader for a FileSlice.
///
/// Reads the underlying `FileSlice` in large, sequential chunks to amortize
/// the cost of `read_bytes` calls, while keeping peak memory usage under control.
///
/// TODO: Rather than wrapping a `FileSlice` in buffering, it will usually be better to adjust a
/// `FileHandle` to directly handle buffering itself.
/// TODO: See: https://github.com/paradedb/paradedb/issues/3374
pub struct BufferedFileSlice {
file_slice: FileSlice,
buffer: RefCell<OwnedBytes>,
buffer_range: RefCell<Range<u64>>,
buffer_max_size: usize,
}
impl BufferedFileSlice {
/// Creates a new `BufferedFileSlice`.
///
/// The `buffer_max_size` is the amount of data that will be read from the
/// `FileSlice` on a buffer miss.
pub fn new(file_slice: FileSlice, buffer_max_size: usize) -> Self {
Self {
file_slice,
buffer: RefCell::new(OwnedBytes::empty()),
buffer_range: RefCell::new(0..0),
buffer_max_size,
}
}
/// Creates a new `BufferedFileSlice` with a default buffer max size.
pub fn new_with_default_buffer_size(file_slice: FileSlice) -> Self {
Self::new(file_slice, DEFAULT_BUFFER_MAX_SIZE)
}
/// Creates an empty `BufferedFileSlice`.
pub fn empty() -> Self {
Self::new(FileSlice::empty(), 0)
}
/// Returns an `OwnedBytes` corresponding to the given `required_range`.
///
/// If the requested range is not in the buffer, this will trigger a read
/// from the underlying `FileSlice`.
///
/// If the requested range is larger than the buffer_max_size, it will be read directly from the
/// source without buffering.
///
/// # Errors
///
/// Returns an `io::Error` if the underlying read fails or the range is
/// out of bounds.
pub fn get_bytes(&self, required_range: Range<u64>) -> io::Result<OwnedBytes> {
let buffer_range = self.buffer_range.borrow();
// Cache miss condition: the required range is not fully contained in the current buffer.
if required_range.start < buffer_range.start || required_range.end > buffer_range.end {
drop(buffer_range); // release borrow before mutating
if required_range.end > self.file_slice.len() as u64 {
return Err(io::Error::new(
io::ErrorKind::UnexpectedEof,
"Requested range extends beyond the end of the file slice.",
));
}
if (required_range.end - required_range.start) as usize > self.buffer_max_size {
// This read is larger than our buffer max size.
// Read it directly and bypass the buffer to avoid churning.
return self
.file_slice
.read_bytes_slice(required_range.start as usize..required_range.end as usize);
}
let new_buffer_start = required_range.start;
let new_buffer_end = min(
new_buffer_start + self.buffer_max_size as u64,
self.file_slice.len() as u64,
);
let read_range = new_buffer_start..new_buffer_end;
let new_buffer = self
.file_slice
.read_bytes_slice(read_range.start as usize..read_range.end as usize)?;
self.buffer.replace(new_buffer);
self.buffer_range.replace(read_range);
}
// Now the data is guaranteed to be in the buffer.
let buffer = self.buffer.borrow();
let buffer_range = self.buffer_range.borrow();
let local_start = (required_range.start - buffer_range.start) as usize;
let local_end = (required_range.end - buffer_range.start) as usize;
Ok(buffer.slice(local_start..local_end))
}
}

23
common/src/file_slice.rs
View File

@@ -1,7 +1,7 @@
use std::fs::File;
use std::ops::{Deref, Range, RangeBounds};
use std::path::Path;
use std::sync::Arc;
use std::sync::{Arc, OnceLock};
use std::{fmt, io};
use async_trait::async_trait;
@@ -339,6 +339,27 @@ impl FileHandle for OwnedBytes {
}
}
pub struct DeferredFileSlice {
opener: Arc<dyn Fn() -> io::Result<FileSlice> + Send + Sync + 'static>,
file_slice: OnceLock<std::io::Result<FileSlice>>,
}
impl DeferredFileSlice {
pub fn new(opener: impl Fn() -> io::Result<FileSlice> + Send + Sync + 'static) -> Self {
DeferredFileSlice {
opener: Arc::new(opener),
file_slice: OnceLock::default(),
}
}
pub fn open(&self) -> io::Result<&FileSlice> {
match self.file_slice.get_or_init(|| (self.opener)()) {
Ok(file_slice) => Ok(file_slice),
Err(e) => Err(io::Error::new(io::ErrorKind::Other, e.to_string())),
}
}
}
#[cfg(test)]
mod tests {
use std::io;

1
common/src/lib.rs
View File

@@ -6,6 +6,7 @@ pub use byteorder::LittleEndian as Endianness;
mod bitset;
pub mod bounds;
pub mod buffered_file_slice;
mod byte_count;
mod datetime;
pub mod file_slice;

32
common/src/vint.rs
View File

@@ -28,7 +28,9 @@ impl BinarySerializable for VIntU128 {
writer.write_all(&buffer)
}
#[allow(clippy::unbuffered_bytes)]
fn deserialize<R: Read>(reader: &mut R) -> io::Result<Self> {
#[allow(clippy::unbuffered_bytes)]
let mut bytes = reader.bytes();
let mut result = 0u128;
let mut shift = 0u64;
@@ -56,6 +58,33 @@ impl BinarySerializable for VIntU128 {
#[derive(Clone, Copy, Debug, Eq, PartialEq)]
pub struct VInt(pub u64);
impl VInt {
pub fn deserialize_with_size<R: Read>(reader: &mut R) -> io::Result<(Self, usize)> {
let mut nbytes = 0;
let mut bytes = reader.bytes();
let mut result = 0u64;
let mut shift = 0u64;
loop {
match bytes.next() {
Some(Ok(b)) => {
nbytes += 1;
result |= u64::from(b % 128u8) << shift;
if b >= STOP_BIT {
return Ok((VInt(result), nbytes));
}
shift += 7;
}
_ => {
return Err(io::Error::new(
io::ErrorKind::InvalidData,
"Reach end of buffer while reading VInt",
));
}
}
}
}
}
const STOP_BIT: u8 = 128;
#[inline]
@@ -195,7 +224,9 @@ impl BinarySerializable for VInt {
writer.write_all(&buffer[0..num_bytes])
}
#[allow(clippy::unbuffered_bytes)]
fn deserialize<R: Read>(reader: &mut R) -> io::Result<Self> {
#[allow(clippy::unbuffered_bytes)]
let mut bytes = reader.bytes();
let mut result = 0u64;
let mut shift = 0u64;
@@ -221,7 +252,6 @@ impl BinarySerializable for VInt {
#[cfg(test)]
mod tests {
use super::{BinarySerializable, VInt, serialize_vint_u32};
fn aux_test_vint(val: u64) {

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

Binary file not shown.
Side by Side

Before

Width:  |  Height:  |  Size: 653 KiB

4
examples/basic_search.rs
View File

@@ -208,7 +208,7 @@ fn main() -> tantivy::Result<()> {
// is the role of the `TopDocs` collector.
// We can now perform our query.
let top_docs = searcher.search(&query, &TopDocs::with_limit(10))?;
let top_docs = searcher.search(&query, &TopDocs::with_limit(10).order_by_score())?;
// The actual documents still need to be
// retrieved from Tantivy's store.
@@ -226,7 +226,7 @@ fn main() -> tantivy::Result<()> {
let query = query_parser.parse_query("title:sea^20 body:whale^70")?;
let (_score, doc_address) = searcher
.search(&query, &TopDocs::with_limit(1))?
.search(&query, &TopDocs::with_limit(1).order_by_score())?
.into_iter()
.next()
.unwrap();

2
examples/custom_tokenizer.rs
View File

@@ -100,7 +100,7 @@ fn main() -> tantivy::Result<()> {
// here we want to get a hit on the 'ken' in Frankenstein
let query = query_parser.parse_query("ken")?;
let top_docs = searcher.search(&query, &TopDocs::with_limit(10))?;
let top_docs = searcher.search(&query, &TopDocs::with_limit(10).order_by_score())?;
for (_, doc_address) in top_docs {
let retrieved_doc: TantivyDocument = searcher.doc(doc_address)?;

4
examples/date_time_field.rs
View File

@@ -50,14 +50,14 @@ fn main() -> tantivy::Result<()> {
{
// Simple exact search on the date
let query = query_parser.parse_query("occurred_at:\"2022-06-22T12:53:50.53Z\"")?;
let count_docs = searcher.search(&*query, &TopDocs::with_limit(5))?;
let count_docs = searcher.search(&*query, &TopDocs::with_limit(5).order_by_score())?;
assert_eq!(count_docs.len(), 1);
}
{
// Range query on the date field
let query = query_parser
.parse_query(r#"occurred_at:[2022-06-22T12:58:00Z TO 2022-06-23T00:00:00Z}"#)?;
let count_docs = searcher.search(&*query, &TopDocs::with_limit(4))?;
let count_docs = searcher.search(&*query, &TopDocs::with_limit(4).order_by_score())?;
assert_eq!(count_docs.len(), 1);
for (_score, doc_address) in count_docs {
let retrieved_doc = searcher.doc::<TantivyDocument>(doc_address)?;

2
examples/deleting_updating_documents.rs
View File

@@ -28,7 +28,7 @@ fn extract_doc_given_isbn(
// The second argument is here to tell we don't care about decoding positions,
// or term frequencies.
let term_query = TermQuery::new(isbn_term.clone(), IndexRecordOption::Basic);
let top_docs = searcher.search(&term_query, &TopDocs::with_limit(1))?;
let top_docs = searcher.search(&term_query, &TopDocs::with_limit(1).order_by_score())?;
if let Some((_score, doc_address)) = top_docs.first() {
let doc = searcher.doc(*doc_address)?;

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(())
}

3
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
//
@@ -146,7 +145,7 @@ fn main() -> tantivy::Result<()> {
let query = FuzzyTermQuery::new(term, 2, true);
let (top_docs, count) = searcher
.search(&query, &(TopDocs::with_limit(5), Count))
.search(&query, &(TopDocs::with_limit(5).order_by_score(), Count))
.unwrap();
assert_eq!(count, 3);
assert_eq!(top_docs.len(), 3);

8
examples/ip_field.rs
View File

@@ -69,25 +69,25 @@ fn main() -> tantivy::Result<()> {
{
// Inclusive range queries
let query = query_parser.parse_query("ip:[192.168.0.80 TO 192.168.0.100]")?;
let count_docs = searcher.search(&*query, &TopDocs::with_limit(5))?;
let count_docs = searcher.search(&*query, &TopDocs::with_limit(5).order_by_score())?;
assert_eq!(count_docs.len(), 1);
}
{
// Exclusive range queries
let query = query_parser.parse_query("ip:{192.168.0.80 TO 192.168.1.100]")?;
let count_docs = searcher.search(&*query, &TopDocs::with_limit(2))?;
let count_docs = searcher.search(&*query, &TopDocs::with_limit(2).order_by_score())?;
assert_eq!(count_docs.len(), 0);
}
{
// Find docs with IP addresses smaller equal 192.168.1.100
let query = query_parser.parse_query("ip:[* TO 192.168.1.100]")?;
let count_docs = searcher.search(&*query, &TopDocs::with_limit(2))?;
let count_docs = searcher.search(&*query, &TopDocs::with_limit(2).order_by_score())?;
assert_eq!(count_docs.len(), 2);
}
{
// Find docs with IP addresses smaller than 192.168.1.100
let query = query_parser.parse_query("ip:[* TO 192.168.1.100}")?;
let count_docs = searcher.search(&*query, &TopDocs::with_limit(2))?;
let count_docs = searcher.search(&*query, &TopDocs::with_limit(2).order_by_score())?;
assert_eq!(count_docs.len(), 2);
}

14
examples/json_field.rs
View File

@@ -59,12 +59,12 @@ fn main() -> tantivy::Result<()> {
let query_parser = QueryParser::for_index(&index, vec![event_type, attributes]);
{
let query = query_parser.parse_query("target:submit-button")?;
let count_docs = searcher.search(&*query, &TopDocs::with_limit(2))?;
let count_docs = searcher.search(&*query, &TopDocs::with_limit(2).order_by_score())?;
assert_eq!(count_docs.len(), 2);
}
{
let query = query_parser.parse_query("target:submit")?;
let count_docs = searcher.search(&*query, &TopDocs::with_limit(2))?;
let count_docs = searcher.search(&*query, &TopDocs::with_limit(2).order_by_score())?;
assert_eq!(count_docs.len(), 2);
}
{
@@ -74,33 +74,33 @@ fn main() -> tantivy::Result<()> {
}
{
let query = query_parser.parse_query("click AND cart.product_id:133")?;
let hits = searcher.search(&*query, &TopDocs::with_limit(2))?;
let hits = searcher.search(&*query, &TopDocs::with_limit(2).order_by_score())?;
assert_eq!(hits.len(), 1);
}
{
// The sub-fields in the json field marked as default field still need to be explicitly
// addressed
let query = query_parser.parse_query("click AND 133")?;
let hits = searcher.search(&*query, &TopDocs::with_limit(2))?;
let hits = searcher.search(&*query, &TopDocs::with_limit(2).order_by_score())?;
assert_eq!(hits.len(), 0);
}
{
// Default json fields are ignored if they collide with the schema
let query = query_parser.parse_query("event_type:holiday-sale")?;
let hits = searcher.search(&*query, &TopDocs::with_limit(2))?;
let hits = searcher.search(&*query, &TopDocs::with_limit(2).order_by_score())?;
assert_eq!(hits.len(), 0);
}
// # Query via full attribute path
{
// This only searches in our schema's `event_type` field
let query = query_parser.parse_query("event_type:click")?;
let hits = searcher.search(&*query, &TopDocs::with_limit(2))?;
let hits = searcher.search(&*query, &TopDocs::with_limit(2).order_by_score())?;
assert_eq!(hits.len(), 2);
}
{
// Default json fields can still be accessed by full path
let query = query_parser.parse_query("attributes.event_type:holiday-sale")?;
let hits = searcher.search(&*query, &TopDocs::with_limit(2))?;
let hits = searcher.search(&*query, &TopDocs::with_limit(2).order_by_score())?;
assert_eq!(hits.len(), 1);
}
Ok(())

86
examples/multiple_snippets.rs Normal file
View File

@@ -0,0 +1,86 @@
// # Multiple Snippets Example
//
// This example demonstrates how to return multiple text fragments
// from a document, useful for long documents with matches in different locations.
use tantivy::collector::TopDocs;
use tantivy::query::QueryParser;
use tantivy::schema::*;
use tantivy::snippet::SnippetGenerator;
use tantivy::{doc, Index, IndexWriter};
use tempfile::TempDir;
fn main() -> tantivy::Result<()> {
let index_path = TempDir::new()?;
// Define the schema
let mut schema_builder = Schema::builder();
let title = schema_builder.add_text_field("title", TEXT | STORED);
let body = schema_builder.add_text_field("body", TEXT | STORED);
let schema = schema_builder.build();
// Create the index
let index = Index::create_in_dir(&index_path, schema)?;
let mut index_writer: IndexWriter = index.writer(50_000_000)?;
// Index a long document with multiple occurrences of "rust"
index_writer.add_document(doc!(
title => "The Rust Programming Language",
body => "Rust is a systems programming language that runs blazingly fast, prevents \
segfaults, and guarantees thread safety. Lorem ipsum dolor sit amet, \
consectetur adipiscing elit. Sed do eiusmod tempor incididunt ut labore. \
Rust empowers everyone to build reliable and efficient software. More filler \
text to create distance between matches. Ut enim ad minim veniam, quis nostrud \
exercitation ullamco laboris nisi ut aliquip ex ea commodo consequat. \
The Rust compiler is known for its helpful error messages. Duis aute irure \
dolor in reprehenderit in voluptate velit esse cillum dolore eu fugiat nulla \
pariatur. Rust has a strong type system and ownership model."
))?;
index_writer.commit()?;
let reader = index.reader()?;
let searcher = reader.searcher();
let query_parser = QueryParser::for_index(&index, vec![title, body]);
let query = query_parser.parse_query("rust")?;
let top_docs = searcher.search(&query, &TopDocs::with_limit(10).order_by_score())?;
// Create snippet generator
let mut snippet_generator = SnippetGenerator::create(&searcher, &*query, body)?;
println!("=== Single Snippet (Default Behavior) ===\n");
for (score, doc_address) in &top_docs {
let doc = searcher.doc::<TantivyDocument>(*doc_address)?;
let snippet = snippet_generator.snippet_from_doc(&doc);
println!("Document score: {}", score);
println!("Title: {}", doc.get_first(title).unwrap().as_str().unwrap());
println!("Single snippet: {}\n", snippet.to_html());
}
println!("\n=== Multiple Snippets (New Feature) ===\n");
// Configure to return multiple snippets
// Get up to 3 snippets
snippet_generator.set_snippets_limit(3);
// Smaller fragments
snippet_generator.set_max_num_chars(80);
// By default, multiple snippets are sorted by score. You can change this to sort by position.
// snippet_generator.set_sort_order(SnippetSortOrder::Position);
for (score, doc_address) in top_docs {
let doc = searcher.doc::<TantivyDocument>(doc_address)?;
let snippets = snippet_generator.snippets_from_doc(&doc);
println!("Document score: {}", score);
println!("Title: {}", doc.get_first(title).unwrap().as_str().unwrap());
println!("Found {} snippets:", snippets.len());
for (i, snippet) in snippets.iter().enumerate() {
println!(" Snippet {}: {}", i + 1, snippet.to_html());
}
println!();
}
Ok(())
}

2
examples/phrase_prefix_search.rs
View File

@@ -63,7 +63,7 @@ fn main() -> Result<()> {
// but not "in the Gulf Stream".
let query = query_parser.parse_query("\"in the su\"*")?;
let top_docs = searcher.search(&query, &TopDocs::with_limit(10))?;
let top_docs = searcher.search(&query, &TopDocs::with_limit(10).order_by_score())?;
let mut titles = top_docs
.into_iter()
.map(|(_score, doc_address)| {

6
examples/pre_tokenized_text.rs
View File

@@ -107,7 +107,8 @@ fn main() -> tantivy::Result<()> {
IndexRecordOption::Basic,
);
let (top_docs, count) = searcher.search(&query, &(TopDocs::with_limit(2), Count))?;
let (top_docs, count) =
searcher.search(&query, &(TopDocs::with_limit(2).order_by_score(), Count))?;
assert_eq!(count, 2);
@@ -128,7 +129,8 @@ fn main() -> tantivy::Result<()> {
IndexRecordOption::Basic,
);
let (_top_docs, count) = searcher.search(&query, &(TopDocs::with_limit(2), Count))?;
let (_top_docs, count) =
searcher.search(&query, &(TopDocs::with_limit(2).order_by_score(), Count))?;
assert_eq!(count, 0);

2
examples/snippet.rs
View File

@@ -50,7 +50,7 @@ fn main() -> tantivy::Result<()> {
let query_parser = QueryParser::for_index(&index, vec![title, body]);
let query = query_parser.parse_query("sycamore spring")?;
let top_docs = searcher.search(&query, &TopDocs::with_limit(10))?;
let top_docs = searcher.search(&query, &TopDocs::with_limit(10).order_by_score())?;
let snippet_generator = SnippetGenerator::create(&searcher, &*query, body)?;

2
examples/stop_words.rs
View File

@@ -102,7 +102,7 @@ fn main() -> tantivy::Result<()> {
// stop words are applied on the query as well.
// The following will be equivalent to `title:frankenstein`
let query = query_parser.parse_query("title:\"the Frankenstein\"")?;
let top_docs = searcher.search(&query, &TopDocs::with_limit(10))?;
let top_docs = searcher.search(&query, &TopDocs::with_limit(10).order_by_score())?;
for (score, doc_address) in top_docs {
let retrieved_doc: TantivyDocument = searcher.doc(doc_address)?;

2
examples/warmer.rs
View File

@@ -164,7 +164,7 @@ fn main() -> tantivy::Result<()> {
move |doc_id: DocId| Reverse(price[doc_id as usize])
};
let most_expensive_first = TopDocs::with_limit(10).custom_score(score_by_price);
let most_expensive_first = TopDocs::with_limit(10).order_by(score_by_price);
let hits = searcher.search(&query, &most_expensive_first)?;
assert_eq!(

2
query-grammar/Cargo.toml
View File

@@ -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() {

195
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;
@@ -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)
}
@@ -696,7 +758,17 @@ fn negate(expr: UserInputAst) -> UserInputAst {
fn leaf(inp: &str) -> IResult<&str, UserInputAst> {
alt((
delimited(char('('), ast, char(')')),
map(char('*'), |_| UserInputAst::from(UserInputLeaf::All)),
map(
terminated(
char('*'),
peek(alt((
value((), multispace1),
value((), char(')')),
value((), eof),
))),
),
|_| UserInputAst::from(UserInputLeaf::All),
),
map(preceded(tuple((tag("NOT"), multispace1)), leaf), negate),
literal,
))(inp)
@@ -717,7 +789,17 @@ fn leaf_infallible(inp: &str) -> JResult<&str, Option<UserInputAst>> {
),
),
(
value((), char('*')),
value(
(),
terminated(
char('*'),
peek(alt((
value((), multispace1),
value((), char(')')),
value((), eof),
))),
),
),
map(nothing, |_| {
(Some(UserInputAst::from(UserInputLeaf::All)), Vec::new())
}),
@@ -753,7 +835,7 @@ 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 +847,7 @@ 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 +1098,25 @@ 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);
}
}
@@ -1596,6 +1691,21 @@ mod test {
test_parse_query_to_ast_helper("abc:a b", "(*\"abc\":a *b)");
test_parse_query_to_ast_helper("abc:\"a b\"", "\"abc\":\"a b\"");
test_parse_query_to_ast_helper("foo:[1 TO 5]", "\"foo\":[\"1\" TO \"5\"]");
// Phrase prefixed with *
test_parse_query_to_ast_helper("foo:(*A)", "\"foo\":*A");
test_parse_query_to_ast_helper("*A", "*A");
test_parse_query_to_ast_helper("(*A)", "*A");
test_parse_query_to_ast_helper("foo:(A OR B)", "(?\"foo\":A ?\"foo\":B)");
test_parse_query_to_ast_helper("foo:(A* OR B*)", "(?\"foo\":A* ?\"foo\":B*)");
test_parse_query_to_ast_helper("foo:(*A OR *B)", "(?\"foo\":*A ?\"foo\":*B)");
}
#[test]
fn test_parse_query_all() {
test_parse_query_to_ast_helper("*", "*");
test_parse_query_to_ast_helper("(*)", "*");
test_parse_query_to_ast_helper("(* )", "*");
}
#[test]
@@ -1694,6 +1804,63 @@ mod test {
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"#);

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!(

3
runtests.sh Executable file
View File

@@ -0,0 +1,3 @@
#! /bin/bash
cargo +stable nextest run --features quickwit,mmap,stopwords,lz4-compression,zstd-compression,failpoints --verbose --workspace

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.

105
src/aggregation/accessor_helpers.rs Normal file
View File

@@ -0,0 +1,105 @@
//! 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,
column_max_value: u64,
missing: &Key,
field_name: &str,
) -> crate::Result<Option<u64>> {
let missing_val = match missing {
Key::Str(_) if column_type == ColumnType::Str => Some(column_max_value + 1),
// Allow fallback to number on text fields
Key::F64(_) if column_type == ColumnType::Str => Some(column_max_value + 1),
Key::U64(_) if column_type == ColumnType::Str => Some(column_max_value + 1),
Key::I64(_) if column_type == ColumnType::Str => Some(column_max_value + 1),
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)
}

1095
src/aggregation/agg_data.rs Normal file
View File

File diff suppressed because it is too large Load Diff

3
src/aggregation/agg_limits.rs
View File

@@ -35,6 +35,7 @@ pub struct AggregationLimitsGuard {
/// Allocated memory with this guard.
allocated_with_the_guard: u64,
}
impl Clone for AggregationLimitsGuard {
fn clone(&self) -> Self {
Self {
@@ -70,7 +71,7 @@ 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)
}

31
src/aggregation/agg_result.rs
View File

@@ -16,7 +16,7 @@ use super::{AggregationError, Key};
use crate::TantivyError;
#[derive(Clone, Default, Debug, PartialEq, Serialize, Deserialize)]
/// The final aggegation result.
/// The final aggregation result.
pub struct AggregationResults(pub FxHashMap<String, AggregationResult>);
impl AggregationResults {
@@ -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,
}

45
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();
@@ -155,7 +152,7 @@ fn test_aggregation_flushing(
searcher.search(&AllQuery, &collector).unwrap()
};
let res: Value = serde_json::to_value(&agg_res)?;
let res: Value = serde_json::from_str(&serde_json::to_string(&agg_res)?)?;
assert_eq!(res["bucketsL1"]["buckets"][0]["doc_count"], 3);
assert_eq!(
@@ -270,7 +267,7 @@ fn test_aggregation_level1_simple() -> crate::Result<()> {
let searcher = reader.searcher();
let agg_res: AggregationResults = searcher.search(&term_query, &collector).unwrap();
let res: Value = serde_json::to_value(&agg_res)?;
let res: Value = serde_json::from_str(&serde_json::to_string(&agg_res)?)?;
assert_eq!(res["average"]["value"], 12.142857142857142);
assert_eq!(
res["range"]["buckets"],
@@ -304,29 +301,6 @@ fn test_aggregation_level1_simple() -> crate::Result<()> {
Ok(())
}
#[test]
fn test_aggregation_term_truncate_sum_other_doc_count() {
let index = get_test_index_2_segments(true).unwrap();
let reader = index.reader().unwrap();
let count_per_text: Aggregation = serde_json::from_value(json!({ "terms": { "field": "text", "size": 1 } })).unwrap();
let aggs: Aggregations = vec![("group_by_term_truncate".to_string(), count_per_text)]
.into_iter()
.collect();
let collector = get_collector(aggs);
let searcher = reader.searcher();
let agg_res: AggregationResults = searcher.search(&AllQuery, &collector).unwrap();
let res: Value = serde_json::to_value(&agg_res).unwrap();
assert_eq!(res, serde_json::json!({
"group_by_term_truncate": {
"buckets": [{ "doc_count": 7, "key": "cool" }],
"doc_count_error_upper_bound": 0,
"sum_other_doc_count": 2,
},
}));
}
#[test]
fn test_aggregation_level1() -> crate::Result<()> {
let index = get_test_index_2_segments(true)?;
@@ -365,7 +339,7 @@ fn test_aggregation_level1() -> crate::Result<()> {
let searcher = reader.searcher();
let agg_res: AggregationResults = searcher.search(&term_query, &collector).unwrap();
let res: Value = serde_json::to_value(&agg_res)?;
let res: Value = serde_json::from_str(&serde_json::to_string(&agg_res)?)?;
assert_eq!(res["average"]["value"], 12.142857142857142);
assert_eq!(res["average_f64"]["value"], 12.214285714285714);
assert_eq!(res["average_i64"]["value"], 12.142857142857142);
@@ -420,7 +394,7 @@ fn test_aggregation_level2(
IndexRecordOption::Basic,
);
let elasticsearch_compatible_json_req = serde_json::json!(
let elasticsearch_compatible_json_req = r#"
{
"rangef64": {
"range": {
@@ -473,8 +447,9 @@ fn test_aggregation_level2(
"term_agg": { "terms": { "field": "text" } }
}
}
});
let agg_req: Aggregations = serde_json::from_value(elasticsearch_compatible_json_req).unwrap();
}
"#;
let agg_req: Aggregations = serde_json::from_str(elasticsearch_compatible_json_req).unwrap();
let agg_res: AggregationResults = if use_distributed_collector {
let collector =
@@ -491,7 +466,7 @@ fn test_aggregation_level2(
searcher.search(&term_query, &collector).unwrap()
};
let res: Value = serde_json::to_value(agg_res)?;
let res: Value = serde_json::from_str(&serde_json::to_string(&agg_res)?)?;
assert_eq!(res["range"]["buckets"][1]["key"], "3-7");
assert_eq!(res["range"]["buckets"][1]["doc_count"], 2u64);

1754
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]

1084
src/aggregation/bucket/term_agg.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(())
}

28
src/aggregation/buf_collector.rs
View File

@@ -1,9 +1,14 @@
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;
#[cfg(test)]
pub(crate) const DOC_BLOCK_SIZE: usize = 64;
#[cfg(not(test))]
pub(crate) const DOC_BLOCK_SIZE: usize = 256;
pub(crate) type DocBlock = [DocId; DOC_BLOCK_SIZE];
/// BufAggregationCollector buffers documents before calling collect_block().
@@ -15,7 +20,7 @@ pub(crate) struct BufAggregationCollector {
}
impl std::fmt::Debug for BufAggregationCollector {
fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
fn fmt(&self, f: &mut std::fmt::Formatter) -> std::fmt::Result {
f.debug_struct("SegmentAggregationResultsCollector")
.field("staged_docs", &&self.staged_docs[..self.num_staged_docs])
.field("num_staged_docs", &self.num_staged_docs)
@@ -37,23 +42,23 @@ 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 +68,19 @@ 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()

37
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};
@@ -63,7 +62,7 @@ impl ExtendedStatsAggregation {
/// Extended stats contains a collection of statistics
/// they extends stats adding variance, standard deviation
/// and bound informations
/// and bound information
#[derive(Clone, Debug, PartialEq, Serialize, Deserialize)]
pub struct ExtendedStats {
/// The number of documents.
@@ -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(())
}
}

134
src/aggregation/metric/top_hits.rs
View File

@@ -1,7 +1,8 @@
use std::cmp::Ordering;
use std::collections::HashMap;
use std::net::Ipv6Addr;
use columnar::{Column, ColumnType, ColumnarReader, DynamicColumn};
use columnar::{Column, ColumnType, ColumnarReader, DynamicColumn, ValueRange};
use common::json_path_writer::JSON_PATH_SEGMENT_SEP_STR;
use common::DateTime;
use regex::Regex;
@@ -9,15 +10,41 @@ 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,
};
use crate::aggregation::segment_agg_result::SegmentAggregationCollector;
use crate::aggregation::AggregationError;
use crate::collector::sort_key::{Comparator, ReverseComparator};
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
///
@@ -357,7 +384,7 @@ impl From<FastFieldValue> for OwnedValue {
/// Holds a fast field value in its u64 representation, and the order in which it should be sorted.
#[derive(Clone, Serialize, Deserialize, Debug)]
struct DocValueAndOrder {
pub(crate) struct DocValueAndOrder {
/// A fast field value in its u64 representation.
value: Option<u64>,
/// Sort order for the value
@@ -429,11 +456,42 @@ impl PartialEq for DocSortValuesAndFields {
impl Eq for DocSortValuesAndFields {}
impl Comparator<DocSortValuesAndFields> for ReverseComparator {
#[inline(always)]
fn compare(&self, lhs: &DocSortValuesAndFields, rhs: &DocSortValuesAndFields) -> Ordering {
rhs.cmp(lhs)
}
fn threshold_to_valuerange(
&self,
threshold: DocSortValuesAndFields,
) -> ValueRange<DocSortValuesAndFields> {
ValueRange::LessThan(threshold, true)
}
}
#[derive(Clone, Debug, PartialEq, Eq, PartialOrd, Ord)]
pub(crate) struct TopHitsSegmentSortKey(pub Vec<DocValueAndOrder>);
impl Comparator<TopHitsSegmentSortKey> for ReverseComparator {
#[inline(always)]
fn compare(&self, lhs: &TopHitsSegmentSortKey, rhs: &TopHitsSegmentSortKey) -> Ordering {
rhs.cmp(lhs)
}
fn threshold_to_valuerange(
&self,
threshold: TopHitsSegmentSortKey,
) -> ValueRange<TopHitsSegmentSortKey> {
ValueRange::LessThan(threshold, true)
}
}
/// The TopHitsCollector used for collecting over segments and merging results.
#[derive(Clone, Serialize, Deserialize, Debug)]
pub struct TopHitsTopNComputer {
req: TopHitsAggregationReq,
top_n: TopNComputer<DocSortValuesAndFields, DocAddress, false>,
top_n: TopNComputer<DocSortValuesAndFields, DocAddress, ReverseComparator>,
}
impl std::cmp::PartialEq for TopHitsTopNComputer {
@@ -457,7 +515,7 @@ impl TopHitsTopNComputer {
pub(crate) fn merge_fruits(&mut self, other_fruit: Self) -> crate::Result<()> {
for doc in other_fruit.top_n.into_vec() {
self.collect(doc.feature, doc.doc);
self.collect(doc.sort_key, doc.doc);
}
Ok(())
}
@@ -469,9 +527,9 @@ impl TopHitsTopNComputer {
.into_sorted_vec()
.into_iter()
.map(|doc| TopHitsVecEntry {
sort: doc.feature.sorts.iter().map(|f| f.value).collect(),
sort: doc.sort_key.sorts.iter().map(|f| f.value).collect(),
doc_value_fields: doc
.feature
.sort_key
.doc_value_fields
.into_iter()
.map(|(k, v)| (k, v.into()))
@@ -492,7 +550,7 @@ impl TopHitsTopNComputer {
pub(crate) struct TopHitsSegmentCollector {
segment_ordinal: SegmentOrdinal,
accessor_idx: usize,
top_n: TopNComputer<Vec<DocValueAndOrder>, DocAddress, false>,
top_n: TopNComputer<TopHitsSegmentSortKey, DocAddress, ReverseComparator>,
}
impl TopHitsSegmentCollector {
@@ -513,13 +571,15 @@ impl TopHitsSegmentCollector {
req: &TopHitsAggregationReq,
) -> TopHitsTopNComputer {
let mut top_hits_computer = TopHitsTopNComputer::new(req);
// Map TopHitsSegmentSortKey back to Vec<DocValueAndOrder> if needed or use directly
// The TopNComputer here stores TopHitsSegmentSortKey.
let top_results = self.top_n.into_vec();
for res in top_results {
let doc_value_fields = req.get_document_field_data(value_accessors, res.doc.doc_id);
top_hits_computer.collect(
DocSortValuesAndFields {
sorts: res.feature,
sorts: res.sort_key.0,
doc_value_fields,
},
res.doc,
@@ -553,7 +613,7 @@ impl TopHitsSegmentCollector {
.collect();
self.top_n.push(
sorts,
TopHitsSegmentSortKey(sorts),
DocAddress {
segment_ord: self.segment_ordinal,
doc_id,
@@ -566,23 +626,18 @@ 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),
)
}
@@ -591,32 +646,22 @@ 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(())
}
@@ -635,6 +680,7 @@ mod tests {
use crate::aggregation::bucket::tests::get_test_index_from_docs;
use crate::aggregation::tests::get_test_index_from_values;
use crate::aggregation::AggregationCollector;
use crate::collector::sort_key::ReverseComparator;
use crate::collector::ComparableDoc;
use crate::query::AllQuery;
use crate::schema::OwnedValue;
@@ -650,7 +696,7 @@ mod tests {
fn collector_with_capacity(capacity: usize) -> super::TopHitsTopNComputer {
super::TopHitsTopNComputer {
top_n: super::TopNComputer::new(capacity),
top_n: super::TopNComputer::new_with_comparator(capacity, ReverseComparator),
req: Default::default(),
}
}
@@ -764,12 +810,12 @@ mod tests {
#[test]
fn test_top_hits_collector_single_feature() -> crate::Result<()> {
let docs = vec![
ComparableDoc::<_, _, false> {
ComparableDoc::<_, _> {
doc: crate::DocAddress {
segment_ord: 0,
doc_id: 0,
},
feature: DocSortValuesAndFields {
sort_key: DocSortValuesAndFields {
sorts: vec![DocValueAndOrder {
value: Some(1),
order: Order::Asc,
@@ -782,7 +828,7 @@ mod tests {
segment_ord: 0,
doc_id: 2,
},
feature: DocSortValuesAndFields {
sort_key: DocSortValuesAndFields {
sorts: vec![DocValueAndOrder {
value: Some(3),
order: Order::Asc,
@@ -795,7 +841,7 @@ mod tests {
segment_ord: 0,
doc_id: 1,
},
feature: DocSortValuesAndFields {
sort_key: DocSortValuesAndFields {
sorts: vec![DocValueAndOrder {
value: Some(5),
order: Order::Asc,
@@ -807,7 +853,7 @@ mod tests {
let mut collector = collector_with_capacity(3);
for doc in docs.clone() {
collector.collect(doc.feature, doc.doc);
collector.collect(doc.sort_key, doc.doc);
}
let res = collector.into_final_result();
@@ -817,15 +863,15 @@ mod tests {
super::TopHitsMetricResult {
hits: vec![
super::TopHitsVecEntry {
sort: vec![docs[0].feature.sorts[0].value],
sort: vec![docs[0].sort_key.sorts[0].value],
doc_value_fields: Default::default(),
},
super::TopHitsVecEntry {
sort: vec![docs[1].feature.sorts[0].value],
sort: vec![docs[1].sort_key.sorts[0].value],
doc_value_fields: Default::default(),
},
super::TopHitsVecEntry {
sort: vec![docs[2].feature.sorts[0].value],
sort: vec![docs[2].sort_key.sorts[0].value],
doc_value_fields: Default::default(),
},
]

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();

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