rust/tantivy
1
0
Fork 0
mirror of https://github.com/quickwit-oss/tantivy.git synced 2025-12-28 13:02:55 +00:00
Code Issues Packages Projects Releases Wiki Activity

Compare commits

base: rust:remove_dyn
Branches Tags
rust:main rust:stuhood.upstream-erased-sort-computer rust:trinity.pointard/ff-query rust:bucket_id_agg rust:paradedb/fix-agg-validation rust:flatheadmill-geo rust:paradedb-paradedb/fix-overflow-issue rust:low_card_optimization rust:paul.masurel/clippy rust:fix_size_hint 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:0.22.0
Branches Tags
rust:stuhood.upstream-erased-sort-computer rust:main rust:trinity.pointard/ff-query rust:bucket_id_agg rust:paradedb/fix-agg-validation rust:flatheadmill-geo rust:paradedb-paradedb/fix-overflow-issue rust:low_card_optimization rust:paul.masurel/clippy rust:fix_size_hint 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

269 Commits
remove_dyn ... 0.22.0

Author SHA1 Message Date
PSeitz
17d5869ad6 update CHANGELOG, use github API in cliff (#2354) 
* update CHANGELOG, use github API in cliff

* reset version to 0.21.1, before release

* chore: Release

* remove unreleased from CHANGELOG
 2024-04-15 10:07:20 +02:00
PSeitz
dfa3aed32d check unsupported parameters top_hits (#2351) 
* check unsupported parameters top_hits

* move to function
 2024-04-10 08:20:52 +02:00
PSeitz
398817ce7b add index sorting deprecation warning (#2353) 
* add index sorting deprecation warning

* remove deprecated IntOptions and DatePrecision
 2024-04-10 08:09:09 +02:00
PSeitz
74940e9345 clippy (#2349) 
* fix clippy

* fix clippy

* fix duplicate imports
 2024-04-09 07:54:44 +02:00
PSeitz
1e9fc51535 update ahash (#2344) 2024-04-09 06:35:39 +02:00
PSeitz
92c32979d2 fix postcard compatibility for top_hits, add postcard test (#2346) 
* fix postcard compatibility for top_hits, add postcard test

* fix top_hits naming, delay data fetch

closes #2347

* fix import
 2024-04-09 06:17:25 +02:00
PSeitz
b644d78a32 fix null byte handling in JSON paths (#2345) 
* fix null byte handling in JSON paths

closes https://github.com/quickwit-oss/tantivy/issues/2193
closes https://github.com/quickwit-oss/tantivy/issues/2340

* avoid repeated term truncation

* fix test

* Apply suggestions from code review

Co-authored-by: Paul Masurel <paul@quickwit.io>

* add comment

---------

Co-authored-by: Paul Masurel <paul@quickwit.io>
 2024-04-05 09:53:35 +02:00
PSeitz
4e79e11007 add collect_block to BoxableSegmentCollector (#2331) 2024-03-21 09:10:25 +01:00
PSeitz
67ebba3c3c expose collect_block buffer size (#2326) 
* expose buffer of collect_block

* flip shard_size segment_size
 2024-03-15 08:02:08 +01:00
PSeitz
7ce950f141 add method to fetch block of first vals in columnar (#2330) 
* add method to fetch block of first vals in columnar

add method to fetch block of first vals in columnar (this is way faster
than single calls for full columns)
add benchmark
fix import warnings

```
test bench_get_block_first_on_full_column                  ... bench:          56 ns/iter (+/- 26)
test bench_get_block_first_on_full_column_single_calls     ... bench:         311 ns/iter (+/- 6)
test bench_get_block_first_on_multi_column                 ... bench:         378 ns/iter (+/- 15)
test bench_get_block_first_on_multi_column_single_calls    ... bench:         546 ns/iter (+/- 13)
test bench_get_block_first_on_optional_column              ... bench:         291 ns/iter (+/- 6)
test bench_get_block_first_on_optional_column_single_calls ... bench:         362 ns/iter (+/- 8)
```

* use remainder
 2024-03-15 08:01:47 +01:00
dependabot[bot]
0cffe5fb09 Update base64 requirement from 0.21.0 to 0.22.0 (#2324) 
Updates the requirements on [base64](https://github.com/marshallpierce/rust-base64) to permit the latest version.
- [Changelog](https://github.com/marshallpierce/rust-base64/blob/master/RELEASE-NOTES.md)
- [Commits](https://github.com/marshallpierce/rust-base64/compare/v0.21.0...v0.22.0)

---
updated-dependencies:
- dependency-name: base64
  dependency-type: direct:production
...

Signed-off-by: dependabot[bot] <support@github.com>
Co-authored-by: dependabot[bot] <49699333+dependabot[bot]@users.noreply.github.com>
 2024-03-15 15:50:34 +09:00
PSeitz
b0e65560a1 handle ip adresses in term aggregation (#2319) 
* handle ip adresses in term aggregation

Stores IpAdresses during the segment term aggregation via u64 representation
and convert to u128(IpV6Adress) via downcast when converting to intermediate results.

Enable Downcasting on `ColumnValues`
Expose u64 variant for u128 encoded data via `open_u64_lenient` method.
Remove lifetime in VecColumn, to avoid 'static lifetime requirement coming
from downcast trait.

* rename method
 2024-03-14 09:41:18 +01:00
PSeitz
ec37295b2f add fast path for full columns in fetch_block (#2328) 
Spotted in `range_date_histogram` query in quickwit benchmark:
5% of time copying docs around, which is not needed in the full index case

remove Column to ColumnIndex deref
 2024-03-14 04:07:11 +01:00
trinity-1686a
f6b0cc1aab allow some mixing of occur and bool in strict query parser (#2323) 
* allow some mixing of occur and bool in strict query parser

* allow all mixing of binary and occur in strict parser
 2024-03-07 15:17:48 +01:00
PSeitz
7e41d31c6e agg: support to deserialize f64 from string (#2311) 
* agg: support to deserialize f64 from string

* remove visit_string

* disallow NaN
 2024-03-05 05:49:41 +01:00
Adam Reichold
40aa4abfe5 Make FacetCounts defaultable and cloneable. (#2322) 2024-03-05 04:11:11 +01:00
dependabot[bot]
2650317622 Update fs4 requirement from 0.7.0 to 0.8.0 (#2321) 
Updates the requirements on [fs4](https://github.com/al8n/fs4-rs) to permit the latest version.
- [Release notes](https://github.com/al8n/fs4-rs/releases)
- [Commits](https://github.com/al8n/fs4-rs/commits)

---
updated-dependencies:
- dependency-name: fs4
  dependency-type: direct:production
...

Signed-off-by: dependabot[bot] <support@github.com>
Co-authored-by: dependabot[bot] <49699333+dependabot[bot]@users.noreply.github.com>
 2024-02-27 03:38:04 +01:00
Paul Masurel
6739357314 Removing split_size and adding split_size and shard_size as segmnet_size (#2320) 
aliases.
 2024-02-26 11:35:22 +01:00
PSeitz
d57622d54b support bool type in term aggregation (#2318) 
* support bool type in term aggregation

* add Bool to Intermediate Key
 2024-02-20 03:22:22 +01:00
PSeitz
f745dbc054 fix Clone for TopNComputer, add top_hits bench (#2315) 
* fix Clone for TopNComputer, add top_hits bench

add top_hits agg bench

test aggregation::agg_bench::bench::bench_aggregation_terms_many_with_sub_agg                                            ... bench: 123,475,175 ns/iter (+/- 30,608,889)
test aggregation::agg_bench::bench::bench_aggregation_terms_many_with_sub_agg_multi                                      ... bench: 194,170,414 ns/iter (+/- 36,495,516)
test aggregation::agg_bench::bench::bench_aggregation_terms_many_with_sub_agg_opt                                        ... bench: 179,742,809 ns/iter (+/- 29,976,507)
test aggregation::agg_bench::bench::bench_aggregation_terms_many_with_sub_agg_sparse                                     ... bench:  27,592,534 ns/iter (+/- 2,672,370)
test aggregation::agg_bench::bench::bench_aggregation_terms_many_with_top_hits_agg                                       ... bench: 552,851,227 ns/iter (+/- 71,975,886)
test aggregation::agg_bench::bench::bench_aggregation_terms_many_with_top_hits_agg_multi                                 ... bench: 558,616,384 ns/iter (+/- 100,890,124)
test aggregation::agg_bench::bench::bench_aggregation_terms_many_with_top_hits_agg_opt                                   ... bench: 554,031,368 ns/iter (+/- 165,452,650)
test aggregation::agg_bench::bench::bench_aggregation_terms_many_with_top_hits_agg_sparse                                ... bench:  46,435,919 ns/iter (+/- 13,681,935)

* add comment
 2024-02-20 03:22:00 +01:00
PSeitz
79b041f81f clippy (#2314) 2024-02-13 05:56:31 +01:00
PSeitz
0e16ed9ef7 Fix serde for TopNComputer (#2313) 
* Fix serde for TopNComputer

The top hits aggregation changed the TopNComputer to be serializable,
but capacity needs to be carried over, as it contains logic which is
checked against when pushing elements (capacity == 0 is not allowed).

* use serde from deser

* remove pub, clippy
 2024-02-07 12:52:06 +01:00
mochi
88a3275dbb add shared search executor (#2312) 2024-02-05 09:33:00 +01:00
PSeitz
1223a87eb2 add fuzz test for hashmap (#2310) 2024-01-31 10:30:21 +01:00
PSeitz
48630ceec9 move into new index module (#2259) 
move core modules to index module
 2024-01-31 10:30:04 +01:00
Adam Reichold
72002e8a89 Make test builds Clippy clean. (#2277) 2024-01-31 02:47:06 +01:00
trinity-1686a
3c9297dd64 report if posting list was actually loaded when warming it up (#2309) 2024-01-29 15:23:16 +01:00
Tushar
0e04ec3136 feat(aggregators/metric): Add a top_hits aggregator (#2198) 
* feat(aggregators/metric): Implement a top_hits aggregator

* fix: Expose get_fields

* fix: Serializer for top_hits request

Also removes extraneous the extraneous third-party
serialization helper.

* chore: Avert panick on parsing invalid top_hits query

* refactor: Allow multiple field names from aggregations

* perf: Replace binary heap with TopNComputer

* fix: Avoid comparator inversion by ComparableDoc

* fix: Rank missing field values lower than present values

* refactor: Make KeyOrder a struct

* feat: Rough attempt at docvalue_fields

* feat: Complete stab at docvalue_fields

- Rename "SearchResult*" => "Retrieval*"
- Revert Vec => HashMap for aggregation accessors.
- Split accessors for core aggregation and field retrieval.
- Resolve globbed field names in docvalue_fields retrieval.
- Handle strings/bytes and other column types with DynamicColumn

* test(unit): Add tests for top_hits aggregator

* fix: docfield_value field globbing

* test(unit): Include dynamic fields

* fix: Value -> OwnedValue

* fix: Use OwnedValue's native Null variant

* chore: Improve readability of test asserts

* chore: Remove DocAddress from top_hits result

* docs: Update aggregator doc

* revert: accidental doc test

* chore: enable time macros only for tests

* chore: Apply suggestions from review

* chore: Apply suggestions from review

* fix: Retrieve all values for fields

* test(unit): Update for multi-value retrieval

* chore: Assert term existence

* feat: Include all columns for a column name

Since a (name, type) constitutes a unique column.

* fix: Resolve json fields

Introduces a translation step to bridge the difference between
ColumnarReaders null `\0` separated json field keys to the common
`.` separated used by SegmentReader. Although, this should probably
be the default behavior for ColumnarReader's public API perhaps.

* chore: Address review on mutability

* chore: s/segment_id/segment_ordinal instances of SegmentOrdinal

* chore: Revert erroneous grammar change
 2024-01-26 16:46:41 +01:00
Paul Masurel
9b7f3a55cf Bumped census version 2024-01-26 19:32:02 +09:00
PSeitz
1dacdb6c85 add histogram agg test on empty index (#2306) 2024-01-23 16:27:34 +01:00
François Massot
30483310ca Minor improvement of README.md (#2305) 
* Update README.md

* Remove useless paragraph

* Wording.
 2024-01-19 17:46:48 +09:00
Tushar
e1d18b5114 chore: Expose TopDocs::order_by_u64_field again (#2282) 2024-01-18 05:58:24 +01:00
trinity-1686a
108f30ba23 allow newline where we allow space in query parser (#2302) 
fix regression from the new parser
 2024-01-17 14:38:35 +01:00
PSeitz
5943ee46bd Truncate keys to u16::MAX in term hashmap (#2299) 
Truncate keys to u16::MAX, instead e.g. storing 0 bytes for keys with length u16::MAX + 1

The term hashmap has a hidden API contract to only accept terms with lenght up u16::MAX.
 2024-01-11 10:19:12 +01:00
PSeitz
f95a76293f add memory arena test (#2298) 
* add memory arena test

* add assert

* Update stacker/src/memory_arena.rs

Co-authored-by: Paul Masurel <paul@quickwit.io>

---------

Co-authored-by: Paul Masurel <paul@quickwit.io>
 2024-01-11 07:18:48 +01:00
Paul Masurel
014328e378 Fix bug that can cause get_docids_for_value_range to panic. (#2295) 
* Fix bug that can cause `get_docids_for_value_range` to panic.

When `selected_docid_range.end == num_rows`, we would get a panic
as we try to access a non-existing blockmeta.

This PR accepts calls to rank with any value.
For any value above num_rows we simply return non_null_rows.

Fixes #2293

* add tests, merge variables

---------

Co-authored-by: Pascal Seitz <pascal.seitz@gmail.com>
 2024-01-09 14:52:20 +01:00
Adam Reichold
53f2fe1fbe Forward regex parser errors to enable understandin their reason. (#2288) 2023-12-22 11:01:10 +01:00
PSeitz
9c75942aaf fix merge panic for JSON fields (#2284) 
Root cause was the positions buffer had residue positions from the
previous term, when the terms were alternating between having and not
having positions in JSON (terms have positions, but not numerics).

Fixes #2283
 2023-12-21 11:05:34 +01:00
PSeitz
bff7c58497 improve indexing benchmark (#2275) 2023-12-11 09:04:42 +01:00
trinity-1686a
9ebc5ed053 use fst for sstable index (#2268) 
* read path for new fst based index

* implement BlockAddrStoreWriter

* extract slop/derivation computation

* use better linear approximator and allow negative correction to approximator

* document format and reorder some fields

* optimize single block sstable size

* plug backward compat
 2023-12-04 15:13:15 +01:00
PSeitz
0b56c88e69 Revert "Preparing for 0.21.2 release." (#2258) 
* Revert "Preparing for 0.21.2 release. (#2256)"

This reverts commit 9caab45136.

* bump version to 0.21.1

* set version to 0.22.0-dev
 2023-12-01 13:46:12 +01:00
PSeitz
24841f0b2a update bitpacker dep (#2269) 2023-12-01 13:45:52 +01:00
PSeitz
1a9fc10be9 add fields_metadata to SegmentReader, add columnar docs (#2222) 
* add fields_metadata to SegmentReader, add columnar docs

* use schema to resolve field, add test

* normalize paths

* merge for FieldsMetadata, add fields_metadata on Index

* Update src/core/segment_reader.rs

Co-authored-by: Paul Masurel <paul@quickwit.io>

* merge code paths

* add Hash

* move function oustide

---------

Co-authored-by: Paul Masurel <paul@quickwit.io>
 2023-11-22 12:29:53 +01:00
PSeitz
07573a7f19 update fst (#2267) 
update fst to 0.5 (deduplicates regex-syntax in the dep tree)
deps cleanup
 2023-11-21 16:06:57 +01:00
BlackHoleFox
daad2dc151 Take string references instead of owned values building Facet paths (#2265) 2023-11-20 09:40:44 +01:00
PSeitz
054f49dc31 support escaped dot, add agg test (#2250) 
add agg test for nested JSON
allow escaping of dot
 2023-11-20 03:00:57 +01:00
PSeitz
47009ed2d3 remove unused deps (#2264) 
found with cargo machete
remove pprof (doesn't work)
 2023-11-20 02:59:59 +01:00
PSeitz
0aae31d7d7 reduce number of allocations (#2257) 
* reduce number of allocations

Explanation makes up around 50% of all allocations (numbers not perf).
It's created during serialization but not called.

- Make Explanation optional in BM25
- Avoid allocations when using Explanation

* use Cow
 2023-11-16 13:47:36 +01:00
Paul Masurel
9caab45136 Preparing for 0.21.2 release. (#2256) 2023-11-15 10:43:36 +09:00
Chris Tam
6d9a7b7eb0 Derive Debug for SchemaBuilder (#2254) 2023-11-15 01:03:44 +01:00
dependabot[bot]
7a2c5804b1 Update itertools requirement from 0.11.0 to 0.12.0 (#2255) 
Updates the requirements on [itertools](https://github.com/rust-itertools/itertools) to permit the latest version.
- [Changelog](https://github.com/rust-itertools/itertools/blob/master/CHANGELOG.md)
- [Commits](https://github.com/rust-itertools/itertools/compare/v0.11.0...v0.12.0)

---
updated-dependencies:
- dependency-name: itertools
  dependency-type: direct:production
...

Signed-off-by: dependabot[bot] <support@github.com>
Co-authored-by: dependabot[bot] <49699333+dependabot[bot]@users.noreply.github.com>
 2023-11-15 01:03:08 +01:00
François Massot
5319977171 Merge pull request #2253 from quickwit-oss/issue/2251-bug-merge-json-object-with-number 
Fix bug occuring when merging JSON object indexed with positions.
 2023-11-14 17:28:29 +01:00
trinity-1686a
828632e8c4 rustfmt 2023-11-14 15:05:16 +01:00
Paul Masurel
6b59ec6fd5 Fix bug occuring when merging JSON object indexed with positions. 
In JSON Object field the presence of term frequencies depend on the
field.
Typically, a string with postiions indexed will have positions
while numbers won't.

The presence or absence of term freqs for a given term is unfortunately
encoded in a very passive way.

It is given by the presence of extra information in the skip info, or
the lack of term freqs after decoding vint blocks.

Before, after writing a segment, we would encode the segment correctly
(without any term freq for number in json object field).
However during merge, we would get the default term freq=1 value.
(this is default in the absence of encoded term freqs)

The merger would then proceed and attempt to decode 1 position when
there are in fact none.

This PR requires to explictly tell the posting serialize whether
term frequencies should be serialized for each new term.

Closes #2251
 2023-11-14 22:41:48 +09:00
PSeitz
b60d862150 docid deltas while indexing (#2249) 
* docid deltas while indexing

storing deltas is especially helpful for repetitive data like logs.
In those cases, recording a doc on a term costed 4 bytes instead of 1
byte now.

HDFS Indexing 1.1GB Total memory consumption:
Before:  760 MB
Now:     590 MB

* use scan for delta decoding
 2023-11-13 05:14:27 +01:00
PSeitz
4837c7811a add missing inlines (#2245) 2023-11-10 08:00:42 +01:00
PSeitz
5a2397d57e add sstable ord_to_term benchmark (#2242) 2023-11-10 07:27:48 +01:00
PSeitz
927b4432c9 Perf: use term hashmap in fastfield (#2243) 
* add shared arena hashmap

* bench fastfield indexing

* use shared arena hashmap in columnar

lower minimum resize in hashtable

* clippy

* add comments
 2023-11-09 13:44:02 +01:00
trinity-1686a
7a0064db1f bump index version (#2237) 
* bump index version

and add constant for lowest supported version

* use range instead of handcoded bounds
 2023-11-06 19:02:37 +01:00
PSeitz
2e7327205d fix coverage run (#2232) 
coverage run uses the compare_hash_only feature which is not compativle
with the test_hashmap_size test
 2023-11-06 11:18:38 +00:00
Paul Masurel
7bc5bf78e2 Fixing functional tests. (#2239) 2023-11-05 18:18:39 +09:00
giovannicuccu
ef603c8c7e rename ReloadPolicy onCommit to onCommitWithDelay (#2235) 
* rename ReloadPolicy onCommit to onCommitWithDelay

* fix format issues

---------

Co-authored-by: Giovanni Cuccu <gcuccu@imolainformatica.it>
 2023-11-03 12:22:10 +01:00
PSeitz
28dd6b6546 collect json paths in indexing (#2231) 
* collect json paths in indexing

* remove unsafe iter_mut_keys
 2023-11-01 11:25:17 +01:00
trinity-1686a
1dda2bb537 handle * inside term in query parser (#2228) 2023-10-27 08:57:02 +02:00
PSeitz
bf6544cf28 fix mmap::Advice reexport (#2230) 2023-10-27 14:09:25 +09:00
PSeitz
ccecf946f7 tantivy 0.21.1 (#2227) 2023-10-27 05:01:44 +02:00
PSeitz
19a859d6fd term hashmap remove copy in is_empty, unused unordered_id (#2229) 2023-10-27 05:01:32 +02:00
PSeitz
83af14caa4 Fix range query (#2226) 
Fix range query end check in advance
Rename vars to reduce ambiguity
add tests

Fixes #2225
 2023-10-25 09:17:31 +02:00
PSeitz
4feeb2323d fix clippy (#2223) 2023-10-24 10:05:22 +02:00
PSeitz
07bf66a197 json path writer (#2224) 
* refactor logic to JsonPathWriter

* use in encode_column_name

* add inlines

* move unsafe block
 2023-10-24 09:45:50 +02:00
trinity-1686a
0d4589219b encode some part of posting list as -1 instead of direct values (#2185) 
* add support for delta-1 encoding posting list

* encode term frequency minus one

* don't emit tf for json integer terms

* make skipreader not pub(crate) mutable
 2023-10-20 16:58:26 +02:00
PSeitz
c2b0469180 improve docs, rework exports (#2220) 
* rework exports

move snippet and advice
make indexer pub, remove indexer reexports

* add deprecation warning

* add architecture overview
 2023-10-18 09:22:24 +02:00
PSeitz
7e1980b218 run coverage only after merge (#2212) 
* run coverage only after merge

coverage is a quite slow step in CI. It can be run only after merging

* Apply suggestions from code review

Co-authored-by: Paul Masurel <paul@quickwit.io>

---------

Co-authored-by: Paul Masurel <paul@quickwit.io>
 2023-10-18 07:19:36 +02:00
PSeitz
ecb9a89a9f add compat mode for JSON (#2219) 2023-10-17 10:00:55 +02:00
PSeitz
5e06e504e6 split into ReferenceValueLeaf (#2217) 2023-10-16 16:31:30 +02:00
PSeitz
182f58cea6 remove Document: DocumentDeserialize dependency (#2211) 
* remove Document: DocumentDeserialize dependency

The dependency requires users to implement an API they may not use.

* remove unnecessary Document bounds
 2023-10-13 07:59:54 +02:00
dependabot[bot]
337ffadefd Update lru requirement from 0.11.0 to 0.12.0 (#2208) 
Updates the requirements on [lru](https://github.com/jeromefroe/lru-rs) to permit the latest version.
- [Changelog](https://github.com/jeromefroe/lru-rs/blob/master/CHANGELOG.md)
- [Commits](https://github.com/jeromefroe/lru-rs/compare/0.11.0...0.12.0)

---
updated-dependencies:
- dependency-name: lru
  dependency-type: direct:production
...

Signed-off-by: dependabot[bot] <support@github.com>
Co-authored-by: dependabot[bot] <49699333+dependabot[bot]@users.noreply.github.com>
 2023-10-12 12:09:56 +02:00
dependabot[bot]
22aa4daf19 Update zstd requirement from 0.12 to 0.13 (#2214) 
Updates the requirements on [zstd](https://github.com/gyscos/zstd-rs) to permit the latest version.
- [Release notes](https://github.com/gyscos/zstd-rs/releases)
- [Commits](https://github.com/gyscos/zstd-rs/compare/v0.12.0...v0.13.0)

---
updated-dependencies:
- dependency-name: zstd
  dependency-type: direct:production
...

Signed-off-by: dependabot[bot] <support@github.com>
Co-authored-by: dependabot[bot] <49699333+dependabot[bot]@users.noreply.github.com>
 2023-10-12 04:24:44 +02:00
PSeitz
493f9b2f2a Read list of JSON fields encoded in dictionary (#2184) 
* Read list of JSON fields encoded in dictionary

add method to get list of fields on InvertedIndexReader

* add field type
 2023-10-09 12:06:22 +02:00
PSeitz
e246e5765d replace ReferenceValue with Self in Value (#2210) 2023-10-06 08:22:15 +02:00
PSeitz
6097235eff fix numeric order, refactor Document (#2209) 
fix numeric order to prefer i64
rename and move Document stuff
 2023-10-05 16:39:56 +02:00
PSeitz
b700c42246 add AsRef, expose object and array iter on Value (#2207) 
add AsRef
expose object and array iter
add to_json on Document
 2023-10-05 03:55:35 +02:00
PSeitz
5b1bf1a993 replace Field with field name (#2196) 2023-10-04 06:21:40 +02:00
PSeitz
041d4fced7 move to_named_doc to Document trait (#2205) 2023-10-04 06:03:07 +02:00
dependabot[bot]
166fc15239 Update memmap2 requirement from 0.7.1 to 0.9.0 (#2204) 
Updates the requirements on [memmap2](https://github.com/RazrFalcon/memmap2-rs) to permit the latest version.
- [Changelog](https://github.com/RazrFalcon/memmap2-rs/blob/master/CHANGELOG.md)
- [Commits](https://github.com/RazrFalcon/memmap2-rs/compare/v0.7.1...v0.9.0)

---
updated-dependencies:
- dependency-name: memmap2
  dependency-type: direct:production
...

Signed-off-by: dependabot[bot] <support@github.com>
Co-authored-by: dependabot[bot] <49699333+dependabot[bot]@users.noreply.github.com>
 2023-10-04 05:00:46 +02:00
PSeitz
514a6e7fef fix bench compile, fix Document reexport (#2203) 2023-10-03 17:28:36 +02:00
dependabot[bot]
82d9127191 Update fs4 requirement from 0.6.3 to 0.7.0 (#2199) 
Updates the requirements on [fs4](https://github.com/al8n/fs4-rs) to permit the latest version.
- [Release notes](https://github.com/al8n/fs4-rs/releases)
- [Commits](https://github.com/al8n/fs4-rs/commits/0.7.0)

---
updated-dependencies:
- dependency-name: fs4
  dependency-type: direct:production
...

Signed-off-by: dependabot[bot] <support@github.com>
Co-authored-by: dependabot[bot] <49699333+dependabot[bot]@users.noreply.github.com>
 2023-10-03 04:43:09 +02:00
PSeitz
03a1f40767 rename DocValue to Value (#2197) 
rename DocValue to Value to avoid confusion with lucene DocValues
rename Value to OwnedValue
 2023-10-02 17:03:00 +02:00
Harrison Burt
1c7c6fd591 POC: Tantivy documents as a trait (#2071) 
* fix windows build (#1)

* Fix windows build

* Add doc traits

* Add field value iter

* Add value and serialization

* Adjust order

* Fix bug

* Correct type

* Fix generic bugs

* Reformat code

* Add generic to index writer which I forgot about

* Fix missing generics on single segment writer

* Add missing type export

* Add default methods for convenience

* Cleanup

* Fix more-like-this query to use standard types

* Update API and fix tests

* Add doc traits

* Add field value iter

* Add value and serialization

* Adjust order

* Fix bug

* Correct type

* Rebase main and fix conflicts

* Reformat code

* Merge upstream

* Fix missing generics on single segment writer

* Add missing type export

* Add default methods for convenience

* Cleanup

* Fix more-like-this query to use standard types

* Update API and fix tests

* Add tokenizer improvements from previous commits

* Add tokenizer improvements from previous commits

* Reformat

* Fix unit tests

* Fix unit tests

* Use enum in changes

* Stage changes

* Add new deserializer logic

* Add serializer integration

* Add document deserializer

* Implement new (de)serialization api for existing types

* Fix bugs and type errors

* Add helper implementations

* Fix errors

* Reformat code

* Add unit tests and some code organisation for serialization

* Add unit tests to deserializer

* Add some small docs

* Add support for deserializing serde values

* Reformat

* Fix typo

* Fix typo

* Change repr of facet

* Remove unused trait methods

* Add child value type

* Resolve comments

* Fix build

* Fix more build errors

* Fix more build errors

* Fix the tests I missed

* Fix examples

* fix numerical order, serialize PreTok Str

* fix coverage

* rename Document to TantivyDocument, rename DocumentAccess to Document

add Binary prefix to binary de/serialization

* fix coverage

---------

Co-authored-by: Pascal Seitz <pascal.seitz@gmail.com>
 2023-10-02 10:01:16 +02:00
PSeitz
b525f653c0 replace BinaryHeap for TopN (#2186) 
* replace BinaryHeap for TopN

replace BinaryHeap for TopN with variant that selects the median with QuickSort,
which runs in O(n) time.

add merge_fruits fast path

* call truncate unconditionally, extend test

* remove special early exit

* add TODO, fmt

* truncate top n instead median, return vec

* simplify code
 2023-09-27 09:25:30 +02:00
ethever.eth
90586bc1e2 chore: remove unused Seek impl for Writers (#2187) (#2189) 
Co-authored-by: famouscat <onismaa@gmail.com>
 2023-09-26 17:03:28 +09:00
PSeitz
832f1633de handle exclusive out of bounds ranges on fastfield range queries (#2174) 
closes https://github.com/quickwit-oss/quickwit/issues/3790
 2023-09-26 08:00:40 +02:00
PSeitz
38db53c465 make column_index pub (#2181) 2023-09-22 08:06:45 +02:00
PSeitz
34920d31f5 Fix DateHistogram bucket gap (#2183) 
* Fix DateHistogram bucket gap

Fixes a computation issue of the number of buckets needed in the
DateHistogram.

This is due to a missing normalization from request values (ms) to fast field
values (ns), when converting an intermediate result to the final result.
This results in a wrong computation by a factor 1_000_000.
The Histogram normalizes values to nanoseconds, to make the user input like
extended_bounds (ms precision) and the values from the fast field (ns precision for date type) compatible.
This normalization happens only for date type fields, as other field types don't have precision settings.
The normalization does not happen due a missing `column_type`, which is not
correctly passed after merging an empty aggregation (which does not have a `column_type` set), with a regular aggregation.

Another related issue is an empty aggregation, which will not have
`column_type` set, will not convert the result to human readable format.

This PR fixes the issue by:
- Limit the allowed field types of DateHistogram to DateType
- Instead of passing the column_type, which is only available on the segment level, we flag the aggregation as `is_date_agg`.
- Fix the merge logic

Add a flag to to normalization only once. This is not an issue
currently, but it could become easily one.

closes https://github.com/quickwit-oss/quickwit/issues/3837

* use older nightly for time crate (breaks build)
 2023-09-21 10:41:35 +02:00
trinity-1686a
0241a05b90 add support for exists query syntax in query parser (#2170) 
* add support for exists query syntax in query parser

* rustfmt

* make Exists require a field
 2023-09-19 11:10:39 +02:00
PSeitz
e125f3b041 fix test (#2178) 2023-09-19 08:21:50 +02:00
PSeitz
c520ac46fc add support for date in term agg (#2172) 
support DateTime in TermsAggregation
Format dates with Rfc3339
 2023-09-14 09:22:18 +02:00
PSeitz
2d7390341c increase min memory to 15MB for indexing (#2176) 
With tantivy 0.20 the minimum memory consumption per SegmentWriter increased to
12MB. 7MB are for the different fast field collectors types (they could be
lazily created). Increase the minimum memory from 3MB to 15MB.

Change memory variable naming from arena to budget.

closes #2156
 2023-09-13 07:38:34 +02:00
dependabot[bot]
03fcdce016 Bump actions/checkout from 3 to 4 (#2171) 
Bumps [actions/checkout](https://github.com/actions/checkout) from 3 to 4.
- [Release notes](https://github.com/actions/checkout/releases)
- [Changelog](https://github.com/actions/checkout/blob/main/CHANGELOG.md)
- [Commits](https://github.com/actions/checkout/compare/v3...v4)

---
updated-dependencies:
- dependency-name: actions/checkout
  dependency-type: direct:production
  update-type: version-update:semver-major
...

Signed-off-by: dependabot[bot] <support@github.com>
Co-authored-by: dependabot[bot] <49699333+dependabot[bot]@users.noreply.github.com>
 2023-09-11 10:47:33 +02:00
Ping Xia
e4e416ac42 extend FuzzyTermQuery to support json field (#2173) 
* extend fuzzy search for json field

* comments

* comments

* fmt fix

* comments
 2023-09-11 05:59:40 +02:00
Igor Motov
19325132b7 Fast-field based implementation of ExistsQuery (#2160) 
Adds an implementation of ExistsQuery that takes advantage of fast fields.

Fixes #2159
 2023-09-07 11:51:49 +09:00
Paul Masurel
389d36f760 Added comments 2023-09-04 11:06:56 +09:00
PSeitz
49448b31c6 chore: Release (#2168) 
* chore: Release

* update CHANGELOG
 2023-09-01 13:58:58 +02:00
PSeitz
ebede0bed7 update CHANGELOG (#2167) 2023-08-31 10:01:44 +02:00
PSeitz
b1d8b072db add missing aggregation part 2 (#2149) 
* add missing aggregation part 2

Add missing support for:
- Mixed types columns
- Key of type string on numerical fields

The special aggregation is slower than the integrated one in TermsAggregation and therefore not
chosen by default, although it can cover all use cases.

* simplify, add num_docs to empty
 2023-08-31 07:55:33 +02:00
ethever.eth
ee6a7c2bbb fix a small typo (#2165) 
Co-authored-by: famouscat <onismaa@gmail.com>
 2023-08-30 20:14:26 +02:00
PSeitz
c4e2708901 fix clippy, fmt (#2162) 2023-08-30 08:04:26 +02:00
PSeitz
5c8cfa50eb add missing parameter for percentiles (#2157) 2023-08-29 13:04:24 +02:00
PSeitz
73cb71762f add missing parameter for stats,min,max,count,sum,avg (#2151) 
* add missing parameter for stats,min,max,count,sum,avg

add missing parameter for stats,min,max,count,sum,avg
closes #1913
partially #1789

* Apply suggestions from code review

Co-authored-by: Paul Masurel <paul@quickwit.io>

---------

Co-authored-by: Paul Masurel <paul@quickwit.io>
 2023-08-28 08:59:51 +02:00
Harrison Burt
267dfe58d7 Fix testing on windows (#2155) 
* Fix missing trait imports

* Fix building tests on windows

* Revert other PR change
 2023-08-27 09:20:44 +09:00
Harrison Burt
131c10d318 Fix missing trait imports (#2154) 2023-08-27 09:20:26 +09:00
Chris Tam
e6cacc40a9 Remove outdated fast field documentation (#2145) 2023-08-24 07:49:49 +02:00
PSeitz
48d4847b38 Improve aggregation error message (#2150) 
* Improve aggregation error message

Improve aggregation error message by wrapping the deserialization with a
custom struct. This deserialization variant is slower, since we need to
keep the deserialized data around twice with this approach.
For now the valid variants list is manually updated. This could be
replaced with a proc macro.
closes #2143

* Simpler implementation

---------

Co-authored-by: Paul Masurel <paul@quickwit.io>
 2023-08-23 20:52:15 +02:00
PSeitz
59460c767f delayed column opening during merge (#2132) 
* lazy columnar merge

This is the first part of addressing #3633
Instead of loading all Column into memory for the merge, only the current column_name
group is loaded. This can be done since the sstable streams the columns lexicographically.

* refactor

* add rustdoc

* replace iterator with BTreeMap
 2023-08-21 08:55:35 +02:00
Paul Masurel
756156beaf Fix doc 2023-08-17 17:47:45 +09:00
PSeitz
480763db0d track memory arena memory usage (#2148) 2023-08-16 18:19:42 +02:00
PSeitz
62ece86f24 track ff dictionary indexing memory consumption (#2147) 2023-08-16 14:00:08 +02:00
Caleb Hattingh
52d9e6f298 Fix doc typos in count aggregation metric (#2127) 2023-08-15 08:50:23 +02:00
Caleb Hattingh
47b315ff18 doc: escape the backslash (#2144) 2023-08-14 19:10:07 +02:00
PSeitz
ed1deee902 fix sort index by date (#2124) 
closes #2112
 2023-08-14 17:36:52 +02:00
PSeitz
2e109018b7 add missing parameter to term agg (#2103) 
* add missing parameter to term agg

* move missing handling to block accessor

* add multivalue test, fix multivalue case, add comments

* add documentation, deactivate special case

* cargo fmt

* resolve merge conflict
 2023-08-14 14:22:18 +02:00
Adam Reichold
22c35b1e00 Fix explanation of boost queries seeking beyond query result. (#2142) 
* Make current nightly Clippy happy.

* Fix explanation of boost queries seeking beyond query result.
 2023-08-14 11:59:11 +09:00
trinity-1686a
b92082b748 implement lenient parser (#2129) 
* move query parser to nom

* add suupport for term grouping

* initial work on infallible parser

* fmt

* add tests and fix minor parsing bugs

* address review comments

* add support for lenient queries in tantivy

* make lenient parser report errors

* allow mixing occur and bool in query
 2023-08-08 15:41:29 +02:00
PSeitz
c2be6603a2 alternative mixed field aggregation collection (#2135) 
* alternative mixed field aggregation collection

instead of having multiple accessor in one AggregationWithAccessor split it into
multiple independent AggregationWithAccessor

* Update src/aggregation/agg_req_with_accessor.rs

Co-authored-by: Paul Masurel <paul@quickwit.io>

---------

Co-authored-by: Paul Masurel <paul@quickwit.io>
 2023-07-27 12:25:31 +02:00
Adam Reichold
c805f08ca7 Fix a few more upcoming Clippy lints (#2133) 2023-07-24 17:07:57 +09:00
Adam Reichold
ccc0335158 Minor improvements to OwnedBytes (#2134) 
This makes it obvious where the `StableDerefTrait` is invoked and avoids
`transmute` when only a lifetime needs to be extended. Furthermore, it makes use
of `slice::split_at` where that seemed appropriate.
 2023-07-24 17:06:33 +09:00
Adam Reichold
42acd334f4 Fixes the new deny-by-default incorrect_partial_ord_impl_on_ord_type Clippy lint (#2131) 2023-07-21 11:36:17 +09:00
Adam Reichold
820f126075 Remove support for Brotli and Snappy compression (#2123) 
LZ4 provides fast and simple compression whereas Zstd is exceptionally flexible
so that the additional support for Brotli and Snappy does not really add
any distinct functionality on top of those two algorithms.

Removing them reduces our maintenance burden and reduces the number of choices
users have to make when setting up their project based on Tantivy.
 2023-07-14 16:54:59 +09:00
Adam Reichold
7e6c4a1856 Include only built-in compression algorithms as enum variants (#2121) 
* Include only built-in compression algorithms as enum variants

This enables compile-time errors when a compression algorithm is requested which
is not actually enabled for the current Cargo project. The cost is that indexes
using other compression algorithms cannot even be loaded (even though they
are not fully accessible in any case).

As a drive-by, this also fixes `--no-default-features` on `cfg(unix)`.

* Provide more instructive error messages for unsupported, but not unknown compression variants.
 2023-07-14 11:02:49 +09:00
Adam Reichold
5fafe4b1ab Add missing query_terms impl for TermSetQuery. (#2120) 2023-07-13 14:54:29 +02:00
PSeitz
1e7cd48cfa remove allocations in split compound words (#2080) 
* remove allocations in split compound words

* clear reused data
 2023-07-13 09:43:02 +09:00
dependabot[bot]
7f51d85bbd Update lru requirement from 0.10.0 to 0.11.0 (#2117) 
Updates the requirements on [lru](https://github.com/jeromefroe/lru-rs) to permit the latest version.
- [Changelog](https://github.com/jeromefroe/lru-rs/blob/master/CHANGELOG.md)
- [Commits](https://github.com/jeromefroe/lru-rs/compare/0.10.0...0.11.0)

---
updated-dependencies:
- dependency-name: lru
  dependency-type: direct:production
...

Signed-off-by: dependabot[bot] <support@github.com>
Co-authored-by: dependabot[bot] <49699333+dependabot[bot]@users.noreply.github.com>
 2023-07-13 09:42:21 +09:00
PSeitz
ad76e32398 Update CHANGELOG.md (#2091) 
* Update CHANGELOG.md

* Update CHANGELOG.md
 2023-07-11 13:58:49 +08:00
dependabot[bot]
7575f9bf1c Update itertools requirement from 0.10.3 to 0.11.0 (#2098) 
Updates the requirements on [itertools](https://github.com/rust-itertools/itertools) to permit the latest version.
- [Changelog](https://github.com/rust-itertools/itertools/blob/master/CHANGELOG.md)
- [Commits](https://github.com/rust-itertools/itertools/compare/v0.10.5...v0.11.0)

---
updated-dependencies:
- dependency-name: itertools
  dependency-type: direct:production
...

Signed-off-by: dependabot[bot] <support@github.com>
Co-authored-by: dependabot[bot] <49699333+dependabot[bot]@users.noreply.github.com>
 2023-07-07 11:14:46 +02:00
Naveen Aiathurai
67bdf3f5f6 fixes order_by_u64_field and order_by_fast_field should allow sorting in ascending order #1676 (#2111) 
* feat: order_by_fast_field allows sorting using parameter order

* chore: change the corresponding values to original one

* chore: fix formatting issues

* fix: first_or_default_col should also sort by order

* chore: empty doc to testcase and docstest fixes

* chore: fix failure tests

* core: add empty document without fastfield

* chore: fix fmt

* chore: change variable name
 2023-07-06 05:10:10 +02:00
François Massot
3c300666ad Merge pull request #2110 from quickwit-oss/fulmicoton/dynamic-follow-up 
Add dynamic filters to text analyzer builder.
 2023-07-03 21:49:24 +02:00
François Massot
b91d3f6be4 Clean comment on 'TextAnalyzerBuilder::filter_dynamic' method. 2023-07-03 18:45:59 +02:00
François Massot
a8e76513bb Remove useless clone. 2023-07-03 22:05:11 +09:00
François Massot
0a23201338 Fix stackoverflow and add docs. 2023-07-03 22:05:11 +09:00
François Massot
81330aaf89 WIP 2023-07-03 22:05:10 +09:00
Paul Masurel
98a3b01992 Removing the BoxedTokenizer 2023-07-03 22:05:10 +09:00
Paul Masurel
d341520938 Dynamic follow up 2023-07-03 22:05:10 +09:00
François Massot
5c9af73e41 Followup fulmicoton poc. 2023-07-03 22:05:10 +09:00
Paul Masurel
ad4c940fa3 proof of concept for dynamic tokenizer. 2023-07-03 22:05:10 +09:00
Paul Masurel
910b0b0c61 Cargo fmt 2023-07-03 22:03:31 +09:00
PSeitz
3fef052bf1 fix flaky test (#2107) 
closes #2099
 2023-06-29 14:30:56 +08:00
PSeitz
040554f2f9 Update to lz4_flex 0.11 (#2106) 2023-06-29 14:16:00 +08:00
PSeitz
17186ca9c9 improve docs (#2105) 2023-06-27 13:37:14 +08:00
François Massot
212d59c9ab Merge pull request #2102 from quickwit-oss/fmassot/ngram-new-should-return-error 
Ngram tokenizer now returns an error with invalid arguments.
 2023-06-27 05:36:09 +02:00
dependabot[bot]
1a1f252a3f Update memmap2 requirement from 0.6.0 to 0.7.1 (#2104) 
Updates the requirements on [memmap2](https://github.com/RazrFalcon/memmap2-rs) to permit the latest version.
- [Changelog](https://github.com/RazrFalcon/memmap2-rs/blob/master/CHANGELOG.md)
- [Commits](https://github.com/RazrFalcon/memmap2-rs/compare/v0.6.0...v0.7.1)

---
updated-dependencies:
- dependency-name: memmap2
  dependency-type: direct:production
...

Signed-off-by: dependabot[bot] <support@github.com>
Co-authored-by: dependabot[bot] <49699333+dependabot[bot]@users.noreply.github.com>
 2023-06-27 05:15:43 +02:00
François Massot
d73706dede Ngram tokenizer now returns an error with invalid arguments. 2023-06-25 20:13:24 +02:00
PSeitz
44850e1036 move fail dep to dev only (#2094) 
wasm compilation fails with dep only
 2023-06-22 06:59:11 +02:00
Adam Reichold
3b0cbf8102 Cosmetic updates to the warmer example. (#2095) 
Just some cosmetic tweaks to make the example easier on the eyes as a colleague
was staring at this for quite some time this week.
 2023-06-22 11:25:01 +09:00
Adam Reichold
4aa131c3db Make TextAnalyzerBuilder publically accessible (#2097) 
This way, client code can name the type to e.g. store it inside structs without
resorting to generics and it means that its documentation is part of the crate
documentation generated by `cargo doc`.
 2023-06-22 11:24:21 +09:00
Naveen Aiathurai
59962097d0 fix: #2078 return error when tokenizer not found while indexing (#2093) 
* fix: #2078 return error when tokenizer not found while indexing

* chore: formatting issues

* chore: fix review comments
 2023-06-16 04:33:55 +02:00
Adam Reichold
ebc78127f3 Add BytesFilterCollector to support filtering based on a bytes fast field (#2075) 
* Do some Clippy- and Cargo-related boy-scouting.

* Add BytesFilterCollector to support filtering based on a bytes fast field

This is basically a copy of the existing FilterCollector but modified and
specialised to work on a bytes fast field.

* Changed semantics of filter collectors to consider multi-valued fields
 2023-06-13 14:19:58 +09:00
PSeitz
8199aa7de7 bump version to 0.20.2 (#2089) 2023-06-12 18:56:54 +08:00
PSeitz
657f0cd3bd add missing Bytes validation to term_agg (#2077) 
returns empty for now instead of failing like before
 2023-06-12 16:38:07 +08:00
Adam Reichold
3a82ef2560 Fix is_child_of function not considering the root facet. (#2086) 2023-06-12 08:35:18 +02:00
PSeitz
3546e7fc63 small agg limit docs improvement (#2073) 
small docs improvement as follow up on bug https://github.com/quickwit-oss/quickwit/issues/3503
 2023-06-12 10:55:24 +09:00
PSeitz
862f367f9e release without Alice in Wonderland, bump version to 0.20.1 (#2087) 
* Release without Alice in Wonderland

* bump version to 0.20.1
 2023-06-12 10:54:03 +09:00
PSeitz
14137d91c4 Update CHANGELOG.md (#2081) 2023-06-12 10:53:40 +09:00
François Massot
924fc70cb5 Merge pull request #2088 from quickwit-oss/fmassot/align-type-priorities-for-json-numbers 
Align numerical type priority order on the search side.
 2023-06-11 22:04:54 +02:00
François Massot
07023948aa Add test that indexes and searches a JSON field. 2023-06-11 21:47:52 +02:00
François Massot
0cb53207ec Fix tests. 2023-06-11 12:13:35 +02:00
François Massot
17c783b4db Align numerical type priority order on the search side. 2023-06-11 11:49:27 +02:00
Harrison Burt
7220df8a09 Fix building on windows with mmap (#2070) 
* Fix windows build

* Make pub

* Update docs

* Re arrange

* Fix compilation error on unix

* Fix unix borrows

* Revert "Fix unix borrows"

This reverts commit c1d94fd12b.

* Fix unix borrows and revert original change

* Fix warning

* Cleaner code.

---------

Co-authored-by: Paul Masurel <paul@quickwit.io>
 2023-06-10 18:32:39 +02:00
PSeitz
e3eacb4388 release tantivy (#2083) 
* prerelease

* chore: Release
 2023-06-09 10:47:46 +02:00
PSeitz
fdecb79273 tokenizer-api: reduce Tokenizer overhead (#2062) 
* tokenizer-api: reduce Tokenizer overhead

Previously a new `Token` for each text encountered was created, which
contains `String::with_capacity(200)`
In the new API the token_stream gets mutable access to the tokenizer,
this allows state to be shared (in this PR Token is shared).
Ideally the allocation for the BoxTokenStream would also be removed, but
this may require some lifetime tricks.

* simplify api

* move lowercase and ascii folding buffer to global

* empty Token text as default
 2023-06-08 18:37:58 +08:00
PSeitz
27f202083c Improve Termmap Indexing Performance +~30% (#2058) 
* update benchmark

* Improve Termmap Indexing Performance +~30%

This contains many small changes to improve Termmap performance.
Most notably:
* Specialized byte compare and equality versions, instead of glibc calls.
* ExpUnrolledLinkedList to not contain inline items.

Allow compare hash only via a feature flag compare_hash_only:
64bits should be enough with a good hash function to compare strings by
their hashes instead of comparing the strings. Disabled by default

CreateHashMap/alice/174693
                        time:   [642.23 µs 643.80 µs 645.24 µs]
                        thrpt:  [258.20 MiB/s 258.78 MiB/s 259.41 MiB/s]
                 change:
                        time:   [-14.429% -13.303% -12.348%] (p = 0.00 < 0.05)
                        thrpt:  [+14.088% +15.344% +16.862%]
                        Performance has improved.
CreateHashMap/alice_expull/174693
                        time:   [877.03 µs 880.44 µs 884.67 µs]
                        thrpt:  [188.32 MiB/s 189.22 MiB/s 189.96 MiB/s]
                 change:
                        time:   [-26.460% -26.274% -26.091%] (p = 0.00 < 0.05)
                        thrpt:  [+35.301% +35.637% +35.981%]
                        Performance has improved.
CreateHashMap/numbers_zipf/8000000
                        time:   [9.1198 ms 9.1573 ms 9.1961 ms]
                        thrpt:  [829.64 MiB/s 833.15 MiB/s 836.57 MiB/s]
                 change:
                        time:   [-35.229% -34.828% -34.384%] (p = 0.00 < 0.05)
                        thrpt:  [+52.403% +53.440% +54.390%]
                        Performance has improved.

* clippy

* add bench for ids

* inline(always) to inline whole block with bounds checks

* cleanup
 2023-06-08 11:13:52 +02:00
PSeitz
ccb09aaa83 allow histogram bounds to be passed as Rfc3339 (#2076) 2023-06-08 09:07:08 +02:00
Valerii
4b7c485a08 feat: add stop words for Hungarian language (#2069) 2023-06-02 07:26:03 +02:00
PSeitz
3942fc6d2b update CHANGELOG (#2068) 2023-06-02 05:00:12 +02:00
Adam Reichold
b325d569ad Expose phrase-prefix queries via the built-in query parser (#2044) 
* Expose phrase-prefix queries via the built-in query parser

This proposes the less-than-imaginative syntax `field:"phrase ter"*` to
perform a phrase prefix query against `field` using `phrase` and `ter` as the
terms. The aim of this is to make this type of query more discoverable and
simplify manual testing.

I did consider exposing the `max_expansions` parameter similar to how slop is
handled, but I think that this is rather something that should be configured via
the querser parser (similar to `set_field_boost` and `set_field_fuzzy`) as
choosing it requires rather intimiate knowledge of the backing index.

* Prevent construction of zero or one term phrase-prefix queries via the query parser.

* Add example using phrase-prefix search via surface API to improve feature discoverability.
 2023-06-01 13:03:16 +02:00
Paul Masurel
7ee78bda52 Readding s in datetime precision variant names (#2065) 
There is no clear win and it change some serialization in quickwit.
 2023-06-01 06:39:46 +02:00
Paul Masurel
184a9daa8a Cancels concurrently running actions for the same PR. (#2067) 2023-06-01 12:57:38 +09:00
Paul Masurel
47e01b345b Simplified linear probing code (#2066) 2023-06-01 04:58:42 +02:00
PSeitz
3af456972e Fix min doc_count empty merge bug (#2057) 
This fixes an issue when min_doc==0 loads terms from the dictionary from
one segment and merges the same term with a subaggregation from another
segment.
Previously the empty structure was not correctly initialized to contain
the subaggregation so the merge was incorrect.
 2023-05-29 14:20:50 +08:00
PSeitz
e56addc63e enable tokenizer on json fields (#2053) 
* enable tokenizer on json fields

enable tokenizer on json fields for type text

* Avoid making the tokenizer within the TextAnalyzer pub(crate)

* Moving BoxableTokenizer to tantivy.

---------

Co-authored-by: Paul Masurel <paul@quickwit.io>
 2023-05-24 10:47:39 +02:00
dependabot[bot]
4be6f83b0a Update criterion requirement from 0.4 to 0.5 (#2056) 
Updates the requirements on [criterion](https://github.com/bheisler/criterion.rs) to permit the latest version.
- [Changelog](https://github.com/bheisler/criterion.rs/blob/master/CHANGELOG.md)
- [Commits](https://github.com/bheisler/criterion.rs/compare/0.4.0...0.5.0)

---
updated-dependencies:
- dependency-name: criterion
  dependency-type: direct:production
...

Signed-off-by: dependabot[bot] <support@github.com>
Co-authored-by: dependabot[bot] <49699333+dependabot[bot]@users.noreply.github.com>
 2023-05-24 15:59:51 +09:00
Adrien Guillo
a789ad9aee Rename DatePrecision to DateTimePrecision (#2051) 2023-05-23 17:09:11 +02:00
Sergei Lavrentev
8cf26da4b2 Add possibility to set up highlighten prefix and postfix for snippet (#1422) 
* add possibility to change highlight prefix and postfix

* add comment to Snippet::new

* add test for highlighten elements

* add default highlight prefix and postfix constants

* fix spelling

* fix tests

* fix spelling

* do fixes after code review

* reduce test_snippet_generator_custom_highlighted_elements code

* fix fmt

* change names to more convenient

---------

Co-authored-by: Sergei Lavrentev <23312691+lavrxxx@users.noreply.github.com>
 2023-05-23 15:09:24 +02:00
trinity-1686a
a3f001360f add support for warming up range of terms (#2042) 
* add support for warming up range of terms

* simplify handling of limit
 2023-05-22 14:29:35 +02:00
trinity-1686a
6564e0c467 fix phrase prefix query (#2043) 
* fix phrase prefix query

it would fail spectacularly when no doc in the segment would match the phrase part of the query

* clippy
 2023-05-22 12:36:20 +02:00
Paul Masurel
d7e97331e5 Minor refactoring find field (#2055) 
* Minor refactoring

Moving find_field_with_default to Schema.

* Clippy comments
 2023-05-22 15:00:48 +09:00
Paul Masurel
4417be165d Minor refactoring (#2054) 
Moving find_field_with_default to Schema.
 2023-05-22 14:56:38 +09:00
PSeitz
6239697a02 switch to ms in histogram for date type (#2045) 
* switch to ms in histogram for date type

switch to ms in histogram, by adding a normalization step that converts
to nanoseconds precision when creating the collector.

closes #2028
related to #2026

* add missing unit long variants

* use single thread to avoid handling test case

* fix docs

* revert CI

* cleanup

* improve docs

* Update src/aggregation/bucket/histogram/histogram.rs

Co-authored-by: Paul Masurel <paul@quickwit.io>

---------

Co-authored-by: Paul Masurel <paul@quickwit.io>
 2023-05-19 08:15:44 +02:00
Paul Masurel
62709b8094 Change in the query grammar. (#2050) 
* Change in the query grammar.

Quotation mark can now be used for phrase queries.
The delimiter is part of the `UserInputLeaf`.
That information is meant to be used in Quickwit to solve #3364.

This PR also adds support for quotation marks escaping in phrase
queries.

* Apply suggestions from code review
 2023-05-19 12:07:10 +09:00
PSeitz
04562c0318 add fastfield tokenizer to IndexBuilder (#2046) 2023-05-18 04:33:42 +02:00
PSeitz
2dfe37940d handle multiple types in term aggregation (#2041) 2023-05-15 11:57:38 +02:00
Denis Bazhenov
e248a4959f Enforcing "NOT" and "-" queries consistency in UserInputAst (#1609) 
* Enforcing "NOT" and "-" queries consistency in UserInputAst

* Mutable implementation if rewrite_ast_clause()
 2023-05-13 00:27:48 +09:00
PSeitz
00c5df610c update termmap benchmark (#2040) 2023-05-12 07:35:06 +02:00
Adam Reichold
fedd9559e7 Expose create a query from a user input AST. (#2039) 2023-05-11 21:53:18 +09:00
Paul Masurel
fe3ecf9567 Added support for madvise (#2036) 
Added support for madvise
 2023-05-11 05:39:17 +02:00
PSeitz
ba3a885a3b handle multiple agg results (#2035) 
handle multiple intermediate aggregation results with the same name.
 2023-05-10 15:00:38 +02:00
PSeitz
d1988be8e9 fix and extend benchmark (#2030) 
* add benchmark, add missing inlines

* fix stacker bench

* add wiki benchmark

* move line split out of bench
 2023-05-10 13:01:56 +02:00
PSeitz
0eafbaab8e fix slop (#2031) 
Fix slop by carrying slop so far for multiterms.
Define slop contract in the API
 2023-05-10 11:45:14 +02:00
PSeitz
d3357a8426 fix ArenaHashMap default (#2034) 
an empty ArenaHashMap is invalid and causes a panic when combined with `get`
 2023-05-10 11:39:47 +02:00
Yuri Astrakhan
74275b76a6 Inline format arguments where makes sense (#2038) 
Applied this command to the code, making it a bit shorter and slightly
more readable.

```
cargo +nightly clippy --all-features --benches --tests --workspace --fix -- -A clippy::all -W clippy::uninlined_format_args
cargo +nightly fmt --all
```
 2023-05-10 18:03:59 +09:00
dependabot[bot]
f479840a1b Update memmap2 requirement from 0.5.3 to 0.6.0 (#2033) 
Updates the requirements on [memmap2](https://github.com/RazrFalcon/memmap2-rs) to permit the latest version.
- [Changelog](https://github.com/RazrFalcon/memmap2-rs/blob/master/CHANGELOG.md)
- [Commits](https://github.com/RazrFalcon/memmap2-rs/compare/v0.5.3...v0.6.0)

---
updated-dependencies:
- dependency-name: memmap2
  dependency-type: direct:production
...

Signed-off-by: dependabot[bot] <support@github.com>
Co-authored-by: dependabot[bot] <49699333+dependabot[bot]@users.noreply.github.com>
 2023-05-10 03:50:14 +02:00
PSeitz
4ee1b5cda0 add seperate tokenizer manager for fast fields (#2019) 
* add seperate tokenizer manager for fast fields

* rename
 2023-05-08 11:22:31 +02:00
PSeitz
45ff0e3c5c clear memory consumption in AggregationLimits (#2022) 
* clear memory consumption in AggregationLimits

clear memory consumption in AggregationLimits at the end of segment collection

* switch to ResourceLimitGuard

* unduplicate code

* merge methods

* Apply suggestions from code review

Co-authored-by: Paul Masurel <paul@quickwit.io>

---------

Co-authored-by: Paul Masurel <paul@quickwit.io>
 2023-05-08 10:15:09 +02:00
PSeitz
4c58b0086d allow slop in both directions (#2020) 
* allow slop in both directions

allow slop in both directions
so "big wolf"~3 can also match "wolf big"

This also fixes #1934, when the docsets were reordered by size and didn't
match the terms.

* remove count

* add test for repeating tokens, unduplicate tests
 2023-05-07 12:05:21 +09:00
Tomoko Uchida
85df322ceb fix typo in the architecture doc (#2009) 2023-05-07 12:04:07 +09:00
François Massot
38c863830f Merge pull request #2027 from quickwit-oss/fmassot/fix-date-histogram 
Fix date histogram bounds and field name.
 2023-05-05 13:03:25 +02:00
François Massot
992f755298 Fix clippy. 2023-05-05 10:51:29 +02:00
François Massot
c8df843f96 Fix date histogram bounds and field name. 2023-05-05 00:52:55 +02:00
Paul Masurel
f28ddb711e Exposing u64-based FastFieldRangeWeight (#2024) 2023-05-03 18:32:00 +09:00
tottoto
73452284ae Remove unused crates from dependencies (#2018) 
* Remove unused crates from dependencies

* Revert rand to columnar

* Revert criterion to stacker
 2023-05-02 12:34:20 +02:00
PSeitz
ba309e18a1 switch to nanosecond precision (#2016) 2023-05-01 03:32:20 +02:00
PSeitz
cbf2bdc75b change bucket count type (#2013) 
* change bucket count type

closes #2012

* Update src/aggregation/agg_limits.rs

Co-authored-by: Paul Masurel <paul@quickwit.io>

* Update src/directory/managed_directory.rs

Co-authored-by: Paul Masurel <paul@quickwit.io>

* fix test

---------

Co-authored-by: Paul Masurel <paul@quickwit.io>
 2023-04-27 15:47:31 +08:00
PSeitz
1f06997d04 fix single collector special case (#2014) 2023-04-27 09:30:19 +02:00
PSeitz
c599bf3b6c chore!:drop JSON support on intermediate agg result (#1992) 
* chore!:drop JSON support on intermediate agg result

add support for other formats by removing skip_serialize and untagged
JSON support is broken anyway due it's lack on f64::INF etc. handling

* Update src/aggregation/intermediate_agg_result.rs

Co-authored-by: Paul Masurel <paul@quickwit.io>

* move from impl

---------

Co-authored-by: Paul Masurel <paul@quickwit.io>
 2023-04-26 13:05:16 +02:00
PSeitz
80df1d9835 Handle error for exists on MMapDirectory (#1988) 
`exists` will return false in case of other io errors, like permission denied
 2023-04-25 09:20:33 +02:00
PSeitz
2e369db936 switch to Aggregation without serde_untagged (#2003) 
* refactor result handling

* remove Internal stuff

* merge different accessors

* switch to Aggregation without serde_untagged

* fix doctests
 2023-04-25 08:54:51 +02:00
PSeitz
7b31100208 refactor vint (#2010) 
- improve performance of vint
vint serialization shows up in performance profiles during indexing.
It would also make sense to limit the value space to u29 and operate on 4 bytes only.
- remove unused code
- add missing inlines
- fix regex test
 2023-04-25 08:49:36 +02:00
trinity-1686a
9c93bfeb51 optimise warmup code path (#2007) 
* optimise warmup code path

* better function naming
 2023-04-21 11:23:09 +02:00
PSeitz
74f9eafefc refactor Term (#2006) 
* refactor Term

add ValueBytes for serialized term values
add missing debug for ip
skip unnecessary json path validation
remove code duplication
add DATE_TIME_PRECISION_INDEXED constant
add missing Term clarification
remove weird value_bytes_mut() API

* fix naming
 2023-04-20 15:31:43 +02:00
RT_Enzyme
ff3d3313c4 fix BooleanQuery document (#1999) 
* fix BooleanQuery document

* Update src/query/boolean_query/boolean_query.rs

---------

Co-authored-by: Paul Masurel <paul@quickwit.io>
 2023-04-20 11:37:20 +02:00
Paul Masurel
fbda511a1a Making more things public for quickwit. (#2005) 2023-04-20 11:37:45 +09:00
Adam Reichold
c1defdda05 Bump aho-corasick dependency to version 1.0 and adjust to API changes (#2002) 
* Drop additional Arc-layer as the automaton itself is now cheap-to-clone.
* Drop state ID type parameter as it is not exposed by the library any more.
 2023-04-18 07:34:30 +02:00
PSeitz
e522163a1c use json in agg tests (#1998) 
* switch to JSON in tests, add flat aggregation types

* use method

* clippy

* remove commented file
 2023-04-17 14:08:48 +02:00
PSeitz
e83abbfe4a perf: faster term hash map (#1940) 
* add term hashmap benchmark

* refactor arena hashmap

add inlines
remove occupied array and use table_entry.is_empty instead (saves 4 bytes per entry)
reduce saturation threshold from 1/3 to 1/2 to reduce memory
use u32 for UnorderedId (we have the 4billion limit anyways on the Columnar stuff)
fix naming LinearProbing
remove byteorder dependency

memory consumption went down from 2Gb to 1.8GB on indexing wikipedia dataset in tantivy

* Update stacker/src/arena_hashmap.rs

Co-authored-by: Paul Masurel <paul@quickwit.io>

---------

Co-authored-by: Paul Masurel <paul@quickwit.io>
 2023-04-17 09:07:33 +02:00
trinity-1686a
780e26331d sstable compression (#1946) 
* compress sstable with zstd

* add some details to sstable readme

* compress only block which benefit from it

* multiple changes to sstable

make compression optional
use OwnedBytes instead of impl Read in sstable, required for next point
use zstd bulk api, which is much faster on small records

* cleanup and use bulk api for compression

* use dedicated byte for compression

* switch block len and compression flag

* change default zstd level in sstable
 2023-04-14 16:25:50 +02:00
trinity-1686a
0286ecea09 re-export a few sstable functions on dicitonary (#1996) 
* re-export a few sstable functions on dicitonary

* Update documentation

Co-authored-by: François Massot <francois.massot@gmail.com>

---------

Co-authored-by: François Massot <francois.massot@gmail.com>
 2023-04-14 11:13:48 +02:00
PSeitz
b0ef9a6252 use crates.io dependency (#1990) 2023-04-14 09:35:20 +08:00
François Massot
36138c493b Merge pull request #1994 from quickwit-oss/fmassot/expose-simple-token-stream 
Expose `SimpleTokenStream` to use it in quickwit for the multilanguage tokenizer
 2023-04-13 18:55:02 +02:00
François Massot
64bce340b2 Expose to use it in quickwit. 2023-04-13 18:28:53 +02:00
trinity-1686a
205e8a0a92 encode dictionary type in fst footer (#1968) 
* encode additional footer for dictionary kind in fst
 2023-04-12 09:43:01 +02:00
Paul Masurel
4b01cc4c49 Made BooleanWeight and BoostWeight public (#1991) 2023-04-12 10:26:30 +09:00
PSeitz
0ed13eeea8 add sparse to agg benchmark (#1986) 
* add sparse to agg benchmark

* Update src/aggregation/agg_bench.rs

Co-authored-by: Paul Masurel <paul@quickwit.io>

---------

Co-authored-by: Paul Masurel <paul@quickwit.io>
 2023-04-11 08:13:32 +02:00
Tony-X
91a38058fe Fix typo in READEME.md (#1989) 2023-04-11 12:07:20 +09:00
PSeitz
41af70799d add percentiles aggregations (#1984) 
* add percentiles aggregations

add percentiles aggregation
fix disabled agg benchmark

* Update src/aggregation/metric/percentiles.rs

Co-authored-by: Paul Masurel <paul@quickwit.io>

* Apply suggestions from code review

Co-authored-by: Paul Masurel <paul@quickwit.io>

* fix import

* fix import

---------

Co-authored-by: Paul Masurel <paul@quickwit.io>
 2023-04-07 07:18:28 +02:00
Paul Masurel
f853bf204b Align the numerical type priority order with columnar. (#1978) 
Closes #1956
 2023-04-07 10:07:54 +09:00
Tony-X
11ae48d3bc Update benchmarks section in READEME.md to link to the bench repo (#1985) 
* Update benchmarks section in READEME.md to link to the bench repo

* Apply suggestions from code review

---------

Co-authored-by: Paul Masurel <paul@quickwit.io>
 2023-04-07 10:07:06 +09:00
Paul Masurel
5eb12173d6 Proptest merge columnar (#1976) 
* Added proptest on columnar merge with a shuffle

Made column serialization more explicit.
Bugfix when a bytes column is missing, and with a shuffle.
Improved the cardinality detection logic / column detection.

* Code review

* CR comments

* Following CR
 2023-04-04 11:28:42 +09:00
PSeitz
5c4ea6a708 tokenizer option on text fastfield (#1945) 
* tokenizer option on text fastfield

allow to set tokenizer option on text fastfield (fixes #1901)
handle PreTokenized strings in fast field

* change visibility

* remove custom de/serialization
 2023-03-31 10:03:38 +02:00
PSeitz
4cf93dab7d fix build (#1973) 2023-03-31 13:54:03 +09:00
PSeitz
5c380b76e7 Better mixed types support in aggs and fix serialization issue (#1971) 
* Better mixed types support in aggs and fix serialization issue

- Improve support for mixed types in JSON field aggregations (pick the right field, #1913)
- Resolve the issue with JSON serialization for numeric keys (fixes #1967)
- Add JSON round-trip test for term buckets
- Remove `u64_lenient`, as this is a footgun without the type
- move aggregation benchmarks

* remove shadowing
 2023-03-31 05:52:11 +02:00
PSeitz
571735c5f7 Fix index sort by on optional/multicolumn (#1972) 
Fix index sort by on optional/multicolumn
add optional columns to proptest
extend proptests for sort
add columnar sort tests
 2023-03-31 04:24:11 +02:00
zhouhui
8e92f960d3 Fix comment: change max_merge_size to max_docs_before_merge. (#1970) 2023-03-28 22:49:00 +09:00
Paul Masurel
057211c3d8 Fixing build on arm (#1966) 2023-03-27 22:42:57 +09:00
Paul Masurel
059fc767ea Added ::MIN ::MAX DateTime. (#1965) 2023-03-27 15:32:53 +09:00
Paul Masurel
694a056255 Faster range (#1954) 
* Faster range queries

This PR does several changes
- ip compact space now uses u32
- the bitunpacker now gets a get_batch function
- we push down range filtering, removing GCD / shift in the bitpacking
  codec.
- we rely on AVX2 routine to do the filtering.

* Apply suggestions from code review

* Apply suggestions from code review

* CR comments
 2023-03-27 14:56:32 +09:00
Paul Masurel
2955e34452 Added proptests for building/merging columnar. (#1963) 2023-03-27 14:56:02 +09:00
Paul Masurel
821208480b Adding Debug/Display impl. Refining the ColumnIndex::get_cardinality 2023-03-26 14:40:37 +09:00
Paul Masurel
a2e3c2ed5b Renaming Column::idx -> Column::index (#1961) 
There was some variable name ghosting happening.
 2023-03-26 13:58:50 +09:00
PSeitz
835f228bfa fix cardinality when merging empty columns (#1960) 
fixes #1958
 2023-03-25 15:58:15 +09:00
Paul Masurel
2b6a4da640 Exposing empty column builder. (#1959) 2023-03-24 16:34:41 +09:00
PSeitz
d6a95381ee add memory check for term agg (#1957) 2023-03-24 06:47:45 +01:00
PSeitz
da2804644f fetch blocks of vals in aggregation for all cardinality (#1950) 
* fetch blocks of vals in aggregation for all cardinality

* move caching in common accessor
 2023-03-23 08:41:11 +01:00
PSeitz
5504cfd012 remove IterColumn (#1955) 
fixes #1658
 2023-03-23 06:43:17 +01:00
trinity-1686a
482b4155e8 fix bug with new sstable index format (#1953) 2023-03-22 10:22:36 +01:00
Till Wegmüller
1a35f6573d Switch fs2 to fs4 as it is now unmaintained and does not support illumos (#1944) 
Signed-off-by: Till Wegmueller <toasterson@gmail.com>
 2023-03-22 13:48:49 +09:00
trinity-1686a
e5e50603a8 new sstable format (#1943) 
* document a new sstable format

* add support for changing target block size

* use new format for sstable index

* handle sstable version errror

* use very small blocks for proptests

* add a footer structure
 2023-03-21 15:03:52 +01:00
PSeitz
8f7f1d6be4 add Display for ByteCount (#1949) 
* add Display for ByteCount

* export missing AggregationLimits
 2023-03-21 08:02:35 +01:00
PSeitz
6a7a1106d6 work in batches of docs (#1937) 
* work in batches of docs

* add fill_buffer test
 2023-03-21 06:57:44 +01:00
PSeitz
9e2faecf5b add memory limit for aggregations (#1942) 
* add memory limit for aggregations

introduce AggregationLimits to set memory consumption limit and bucket limits
memory limit is checked during aggregation, bucket limit is checked before returning the aggregation request.

* Apply suggestions from code review

Co-authored-by: Paul Masurel <paul@quickwit.io>

* add ByteCount with human readable format

---------

Co-authored-by: Paul Masurel <paul@quickwit.io>
 2023-03-16 06:21:07 +01:00
PSeitz
b6703f1b3c fix validation in date histogram (#1936) 
fix validation in date histogram for parameters interval and date_interval
 2023-03-15 06:10:43 +01:00
PSeitz
2fb3740cb0 handle missing column for aggs (#1920) 
* handle missing column for aggs

add empty column fallback for missing column in aggs.
Fix sort for term agg on sub-agg with missing value (null is smallest)

* add error when field is not fast
 2023-03-15 06:09:59 +01:00
PSeitz
8459efa32c split term collection count and sub_agg (#1921) 
use unrolled ColumnValues::get_vals
 2023-03-13 04:37:41 +01:00
PSeitz
61cfd8dc57 fix clippy (#1927) 2023-03-13 03:12:02 +01:00
trinity-1686a
064518156f refactor tokenization pipeline to use GATs (#1924) 
* refactor tokenization pipeline to use GATs

* fix doctests

* fix clippy lints

* remove commented code
 2023-03-09 09:39:37 +01:00
PSeitz
a42a96f470 fix panic in dict column merge (#1930) 
* fix panic in dict column merge

* Bugfix and added unit test

---------

Co-authored-by: Paul Masurel <paul@quickwit.io>
 2023-03-08 22:04:37 +09:00
trinity-1686a
fcf5a25d93 use DeltaReader directly to implement Dictionnary::ord_to_term (#1928) 2023-03-08 11:15:56 +09:00
dependabot[bot]
c0a5b28fd3 Update lru requirement from 0.9.0 to 0.10.0 (#1932) 
Updates the requirements on [lru](https://github.com/jeromefroe/lru-rs) to permit the latest version.
- [Release notes](https://github.com/jeromefroe/lru-rs/releases)
- [Changelog](https://github.com/jeromefroe/lru-rs/blob/master/CHANGELOG.md)
- [Commits](https://github.com/jeromefroe/lru-rs/compare/0.9.0...0.10.0)

---
updated-dependencies:
- dependency-name: lru
  dependency-type: direct:production
...

Signed-off-by: dependabot[bot] <support@github.com>
Co-authored-by: dependabot[bot] <49699333+dependabot[bot]@users.noreply.github.com>
 2023-03-07 15:09:02 +09:00
trinity-1686a
a4f7ca8309 use DeltaReader directly to implement Dictionnary::term_ord (#1925) 
* use DeltaReader directly to implement Dictionnary::term_ord

* add some additional test case for Dictionary::term_ord
 2023-03-06 09:45:22 +01:00
Paul Masurel
364e321415 Clippy fix (#1926) 2023-03-06 10:37:17 +09:00
Paul Masurel
ed5a3b3172 Bumped murmurhash version 2023-03-03 21:24:32 +09:00
349 changed files with 31031 additions and 9988 deletions
Expand all files Collapse all files

13
.github/workflows/coverage.yml vendored
View File

@@ -3,20 +3,23 @@ name: Coverage
on:
push:
branches: [main]
pull_request:
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@v3
- uses: actions/checkout@v4
- name: Install Rust
run: rustup toolchain install nightly --profile minimal --component llvm-tools-preview
run: rustup toolchain install nightly-2023-09-10 --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 llvm-cov --all-features --workspace --doctests --lcov --output-path lcov.info
run: cargo +nightly-2023-09-10 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

7
.github/workflows/long_running.yml vendored
View File

@@ -8,13 +8,18 @@ env:
CARGO_TERM_COLOR: always
NUM_FUNCTIONAL_TEST_ITERATIONS: 20000
# 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:
test:
runs-on: ubuntu-latest
steps:
- uses: actions/checkout@v3
- uses: actions/checkout@v4
- name: Install stable
uses: actions-rs/toolchain@v1
with:

18
.github/workflows/test.yml vendored
View File

@@ -9,13 +9,18 @@ on:
env:
CARGO_TERM_COLOR: always
# 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:
check:
runs-on: ubuntu-latest
steps:
- uses: actions/checkout@v3
- uses: actions/checkout@v4
- name: Install nightly
uses: actions-rs/toolchain@v1
@@ -34,6 +39,13 @@ jobs:
- name: Check Formatting
run: cargo +nightly fmt --all -- --check
- name: Check Stable Compilation
run: cargo build --all-features
- name: Check Bench Compilation
run: cargo +nightly bench --no-run --profile=dev --all-features
- uses: actions-rs/clippy-check@v1
with:
@@ -48,14 +60,14 @@ jobs:
strategy:
matrix:
features: [
{ label: "all", flags: "mmap,stopwords,brotli-compression,lz4-compression,snappy-compression,zstd-compression,failpoints" },
{ label: "all", flags: "mmap,stopwords,lz4-compression,zstd-compression,failpoints" },
{ label: "quickwit", flags: "mmap,quickwit,failpoints" }
]
name: test-${{ matrix.features.label}}
steps:
- uses: actions/checkout@v3
- uses: actions/checkout@v4
- name: Install stable
uses: actions-rs/toolchain@v1

2
.gitignore vendored
View File

@@ -13,3 +13,5 @@ benchmark
.idea
trace.dat
cargo-timing*
control
variable

2
ARCHITECTURE.md
View File

@@ -254,7 +254,7 @@ The token positions of all of the terms are then stored in a separate file with
The [TermInfo](src/postings/term_info.rs) gives an offset (expressed in position this time) in this file. As we iterate through the docset,
we advance the position reader by the number of term frequencies of the current document.
## [fieldnorms/](src/fieldnorms): Here is my doc, how many tokens in this field?
## [fieldnorm/](src/fieldnorm): Here is my doc, how many tokens in this field?
The [BM25](https://en.wikipedia.org/wiki/Okapi_BM25) formula also requires to know the number of tokens stored in a specific field for a given document. We store this information on one byte per document in the fieldnorm.
The fieldnorm is therefore compressed. Values up to 40 are encoded unchanged.

184
CHANGELOG.md
View File

@@ -1,3 +1,187 @@
Tantivy 0.22
================================
Tantivy 0.22 will be able to read indices created with Tantivy 0.21.
#### Bugfixes
- Fix null byte handling in JSON paths (null bytes in json keys caused panic during indexing) [#2345](https://github.com/quickwit-oss/tantivy/pull/2345)(@PSeitz)
- Fix bug that can cause `get_docids_for_value_range` to panic. [#2295](https://github.com/quickwit-oss/tantivy/pull/2295)(@fulmicoton)
- Avoid 1 document indices by increase min memory to 15MB for indexing [#2176](https://github.com/quickwit-oss/tantivy/pull/2176)(@PSeitz)
- Fix merge panic for JSON fields [#2284](https://github.com/quickwit-oss/tantivy/pull/2284)(@PSeitz)
- Fix bug occuring when merging JSON object indexed with positions. [#2253](https://github.com/quickwit-oss/tantivy/pull/2253)(@fulmicoton)
- Fix empty DateHistogram gap bug [#2183](https://github.com/quickwit-oss/tantivy/pull/2183)(@PSeitz)
- Fix range query end check (fields with less than 1 value per doc are affected) [#2226](https://github.com/quickwit-oss/tantivy/pull/2226)(@PSeitz)
- Handle exclusive out of bounds ranges on fastfield range queries [#2174](https://github.com/quickwit-oss/tantivy/pull/2174)(@PSeitz)
#### Breaking API Changes
- rename ReloadPolicy onCommit to onCommitWithDelay [#2235](https://github.com/quickwit-oss/tantivy/pull/2235)(@giovannicuccu)
- Move exports from the root into modules [#2220](https://github.com/quickwit-oss/tantivy/pull/2220)(@PSeitz)
- Accept field name instead of `Field` in FilterCollector [#2196](https://github.com/quickwit-oss/tantivy/pull/2196)(@PSeitz)
- remove deprecated IntOptions and DateTime [#2353](https://github.com/quickwit-oss/tantivy/pull/2353)(@PSeitz)
#### Features/Improvements
- Tantivy documents as a trait: Index data directly without converting to tantivy types first [#2071](https://github.com/quickwit-oss/tantivy/pull/2071)(@ChillFish8)
- encode some part of posting list as -1 instead of direct values (smaller inverted indices) [#2185](https://github.com/quickwit-oss/tantivy/pull/2185)(@trinity-1686a)
- **Aggregation**
- Support to deserialize f64 from string [#2311](https://github.com/quickwit-oss/tantivy/pull/2311)(@PSeitz)
- Add a top_hits aggregator [#2198](https://github.com/quickwit-oss/tantivy/pull/2198)(@ditsuke)
- Support bool type in term aggregation [#2318](https://github.com/quickwit-oss/tantivy/pull/2318)(@PSeitz)
- Support ip adresses in term aggregation [#2319](https://github.com/quickwit-oss/tantivy/pull/2319)(@PSeitz)
- Support date type in term aggregation [#2172](https://github.com/quickwit-oss/tantivy/pull/2172)(@PSeitz)
- Support escaped dot when addressing field [#2250](https://github.com/quickwit-oss/tantivy/pull/2250)(@PSeitz)
- Add ExistsQuery to check documents that have a value [#2160](https://github.com/quickwit-oss/tantivy/pull/2160)(@imotov)
- Expose TopDocs::order_by_u64_field again [#2282](https://github.com/quickwit-oss/tantivy/pull/2282)(@ditsuke)
- **Memory/Performance**
- Faster TopN: replace BinaryHeap with TopNComputer [#2186](https://github.com/quickwit-oss/tantivy/pull/2186)(@PSeitz)
- reduce number of allocations during indexing [#2257](https://github.com/quickwit-oss/tantivy/pull/2257)(@PSeitz)
- Less Memory while indexing: docid deltas while indexing [#2249](https://github.com/quickwit-oss/tantivy/pull/2249)(@PSeitz)
- Faster indexing: use term hashmap in fastfield [#2243](https://github.com/quickwit-oss/tantivy/pull/2243)(@PSeitz)
- term hashmap remove copy in is_empty, unused unordered_id [#2229](https://github.com/quickwit-oss/tantivy/pull/2229)(@PSeitz)
- add method to fetch block of first values in columnar [#2330](https://github.com/quickwit-oss/tantivy/pull/2330)(@PSeitz)
- Faster aggregations: add fast path for full columns in fetch_block [#2328](https://github.com/quickwit-oss/tantivy/pull/2328)(@PSeitz)
- Faster sstable loading: use fst for sstable index [#2268](https://github.com/quickwit-oss/tantivy/pull/2268)(@trinity-1686a)
- **QueryParser**
- allow newline where we allow space in query parser [#2302](https://github.com/quickwit-oss/tantivy/pull/2302)(@trinity-1686a)
- allow some mixing of occur and bool in strict query parser [#2323](https://github.com/quickwit-oss/tantivy/pull/2323)(@trinity-1686a)
- handle * inside term in lenient query parser [#2228](https://github.com/quickwit-oss/tantivy/pull/2228)(@trinity-1686a)
- add support for exists query syntax in query parser [#2170](https://github.com/quickwit-oss/tantivy/pull/2170)(@trinity-1686a)
- Add shared search executor [#2312](https://github.com/quickwit-oss/tantivy/pull/2312)(@MochiXu)
- Truncate keys to u16::MAX in term hashmap [#2299](https://github.com/quickwit-oss/tantivy/pull/2299)(@PSeitz)
- report if a term matched when warming up posting list [#2309](https://github.com/quickwit-oss/tantivy/pull/2309)(@trinity-1686a)
- Support json fields in FuzzyTermQuery [#2173](https://github.com/quickwit-oss/tantivy/pull/2173)(@PingXia-at)
- Read list of fields encoded in term dictionary for JSON fields [#2184](https://github.com/quickwit-oss/tantivy/pull/2184)(@PSeitz)
- add collect_block to BoxableSegmentCollector [#2331](https://github.com/quickwit-oss/tantivy/pull/2331)(@PSeitz)
- expose collect_block buffer size [#2326](https://github.com/quickwit-oss/tantivy/pull/2326)(@PSeitz)
- Forward regex parser errors [#2288](https://github.com/quickwit-oss/tantivy/pull/2288)(@adamreichold)
- Make FacetCounts defaultable and cloneable. [#2322](https://github.com/quickwit-oss/tantivy/pull/2322)(@adamreichold)
- Derive Debug for SchemaBuilder [#2254](https://github.com/quickwit-oss/tantivy/pull/2254)(@GodTamIt)
- add missing inlines to tantivy options [#2245](https://github.com/quickwit-oss/tantivy/pull/2245)(@PSeitz)
Tantivy 0.21.1
================================
#### Bugfixes
- Range queries on fast fields with less values on that field than documents had an invalid end condition, leading to missing results. [#2226](https://github.com/quickwit-oss/tantivy/issues/2226)(@appaquet @PSeitz)
- Increase the minimum memory budget from 3MB to 15MB to avoid single doc segments (API fix). [#2176](https://github.com/quickwit-oss/tantivy/issues/2176)(@PSeitz)
Tantivy 0.21
================================
#### Bugfixes
- Fix track fast field memory consumption, which led to higher memory consumption than the budget allowed during indexing [#2148](https://github.com/quickwit-oss/tantivy/issues/2148)[#2147](https://github.com/quickwit-oss/tantivy/issues/2147)(@PSeitz)
- Fix a regression from 0.20 where sort index by date wasn't working anymore [#2124](https://github.com/quickwit-oss/tantivy/issues/2124)(@PSeitz)
- Fix getting the root facet on the `FacetCollector`. [#2086](https://github.com/quickwit-oss/tantivy/issues/2086)(@adamreichold)
- Align numerical type priority order of columnar and query. [#2088](https://github.com/quickwit-oss/tantivy/issues/2088)(@fmassot)
#### Breaking Changes
- Remove support for Brotli and Snappy compression [#2123](https://github.com/quickwit-oss/tantivy/issues/2123)(@adamreichold)
#### Features/Improvements
- Implement lenient query parser [#2129](https://github.com/quickwit-oss/tantivy/pull/2129)(@trinity-1686a)
- order_by_u64_field and order_by_fast_field allow sorting in ascending and descending order [#2111](https://github.com/quickwit-oss/tantivy/issues/2111)(@naveenann)
- Allow dynamic filters in text analyzer builder [#2110](https://github.com/quickwit-oss/tantivy/issues/2110)(@fulmicoton @fmassot)
- **Aggregation**
- Add missing parameter for term aggregation [#2149](https://github.com/quickwit-oss/tantivy/issues/2149)[#2103](https://github.com/quickwit-oss/tantivy/issues/2103)(@PSeitz)
- Add missing parameter for percentiles [#2157](https://github.com/quickwit-oss/tantivy/issues/2157)(@PSeitz)
- Add missing parameter for stats,min,max,count,sum,avg [#2151](https://github.com/quickwit-oss/tantivy/issues/2151)(@PSeitz)
- Improve aggregation deserialization error message [#2150](https://github.com/quickwit-oss/tantivy/issues/2150)(@PSeitz)
- Add validation for type Bytes to term_agg [#2077](https://github.com/quickwit-oss/tantivy/issues/2077)(@PSeitz)
- Alternative mixed field collection [#2135](https://github.com/quickwit-oss/tantivy/issues/2135)(@PSeitz)
- Add missing query_terms impl for TermSetQuery. [#2120](https://github.com/quickwit-oss/tantivy/issues/2120)(@adamreichold)
- Minor improvements to OwnedBytes [#2134](https://github.com/quickwit-oss/tantivy/issues/2134)(@adamreichold)
- Remove allocations in split compound words [#2080](https://github.com/quickwit-oss/tantivy/issues/2080)(@PSeitz)
- Ngram tokenizer now returns an error with invalid arguments [#2102](https://github.com/quickwit-oss/tantivy/issues/2102)(@fmassot)
- Make TextAnalyzerBuilder public [#2097](https://github.com/quickwit-oss/tantivy/issues/2097)(@adamreichold)
- Return an error when tokenizer is not found while indexing [#2093](https://github.com/quickwit-oss/tantivy/issues/2093)(@naveenann)
- Delayed column opening during merge [#2132](https://github.com/quickwit-oss/tantivy/issues/2132)(@PSeitz)
Tantivy 0.20.2
================================
- Align numerical type priority order on the search side. [#2088](https://github.com/quickwit-oss/tantivy/issues/2088) (@fmassot)
- Fix is_child_of function not considering the root facet. [#2086](https://github.com/quickwit-oss/tantivy/issues/2086) (@adamreichhold)
Tantivy 0.20.1
================================
- Fix building on windows with mmap [#2070](https://github.com/quickwit-oss/tantivy/issues/2070) (@ChillFish8)
Tantivy 0.20
================================
#### Bugfixes
- Fix phrase queries with slop (slop supports now transpositions, algorithm that carries slop so far for num terms > 2) [#2031](https://github.com/quickwit-oss/tantivy/issues/2031)[#2020](https://github.com/quickwit-oss/tantivy/issues/2020)(@PSeitz)
- Handle error for exists on MMapDirectory [#1988](https://github.com/quickwit-oss/tantivy/issues/1988) (@PSeitz)
- Aggregation
- Fix min doc_count empty merge bug [#2057](https://github.com/quickwit-oss/tantivy/issues/2057) (@PSeitz)
- Fix: Sort order for term aggregations (sort order on key was inverted) [#1858](https://github.com/quickwit-oss/tantivy/issues/1858) (@PSeitz)
#### Features/Improvements
- Add PhrasePrefixQuery [#1842](https://github.com/quickwit-oss/tantivy/issues/1842) (@trinity-1686a)
- Add `coerce` option for text and numbers types (convert the value instead of returning an error during indexing) [#1904](https://github.com/quickwit-oss/tantivy/issues/1904) (@PSeitz)
- Add regex tokenizer [#1759](https://github.com/quickwit-oss/tantivy/issues/1759)(@mkleen)
- Move tokenizer API to seperate crate. Having a seperate crate with a stable API will allow us to use tokenizers with different tantivy versions. [#1767](https://github.com/quickwit-oss/tantivy/issues/1767) (@PSeitz)
- **Columnar crate**: New fast field handling (@fulmicoton @PSeitz) [#1806](https://github.com/quickwit-oss/tantivy/issues/1806)[#1809](https://github.com/quickwit-oss/tantivy/issues/1809)
- Support for fast fields with optional values. Previously tantivy supported only single-valued and multi-value fast fields. The encoding of optional fast fields is now very compact.
- Fast field Support for JSON (schemaless fast fields). Support multiple types on the same column. [#1876](https://github.com/quickwit-oss/tantivy/issues/1876) (@fulmicoton)
- Unified access for fast fields over different cardinalities.
- Unified storage for typed and untyped fields.
- Move fastfield codecs into columnar. [#1782](https://github.com/quickwit-oss/tantivy/issues/1782) (@fulmicoton)
- Sparse dense index for optional values [#1716](https://github.com/quickwit-oss/tantivy/issues/1716) (@PSeitz)
- Switch to nanosecond precision in DateTime fastfield [#2016](https://github.com/quickwit-oss/tantivy/issues/2016) (@PSeitz)
- **Aggregation**
- Add `date_histogram` aggregation (only `fixed_interval` for now) [#1900](https://github.com/quickwit-oss/tantivy/issues/1900) (@PSeitz)
- Add `percentiles` aggregations [#1984](https://github.com/quickwit-oss/tantivy/issues/1984) (@PSeitz)
- [**breaking**] Drop JSON support on intermediate agg result (we use postcard as format in `quickwit` to send intermediate results) [#1992](https://github.com/quickwit-oss/tantivy/issues/1992) (@PSeitz)
- Set memory limit in bytes for aggregations after which they abort (Previously there was only the bucket limit) [#1942](https://github.com/quickwit-oss/tantivy/issues/1942)[#1957](https://github.com/quickwit-oss/tantivy/issues/1957)(@PSeitz)
- Add support for u64,i64,f64 fields in term aggregation [#1883](https://github.com/quickwit-oss/tantivy/issues/1883) (@PSeitz)
- Allow histogram bounds to be passed as Rfc3339 [#2076](https://github.com/quickwit-oss/tantivy/issues/2076) (@PSeitz)
- Add count, min, max, and sum aggregations [#1794](https://github.com/quickwit-oss/tantivy/issues/1794) (@guilload)
- Switch to Aggregation without serde_untagged => better deserialization errors. [#2003](https://github.com/quickwit-oss/tantivy/issues/2003) (@PSeitz)
- Switch to ms in histogram for date type (ES compatibility) [#2045](https://github.com/quickwit-oss/tantivy/issues/2045) (@PSeitz)
- Reduce term aggregation memory consumption [#2013](https://github.com/quickwit-oss/tantivy/issues/2013) (@PSeitz)
- Reduce agg memory consumption: Replace generic aggregation collector (which has a high memory requirement per instance) in aggregation tree with optimized versions behind a trait.
- Split term collection count and sub_agg (Faster term agg with less memory consumption for cases without sub-aggs) [#1921](https://github.com/quickwit-oss/tantivy/issues/1921) (@PSeitz)
- Schemaless aggregations: In combination with stacker tantivy supports now schemaless aggregations via the JSON type.
- Add aggregation support for JSON type [#1888](https://github.com/quickwit-oss/tantivy/issues/1888) (@PSeitz)
- Mixed types support on JSON fields in aggs [#1971](https://github.com/quickwit-oss/tantivy/issues/1971) (@PSeitz)
- Perf: Fetch blocks of vals in aggregation for all cardinality [#1950](https://github.com/quickwit-oss/tantivy/issues/1950) (@PSeitz)
- Allow histogram bounds to be passed as Rfc3339 [#2076](https://github.com/quickwit-oss/tantivy/issues/2076) (@PSeitz)
- `Searcher` with disabled scoring via `EnableScoring::Disabled` [#1780](https://github.com/quickwit-oss/tantivy/issues/1780) (@shikhar)
- Enable tokenizer on json fields [#2053](https://github.com/quickwit-oss/tantivy/issues/2053) (@PSeitz)
- Enforcing "NOT" and "-" queries consistency in UserInputAst [#1609](https://github.com/quickwit-oss/tantivy/issues/1609) (@bazhenov)
- Faster indexing
- Refactor tokenization pipeline to use GATs [#1924](https://github.com/quickwit-oss/tantivy/issues/1924) (@trinity-1686a)
- Faster term hash map [#2058](https://github.com/quickwit-oss/tantivy/issues/2058)[#1940](https://github.com/quickwit-oss/tantivy/issues/1940) (@PSeitz)
- tokenizer-api: reduce Tokenizer allocation overhead [#2062](https://github.com/quickwit-oss/tantivy/issues/2062) (@PSeitz)
- Refactor vint [#2010](https://github.com/quickwit-oss/tantivy/issues/2010) (@PSeitz)
- Faster search
- Work in batches of docs on the SegmentCollector (Only for cases without score for now) [#1937](https://github.com/quickwit-oss/tantivy/issues/1937) (@PSeitz)
- Faster fast field range queries using SIMD [#1954](https://github.com/quickwit-oss/tantivy/issues/1954) (@fulmicoton)
- Improve fast field range query performance [#1864](https://github.com/quickwit-oss/tantivy/issues/1864) (@PSeitz)
- Make BM25 scoring more flexible [#1855](https://github.com/quickwit-oss/tantivy/issues/1855) (@alexcole)
- Switch fs2 to fs4 as it is now unmaintained and does not support illumos [#1944](https://github.com/quickwit-oss/tantivy/issues/1944) (@Toasterson)
- Made BooleanWeight and BoostWeight public [#1991](https://github.com/quickwit-oss/tantivy/issues/1991) (@fulmicoton)
- Make index compatible with virtual drives on Windows [#1843](https://github.com/quickwit-oss/tantivy/issues/1843) (@gyk)
- Add stop words for Hungarian language [#2069](https://github.com/quickwit-oss/tantivy/issues/2069) (@tnxbutno)
- Auto downgrade index record option, instead of vint error [#1857](https://github.com/quickwit-oss/tantivy/issues/1857) (@PSeitz)
- Enable range query on fast field for u64 compatible types [#1762](https://github.com/quickwit-oss/tantivy/issues/1762) (@PSeitz) [#1876]
- sstable
- Isolating sstable and stacker in independant crates. [#1718](https://github.com/quickwit-oss/tantivy/issues/1718) (@fulmicoton)
- New sstable format [#1943](https://github.com/quickwit-oss/tantivy/issues/1943)[#1953](https://github.com/quickwit-oss/tantivy/issues/1953) (@trinity-1686a)
- Use DeltaReader directly to implement Dictionnary::ord_to_term [#1928](https://github.com/quickwit-oss/tantivy/issues/1928) (@trinity-1686a)
- Use DeltaReader directly to implement Dictionnary::term_ord [#1925](https://github.com/quickwit-oss/tantivy/issues/1925) (@trinity-1686a)
- Add seperate tokenizer manager for fast fields [#2019](https://github.com/quickwit-oss/tantivy/issues/2019) (@PSeitz)
- Make construction of LevenshteinAutomatonBuilder for FuzzyTermQuery instances lazy. [#1756](https://github.com/quickwit-oss/tantivy/issues/1756) (@adamreichold)
- Added support for madvise when opening an mmaped Index [#2036](https://github.com/quickwit-oss/tantivy/issues/2036) (@fulmicoton)
- Rename `DatePrecision` to `DateTimePrecision` [#2051](https://github.com/quickwit-oss/tantivy/issues/2051) (@guilload)
- Query Parser
- Quotation mark can now be used for phrase queries. [#2050](https://github.com/quickwit-oss/tantivy/issues/2050) (@fulmicoton)
- PhrasePrefixQuery is supported in the query parser via: `field:"phrase ter"*` [#2044](https://github.com/quickwit-oss/tantivy/issues/2044) (@adamreichold)
- Docs
- Update examples for literate docs [#1880](https://github.com/quickwit-oss/tantivy/issues/1880) (@PSeitz)
- Add ip field example [#1775](https://github.com/quickwit-oss/tantivy/issues/1775) (@PSeitz)
- Fix doc store cache documentation [#1821](https://github.com/quickwit-oss/tantivy/issues/1821) (@PSeitz)
- Fix BooleanQuery document [#1999](https://github.com/quickwit-oss/tantivy/issues/1999) (@RT_Enzyme)
- Update comments in the faceted search example [#1737](https://github.com/quickwit-oss/tantivy/issues/1737) (@DawChihLiou)
Tantivy 0.19
================================
#### Bugfixes

83
Cargo.toml
View File

@@ -1,6 +1,6 @@
[package]
name = "tantivy"
version = "0.19.0"
version = "0.22.0"
authors = ["Paul Masurel <paul.masurel@gmail.com>"]
license = "MIT"
categories = ["database-implementations", "data-structures"]
@@ -11,57 +11,57 @@ repository = "https://github.com/quickwit-oss/tantivy"
readme = "README.md"
keywords = ["search", "information", "retrieval"]
edition = "2021"
rust-version = "1.62"
rust-version = "1.63"
exclude = ["benches/*.json", "benches/*.txt"]
[dependencies]
oneshot = "0.1.5"
base64 = "0.21.0"
base64 = "0.22.0"
byteorder = "1.4.3"
crc32fast = "1.3.2"
once_cell = "1.10.0"
regex = { version = "1.5.5", default-features = false, features = ["std", "unicode"] }
aho-corasick = "0.7"
tantivy-fst = "0.4.0"
memmap2 = { version = "0.5.3", optional = true }
lz4_flex = { version = "0.10", default-features = false, features = ["checked-decode"], optional = true }
brotli = { version = "3.3.4", optional = true }
zstd = { version = "0.12", optional = true, default-features = false }
snap = { version = "1.0.5", optional = true }
aho-corasick = "1.0"
tantivy-fst = "0.5"
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 }
tempfile = { version = "3.3.0", optional = true }
log = "0.4.16"
serde = { version = "1.0.136", features = ["derive"] }
serde_json = "1.0.79"
num_cpus = "1.13.1"
fs2 = { version = "0.4.3", optional = true }
fs4 = { version = "0.8.0", optional = true }
levenshtein_automata = "0.2.1"
uuid = { version = "1.0.0", features = ["v4", "serde"] }
crossbeam-channel = "0.5.4"
rust-stemmers = "1.2.0"
downcast-rs = "1.2.0"
bitpacking = { version = "0.8.4", default-features = false, features = ["bitpacker4x"] }
census = "0.4.0"
bitpacking = { version = "0.9.2", default-features = false, features = ["bitpacker4x"] }
census = "0.4.2"
rustc-hash = "1.1.0"
thiserror = "1.0.30"
htmlescape = "0.3.1"
fail = "0.5.0"
murmurhash32 = "0.2.0"
fail = { version = "0.5.0", optional = true }
time = { version = "0.3.10", features = ["serde-well-known"] }
smallvec = "1.8.0"
rayon = "1.5.2"
lru = "0.9.0"
lru = "0.12.0"
fastdivide = "0.4.0"
itertools = "0.10.3"
itertools = "0.12.0"
measure_time = "0.8.2"
async-trait = "0.1.53"
arc-swap = "1.5.0"
columnar = { version="0.1", path="./columnar", package ="tantivy-columnar" }
sstable = { version="0.1", path="./sstable", package ="tantivy-sstable", optional = true }
stacker = { version="0.1", path="./stacker", package ="tantivy-stacker" }
query-grammar = { version= "0.19.0", path="./query-grammar", package = "tantivy-query-grammar" }
tantivy-bitpacker = { version= "0.3", path="./bitpacker" }
common = { version= "0.5", path = "./common/", package = "tantivy-common" }
tokenizer-api = { version="0.1", path="./tokenizer-api", package="tantivy-tokenizer-api" }
columnar = { version= "0.3", path="./columnar", package ="tantivy-columnar" }
sstable = { version= "0.3", path="./sstable", package ="tantivy-sstable", optional = true }
stacker = { version= "0.3", path="./stacker", package ="tantivy-stacker" }
query-grammar = { version= "0.22.0", path="./query-grammar", package = "tantivy-query-grammar" }
tantivy-bitpacker = { version= "0.6", path="./bitpacker" }
common = { version= "0.7", path = "./common/", package = "tantivy-common" }
tokenizer-api = { version= "0.3", path="./tokenizer-api", package="tantivy-tokenizer-api" }
sketches-ddsketch = { version = "0.2.1", features = ["use_serde"] }
futures-util = { version = "0.3.28", optional = true }
fnv = "1.0.7"
[target.'cfg(windows)'.dependencies]
winapi = "0.3.9"
@@ -72,12 +72,18 @@ maplit = "1.0.2"
matches = "0.1.9"
pretty_assertions = "1.2.1"
proptest = "1.0.0"
criterion = "0.4"
test-log = "0.2.10"
env_logger = "0.10.0"
pprof = { version = "0.11.0", features = ["flamegraph", "criterion"] }
futures = "0.3.21"
paste = "1.0.11"
more-asserts = "0.3.1"
rand_distr = "0.4.3"
time = { version = "0.3.10", features = ["serde-well-known", "macros"] }
postcard = { version = "1.0.4", features = [
"use-std",
], default-features = false }
[target.'cfg(not(windows))'.dev-dependencies]
criterion = { version = "0.5", default-features = false }
[dev-dependencies.fail]
version = "0.5.0"
@@ -88,24 +94,32 @@ opt-level = 3
debug = false
debug-assertions = false
[profile.bench]
opt-level = 3
debug = true
debug-assertions = false
[profile.test]
debug-assertions = true
overflow-checks = true
[features]
default = ["mmap", "stopwords", "lz4-compression"]
mmap = ["fs2", "tempfile", "memmap2"]
mmap = ["fs4", "tempfile", "memmap2"]
stopwords = []
brotli-compression = ["brotli"]
lz4-compression = ["lz4_flex"]
snappy-compression = ["snap"]
zstd-compression = ["zstd"]
failpoints = ["fail/failpoints"]
failpoints = ["fail", "fail/failpoints"]
unstable = [] # useful for benches.
quickwit = ["sstable"]
quickwit = ["sstable", "futures-util"]
# Compares only the hash of a string when indexing data.
# Increases indexing speed, but may lead to extremely rare missing terms, when there's a hash collision.
# Uses 64bit ahash.
compare_hash_only = ["stacker/compare_hash_only"]
[workspace]
members = ["query-grammar", "bitpacker", "common", "ownedbytes", "stacker", "sstable", "tokenizer-api", "columnar"]
@@ -120,7 +134,7 @@ members = ["query-grammar", "bitpacker", "common", "ownedbytes", "stacker", "sst
[[test]]
name = "failpoints"
path = "tests/failpoints/mod.rs"
required-features = ["fail/failpoints"]
required-features = ["failpoints"]
[[bench]]
name = "analyzer"
@@ -129,4 +143,3 @@ harness = false
[[bench]]
name = "index-bench"
harness = false

2
Makefile
View File

@@ -1,5 +1,5 @@
test:
echo "Run test only... No examples."
@echo "Run test only... No examples."
cargo test --tests --lib
fmt:

37
README.md
View File

@@ -5,19 +5,18 @@
[![License: MIT](https://img.shields.io/badge/License-MIT-yellow.svg)](https://opensource.org/licenses/MIT)
[![Crates.io](https://img.shields.io/crates/v/tantivy.svg)](https://crates.io/crates/tantivy)
![Tantivy](https://tantivy-search.github.io/logo/tantivy-logo.png)
<img src="https://tantivy-search.github.io/logo/tantivy-logo.png" alt="Tantivy, the fastest full-text search engine library written in Rust" height="250">
**Tantivy** is a **full-text search engine library** written in Rust.
## Fast full-text search engine library written in Rust
It is closer to [Apache Lucene](https://lucene.apache.org/) than to [Elasticsearch](https://www.elastic.co/products/elasticsearch) or [Apache Solr](https://lucene.apache.org/solr/) in the sense it is not
an off-the-shelf search engine server, but rather a crate that can be used
to build such a search engine.
**If you are looking for an alternative to Elasticsearch or Apache Solr, check out [Quickwit](https://github.com/quickwit-oss/quickwit), our distributed search engine built on top of Tantivy.**
Tantivy is closer to [Apache Lucene](https://lucene.apache.org/) than to [Elasticsearch](https://www.elastic.co/products/elasticsearch) or [Apache Solr](https://lucene.apache.org/solr/) in the sense it is not
an off-the-shelf search engine server, but rather a crate that can be used to build such a search engine.
Tantivy is, in fact, strongly inspired by Lucene's design.
If you are looking for an alternative to Elasticsearch or Apache Solr, check out [Quickwit](https://github.com/quickwit-oss/quickwit), our search engine built on top of Tantivy.
# Benchmark
## Benchmark
The following [benchmark](https://tantivy-search.github.io/bench/) breakdowns
performance for different types of queries/collections.
@@ -26,7 +25,9 @@ Your mileage WILL vary depending on the nature of queries and their load.
<img src="doc/assets/images/searchbenchmark.png">
# Features
Details about the benchmark can be found at this [repository](https://github.com/quickwit-oss/search-benchmark-game).
## Features
- Full-text search
- Configurable tokenizer (stemming available for 17 Latin languages) with third party support for Chinese ([tantivy-jieba](https://crates.io/crates/tantivy-jieba) and [cang-jie](https://crates.io/crates/cang-jie)), Japanese ([lindera](https://github.com/lindera-morphology/lindera-tantivy), [Vaporetto](https://crates.io/crates/vaporetto_tantivy), and [tantivy-tokenizer-tiny-segmenter](https://crates.io/crates/tantivy-tokenizer-tiny-segmenter)) and Korean ([lindera](https://github.com/lindera-morphology/lindera-tantivy) + [lindera-ko-dic-builder](https://github.com/lindera-morphology/lindera-ko-dic-builder))
@@ -42,7 +43,7 @@ Your mileage WILL vary depending on the nature of queries and their load.
- Single valued and multivalued u64, i64, and f64 fast fields (equivalent of doc values in Lucene)
- `&[u8]` fast fields
- Text, i64, u64, f64, dates, ip, bool, and hierarchical facet fields
- Compressed document store (LZ4, Zstd, None, Brotli, Snap)
- Compressed document store (LZ4, Zstd, None)
- Range queries
- Faceted search
- Configurable indexing (optional term frequency and position indexing)
@@ -52,11 +53,11 @@ Your mileage WILL vary depending on the nature of queries and their load.
- Searcher Warmer API
- Cheesy logo with a horse
## Non-features
### Non-features
Distributed search is out of the scope of Tantivy, but if you are looking for this feature, check out [Quickwit](https://github.com/quickwit-oss/quickwit/).
# Getting started
## Getting started
Tantivy works on stable Rust and supports Linux, macOS, and Windows.
@@ -66,7 +67,7 @@ index documents, and search via the CLI or a small server with a REST API.
It walks you through getting a Wikipedia search engine up and running in a few minutes.
- [Reference doc for the last released version](https://docs.rs/tantivy/)
# How can I support this project?
## How can I support this project?
There are many ways to support this project.
@@ -77,16 +78,16 @@ There are many ways to support this project.
- Contribute code (you can join [our Discord server](https://discord.gg/MT27AG5EVE))
- Talk about Tantivy around you
# Contributing code
## Contributing code
We use the GitHub Pull Request workflow: reference a GitHub ticket and/or include a comprehensive commit message when opening a PR.
Feel free to update CHANGELOG.md with your contribution.
## Tokenizer
### Tokenizer
When implementing a tokenizer for tantivy depend on the `tantivy-tokenizer-api` crate.
## Clone and build locally
### Clone and build locally
Tantivy compiles on stable Rust.
To check out and run tests, you can simply run:
@@ -97,7 +98,7 @@ cd tantivy
cargo test
```
# Companies Using Tantivy
## Companies Using Tantivy
<p align="left">
<img align="center" src="doc/assets/images/etsy.png" alt="Etsy" height="25" width="auto" />&nbsp;
@@ -109,7 +110,7 @@ cargo test
<img align="center" src="doc/assets/images/element-dark-theme.png#gh-dark-mode-only" alt="Element.io" height="25" width="auto" />
</p>
# FAQ
## FAQ
### Can I use Tantivy in other languages?

21
RELEASE.md Normal file
View File

@@ -0,0 +1,21 @@
# Release a new Tantivy Version
## Steps
1. Identify new packages in workspace since last release
2. Identify changed packages in workspace since last release
3. Bump version in `Cargo.toml` and their dependents for all changed packages
4. Update version of root `Cargo.toml`
5. Publish version starting with leaf nodes
6. Set git tag with new version
In conjucation with `cargo-release` Steps 1-4 (I'm not sure if the change detection works):
Set new packages to version 0.0.0
Replace prev-tag-name
```bash
cargo release --workspace --no-publish -v --prev-tag-name 0.19 --push-remote origin minor --no-tag --execute
```
no-tag or it will create tags for all the subpackages

23
appveyor.yml
View File

@@ -1,23 +0,0 @@
# Appveyor configuration template for Rust using rustup for Rust installation
# https://github.com/starkat99/appveyor-rust
os: Visual Studio 2015
environment:
matrix:
- channel: stable
target: x86_64-pc-windows-msvc
install:
- appveyor DownloadFile https://win.rustup.rs/ -FileName rustup-init.exe
- rustup-init -yv --default-toolchain %channel% --default-host %target%
- set PATH=%PATH%;%USERPROFILE%\.cargo\bin
- if defined msys_bits set PATH=%PATH%;C:\msys64\mingw%msys_bits%\bin
- rustc -vV
- cargo -vV
build: false
test_script:
- REM SET RUST_LOG=tantivy,test & cargo test --all --verbose --no-default-features --features lz4-compression --features mmap
- REM SET RUST_LOG=tantivy,test & cargo test test_store --verbose --no-default-features --features lz4-compression --features snappy-compression --features brotli-compression --features mmap
- REM SET RUST_BACKTRACE=1 & cargo build --examples

27
benches/analyzer.rs
View File

@@ -1,11 +1,13 @@
use criterion::{criterion_group, criterion_main, Criterion};
use tantivy::tokenizer::TokenizerManager;
use tantivy::tokenizer::{
LowerCaser, RemoveLongFilter, SimpleTokenizer, TextAnalyzer, TokenizerManager,
};
const ALICE_TXT: &str = include_str!("alice.txt");
pub fn criterion_benchmark(c: &mut Criterion) {
let tokenizer_manager = TokenizerManager::default();
let tokenizer = tokenizer_manager.get("default").unwrap();
let mut tokenizer = tokenizer_manager.get("default").unwrap();
c.bench_function("default-tokenize-alice", |b| {
b.iter(|| {
let mut word_count = 0;
@@ -16,7 +18,26 @@ pub fn criterion_benchmark(c: &mut Criterion) {
assert_eq!(word_count, 30_731);
})
});
let mut dynamic_analyzer = TextAnalyzer::builder(SimpleTokenizer::default())
.dynamic()
.filter_dynamic(RemoveLongFilter::limit(40))
.filter_dynamic(LowerCaser)
.build();
c.bench_function("dynamic-tokenize-alice", |b| {
b.iter(|| {
let mut word_count = 0;
let mut token_stream = dynamic_analyzer.token_stream(ALICE_TXT);
while token_stream.advance() {
word_count += 1;
}
assert_eq!(word_count, 30_731);
})
});
}
criterion_group!(benches, criterion_benchmark);
criterion_group! {
name = benches;
config = Criterion::default().sample_size(200);
targets = criterion_benchmark
}
criterion_main!(benches);

1000
benches/gh.json Normal file
View File

File diff suppressed because one or more lines are too long

267
benches/index-bench.rs
View File

@@ -1,10 +1,100 @@
use criterion::{criterion_group, criterion_main, Criterion};
use pprof::criterion::{Output, PProfProfiler};
use tantivy::schema::{INDEXED, STORED, STRING, TEXT};
use tantivy::Index;
use criterion::{criterion_group, criterion_main, BatchSize, Bencher, Criterion, Throughput};
use tantivy::schema::{TantivyDocument, FAST, INDEXED, STORED, STRING, TEXT};
use tantivy::{tokenizer, Index, IndexWriter};
const HDFS_LOGS: &str = include_str!("hdfs.json");
const NUM_REPEATS: usize = 2;
const GH_LOGS: &str = include_str!("gh.json");
const WIKI: &str = include_str!("wiki.json");
fn benchmark(
b: &mut Bencher,
input: &str,
schema: tantivy::schema::Schema,
commit: bool,
parse_json: bool,
is_dynamic: bool,
) {
if is_dynamic {
benchmark_dynamic_json(b, input, schema, commit, parse_json)
} else {
_benchmark(b, input, schema, commit, parse_json, |schema, doc_json| {
TantivyDocument::parse_json(&schema, doc_json).unwrap()
})
}
}
fn get_index(schema: tantivy::schema::Schema) -> Index {
let mut index = Index::create_in_ram(schema.clone());
let ff_tokenizer_manager = tokenizer::TokenizerManager::default();
ff_tokenizer_manager.register(
"raw",
tokenizer::TextAnalyzer::builder(tokenizer::RawTokenizer::default())
.filter(tokenizer::RemoveLongFilter::limit(255))
.build(),
);
index.set_fast_field_tokenizers(ff_tokenizer_manager.clone());
index
}
fn _benchmark(
b: &mut Bencher,
input: &str,
schema: tantivy::schema::Schema,
commit: bool,
include_json_parsing: bool,
create_doc: impl Fn(&tantivy::schema::Schema, &str) -> TantivyDocument,
) {
if include_json_parsing {
let lines: Vec<&str> = input.trim().split('\n').collect();
b.iter(|| {
let index = get_index(schema.clone());
let mut index_writer: IndexWriter =
index.writer_with_num_threads(1, 100_000_000).unwrap();
for doc_json in &lines {
let doc = create_doc(&schema, doc_json);
index_writer.add_document(doc).unwrap();
}
if commit {
index_writer.commit().unwrap();
}
})
} else {
let docs: Vec<_> = input
.trim()
.split('\n')
.map(|doc_json| create_doc(&schema, doc_json))
.collect();
b.iter_batched(
|| docs.clone(),
|docs| {
let index = get_index(schema.clone());
let mut index_writer: IndexWriter =
index.writer_with_num_threads(1, 100_000_000).unwrap();
for doc in docs {
index_writer.add_document(doc).unwrap();
}
if commit {
index_writer.commit().unwrap();
}
},
BatchSize::SmallInput,
)
}
}
fn benchmark_dynamic_json(
b: &mut Bencher,
input: &str,
schema: tantivy::schema::Schema,
commit: bool,
parse_json: bool,
) {
let json_field = schema.get_field("json").unwrap();
_benchmark(b, input, schema, commit, parse_json, |_schema, doc_json| {
let json_val: serde_json::Map<String, serde_json::Value> =
serde_json::from_str(doc_json).unwrap();
tantivy::doc!(json_field=>json_val)
})
}
pub fn hdfs_index_benchmark(c: &mut Criterion) {
let schema = {
@@ -14,7 +104,14 @@ pub fn hdfs_index_benchmark(c: &mut Criterion) {
schema_builder.add_text_field("severity", STRING);
schema_builder.build()
};
let schema_with_store = {
let schema_only_fast = {
let mut schema_builder = tantivy::schema::SchemaBuilder::new();
schema_builder.add_u64_field("timestamp", FAST);
schema_builder.add_text_field("body", FAST);
schema_builder.add_text_field("severity", FAST);
schema_builder.build()
};
let _schema_with_store = {
let mut schema_builder = tantivy::schema::SchemaBuilder::new();
schema_builder.add_u64_field("timestamp", INDEXED | STORED);
schema_builder.add_text_field("body", TEXT | STORED);
@@ -23,99 +120,99 @@ pub fn hdfs_index_benchmark(c: &mut Criterion) {
};
let dynamic_schema = {
let mut schema_builder = tantivy::schema::SchemaBuilder::new();
schema_builder.add_json_field("json", TEXT);
schema_builder.add_json_field("json", TEXT | FAST);
schema_builder.build()
};
let mut group = c.benchmark_group("index-hdfs");
group.throughput(Throughput::Bytes(HDFS_LOGS.len() as u64));
group.sample_size(20);
group.bench_function("index-hdfs-no-commit", |b| {
b.iter(|| {
let index = Index::create_in_ram(schema.clone());
let index_writer = index.writer_with_num_threads(1, 100_000_000).unwrap();
for _ in 0..NUM_REPEATS {
for doc_json in HDFS_LOGS.trim().split('\n') {
let doc = schema.parse_document(doc_json).unwrap();
index_writer.add_document(doc).unwrap();
}
let benches = [
("only-indexed-".to_string(), schema, false),
//("stored-".to_string(), _schema_with_store, false),
("only-fast-".to_string(), schema_only_fast, false),
("dynamic-".to_string(), dynamic_schema, true),
];
for (prefix, schema, is_dynamic) in benches {
for commit in [false, true] {
let suffix = if commit { "with-commit" } else { "no-commit" };
for parse_json in [false] {
// for parse_json in [false, true] {
let suffix = if parse_json {
format!("{}-with-json-parsing", suffix)
} else {
format!("{}", suffix)
};
let bench_name = format!("{}{}", prefix, suffix);
group.bench_function(bench_name, |b| {
benchmark(b, HDFS_LOGS, schema.clone(), commit, parse_json, is_dynamic)
});
}
})
}
}
}
pub fn gh_index_benchmark(c: &mut Criterion) {
let dynamic_schema = {
let mut schema_builder = tantivy::schema::SchemaBuilder::new();
schema_builder.add_json_field("json", TEXT | FAST);
schema_builder.build()
};
let dynamic_schema_fast = {
let mut schema_builder = tantivy::schema::SchemaBuilder::new();
schema_builder.add_json_field("json", FAST);
schema_builder.build()
};
let mut group = c.benchmark_group("index-gh");
group.throughput(Throughput::Bytes(GH_LOGS.len() as u64));
group.bench_function("index-gh-no-commit", |b| {
benchmark_dynamic_json(b, GH_LOGS, dynamic_schema.clone(), false, false)
});
group.bench_function("index-hdfs-with-commit", |b| {
b.iter(|| {
let index = Index::create_in_ram(schema.clone());
let mut index_writer = index.writer_with_num_threads(1, 100_000_000).unwrap();
for _ in 0..NUM_REPEATS {
for doc_json in HDFS_LOGS.trim().split('\n') {
let doc = schema.parse_document(doc_json).unwrap();
index_writer.add_document(doc).unwrap();
}
}
index_writer.commit().unwrap();
})
group.bench_function("index-gh-fast", |b| {
benchmark_dynamic_json(b, GH_LOGS, dynamic_schema_fast.clone(), false, false)
});
group.bench_function("index-hdfs-no-commit-with-docstore", |b| {
b.iter(|| {
let index = Index::create_in_ram(schema_with_store.clone());
let index_writer = index.writer_with_num_threads(1, 100_000_000).unwrap();
for _ in 0..NUM_REPEATS {
for doc_json in HDFS_LOGS.trim().split('\n') {
let doc = schema.parse_document(doc_json).unwrap();
index_writer.add_document(doc).unwrap();
}
}
})
group.bench_function("index-gh-fast-with-commit", |b| {
benchmark_dynamic_json(b, GH_LOGS, dynamic_schema_fast.clone(), true, false)
});
group.bench_function("index-hdfs-with-commit-with-docstore", |b| {
b.iter(|| {
let index = Index::create_in_ram(schema_with_store.clone());
let mut index_writer = index.writer_with_num_threads(1, 100_000_000).unwrap();
for _ in 0..NUM_REPEATS {
for doc_json in HDFS_LOGS.trim().split('\n') {
let doc = schema.parse_document(doc_json).unwrap();
index_writer.add_document(doc).unwrap();
}
}
index_writer.commit().unwrap();
})
}
pub fn wiki_index_benchmark(c: &mut Criterion) {
let dynamic_schema = {
let mut schema_builder = tantivy::schema::SchemaBuilder::new();
schema_builder.add_json_field("json", TEXT | FAST);
schema_builder.build()
};
let mut group = c.benchmark_group("index-wiki");
group.throughput(Throughput::Bytes(WIKI.len() as u64));
group.bench_function("index-wiki-no-commit", |b| {
benchmark_dynamic_json(b, WIKI, dynamic_schema.clone(), false, false)
});
group.bench_function("index-hdfs-no-commit-json-without-docstore", |b| {
b.iter(|| {
let index = Index::create_in_ram(dynamic_schema.clone());
let json_field = dynamic_schema.get_field("json").unwrap();
let mut index_writer = index.writer_with_num_threads(1, 100_000_000).unwrap();
for _ in 0..NUM_REPEATS {
for doc_json in HDFS_LOGS.trim().split('\n') {
let json_val: serde_json::Map<String, serde_json::Value> =
serde_json::from_str(doc_json).unwrap();
let doc = tantivy::doc!(json_field=>json_val);
index_writer.add_document(doc).unwrap();
}
}
index_writer.commit().unwrap();
})
});
group.bench_function("index-hdfs-with-commit-json-without-docstore", |b| {
b.iter(|| {
let index = Index::create_in_ram(dynamic_schema.clone());
let json_field = dynamic_schema.get_field("json").unwrap();
let mut index_writer = index.writer_with_num_threads(1, 100_000_000).unwrap();
for _ in 0..NUM_REPEATS {
for doc_json in HDFS_LOGS.trim().split('\n') {
let json_val: serde_json::Map<String, serde_json::Value> =
serde_json::from_str(doc_json).unwrap();
let doc = tantivy::doc!(json_field=>json_val);
index_writer.add_document(doc).unwrap();
}
}
index_writer.commit().unwrap();
})
group.bench_function("index-wiki-with-commit", |b| {
benchmark_dynamic_json(b, WIKI, dynamic_schema.clone(), true, false)
});
}
criterion_group! {
name = benches;
config = Criterion::default().with_profiler(PProfProfiler::new(100, Output::Flamegraph(None)));
config = Criterion::default();
targets = hdfs_index_benchmark
}
criterion_main!(benches);
criterion_group! {
name = gh_benches;
config = Criterion::default();
targets = gh_index_benchmark
}
criterion_group! {
name = wiki_benches;
config = Criterion::default();
targets = wiki_index_benchmark
}
criterion_main!(benches, gh_benches, wiki_benches);

1000
benches/wiki.json Normal file
View File

File diff suppressed because one or more lines are too long

3
bitpacker/Cargo.toml
View File

@@ -1,6 +1,6 @@
[package]
name = "tantivy-bitpacker"
version = "0.3.0"
version = "0.6.0"
edition = "2021"
authors = ["Paul Masurel <paul.masurel@gmail.com>"]
license = "MIT"
@@ -15,6 +15,7 @@ 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"] }
[dev-dependencies]
rand = "0.8"

178
bitpacker/src/bitpacker.rs
View File

@@ -1,10 +1,13 @@
use std::convert::TryInto;
use std::io;
use std::ops::{Range, RangeInclusive};
use bitpacking::{BitPacker as ExternalBitPackerTrait, BitPacker1x};
pub struct BitPacker {
mini_buffer: u64,
mini_buffer_written: usize,
}
impl Default for BitPacker {
fn default() -> Self {
BitPacker::new()
@@ -118,6 +121,125 @@ impl BitUnpacker {
let val_shifted = val_unshifted_unmasked >> bit_shift;
val_shifted & self.mask
}
// Decodes the range of bitpacked `u32` values with idx
// in [start_idx, start_idx + output.len()).
//
// #Panics
//
// This methods panics if `num_bits` is > 32.
fn get_batch_u32s(&self, start_idx: u32, data: &[u8], output: &mut [u32]) {
assert!(
self.bit_width() <= 32,
"Bitwidth must be <= 32 to use this method."
);
let end_idx = start_idx + output.len() as u32;
let end_bit_read = end_idx * self.num_bits;
let end_byte_read = (end_bit_read + 7) / 8;
assert!(
end_byte_read as usize <= 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;
}
};
// We use an unrolled routine to decode 32 values at once.
// We therefore decompose our range of values to decode into three ranges:
// - Entrance ramp: [start_idx, fast_track_start) (up to 31 values)
// - Highway: [fast_track_start, fast_track_end) (a length multiple of 32s)
// - Exit ramp: [fast_track_end, start_idx + output.len()) (up to 31 values)
// 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 = 8 - (start_idx % 8) % 8;
let highway_start: u32 = start_idx + entrance_ramp_len;
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: 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 * self.num_bits) as usize / 8;
let mut output_cursor = (highway_start - start_idx) as usize;
for _ in 0..num_blocks {
offset += BitPacker1x.decompress(
&data[offset..],
&mut output[output_cursor..],
self.num_bits as u8,
);
output_cursor += 32;
}
// Exit ramp
let highway_end = highway_start + num_blocks * BitPacker1x::BLOCK_LEN as u32;
get_batch_ramp(highway_end, &mut output[output_cursor..]);
}
pub fn get_ids_for_value_range(
&self,
range: RangeInclusive<u64>,
id_range: Range<u32>,
data: &[u8],
positions: &mut Vec<u32>,
) {
if self.bit_width() > 32 {
self.get_ids_for_value_range_slow(range, id_range, 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)
}
}
fn get_ids_for_value_range_slow(
&self,
range: RangeInclusive<u64>,
id_range: Range<u32>,
data: &[u8],
positions: &mut Vec<u32>,
) {
positions.clear();
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);
if range.contains(&val) {
positions.push(i);
}
}
}
fn get_ids_for_value_range_fast(
&self,
value_range: RangeInclusive<u32>,
id_range: Range<u32>,
data: &[u8],
positions: &mut Vec<u32>,
) {
positions.resize(id_range.len(), 0u32);
self.get_batch_u32s(id_range.start, data, positions);
crate::filter_vec::filter_vec_in_place(value_range, id_range.start, positions)
}
}
#[cfg(test)]
@@ -200,4 +322,58 @@ mod test {
test_bitpacker_aux(num_bits, &vals);
}
}
#[test]
#[should_panic]
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[..]);
}
#[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[..]);
}
#[test]
#[should_panic]
fn test_get_batch_panics_when_off_scope() {
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[..]);
}
proptest::proptest! {
#[test]
fn test_get_batch_u32s_proptest(num_bits in 0u8..=32u8) {
let mask =
if num_bits == 32u8 {
u32::MAX
} else {
(1u32 << num_bits) - 1
};
let mut buffer: Vec<u8> = Vec::new();
let mut bitpacker = BitPacker::new();
for val in 0..100 {
bitpacker.write(val & mask as u64, num_bits, &mut buffer).unwrap();
}
bitpacker.flush(&mut buffer).unwrap();
let bitunpacker = BitUnpacker::new(num_bits);
let mut output: Vec<u32> = Vec::new();
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);
for i in 0..len {
let expected = (start_idx + i as u32) & mask;
assert_eq!(output[i], expected);
}
}
}
}
}
}

4
bitpacker/src/blocked_bitpacker.rs
View File

@@ -64,10 +64,8 @@ fn mem_usage<T>(items: &Vec<T>) -> usize {
impl BlockedBitpacker {
pub fn new() -> Self {
let mut compressed_blocks = vec![];
compressed_blocks.resize(8, 0);
Self {
compressed_blocks,
compressed_blocks: vec![0; 8],
buffer: vec![],
offset_and_bits: vec![],
}

365
bitpacker/src/filter_vec/avx2.rs Normal file
View File

@@ -0,0 +1,365 @@
//! SIMD filtering of a vector as described in the following blog post.
//! <https://quickwit.io/blog/filtering%20a%20vector%20with%20simd%20instructions%20avx-2%20and%20avx-512>
use std::arch::x86_64::{
__m256i as DataType, _mm256_add_epi32 as op_add, _mm256_cmpgt_epi32 as op_greater,
_mm256_lddqu_si256 as load_unaligned, _mm256_or_si256 as op_or, _mm256_set1_epi32 as set1,
_mm256_storeu_si256 as store_unaligned, _mm256_xor_si256 as op_xor, *,
};
use std::ops::RangeInclusive;
const NUM_LANES: usize = 8;
const HIGHEST_BIT: u32 = 1 << 31;
#[inline]
fn u32_to_i32(val: u32) -> i32 {
(val ^ HIGHEST_BIT) as 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)
}
pub fn filter_vec_in_place(range: RangeInclusive<u32>, offset: u32, output: &mut Vec<u32>) {
// We use a monotonic mapping from u32 to i32 to make the comparison possible in AVX2.
let range_i32: RangeInclusive<i32> = u32_to_i32(*range.start())..=u32_to_i32(*range.end());
let num_words = output.len() / NUM_LANES;
let mut output_len = unsafe {
filter_vec_avx2_aux(
output.as_ptr() as *const __m256i,
range_i32,
output.as_mut_ptr(),
offset,
num_words,
)
};
let reminder_start = num_words * NUM_LANES;
for i in reminder_start..output.len() {
let val = output[i];
output[output_len] = offset + i as u32;
output_len += if range.contains(&val) { 1 } else { 0 };
}
output.truncate(output_len);
}
#[target_feature(enable = "avx2")]
unsafe fn filter_vec_avx2_aux(
mut input: *const __m256i,
range: RangeInclusive<i32>,
output: *mut u32,
offset: u32,
num_words: usize,
) -> usize {
let mut output_tail = output;
let range_simd = set1(*range.start())..=set1(*range.end());
let mut ids = from_u32x8([
offset,
offset + 1,
offset + 2,
offset + 3,
offset + 4,
offset + 5,
offset + 6,
offset + 7,
]);
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);
}
output_tail.offset_from(output) as usize
}
#[inline]
#[target_feature(enable = "avx2")]
unsafe fn compact(data: DataType, mask: u8) -> DataType {
let vperm_mask = MASK_TO_PERMUTATION[mask as usize];
_mm256_permutevar8x32_epi32(data, vperm_mask)
}
#[inline]
#[target_feature(enable = "avx2")]
unsafe fn compute_filter_bitset(val: __m256i, range: std::ops::RangeInclusive<__m256i>) -> u8 {
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
}
union U8x32 {
vector: DataType,
vals: [u32; NUM_LANES],
}
const fn from_u32x8(vals: [u32; NUM_LANES]) -> DataType {
unsafe { U8x32 { vals }.vector }
}
const MASK_TO_PERMUTATION: [DataType; 256] = [
from_u32x8([0, 0, 0, 0, 0, 0, 0, 0]),
from_u32x8([0, 0, 0, 0, 0, 0, 0, 0]),
from_u32x8([1, 0, 0, 0, 0, 0, 0, 0]),
from_u32x8([0, 1, 0, 0, 0, 0, 0, 0]),
from_u32x8([2, 0, 0, 0, 0, 0, 0, 0]),
from_u32x8([0, 2, 0, 0, 0, 0, 0, 0]),
from_u32x8([1, 2, 0, 0, 0, 0, 0, 0]),
from_u32x8([0, 1, 2, 0, 0, 0, 0, 0]),
from_u32x8([3, 0, 0, 0, 0, 0, 0, 0]),
from_u32x8([0, 3, 0, 0, 0, 0, 0, 0]),
from_u32x8([1, 3, 0, 0, 0, 0, 0, 0]),
from_u32x8([0, 1, 3, 0, 0, 0, 0, 0]),
from_u32x8([2, 3, 0, 0, 0, 0, 0, 0]),
from_u32x8([0, 2, 3, 0, 0, 0, 0, 0]),
from_u32x8([1, 2, 3, 0, 0, 0, 0, 0]),
from_u32x8([0, 1, 2, 3, 0, 0, 0, 0]),
from_u32x8([4, 0, 0, 0, 0, 0, 0, 0]),
from_u32x8([0, 4, 0, 0, 0, 0, 0, 0]),
from_u32x8([1, 4, 0, 0, 0, 0, 0, 0]),
from_u32x8([0, 1, 4, 0, 0, 0, 0, 0]),
from_u32x8([2, 4, 0, 0, 0, 0, 0, 0]),
from_u32x8([0, 2, 4, 0, 0, 0, 0, 0]),
from_u32x8([1, 2, 4, 0, 0, 0, 0, 0]),
from_u32x8([0, 1, 2, 4, 0, 0, 0, 0]),
from_u32x8([3, 4, 0, 0, 0, 0, 0, 0]),
from_u32x8([0, 3, 4, 0, 0, 0, 0, 0]),
from_u32x8([1, 3, 4, 0, 0, 0, 0, 0]),
from_u32x8([0, 1, 3, 4, 0, 0, 0, 0]),
from_u32x8([2, 3, 4, 0, 0, 0, 0, 0]),
from_u32x8([0, 2, 3, 4, 0, 0, 0, 0]),
from_u32x8([1, 2, 3, 4, 0, 0, 0, 0]),
from_u32x8([0, 1, 2, 3, 4, 0, 0, 0]),
from_u32x8([5, 0, 0, 0, 0, 0, 0, 0]),
from_u32x8([0, 5, 0, 0, 0, 0, 0, 0]),
from_u32x8([1, 5, 0, 0, 0, 0, 0, 0]),
from_u32x8([0, 1, 5, 0, 0, 0, 0, 0]),
from_u32x8([2, 5, 0, 0, 0, 0, 0, 0]),
from_u32x8([0, 2, 5, 0, 0, 0, 0, 0]),
from_u32x8([1, 2, 5, 0, 0, 0, 0, 0]),
from_u32x8([0, 1, 2, 5, 0, 0, 0, 0]),
from_u32x8([3, 5, 0, 0, 0, 0, 0, 0]),
from_u32x8([0, 3, 5, 0, 0, 0, 0, 0]),
from_u32x8([1, 3, 5, 0, 0, 0, 0, 0]),
from_u32x8([0, 1, 3, 5, 0, 0, 0, 0]),
from_u32x8([2, 3, 5, 0, 0, 0, 0, 0]),
from_u32x8([0, 2, 3, 5, 0, 0, 0, 0]),
from_u32x8([1, 2, 3, 5, 0, 0, 0, 0]),
from_u32x8([0, 1, 2, 3, 5, 0, 0, 0]),
from_u32x8([4, 5, 0, 0, 0, 0, 0, 0]),
from_u32x8([0, 4, 5, 0, 0, 0, 0, 0]),
from_u32x8([1, 4, 5, 0, 0, 0, 0, 0]),
from_u32x8([0, 1, 4, 5, 0, 0, 0, 0]),
from_u32x8([2, 4, 5, 0, 0, 0, 0, 0]),
from_u32x8([0, 2, 4, 5, 0, 0, 0, 0]),
from_u32x8([1, 2, 4, 5, 0, 0, 0, 0]),
from_u32x8([0, 1, 2, 4, 5, 0, 0, 0]),
from_u32x8([3, 4, 5, 0, 0, 0, 0, 0]),
from_u32x8([0, 3, 4, 5, 0, 0, 0, 0]),
from_u32x8([1, 3, 4, 5, 0, 0, 0, 0]),
from_u32x8([0, 1, 3, 4, 5, 0, 0, 0]),
from_u32x8([2, 3, 4, 5, 0, 0, 0, 0]),
from_u32x8([0, 2, 3, 4, 5, 0, 0, 0]),
from_u32x8([1, 2, 3, 4, 5, 0, 0, 0]),
from_u32x8([0, 1, 2, 3, 4, 5, 0, 0]),
from_u32x8([6, 0, 0, 0, 0, 0, 0, 0]),
from_u32x8([0, 6, 0, 0, 0, 0, 0, 0]),
from_u32x8([1, 6, 0, 0, 0, 0, 0, 0]),
from_u32x8([0, 1, 6, 0, 0, 0, 0, 0]),
from_u32x8([2, 6, 0, 0, 0, 0, 0, 0]),
from_u32x8([0, 2, 6, 0, 0, 0, 0, 0]),
from_u32x8([1, 2, 6, 0, 0, 0, 0, 0]),
from_u32x8([0, 1, 2, 6, 0, 0, 0, 0]),
from_u32x8([3, 6, 0, 0, 0, 0, 0, 0]),
from_u32x8([0, 3, 6, 0, 0, 0, 0, 0]),
from_u32x8([1, 3, 6, 0, 0, 0, 0, 0]),
from_u32x8([0, 1, 3, 6, 0, 0, 0, 0]),
from_u32x8([2, 3, 6, 0, 0, 0, 0, 0]),
from_u32x8([0, 2, 3, 6, 0, 0, 0, 0]),
from_u32x8([1, 2, 3, 6, 0, 0, 0, 0]),
from_u32x8([0, 1, 2, 3, 6, 0, 0, 0]),
from_u32x8([4, 6, 0, 0, 0, 0, 0, 0]),
from_u32x8([0, 4, 6, 0, 0, 0, 0, 0]),
from_u32x8([1, 4, 6, 0, 0, 0, 0, 0]),
from_u32x8([0, 1, 4, 6, 0, 0, 0, 0]),
from_u32x8([2, 4, 6, 0, 0, 0, 0, 0]),
from_u32x8([0, 2, 4, 6, 0, 0, 0, 0]),
from_u32x8([1, 2, 4, 6, 0, 0, 0, 0]),
from_u32x8([0, 1, 2, 4, 6, 0, 0, 0]),
from_u32x8([3, 4, 6, 0, 0, 0, 0, 0]),
from_u32x8([0, 3, 4, 6, 0, 0, 0, 0]),
from_u32x8([1, 3, 4, 6, 0, 0, 0, 0]),
from_u32x8([0, 1, 3, 4, 6, 0, 0, 0]),
from_u32x8([2, 3, 4, 6, 0, 0, 0, 0]),
from_u32x8([0, 2, 3, 4, 6, 0, 0, 0]),
from_u32x8([1, 2, 3, 4, 6, 0, 0, 0]),
from_u32x8([0, 1, 2, 3, 4, 6, 0, 0]),
from_u32x8([5, 6, 0, 0, 0, 0, 0, 0]),
from_u32x8([0, 5, 6, 0, 0, 0, 0, 0]),
from_u32x8([1, 5, 6, 0, 0, 0, 0, 0]),
from_u32x8([0, 1, 5, 6, 0, 0, 0, 0]),
from_u32x8([2, 5, 6, 0, 0, 0, 0, 0]),
from_u32x8([0, 2, 5, 6, 0, 0, 0, 0]),
from_u32x8([1, 2, 5, 6, 0, 0, 0, 0]),
from_u32x8([0, 1, 2, 5, 6, 0, 0, 0]),
from_u32x8([3, 5, 6, 0, 0, 0, 0, 0]),
from_u32x8([0, 3, 5, 6, 0, 0, 0, 0]),
from_u32x8([1, 3, 5, 6, 0, 0, 0, 0]),
from_u32x8([0, 1, 3, 5, 6, 0, 0, 0]),
from_u32x8([2, 3, 5, 6, 0, 0, 0, 0]),
from_u32x8([0, 2, 3, 5, 6, 0, 0, 0]),
from_u32x8([1, 2, 3, 5, 6, 0, 0, 0]),
from_u32x8([0, 1, 2, 3, 5, 6, 0, 0]),
from_u32x8([4, 5, 6, 0, 0, 0, 0, 0]),
from_u32x8([0, 4, 5, 6, 0, 0, 0, 0]),
from_u32x8([1, 4, 5, 6, 0, 0, 0, 0]),
from_u32x8([0, 1, 4, 5, 6, 0, 0, 0]),
from_u32x8([2, 4, 5, 6, 0, 0, 0, 0]),
from_u32x8([0, 2, 4, 5, 6, 0, 0, 0]),
from_u32x8([1, 2, 4, 5, 6, 0, 0, 0]),
from_u32x8([0, 1, 2, 4, 5, 6, 0, 0]),
from_u32x8([3, 4, 5, 6, 0, 0, 0, 0]),
from_u32x8([0, 3, 4, 5, 6, 0, 0, 0]),
from_u32x8([1, 3, 4, 5, 6, 0, 0, 0]),
from_u32x8([0, 1, 3, 4, 5, 6, 0, 0]),
from_u32x8([2, 3, 4, 5, 6, 0, 0, 0]),
from_u32x8([0, 2, 3, 4, 5, 6, 0, 0]),
from_u32x8([1, 2, 3, 4, 5, 6, 0, 0]),
from_u32x8([0, 1, 2, 3, 4, 5, 6, 0]),
from_u32x8([7, 0, 0, 0, 0, 0, 0, 0]),
from_u32x8([0, 7, 0, 0, 0, 0, 0, 0]),
from_u32x8([1, 7, 0, 0, 0, 0, 0, 0]),
from_u32x8([0, 1, 7, 0, 0, 0, 0, 0]),
from_u32x8([2, 7, 0, 0, 0, 0, 0, 0]),
from_u32x8([0, 2, 7, 0, 0, 0, 0, 0]),
from_u32x8([1, 2, 7, 0, 0, 0, 0, 0]),
from_u32x8([0, 1, 2, 7, 0, 0, 0, 0]),
from_u32x8([3, 7, 0, 0, 0, 0, 0, 0]),
from_u32x8([0, 3, 7, 0, 0, 0, 0, 0]),
from_u32x8([1, 3, 7, 0, 0, 0, 0, 0]),
from_u32x8([0, 1, 3, 7, 0, 0, 0, 0]),
from_u32x8([2, 3, 7, 0, 0, 0, 0, 0]),
from_u32x8([0, 2, 3, 7, 0, 0, 0, 0]),
from_u32x8([1, 2, 3, 7, 0, 0, 0, 0]),
from_u32x8([0, 1, 2, 3, 7, 0, 0, 0]),
from_u32x8([4, 7, 0, 0, 0, 0, 0, 0]),
from_u32x8([0, 4, 7, 0, 0, 0, 0, 0]),
from_u32x8([1, 4, 7, 0, 0, 0, 0, 0]),
from_u32x8([0, 1, 4, 7, 0, 0, 0, 0]),
from_u32x8([2, 4, 7, 0, 0, 0, 0, 0]),
from_u32x8([0, 2, 4, 7, 0, 0, 0, 0]),
from_u32x8([1, 2, 4, 7, 0, 0, 0, 0]),
from_u32x8([0, 1, 2, 4, 7, 0, 0, 0]),
from_u32x8([3, 4, 7, 0, 0, 0, 0, 0]),
from_u32x8([0, 3, 4, 7, 0, 0, 0, 0]),
from_u32x8([1, 3, 4, 7, 0, 0, 0, 0]),
from_u32x8([0, 1, 3, 4, 7, 0, 0, 0]),
from_u32x8([2, 3, 4, 7, 0, 0, 0, 0]),
from_u32x8([0, 2, 3, 4, 7, 0, 0, 0]),
from_u32x8([1, 2, 3, 4, 7, 0, 0, 0]),
from_u32x8([0, 1, 2, 3, 4, 7, 0, 0]),
from_u32x8([5, 7, 0, 0, 0, 0, 0, 0]),
from_u32x8([0, 5, 7, 0, 0, 0, 0, 0]),
from_u32x8([1, 5, 7, 0, 0, 0, 0, 0]),
from_u32x8([0, 1, 5, 7, 0, 0, 0, 0]),
from_u32x8([2, 5, 7, 0, 0, 0, 0, 0]),
from_u32x8([0, 2, 5, 7, 0, 0, 0, 0]),
from_u32x8([1, 2, 5, 7, 0, 0, 0, 0]),
from_u32x8([0, 1, 2, 5, 7, 0, 0, 0]),
from_u32x8([3, 5, 7, 0, 0, 0, 0, 0]),
from_u32x8([0, 3, 5, 7, 0, 0, 0, 0]),
from_u32x8([1, 3, 5, 7, 0, 0, 0, 0]),
from_u32x8([0, 1, 3, 5, 7, 0, 0, 0]),
from_u32x8([2, 3, 5, 7, 0, 0, 0, 0]),
from_u32x8([0, 2, 3, 5, 7, 0, 0, 0]),
from_u32x8([1, 2, 3, 5, 7, 0, 0, 0]),
from_u32x8([0, 1, 2, 3, 5, 7, 0, 0]),
from_u32x8([4, 5, 7, 0, 0, 0, 0, 0]),
from_u32x8([0, 4, 5, 7, 0, 0, 0, 0]),
from_u32x8([1, 4, 5, 7, 0, 0, 0, 0]),
from_u32x8([0, 1, 4, 5, 7, 0, 0, 0]),
from_u32x8([2, 4, 5, 7, 0, 0, 0, 0]),
from_u32x8([0, 2, 4, 5, 7, 0, 0, 0]),
from_u32x8([1, 2, 4, 5, 7, 0, 0, 0]),
from_u32x8([0, 1, 2, 4, 5, 7, 0, 0]),
from_u32x8([3, 4, 5, 7, 0, 0, 0, 0]),
from_u32x8([0, 3, 4, 5, 7, 0, 0, 0]),
from_u32x8([1, 3, 4, 5, 7, 0, 0, 0]),
from_u32x8([0, 1, 3, 4, 5, 7, 0, 0]),
from_u32x8([2, 3, 4, 5, 7, 0, 0, 0]),
from_u32x8([0, 2, 3, 4, 5, 7, 0, 0]),
from_u32x8([1, 2, 3, 4, 5, 7, 0, 0]),
from_u32x8([0, 1, 2, 3, 4, 5, 7, 0]),
from_u32x8([6, 7, 0, 0, 0, 0, 0, 0]),
from_u32x8([0, 6, 7, 0, 0, 0, 0, 0]),
from_u32x8([1, 6, 7, 0, 0, 0, 0, 0]),
from_u32x8([0, 1, 6, 7, 0, 0, 0, 0]),
from_u32x8([2, 6, 7, 0, 0, 0, 0, 0]),
from_u32x8([0, 2, 6, 7, 0, 0, 0, 0]),
from_u32x8([1, 2, 6, 7, 0, 0, 0, 0]),
from_u32x8([0, 1, 2, 6, 7, 0, 0, 0]),
from_u32x8([3, 6, 7, 0, 0, 0, 0, 0]),
from_u32x8([0, 3, 6, 7, 0, 0, 0, 0]),
from_u32x8([1, 3, 6, 7, 0, 0, 0, 0]),
from_u32x8([0, 1, 3, 6, 7, 0, 0, 0]),
from_u32x8([2, 3, 6, 7, 0, 0, 0, 0]),
from_u32x8([0, 2, 3, 6, 7, 0, 0, 0]),
from_u32x8([1, 2, 3, 6, 7, 0, 0, 0]),
from_u32x8([0, 1, 2, 3, 6, 7, 0, 0]),
from_u32x8([4, 6, 7, 0, 0, 0, 0, 0]),
from_u32x8([0, 4, 6, 7, 0, 0, 0, 0]),
from_u32x8([1, 4, 6, 7, 0, 0, 0, 0]),
from_u32x8([0, 1, 4, 6, 7, 0, 0, 0]),
from_u32x8([2, 4, 6, 7, 0, 0, 0, 0]),
from_u32x8([0, 2, 4, 6, 7, 0, 0, 0]),
from_u32x8([1, 2, 4, 6, 7, 0, 0, 0]),
from_u32x8([0, 1, 2, 4, 6, 7, 0, 0]),
from_u32x8([3, 4, 6, 7, 0, 0, 0, 0]),
from_u32x8([0, 3, 4, 6, 7, 0, 0, 0]),
from_u32x8([1, 3, 4, 6, 7, 0, 0, 0]),
from_u32x8([0, 1, 3, 4, 6, 7, 0, 0]),
from_u32x8([2, 3, 4, 6, 7, 0, 0, 0]),
from_u32x8([0, 2, 3, 4, 6, 7, 0, 0]),
from_u32x8([1, 2, 3, 4, 6, 7, 0, 0]),
from_u32x8([0, 1, 2, 3, 4, 6, 7, 0]),
from_u32x8([5, 6, 7, 0, 0, 0, 0, 0]),
from_u32x8([0, 5, 6, 7, 0, 0, 0, 0]),
from_u32x8([1, 5, 6, 7, 0, 0, 0, 0]),
from_u32x8([0, 1, 5, 6, 7, 0, 0, 0]),
from_u32x8([2, 5, 6, 7, 0, 0, 0, 0]),
from_u32x8([0, 2, 5, 6, 7, 0, 0, 0]),
from_u32x8([1, 2, 5, 6, 7, 0, 0, 0]),
from_u32x8([0, 1, 2, 5, 6, 7, 0, 0]),
from_u32x8([3, 5, 6, 7, 0, 0, 0, 0]),
from_u32x8([0, 3, 5, 6, 7, 0, 0, 0]),
from_u32x8([1, 3, 5, 6, 7, 0, 0, 0]),
from_u32x8([0, 1, 3, 5, 6, 7, 0, 0]),
from_u32x8([2, 3, 5, 6, 7, 0, 0, 0]),
from_u32x8([0, 2, 3, 5, 6, 7, 0, 0]),
from_u32x8([1, 2, 3, 5, 6, 7, 0, 0]),
from_u32x8([0, 1, 2, 3, 5, 6, 7, 0]),
from_u32x8([4, 5, 6, 7, 0, 0, 0, 0]),
from_u32x8([0, 4, 5, 6, 7, 0, 0, 0]),
from_u32x8([1, 4, 5, 6, 7, 0, 0, 0]),
from_u32x8([0, 1, 4, 5, 6, 7, 0, 0]),
from_u32x8([2, 4, 5, 6, 7, 0, 0, 0]),
from_u32x8([0, 2, 4, 5, 6, 7, 0, 0]),
from_u32x8([1, 2, 4, 5, 6, 7, 0, 0]),
from_u32x8([0, 1, 2, 4, 5, 6, 7, 0]),
from_u32x8([3, 4, 5, 6, 7, 0, 0, 0]),
from_u32x8([0, 3, 4, 5, 6, 7, 0, 0]),
from_u32x8([1, 3, 4, 5, 6, 7, 0, 0]),
from_u32x8([0, 1, 3, 4, 5, 6, 7, 0]),
from_u32x8([2, 3, 4, 5, 6, 7, 0, 0]),
from_u32x8([0, 2, 3, 4, 5, 6, 7, 0]),
from_u32x8([1, 2, 3, 4, 5, 6, 7, 0]),
from_u32x8([0, 1, 2, 3, 4, 5, 6, 7]),
];

165
bitpacker/src/filter_vec/mod.rs Normal file
View File

@@ -0,0 +1,165 @@
use std::ops::RangeInclusive;
#[cfg(target_arch = "x86_64")]
mod avx2;
mod scalar;
#[derive(Clone, Copy, Eq, PartialEq, Debug)]
#[repr(u8)]
enum FilterImplPerInstructionSet {
#[cfg(target_arch = "x86_64")]
AVX2 = 0u8,
Scalar = 1u8,
}
impl FilterImplPerInstructionSet {
#[inline]
pub fn is_available(&self) -> bool {
match *self {
#[cfg(target_arch = "x86_64")]
FilterImplPerInstructionSet::AVX2 => is_x86_feature_detected!("avx2"),
FilterImplPerInstructionSet::Scalar => true,
}
}
}
// List of available implementation in preferred order.
#[cfg(target_arch = "x86_64")]
const IMPLS: [FilterImplPerInstructionSet; 2] = [
FilterImplPerInstructionSet::AVX2,
FilterImplPerInstructionSet::Scalar,
];
#[cfg(not(target_arch = "x86_64"))]
const IMPLS: [FilterImplPerInstructionSet; 1] = [FilterImplPerInstructionSet::Scalar];
impl FilterImplPerInstructionSet {
#[allow(unused_variables)]
#[inline]
fn from(code: u8) -> FilterImplPerInstructionSet {
#[cfg(target_arch = "x86_64")]
if code == FilterImplPerInstructionSet::AVX2 as u8 {
return FilterImplPerInstructionSet::AVX2;
}
FilterImplPerInstructionSet::Scalar
}
#[inline]
fn filter_vec_in_place(self, range: RangeInclusive<u32>, offset: u32, output: &mut Vec<u32>) {
match self {
#[cfg(target_arch = "x86_64")]
FilterImplPerInstructionSet::AVX2 => avx2::filter_vec_in_place(range, offset, output),
FilterImplPerInstructionSet::Scalar => {
scalar::filter_vec_in_place(range, offset, output)
}
}
}
}
#[inline]
fn get_best_available_instruction_set() -> FilterImplPerInstructionSet {
use std::sync::atomic::{AtomicU8, Ordering};
static INSTRUCTION_SET_BYTE: AtomicU8 = AtomicU8::new(u8::MAX);
let instruction_set_byte: u8 = INSTRUCTION_SET_BYTE.load(Ordering::Relaxed);
if instruction_set_byte == u8::MAX {
// Let's initialize the instruction set and cache it.
let instruction_set = IMPLS
.into_iter()
.find(FilterImplPerInstructionSet::is_available)
.unwrap();
INSTRUCTION_SET_BYTE.store(instruction_set as u8, Ordering::Relaxed);
return instruction_set;
}
FilterImplPerInstructionSet::from(instruction_set_byte)
}
pub fn filter_vec_in_place(range: RangeInclusive<u32>, offset: u32, output: &mut Vec<u32>) {
get_best_available_instruction_set().filter_vec_in_place(range, offset, output)
}
#[cfg(test)]
mod tests {
use super::*;
#[test]
fn test_get_best_available_instruction_set() {
// This does not test much unfortunately.
// We just make sure the function returns without crashing and returns the same result.
let instruction_set = get_best_available_instruction_set();
assert_eq!(get_best_available_instruction_set(), instruction_set);
}
#[cfg(target_arch = "x86_64")]
#[test]
fn test_instruction_set_to_code_from_code() {
for instruction_set in [
FilterImplPerInstructionSet::AVX2,
FilterImplPerInstructionSet::Scalar,
] {
let code = instruction_set as u8;
assert_eq!(instruction_set, FilterImplPerInstructionSet::from(code));
}
}
fn test_filter_impl_empty_aux(filter_impl: FilterImplPerInstructionSet) {
let mut output = vec![];
filter_impl.filter_vec_in_place(0..=u32::MAX, 0, &mut output);
assert_eq!(&output, &[]);
}
fn test_filter_impl_simple_aux(filter_impl: FilterImplPerInstructionSet) {
let mut output = vec![3, 2, 1, 5, 11, 2, 5, 10, 2];
filter_impl.filter_vec_in_place(3..=10, 0, &mut output);
assert_eq!(&output, &[0, 3, 6, 7]);
}
fn test_filter_impl_simple_aux_shifted(filter_impl: FilterImplPerInstructionSet) {
let mut output = vec![3, 2, 1, 5, 11, 2, 5, 10, 2];
filter_impl.filter_vec_in_place(3..=10, 10, &mut output);
assert_eq!(&output, &[10, 13, 16, 17]);
}
fn test_filter_impl_simple_outside_i32_range(filter_impl: FilterImplPerInstructionSet) {
let mut output = vec![u32::MAX, i32::MAX as u32 + 1, 0, 1, 3, 1, 1, 1, 1];
filter_impl.filter_vec_in_place(1..=i32::MAX as u32 + 1u32, 0, &mut output);
assert_eq!(&output, &[1, 3, 4, 5, 6, 7, 8]);
}
fn test_filter_impl_test_suite(filter_impl: FilterImplPerInstructionSet) {
test_filter_impl_empty_aux(filter_impl);
test_filter_impl_simple_aux(filter_impl);
test_filter_impl_simple_aux_shifted(filter_impl);
test_filter_impl_simple_outside_i32_range(filter_impl);
}
#[test]
#[cfg(target_arch = "x86_64")]
fn test_filter_implementation_avx2() {
if FilterImplPerInstructionSet::AVX2.is_available() {
test_filter_impl_test_suite(FilterImplPerInstructionSet::AVX2);
}
}
#[test]
fn test_filter_implementation_scalar() {
test_filter_impl_test_suite(FilterImplPerInstructionSet::Scalar);
}
#[cfg(target_arch = "x86_64")]
proptest::proptest! {
#[test]
fn test_filter_compare_scalar_and_avx2_impl_proptest(
start in proptest::prelude::any::<u32>(),
end in proptest::prelude::any::<u32>(),
offset in 0u32..2u32,
mut vals in proptest::collection::vec(0..u32::MAX, 0..30)) {
if FilterImplPerInstructionSet::AVX2.is_available() {
let mut vals_clone = vals.clone();
FilterImplPerInstructionSet::AVX2.filter_vec_in_place(start..=end, offset, &mut vals);
FilterImplPerInstructionSet::Scalar.filter_vec_in_place(start..=end, offset, &mut vals_clone);
assert_eq!(&vals, &vals_clone);
}
}
}
}

13
bitpacker/src/filter_vec/scalar.rs Normal file
View File

@@ -0,0 +1,13 @@
use std::ops::RangeInclusive;
pub fn filter_vec_in_place(range: RangeInclusive<u32>, offset: u32, output: &mut Vec<u32>) {
// We restrict the accepted boundary, because unsigned integers & SIMD don't
// play well.
let mut output_cursor = 0;
for i in 0..output.len() {
let val = output[i];
output[output_cursor] = offset + i as u32;
output_cursor += if range.contains(&val) { 1 } else { 0 };
}
output.truncate(output_cursor);
}

1
bitpacker/src/lib.rs
View File

@@ -1,5 +1,6 @@
mod bitpacker;
mod blocked_bitpacker;
mod filter_vec;
use std::cmp::Ordering;

93
cliff.toml Normal file
View File

@@ -0,0 +1,93 @@
# configuration file for git-cliff{ pattern = "foo", replace = "bar"}
# see https://github.com/orhun/git-cliff#configuration-file
[remote.github]
owner = "quickwit-oss"
repo = "tantivy"
[changelog]
# changelog header
header = """
"""
# template for the changelog body
# https://tera.netlify.app/docs/#introduction
body = """
## What's Changed
{%- if version %} in {{ version }}{%- endif -%}
{% for commit in commits %}
{% if commit.github.pr_title -%}
{%- set commit_message = commit.github.pr_title -%}
{%- else -%}
{%- set commit_message = commit.message -%}
{%- endif -%}
- {{ commit_message | split(pat="\n") | first | trim }}\
{% if commit.github.pr_number %} \
[#{{ commit.github.pr_number }}]({{ self::remote_url() }}/pull/{{ commit.github.pr_number }}){% if commit.github.username %}(@{{ commit.github.username }}){%- endif -%} \
{%- endif %}
{%- endfor -%}
{% if github.contributors | filter(attribute="is_first_time", value=true) | length != 0 %}
{% raw %}\n{% endraw -%}
## New Contributors
{%- endif %}\
{% for contributor in github.contributors | filter(attribute="is_first_time", value=true) %}
* @{{ contributor.username }} made their first contribution
{%- if contributor.pr_number %} in \
[#{{ contributor.pr_number }}]({{ self::remote_url() }}/pull/{{ contributor.pr_number }}) \
{%- endif %}
{%- endfor -%}
{% if version %}
{% if previous.version %}
**Full Changelog**: {{ self::remote_url() }}/compare/{{ previous.version }}...{{ version }}
{% endif %}
{% else -%}
{% raw %}\n{% endraw %}
{% endif %}
{%- macro remote_url() -%}
https://github.com/{{ remote.github.owner }}/{{ remote.github.repo }}
{%- endmacro -%}
"""
# remove the leading and trailing whitespace from the template
trim = true
# changelog footer
footer = """
"""
postprocessors = [
]
[git]
# parse the commits based on https://www.conventionalcommits.org
# This is required or commit.message contains the whole commit message and not just the title
conventional_commits = false
# filter out the commits that are not conventional
filter_unconventional = true
# process each line of a commit as an individual commit
split_commits = false
# regex for preprocessing the commit messages
commit_preprocessors = [
{ pattern = '\((\w+\s)?#([0-9]+)\)', replace = ""},
]
#link_parsers = [
#{ pattern = "#(\\d+)", href = "https://github.com/quickwit-oss/tantivy/pulls/$1"},
#]
# regex for parsing and grouping commits
# protect breaking changes from being skipped due to matching a skipping commit_parser
protect_breaking_commits = false
# filter out the commits that are not matched by commit parsers
filter_commits = false
# glob pattern for matching git tags
tag_pattern = "v[0-9]*"
# regex for skipping tags
skip_tags = "v0.1.0-beta.1"
# regex for ignoring tags
ignore_tags = ""
# sort the tags topologically
topo_order = false
# sort the commits inside sections by oldest/newest order
sort_commits = "newest"
# limit the number of commits included in the changelog.
# limit_commits = 42

24
columnar/Cargo.toml
View File

@@ -1,28 +1,28 @@
[package]
name = "tantivy-columnar"
version = "0.1.0"
version = "0.3.0"
edition = "2021"
license = "MIT"
homepage = "https://github.com/quickwit-oss/tantivy"
repository = "https://github.com/quickwit-oss/tantivy"
description = "column oriented storage for tantivy"
categories = ["database-implementations", "data-structures", "compression"]
[dependencies]
itertools = "0.10.5"
log = "0.4.17"
fnv = "1.0.7"
itertools = "0.12.0"
fastdivide = "0.4.0"
rand = { version = "0.8.5", optional = true }
measure_time = { version = "0.8.2", optional = true }
prettytable-rs = { version = "0.10.0", optional = true }
stacker = { path = "../stacker", package="tantivy-stacker"}
sstable = { path = "../sstable", package = "tantivy-sstable" }
common = { path = "../common", package = "tantivy-common" }
tantivy-bitpacker = { version= "0.3", path = "../bitpacker/" }
stacker = { version= "0.3", path = "../stacker", package="tantivy-stacker"}
sstable = { version= "0.3", path = "../sstable", package = "tantivy-sstable" }
common = { version= "0.7", path = "../common", package = "tantivy-common" }
tantivy-bitpacker = { version= "0.6", path = "../bitpacker/" }
serde = "1.0.152"
downcast-rs = "1.2.0"
[dev-dependencies]
proptest = "1"
more-asserts = "0.3.1"
rand = "0.8.5"
rand = "0.8"
[features]
unstable = []

155
columnar/benches/bench_first_vals.rs Normal file
View File

@@ -0,0 +1,155 @@
#![feature(test)]
extern crate test;
use std::sync::Arc;
use rand::prelude::*;
use tantivy_columnar::column_values::{serialize_and_load_u64_based_column_values, CodecType};
use tantivy_columnar::*;
use test::{black_box, Bencher};
struct Columns {
pub optional: Column,
pub full: Column,
pub multi: Column,
}
fn get_test_columns() -> Columns {
let data = generate_permutation();
let mut dataframe_writer = ColumnarWriter::default();
for (idx, val) in data.iter().enumerate() {
dataframe_writer.record_numerical(idx as u32, "full_values", NumericalValue::U64(*val));
if idx % 2 == 0 {
dataframe_writer.record_numerical(
idx as u32,
"optional_values",
NumericalValue::U64(*val),
);
}
dataframe_writer.record_numerical(idx as u32, "multi_values", NumericalValue::U64(*val));
dataframe_writer.record_numerical(idx as u32, "multi_values", NumericalValue::U64(*val));
}
let mut buffer: Vec<u8> = Vec::new();
dataframe_writer
.serialize(data.len() as u32, None, &mut buffer)
.unwrap();
let columnar = ColumnarReader::open(buffer).unwrap();
let cols: Vec<DynamicColumnHandle> = columnar.read_columns("optional_values").unwrap();
assert_eq!(cols.len(), 1);
let optional = cols[0].open_u64_lenient().unwrap().unwrap();
assert_eq!(optional.index.get_cardinality(), Cardinality::Optional);
let cols: Vec<DynamicColumnHandle> = columnar.read_columns("full_values").unwrap();
assert_eq!(cols.len(), 1);
let column_full = cols[0].open_u64_lenient().unwrap().unwrap();
assert_eq!(column_full.index.get_cardinality(), Cardinality::Full);
let cols: Vec<DynamicColumnHandle> = columnar.read_columns("multi_values").unwrap();
assert_eq!(cols.len(), 1);
let multi = cols[0].open_u64_lenient().unwrap().unwrap();
assert_eq!(multi.index.get_cardinality(), Cardinality::Multivalued);
Columns {
optional,
full: column_full,
multi,
}
}
const NUM_VALUES: u64 = 100_000;
fn generate_permutation() -> Vec<u64> {
let mut permutation: Vec<u64> = (0u64..NUM_VALUES).collect();
permutation.shuffle(&mut StdRng::from_seed([1u8; 32]));
permutation
}
pub fn serialize_and_load(column: &[u64], codec_type: CodecType) -> Arc<dyn ColumnValues<u64>> {
serialize_and_load_u64_based_column_values(&column, &[codec_type])
}
fn run_bench_on_column_full_scan(b: &mut Bencher, column: Column) {
let num_iter = black_box(NUM_VALUES);
b.iter(|| {
let mut sum = 0u64;
for i in 0..num_iter as u32 {
let val = column.first(i);
sum += val.unwrap_or(0);
}
sum
});
}
fn run_bench_on_column_block_fetch(b: &mut Bencher, column: Column) {
let mut block: Vec<Option<u64>> = vec![None; 64];
let fetch_docids = (0..64).collect::<Vec<_>>();
b.iter(move || {
column.first_vals(&fetch_docids, &mut block);
block[0]
});
}
fn run_bench_on_column_block_single_calls(b: &mut Bencher, column: Column) {
let mut block: Vec<Option<u64>> = vec![None; 64];
let fetch_docids = (0..64).collect::<Vec<_>>();
b.iter(move || {
for i in 0..fetch_docids.len() {
block[i] = column.first(fetch_docids[i]);
}
block[0]
});
}
/// Column first method
#[bench]
fn bench_get_first_on_full_column_full_scan(b: &mut Bencher) {
let column = get_test_columns().full;
run_bench_on_column_full_scan(b, column);
}
#[bench]
fn bench_get_first_on_optional_column_full_scan(b: &mut Bencher) {
let column = get_test_columns().optional;
run_bench_on_column_full_scan(b, column);
}
#[bench]
fn bench_get_first_on_multi_column_full_scan(b: &mut Bencher) {
let column = get_test_columns().multi;
run_bench_on_column_full_scan(b, column);
}
/// Block fetch column accessor
#[bench]
fn bench_get_block_first_on_optional_column(b: &mut Bencher) {
let column = get_test_columns().optional;
run_bench_on_column_block_fetch(b, column);
}
#[bench]
fn bench_get_block_first_on_multi_column(b: &mut Bencher) {
let column = get_test_columns().multi;
run_bench_on_column_block_fetch(b, column);
}
#[bench]
fn bench_get_block_first_on_full_column(b: &mut Bencher) {
let column = get_test_columns().full;
run_bench_on_column_block_fetch(b, column);
}
#[bench]
fn bench_get_block_first_on_optional_column_single_calls(b: &mut Bencher) {
let column = get_test_columns().optional;
run_bench_on_column_block_single_calls(b, column);
}
#[bench]
fn bench_get_block_first_on_multi_column_single_calls(b: &mut Bencher) {
let column = get_test_columns().multi;
run_bench_on_column_block_single_calls(b, column);
}
#[bench]
fn bench_get_block_first_on_full_column_single_calls(b: &mut Bencher) {
let column = get_test_columns().full;
run_bench_on_column_block_single_calls(b, column);
}

0
columnar/benches/bench_u128.rs → columnar/benches/bench_values_u128.rs
View File

8
columnar/benches/bench_u64.rs → columnar/benches/bench_values_u64.rs
View File

@@ -16,14 +16,6 @@ fn generate_permutation() -> Vec<u64> {
permutation
}
fn generate_random() -> Vec<u64> {
let mut permutation: Vec<u64> = (0u64..100_000u64)
.map(|el| el + random::<u16>() as u64)
.collect();
permutation.shuffle(&mut StdRng::from_seed([1u8; 32]));
permutation
}
// Warning: this generates the same permutation at each call
fn generate_permutation_gcd() -> Vec<u64> {
let mut permutation: Vec<u64> = (1u64..100_000u64).map(|el| el * 1000).collect();

1
columnar/columnar-cli/Cargo.toml
View File

@@ -8,7 +8,6 @@ license = "MIT"
columnar = {path="../", package="tantivy-columnar"}
serde_json = "1"
serde_json_borrow = {git="https://github.com/PSeitz/serde_json_borrow/"}
serde = "1"
[workspace]
members = []

158
columnar/src/block_accessor.rs Normal file
View File

@@ -0,0 +1,158 @@
use std::cmp::Ordering;
use crate::{Column, DocId, RowId};
#[derive(Debug, Default, Clone)]
pub struct ColumnBlockAccessor<T> {
val_cache: Vec<T>,
docid_cache: Vec<DocId>,
missing_docids_cache: Vec<DocId>,
row_id_cache: Vec<RowId>,
}
impl<T: PartialOrd + Copy + std::fmt::Debug + Send + Sync + 'static + Default>
ColumnBlockAccessor<T>
{
#[inline]
pub fn fetch_block<'a>(&'a mut self, docs: &'a [u32], accessor: &Column<T>) {
if accessor.index.get_cardinality().is_full() {
self.val_cache.resize(docs.len(), T::default());
accessor.values.get_vals(docs, &mut self.val_cache);
} else {
self.docid_cache.clear();
self.row_id_cache.clear();
accessor.row_ids_for_docs(docs, &mut self.docid_cache, &mut self.row_id_cache);
self.val_cache.resize(self.row_id_cache.len(), T::default());
accessor
.values
.get_vals(&self.row_id_cache, &mut self.val_cache);
}
}
#[inline]
pub fn fetch_block_with_missing(&mut self, docs: &[u32], accessor: &Column<T>, missing: T) {
self.fetch_block(docs, accessor);
// no missing values
if accessor.index.get_cardinality().is_full() {
return;
}
// We can compare docid_cache length with docs to find missing docs
// For multi value columns we can't rely on the length and always need to scan
if accessor.index.get_cardinality().is_multivalue() || docs.len() != self.docid_cache.len()
{
self.missing_docids_cache.clear();
find_missing_docs(docs, &self.docid_cache, |doc| {
self.missing_docids_cache.push(doc);
self.val_cache.push(missing);
});
self.docid_cache
.extend_from_slice(&self.missing_docids_cache);
}
}
#[inline]
pub fn iter_vals(&self) -> impl Iterator<Item = T> + '_ {
self.val_cache.iter().cloned()
}
#[inline]
/// Returns an iterator over the docids and values
/// The passed in `docs` slice needs to be the same slice that was passed to `fetch_block` or
/// `fetch_block_with_missing`.
///
/// The docs is used if the column is full (each docs has exactly one value), otherwise the
/// internal docid vec is used for the iterator, which e.g. may contain duplicate docs.
pub fn iter_docid_vals<'a>(
&'a self,
docs: &'a [u32],
accessor: &Column<T>,
) -> impl Iterator<Item = (DocId, T)> + '_ {
if accessor.index.get_cardinality().is_full() {
docs.iter().cloned().zip(self.val_cache.iter().cloned())
} else {
self.docid_cache
.iter()
.cloned()
.zip(self.val_cache.iter().cloned())
}
}
}
/// Given two sorted lists of docids `docs` and `hits`, hits is a subset of `docs`.
/// Return all docs that are not in `hits`.
fn find_missing_docs<F>(docs: &[u32], hits: &[u32], mut callback: F)
where F: FnMut(u32) {
let mut docs_iter = docs.iter();
let mut hits_iter = hits.iter();
let mut doc = docs_iter.next();
let mut hit = hits_iter.next();
while let (Some(&current_doc), Some(&current_hit)) = (doc, hit) {
match current_doc.cmp(&current_hit) {
Ordering::Less => {
callback(current_doc);
doc = docs_iter.next();
}
Ordering::Equal => {
doc = docs_iter.next();
hit = hits_iter.next();
}
Ordering::Greater => {
hit = hits_iter.next();
}
}
}
while let Some(&current_doc) = doc {
callback(current_doc);
doc = docs_iter.next();
}
}
#[cfg(test)]
mod tests {
use super::*;
#[test]
fn test_find_missing_docs() {
let docs: Vec<u32> = vec![1, 2, 3, 4, 5, 6, 7, 8, 9, 10];
let hits: Vec<u32> = vec![2, 4, 6, 8, 10];
let mut missing_docs: Vec<u32> = Vec::new();
find_missing_docs(&docs, &hits, |missing_doc| {
missing_docs.push(missing_doc);
});
assert_eq!(missing_docs, vec![1, 3, 5, 7, 9]);
}
#[test]
fn test_find_missing_docs_empty() {
let docs: Vec<u32> = Vec::new();
let hits: Vec<u32> = vec![2, 4, 6, 8, 10];
let mut missing_docs: Vec<u32> = Vec::new();
find_missing_docs(&docs, &hits, |missing_doc| {
missing_docs.push(missing_doc);
});
assert_eq!(missing_docs, vec![]);
}
#[test]
fn test_find_missing_docs_all_missing() {
let docs: Vec<u32> = vec![1, 2, 3, 4, 5];
let hits: Vec<u32> = Vec::new();
let mut missing_docs: Vec<u32> = Vec::new();
find_missing_docs(&docs, &hits, |missing_doc| {
missing_docs.push(missing_doc);
});
assert_eq!(missing_docs, vec![1, 2, 3, 4, 5]);
}
}

25
columnar/src/column/dictionary_encoded.rs
View File

@@ -1,6 +1,6 @@
use std::io;
use std::ops::Deref;
use std::sync::Arc;
use std::{fmt, io};
use sstable::{Dictionary, VoidSSTable};
@@ -21,7 +21,22 @@ pub struct BytesColumn {
pub(crate) term_ord_column: Column<u64>,
}
impl fmt::Debug for BytesColumn {
fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
f.debug_struct("BytesColumn")
.field("term_ord_column", &self.term_ord_column)
.finish()
}
}
impl BytesColumn {
pub fn empty(num_docs: u32) -> BytesColumn {
BytesColumn {
dictionary: Arc::new(Dictionary::empty()),
term_ord_column: Column::build_empty_column(num_docs),
}
}
/// Fills the given `output` buffer with the term associated to the ordinal `ord`.
///
/// Returns `false` if the term does not exist (e.g. `term_ord` is greater or equal to the
@@ -56,6 +71,12 @@ impl BytesColumn {
#[derive(Clone)]
pub struct StrColumn(BytesColumn);
impl fmt::Debug for StrColumn {
fn fmt(&self, f: &mut fmt::Formatter) -> fmt::Result {
write!(f, "{:?}", self.term_ord_column)
}
}
impl From<StrColumn> for BytesColumn {
fn from(str_column: StrColumn) -> BytesColumn {
str_column.0
@@ -63,7 +84,7 @@ impl From<StrColumn> for BytesColumn {
}
impl StrColumn {
pub(crate) fn wrap(bytes_column: BytesColumn) -> StrColumn {
pub fn wrap(bytes_column: BytesColumn) -> StrColumn {
StrColumn(bytes_column)
}

106
columnar/src/column/mod.rs
View File

@@ -1,29 +1,48 @@
mod dictionary_encoded;
mod serialize;
use std::fmt::Debug;
use std::fmt::{self, Debug};
use std::io::Write;
use std::ops::{Deref, Range, RangeInclusive};
use std::ops::{Range, RangeInclusive};
use std::sync::Arc;
use common::BinarySerializable;
pub use dictionary_encoded::{BytesColumn, StrColumn};
pub use serialize::{
open_column_bytes, open_column_str, open_column_u128, open_column_u64,
serialize_column_mappable_to_u128, serialize_column_mappable_to_u64,
open_column_bytes, open_column_str, open_column_u128, open_column_u128_as_compact_u64,
open_column_u64, serialize_column_mappable_to_u128, serialize_column_mappable_to_u64,
};
use crate::column_index::ColumnIndex;
use crate::column_index::{ColumnIndex, Set};
use crate::column_values::monotonic_mapping::StrictlyMonotonicMappingToInternal;
use crate::column_values::{monotonic_map_column, ColumnValues};
use crate::{Cardinality, MonotonicallyMappableToU64, RowId};
use crate::{Cardinality, DocId, EmptyColumnValues, MonotonicallyMappableToU64, RowId};
#[derive(Clone)]
pub struct Column<T = u64> {
pub idx: ColumnIndex,
pub index: ColumnIndex,
pub values: Arc<dyn ColumnValues<T>>,
}
impl<T: Debug + PartialOrd + Send + Sync + Copy + 'static> Debug for Column<T> {
fn fmt(&self, f: &mut fmt::Formatter) -> fmt::Result {
let num_docs = self.num_docs();
let entries = (0..num_docs)
.map(|i| (i, self.values_for_doc(i).collect::<Vec<_>>()))
.filter(|(_, vals)| !vals.is_empty());
f.debug_map().entries(entries).finish()
}
}
impl<T: PartialOrd + Default> Column<T> {
pub fn build_empty_column(num_docs: u32) -> Column<T> {
Column {
index: ColumnIndex::Empty { num_docs },
values: Arc::new(EmptyColumnValues),
}
}
}
impl<T: MonotonicallyMappableToU64> Column<T> {
pub fn to_u64_monotonic(self) -> Column<u64> {
let values = Arc::new(monotonic_map_column(
@@ -31,7 +50,7 @@ impl<T: MonotonicallyMappableToU64> Column<T> {
StrictlyMonotonicMappingToInternal::<T>::new(),
));
Column {
idx: self.idx,
index: self.index,
values,
}
}
@@ -40,11 +59,11 @@ impl<T: MonotonicallyMappableToU64> Column<T> {
impl<T: PartialOrd + Copy + Debug + Send + Sync + 'static> Column<T> {
#[inline]
pub fn get_cardinality(&self) -> Cardinality {
self.idx.get_cardinality()
self.index.get_cardinality()
}
pub fn num_docs(&self) -> RowId {
match &self.idx {
match &self.index {
ColumnIndex::Empty { num_docs } => *num_docs,
ColumnIndex::Full => self.values.num_vals(),
ColumnIndex::Optional(optional_index) => optional_index.num_docs(),
@@ -64,12 +83,56 @@ impl<T: PartialOrd + Copy + Debug + Send + Sync + 'static> Column<T> {
self.values.max_value()
}
#[inline]
pub fn first(&self, row_id: RowId) -> Option<T> {
self.values_for_doc(row_id).next()
}
pub fn values_for_doc(&self, row_id: RowId) -> impl Iterator<Item = T> + '_ {
self.value_row_ids(row_id)
/// 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 docis to row_ids.
///
/// returns the row_ids and the matching docids on the same index
/// e.g.
/// DocId In: [0, 5, 6]
/// DocId Out: [0, 0, 6, 6]
/// RowId Out: [0, 1, 2, 3]
#[inline]
pub fn row_ids_for_docs(
&self,
doc_ids: &[DocId],
doc_ids_out: &mut Vec<DocId>,
row_ids: &mut Vec<RowId>,
) {
self.index.docids_to_rowids(doc_ids, doc_ids_out, row_ids)
}
pub fn values_for_doc(&self, doc_id: DocId) -> impl Iterator<Item = T> + '_ {
self.index
.value_row_ids(doc_id)
.map(|value_row_id: RowId| self.values.get_val(value_row_id))
}
@@ -82,17 +145,19 @@ impl<T: PartialOrd + Copy + Debug + Send + Sync + 'static> Column<T> {
doc_ids: &mut Vec<u32>,
) {
// convert passed docid range to row id range
let rowid_range = self.idx.docid_range_to_rowids(selected_docid_range.clone());
let rowid_range = self
.index
.docid_range_to_rowids(selected_docid_range.clone());
// Load rows
self.values
.get_row_ids_for_value_range(value_range, rowid_range, doc_ids);
// Convert rows to docids
self.idx
self.index
.select_batch_in_place(selected_docid_range.start, doc_ids);
}
/// Fils the output vector with the (possibly multiple values that are associated_with
/// Fills the output vector with the (possibly multiple values that are associated_with
/// `row_id`.
///
/// This method clears the `output` vector.
@@ -109,14 +174,6 @@ impl<T: PartialOrd + Copy + Debug + Send + Sync + 'static> Column<T> {
}
}
impl<T> Deref for Column<T> {
type Target = ColumnIndex;
fn deref(&self) -> &Self::Target {
&self.idx
}
}
impl BinarySerializable for Cardinality {
fn serialize<W: Write + ?Sized>(&self, writer: &mut W) -> std::io::Result<()> {
self.to_code().serialize(writer)
@@ -138,6 +195,7 @@ struct FirstValueWithDefault<T: Copy> {
impl<T: PartialOrd + Debug + Send + Sync + Copy + 'static> ColumnValues<T>
for FirstValueWithDefault<T>
{
#[inline(always)]
fn get_val(&self, idx: u32) -> T {
self.column.first(idx).unwrap_or(self.default_value)
}
@@ -151,7 +209,7 @@ impl<T: PartialOrd + Debug + Send + Sync + Copy + 'static> ColumnValues<T>
}
fn num_vals(&self) -> u32 {
match &self.column.idx {
match &self.column.index {
ColumnIndex::Empty { .. } => 0u32,
ColumnIndex::Full => self.column.values.num_vals(),
ColumnIndex::Optional(optional_idx) => optional_idx.num_docs(),

24
columnar/src/column/serialize.rs
View File

@@ -52,7 +52,7 @@ pub fn open_column_u64<T: MonotonicallyMappableToU64>(bytes: OwnedBytes) -> io::
let column_index = crate::column_index::open_column_index(column_index_data)?;
let column_values = load_u64_based_column_values(column_values_data)?;
Ok(Column {
idx: column_index,
index: column_index,
values: column_values,
})
}
@@ -71,7 +71,27 @@ pub fn open_column_u128<T: MonotonicallyMappableToU128>(
let column_index = crate::column_index::open_column_index(column_index_data)?;
let column_values = crate::column_values::open_u128_mapped(column_values_data)?;
Ok(Column {
idx: column_index,
index: column_index,
values: column_values,
})
}
/// Open the column as u64.
///
/// See [`open_u128_as_compact_u64`] for more details.
pub fn open_column_u128_as_compact_u64(bytes: OwnedBytes) -> io::Result<Column<u64>> {
let (body, column_index_num_bytes_payload) = bytes.rsplit(4);
let column_index_num_bytes = u32::from_le_bytes(
column_index_num_bytes_payload
.as_slice()
.try_into()
.unwrap(),
);
let (column_index_data, column_values_data) = body.split(column_index_num_bytes as usize);
let column_index = crate::column_index::open_column_index(column_index_data)?;
let column_values = crate::column_values::open_u128_as_compact_u64(column_values_data)?;
Ok(Column {
index: column_index,
values: column_values,
})
}

134
columnar/src/column_index/merge/mod.rs
View File

@@ -1,29 +1,82 @@
mod shuffled;
mod stacked;
use common::ReadOnlyBitSet;
use shuffled::merge_column_index_shuffled;
use stacked::merge_column_index_stacked;
use crate::column_index::SerializableColumnIndex;
use crate::{Cardinality, ColumnIndex, MergeRowOrder};
// For simplification, we never have cardinality go down due to deletes.
fn detect_cardinality(columns: &[Option<ColumnIndex>]) -> Cardinality {
columns
.iter()
.flatten()
.map(ColumnIndex::get_cardinality)
.max()
.unwrap_or(Cardinality::Full)
fn detect_cardinality_single_column_index(
column_index: &ColumnIndex,
alive_bitset_opt: &Option<ReadOnlyBitSet>,
) -> Cardinality {
let Some(alive_bitset) = alive_bitset_opt else {
return column_index.get_cardinality();
};
let cardinality_before_deletes = column_index.get_cardinality();
if cardinality_before_deletes == Cardinality::Full {
// The columnar cardinality can only become more restrictive in the presence of deletes
// (where cardinality sorted from the more restrictive to the least restrictive are Full,
// Optional, Multivalued)
//
// If we are already "Full", we are guaranteed to stay "Full" after deletes.
return Cardinality::Full;
}
let mut cardinality_so_far = Cardinality::Full;
for doc_id in alive_bitset.iter() {
let num_values = column_index.value_row_ids(doc_id).len();
let row_cardinality = match num_values {
0 => Cardinality::Optional,
1 => Cardinality::Full,
_ => Cardinality::Multivalued,
};
cardinality_so_far = cardinality_so_far.max(row_cardinality);
if cardinality_so_far >= cardinality_before_deletes {
// There won't be any improvement in the cardinality.
// We can early exit.
return cardinality_before_deletes;
}
}
cardinality_so_far
}
fn detect_cardinality(
column_indexes: &[ColumnIndex],
merge_row_order: &MergeRowOrder,
) -> Cardinality {
match merge_row_order {
MergeRowOrder::Stack(_) => column_indexes
.iter()
.map(ColumnIndex::get_cardinality)
.max()
.unwrap_or(Cardinality::Full),
MergeRowOrder::Shuffled(shuffle_merge_order) => {
let mut merged_cardinality = Cardinality::Full;
for (column_index, alive_bitset_opt) in column_indexes
.iter()
.zip(shuffle_merge_order.alive_bitsets.iter())
{
let cardinality: Cardinality =
detect_cardinality_single_column_index(column_index, alive_bitset_opt);
if cardinality == Cardinality::Multivalued {
return cardinality;
}
merged_cardinality = merged_cardinality.max(cardinality);
}
merged_cardinality
}
}
}
pub fn merge_column_index<'a>(
columns: &'a [Option<ColumnIndex>],
columns: &'a [ColumnIndex],
merge_row_order: &'a MergeRowOrder,
) -> SerializableColumnIndex<'a> {
// For simplification, we do not try to detect whether the cardinality could be
// downgraded thanks to deletes.
let cardinality_after_merge = detect_cardinality(columns);
let cardinality_after_merge = detect_cardinality(columns, merge_row_order);
match merge_row_order {
MergeRowOrder::Stack(stack_merge_order) => {
merge_column_index_stacked(columns, cardinality_after_merge, stack_merge_order)
@@ -45,42 +98,61 @@ mod tests {
use crate::column_index::merge::detect_cardinality;
use crate::column_index::multivalued_index::MultiValueIndex;
use crate::column_index::{merge_column_index, OptionalIndex, SerializableColumnIndex};
use crate::{Cardinality, ColumnIndex, MergeRowOrder, RowAddr, RowId, ShuffleMergeOrder};
use crate::{
Cardinality, ColumnIndex, MergeRowOrder, RowAddr, RowId, ShuffleMergeOrder, StackMergeOrder,
};
#[test]
fn test_detect_cardinality() {
assert_eq!(detect_cardinality(&[]), Cardinality::Full);
assert_eq!(
detect_cardinality(&[], &StackMergeOrder::stack_for_test(&[]).into()),
Cardinality::Full
);
let optional_index: ColumnIndex = OptionalIndex::for_test(1, &[]).into();
let multivalued_index: ColumnIndex = MultiValueIndex::for_test(&[0, 1]).into();
assert_eq!(
detect_cardinality(&[Some(optional_index.clone()), None]),
detect_cardinality(
&[optional_index.clone(), ColumnIndex::Empty { num_docs: 0 }],
&StackMergeOrder::stack_for_test(&[1, 0]).into()
),
Cardinality::Optional
);
assert_eq!(
detect_cardinality(&[Some(optional_index.clone()), Some(ColumnIndex::Full)]),
detect_cardinality(
&[optional_index.clone(), ColumnIndex::Full],
&StackMergeOrder::stack_for_test(&[1, 1]).into()
),
Cardinality::Optional
);
assert_eq!(
detect_cardinality(&[Some(multivalued_index.clone()), None]),
detect_cardinality(
&[
multivalued_index.clone(),
ColumnIndex::Empty { num_docs: 0 }
],
&StackMergeOrder::stack_for_test(&[1, 0]).into()
),
Cardinality::Multivalued
);
assert_eq!(
detect_cardinality(&[
Some(multivalued_index.clone()),
Some(optional_index.clone())
]),
detect_cardinality(
&[multivalued_index.clone(), optional_index.clone()],
&StackMergeOrder::stack_for_test(&[1, 1]).into()
),
Cardinality::Multivalued
);
assert_eq!(
detect_cardinality(&[Some(optional_index), Some(multivalued_index)]),
detect_cardinality(
&[optional_index, multivalued_index],
&StackMergeOrder::stack_for_test(&[1, 1]).into()
),
Cardinality::Multivalued
);
}
#[test]
fn test_merge_index_multivalued_sorted() {
let column_indexes: Vec<Option<ColumnIndex>> =
vec![Some(MultiValueIndex::for_test(&[0, 2, 5]).into())];
let column_indexes: Vec<ColumnIndex> = vec![MultiValueIndex::for_test(&[0, 2, 5]).into()];
let merge_row_order: MergeRowOrder = ShuffleMergeOrder::for_test(
&[2],
vec![
@@ -96,18 +168,19 @@ mod tests {
)
.into();
let merged_column_index = merge_column_index(&column_indexes[..], &merge_row_order);
let SerializableColumnIndex::Multivalued(start_index_iterable) = merged_column_index
else { panic!("Excpected a multivalued index") };
let SerializableColumnIndex::Multivalued(start_index_iterable) = merged_column_index else {
panic!("Excpected a multivalued index")
};
let start_indexes: Vec<RowId> = start_index_iterable.boxed_iter().collect();
assert_eq!(&start_indexes, &[0, 3, 5]);
}
#[test]
fn test_merge_index_multivalued_sorted_several_segment() {
let column_indexes: Vec<Option<ColumnIndex>> = vec![
Some(MultiValueIndex::for_test(&[0, 2, 5]).into()),
None,
Some(MultiValueIndex::for_test(&[0, 1, 4]).into()),
let column_indexes: Vec<ColumnIndex> = vec![
MultiValueIndex::for_test(&[0, 2, 5]).into(),
ColumnIndex::Empty { num_docs: 0 },
MultiValueIndex::for_test(&[0, 1, 4]).into(),
];
let merge_row_order: MergeRowOrder = ShuffleMergeOrder::for_test(
&[2, 0, 2],
@@ -128,8 +201,9 @@ mod tests {
)
.into();
let merged_column_index = merge_column_index(&column_indexes[..], &merge_row_order);
let SerializableColumnIndex::Multivalued(start_index_iterable) = merged_column_index
else { panic!("Excpected a multivalued index") };
let SerializableColumnIndex::Multivalued(start_index_iterable) = merged_column_index else {
panic!("Excpected a multivalued index")
};
let start_indexes: Vec<RowId> = start_index_iterable.boxed_iter().collect();
assert_eq!(&start_indexes, &[0, 3, 5, 6]);
}

61
columnar/src/column_index/merge/shuffled.rs
View File

@@ -5,7 +5,7 @@ use crate::iterable::Iterable;
use crate::{Cardinality, ColumnIndex, RowId, ShuffleMergeOrder};
pub fn merge_column_index_shuffled<'a>(
column_indexes: &'a [Option<ColumnIndex>],
column_indexes: &'a [ColumnIndex],
cardinality_after_merge: Cardinality,
shuffle_merge_order: &'a ShuffleMergeOrder,
) -> SerializableColumnIndex<'a> {
@@ -33,41 +33,41 @@ pub fn merge_column_index_shuffled<'a>(
///
/// In other words the column_indexes passed as argument may NOT be multivalued.
fn merge_column_index_shuffled_optional<'a>(
column_indexes: &'a [Option<ColumnIndex>],
column_indexes: &'a [ColumnIndex],
merge_order: &'a ShuffleMergeOrder,
) -> Box<dyn Iterable<RowId> + 'a> {
Box::new(ShuffledOptionalIndex {
Box::new(ShuffledIndex {
column_indexes,
merge_order,
})
}
struct ShuffledOptionalIndex<'a> {
column_indexes: &'a [Option<ColumnIndex>],
struct ShuffledIndex<'a> {
column_indexes: &'a [ColumnIndex],
merge_order: &'a ShuffleMergeOrder,
}
impl<'a> Iterable<u32> for ShuffledOptionalIndex<'a> {
impl<'a> Iterable<u32> for ShuffledIndex<'a> {
fn boxed_iter(&self) -> Box<dyn Iterator<Item = u32> + '_> {
Box::new(self.merge_order
.iter_new_to_old_row_addrs()
.enumerate()
.filter_map(|(new_row_id, old_row_addr)| {
let Some(column_index) = &self.column_indexes[old_row_addr.segment_ord as usize] else {
return None;
};
let row_id = new_row_id as u32;
if column_index.has_value(old_row_addr.row_id) {
Some(row_id)
} else {
None
}
}))
Box::new(
self.merge_order
.iter_new_to_old_row_addrs()
.enumerate()
.filter_map(|(new_row_id, old_row_addr)| {
let column_index = &self.column_indexes[old_row_addr.segment_ord as usize];
let row_id = new_row_id as u32;
if column_index.has_value(old_row_addr.row_id) {
Some(row_id)
} else {
None
}
}),
)
}
}
fn merge_column_index_shuffled_multivalued<'a>(
column_indexes: &'a [Option<ColumnIndex>],
column_indexes: &'a [ColumnIndex],
merge_order: &'a ShuffleMergeOrder,
) -> Box<dyn Iterable<RowId> + 'a> {
Box::new(ShuffledMultivaluedIndex {
@@ -77,19 +77,16 @@ fn merge_column_index_shuffled_multivalued<'a>(
}
struct ShuffledMultivaluedIndex<'a> {
column_indexes: &'a [Option<ColumnIndex>],
column_indexes: &'a [ColumnIndex],
merge_order: &'a ShuffleMergeOrder,
}
fn iter_num_values<'a>(
column_indexes: &'a [Option<ColumnIndex>],
column_indexes: &'a [ColumnIndex],
merge_order: &'a ShuffleMergeOrder,
) -> impl Iterator<Item = u32> + 'a {
merge_order.iter_new_to_old_row_addrs().map(|row_addr| {
let Some(column_index) = &column_indexes[row_addr.segment_ord as usize] else {
// No values in the entire column. It surely means there are 0 values associated to this row.
return 0u32;
};
let column_index = &column_indexes[row_addr.segment_ord as usize];
match column_index {
ColumnIndex::Empty { .. } => 0u32,
ColumnIndex::Full => 1,
@@ -143,7 +140,7 @@ mod tests {
#[test]
fn test_merge_column_index_optional_shuffle() {
let optional_index: ColumnIndex = OptionalIndex::for_test(2, &[0]).into();
let column_indexes = vec![Some(optional_index), Some(ColumnIndex::Full)];
let column_indexes = [optional_index, ColumnIndex::Full];
let row_addrs = vec![
RowAddr {
segment_ord: 0u32,
@@ -160,7 +157,13 @@ mod tests {
Cardinality::Optional,
&shuffle_merge_order,
);
let SerializableColumnIndex::Optional { non_null_row_ids, num_rows } = serializable_index else { panic!() };
let SerializableColumnIndex::Optional {
non_null_row_ids,
num_rows,
} = serializable_index
else {
panic!()
};
assert_eq!(num_rows, 2);
let non_null_rows: Vec<RowId> = non_null_row_ids.boxed_iter().collect();
assert_eq!(&non_null_rows, &[1]);

34
columnar/src/column_index/merge/stacked.rs
View File

@@ -9,7 +9,7 @@ use crate::{Cardinality, ColumnIndex, RowId, StackMergeOrder};
///
/// There are no sort nor deletes involved.
pub fn merge_column_index_stacked<'a>(
columns: &'a [Option<ColumnIndex>],
columns: &'a [ColumnIndex],
cardinality_after_merge: Cardinality,
stack_merge_order: &'a StackMergeOrder,
) -> SerializableColumnIndex<'a> {
@@ -33,7 +33,7 @@ pub fn merge_column_index_stacked<'a>(
}
struct StackedOptionalIndex<'a> {
columns: &'a [Option<ColumnIndex>],
columns: &'a [ColumnIndex],
stack_merge_order: &'a StackMergeOrder,
}
@@ -46,16 +46,16 @@ impl<'a> Iterable<RowId> for StackedOptionalIndex<'a> {
.flat_map(|(columnar_id, column_index_opt)| {
let columnar_row_range = self.stack_merge_order.columnar_range(columnar_id);
let rows_it: Box<dyn Iterator<Item = RowId>> = match column_index_opt {
Some(ColumnIndex::Full) => Box::new(columnar_row_range),
Some(ColumnIndex::Optional(optional_index)) => Box::new(
ColumnIndex::Full => Box::new(columnar_row_range),
ColumnIndex::Optional(optional_index) => Box::new(
optional_index
.iter_rows()
.map(move |row_id: RowId| columnar_row_range.start + row_id),
),
Some(ColumnIndex::Multivalued(_)) => {
ColumnIndex::Multivalued(_) => {
panic!("No multivalued index is allowed when stacking column index");
}
None | Some(ColumnIndex::Empty { .. }) => Box::new(std::iter::empty()),
ColumnIndex::Empty { .. } => Box::new(std::iter::empty()),
};
rows_it
}),
@@ -65,20 +65,18 @@ impl<'a> Iterable<RowId> for StackedOptionalIndex<'a> {
#[derive(Clone, Copy)]
struct StackedMultivaluedIndex<'a> {
columns: &'a [Option<ColumnIndex>],
columns: &'a [ColumnIndex],
stack_merge_order: &'a StackMergeOrder,
}
fn convert_column_opt_to_multivalued_index<'a>(
column_index_opt: Option<&'a ColumnIndex>,
column_index_opt: &'a ColumnIndex,
num_rows: RowId,
) -> Box<dyn Iterator<Item = RowId> + 'a> {
match column_index_opt {
None | Some(ColumnIndex::Empty { .. }) => {
Box::new(iter::repeat(0u32).take(num_rows as usize + 1))
}
Some(ColumnIndex::Full) => Box::new(0..num_rows + 1),
Some(ColumnIndex::Optional(optional_index)) => {
ColumnIndex::Empty { .. } => Box::new(iter::repeat(0u32).take(num_rows as usize + 1)),
ColumnIndex::Full => Box::new(0..num_rows + 1),
ColumnIndex::Optional(optional_index) => {
Box::new(
(0..num_rows)
// TODO optimize
@@ -86,9 +84,7 @@ fn convert_column_opt_to_multivalued_index<'a>(
.chain(std::iter::once(optional_index.num_non_nulls())),
)
}
Some(ColumnIndex::Multivalued(multivalued_index)) => {
multivalued_index.start_index_column.iter()
}
ColumnIndex::Multivalued(multivalued_index) => multivalued_index.start_index_column.iter(),
}
}
@@ -97,7 +93,6 @@ impl<'a> Iterable<RowId> for StackedMultivaluedIndex<'a> {
let multivalued_indexes =
self.columns
.iter()
.map(Option::as_ref)
.enumerate()
.map(|(columnar_id, column_opt)| {
let num_rows =
@@ -116,10 +111,7 @@ fn stack_multivalued_indexes<'a>(
let mut last_row_id = 0;
let mut current_it = multivalued_indexes.next();
Box::new(std::iter::from_fn(move || loop {
let Some(multivalued_index) = current_it.as_mut() else {
return None;
};
if let Some(row_id) = multivalued_index.next() {
if let Some(row_id) = current_it.as_mut()?.next() {
last_row_id = offset + row_id;
return Some(last_row_id);
}

82
columnar/src/column_index/mod.rs
View File

@@ -1,3 +1,8 @@
//! # `column_index`
//!
//! `column_index` provides rank and select operations to associate positions when not all
//! documents have exactly one element.
mod merge;
mod multivalued_index;
mod optional_index;
@@ -12,7 +17,7 @@ pub use serialize::{open_column_index, serialize_column_index, SerializableColum
use crate::column_index::multivalued_index::MultiValueIndex;
use crate::{Cardinality, DocId, RowId};
#[derive(Clone)]
#[derive(Clone, Debug)]
pub enum ColumnIndex {
Empty {
num_docs: u32,
@@ -37,11 +42,15 @@ impl From<MultiValueIndex> for ColumnIndex {
}
impl ColumnIndex {
/// Returns the cardinality of the column index.
///
/// By convention, if the column contains no docs, we consider that it is
/// full.
#[inline]
pub fn get_cardinality(&self) -> Cardinality {
match self {
ColumnIndex::Empty { num_docs: 0 } | ColumnIndex::Full => Cardinality::Full,
ColumnIndex::Empty { .. } => Cardinality::Optional,
ColumnIndex::Full => Cardinality::Full,
ColumnIndex::Optional(_) => Cardinality::Optional,
ColumnIndex::Multivalued(_) => Cardinality::Multivalued,
}
@@ -74,18 +83,57 @@ impl ColumnIndex {
}
}
pub fn docid_range_to_rowids(&self, doc_id: Range<DocId>) -> Range<RowId> {
/// Translates a block of docis to row_ids.
///
/// returns the row_ids and the matching docids on the same index
/// e.g.
/// DocId In: [0, 5, 6]
/// DocId Out: [0, 0, 6, 6]
/// RowId Out: [0, 1, 2, 3]
#[inline]
pub fn docids_to_rowids(
&self,
doc_ids: &[DocId],
doc_ids_out: &mut Vec<DocId>,
row_ids: &mut Vec<RowId>,
) {
match self {
ColumnIndex::Empty { .. } => {}
ColumnIndex::Full => {
doc_ids_out.extend_from_slice(doc_ids);
row_ids.extend_from_slice(doc_ids);
}
ColumnIndex::Optional(optional_index) => {
for doc_id in doc_ids {
if let Some(row_id) = optional_index.rank_if_exists(*doc_id) {
doc_ids_out.push(*doc_id);
row_ids.push(row_id);
}
}
}
ColumnIndex::Multivalued(multivalued_index) => {
for doc_id in doc_ids {
for row_id in multivalued_index.range(*doc_id) {
doc_ids_out.push(*doc_id);
row_ids.push(row_id);
}
}
}
}
}
pub fn docid_range_to_rowids(&self, doc_id_range: Range<DocId>) -> Range<RowId> {
match self {
ColumnIndex::Empty { .. } => 0..0,
ColumnIndex::Full => doc_id,
ColumnIndex::Full => doc_id_range,
ColumnIndex::Optional(optional_index) => {
let row_start = optional_index.rank(doc_id.start);
let row_end = optional_index.rank(doc_id.end);
let row_start = optional_index.rank(doc_id_range.start);
let row_end = optional_index.rank(doc_id_range.end);
row_start..row_end
}
ColumnIndex::Multivalued(multivalued_index) => {
let end_docid = doc_id.end.min(multivalued_index.num_docs() - 1) + 1;
let start_docid = doc_id.start.min(end_docid);
let end_docid = doc_id_range.end.min(multivalued_index.num_docs() - 1) + 1;
let start_docid = doc_id_range.start.min(end_docid);
let row_start = multivalued_index.start_index_column.get_val(start_docid);
let row_end = multivalued_index.start_index_column.get_val(end_docid);
@@ -113,3 +161,21 @@ impl ColumnIndex {
}
}
}
#[cfg(test)]
mod tests {
use crate::{Cardinality, ColumnIndex};
#[test]
fn test_column_index_get_cardinality() {
assert_eq!(
ColumnIndex::Empty { num_docs: 0 }.get_cardinality(),
Cardinality::Full
);
assert_eq!(ColumnIndex::Full.get_cardinality(), Cardinality::Full);
assert_eq!(
ColumnIndex::Empty { num_docs: 1 }.get_cardinality(),
Cardinality::Optional
);
}
}

15
columnar/src/column_index/multivalued_index.rs
View File

@@ -35,6 +35,14 @@ pub struct MultiValueIndex {
pub start_index_column: Arc<dyn crate::ColumnValues<RowId>>,
}
impl std::fmt::Debug for MultiValueIndex {
fn fmt(&self, f: &mut std::fmt::Formatter) -> std::fmt::Result {
f.debug_struct("MultiValuedIndex")
.field("num_rows", &self.start_index_column.num_vals())
.finish_non_exhaustive()
}
}
impl From<Arc<dyn ColumnValues<RowId>>> for MultiValueIndex {
fn from(start_index_column: Arc<dyn ColumnValues<RowId>>) -> Self {
MultiValueIndex { start_index_column }
@@ -106,11 +114,8 @@ impl MultiValueIndex {
#[cfg(test)]
mod tests {
use std::ops::Range;
use std::sync::Arc;
use super::MultiValueIndex;
use crate::column_values::IterColumn;
use crate::{ColumnValues, RowId};
fn index_to_pos_helper(
index: &MultiValueIndex,
@@ -124,9 +129,7 @@ mod tests {
#[test]
fn test_positions_to_docid() {
let offsets: Vec<RowId> = vec![0, 10, 12, 15, 22, 23]; // docid values are [0..10, 10..12, 12..15, etc.]
let column: Arc<dyn ColumnValues<RowId>> = Arc::new(IterColumn::from(offsets.into_iter()));
let index = MultiValueIndex::from(column);
let index = MultiValueIndex::for_test(&[0, 10, 12, 15, 22, 23]);
assert_eq!(index.num_docs(), 5);
let positions = &[10u32, 11, 15, 20, 21, 22];
assert_eq!(index_to_pos_helper(&index, 0..5, positions), vec![1, 3, 4]);

26
columnar/src/column_index/optional_index/mod.rs
View File

@@ -21,8 +21,6 @@ const DENSE_BLOCK_THRESHOLD: u32 =
const ELEMENTS_PER_BLOCK: u32 = u16::MAX as u32 + 1;
const BLOCK_SIZE: RowId = 1 << 16;
#[derive(Copy, Clone, Debug)]
struct BlockMeta {
non_null_rows_before_block: u32,
@@ -88,6 +86,15 @@ pub struct OptionalIndex {
block_metas: Arc<[BlockMeta]>,
}
impl std::fmt::Debug for OptionalIndex {
fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
f.debug_struct("OptionalIndex")
.field("num_rows", &self.num_rows)
.field("num_non_null_rows", &self.num_non_null_rows)
.finish_non_exhaustive()
}
}
/// Splits a value address into lower and upper 16bits.
/// The lower 16 bits are the value in the block
/// The upper 16 bits are the block index
@@ -100,8 +107,8 @@ struct RowAddr {
#[inline(always)]
fn row_addr_from_row_id(row_id: RowId) -> RowAddr {
RowAddr {
block_id: (row_id / BLOCK_SIZE) as u16,
in_block_row_id: (row_id % BLOCK_SIZE) as u16,
block_id: (row_id / ELEMENTS_PER_BLOCK) as u16,
in_block_row_id: (row_id % ELEMENTS_PER_BLOCK) as u16,
}
}
@@ -176,8 +183,13 @@ impl Set<RowId> for OptionalIndex {
}
}
/// Any value doc_id is allowed.
/// In particular, doc_id = num_rows.
#[inline]
fn rank(&self, doc_id: DocId) -> RowId {
if doc_id >= self.num_docs() {
return self.num_non_nulls();
}
let RowAddr {
block_id,
in_block_row_id,
@@ -191,13 +203,15 @@ impl Set<RowId> for OptionalIndex {
block_meta.non_null_rows_before_block + block_offset_row_id
}
/// Any value doc_id is allowed.
/// In particular, doc_id = num_rows.
#[inline]
fn rank_if_exists(&self, doc_id: DocId) -> Option<RowId> {
let RowAddr {
block_id,
in_block_row_id,
} = row_addr_from_row_id(doc_id);
let block_meta = self.block_metas[block_id as usize];
let block_meta = *self.block_metas.get(block_id as usize)?;
let block = self.block(block_meta);
let block_offset_row_id = match block {
Block::Dense(dense_block) => dense_block.rank_if_exists(in_block_row_id),
@@ -482,7 +496,7 @@ fn deserialize_optional_index_block_metadatas(
non_null_rows_before_block += num_non_null_rows;
}
block_metas.resize(
((num_rows + BLOCK_SIZE - 1) / BLOCK_SIZE) as usize,
((num_rows + ELEMENTS_PER_BLOCK - 1) / ELEMENTS_PER_BLOCK) as usize,
BlockMeta {
non_null_rows_before_block,
start_byte_offset,

3
columnar/src/column_index/optional_index/set.rs
View File

@@ -39,7 +39,8 @@ pub trait Set<T> {
///
/// # Panics
///
/// May panic if rank is greater than the number of elements in the Set.
/// May panic if rank is greater or equal to the number of
/// elements in the Set.
fn select(&self, rank: T) -> T;
/// Creates a brand new select cursor.

1
columnar/src/column_index/optional_index/set_block/dense.rs
View File

@@ -1,4 +1,3 @@
use std::convert::TryInto;
use std::io::{self, Write};
use common::BinarySerializable;

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

@@ -1,8 +1,31 @@
use proptest::prelude::{any, prop, *};
use proptest::strategy::Strategy;
use proptest::prelude::*;
use proptest::{prop_oneof, proptest};
use super::*;
use crate::{ColumnarReader, ColumnarWriter, DynamicColumnHandle};
#[test]
fn test_optional_index_bug_2293() {
// tests for panic in docid_range_to_rowids for docid == num_docs
test_optional_index_with_num_docs(ELEMENTS_PER_BLOCK - 1);
test_optional_index_with_num_docs(ELEMENTS_PER_BLOCK);
test_optional_index_with_num_docs(ELEMENTS_PER_BLOCK + 1);
}
fn test_optional_index_with_num_docs(num_docs: u32) {
let mut dataframe_writer = ColumnarWriter::default();
dataframe_writer.record_numerical(100, "score", 80i64);
let mut buffer: Vec<u8> = Vec::new();
dataframe_writer
.serialize(num_docs, None, &mut buffer)
.unwrap();
let columnar = ColumnarReader::open(buffer).unwrap();
assert_eq!(columnar.num_columns(), 1);
let cols: Vec<DynamicColumnHandle> = columnar.read_columns("score").unwrap();
assert_eq!(cols.len(), 1);
let col = cols[0].open().unwrap();
col.column_index().docid_range_to_rowids(0..num_docs);
}
#[test]
fn test_dense_block_threshold() {
@@ -35,7 +58,7 @@ proptest! {
#[test]
fn test_with_random_sets_simple() {
let vals = 10..BLOCK_SIZE * 2;
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();
@@ -171,7 +194,7 @@ fn test_optional_index_rank() {
test_optional_index_rank_aux(&[0u32, 1u32]);
let mut block = Vec::new();
block.push(3u32);
block.extend((0..BLOCK_SIZE).map(|i| i + BLOCK_SIZE + 1));
block.extend((0..ELEMENTS_PER_BLOCK).map(|i| i + ELEMENTS_PER_BLOCK + 1));
test_optional_index_rank_aux(&block);
}
@@ -185,8 +208,8 @@ fn test_optional_index_iter_empty_one() {
fn test_optional_index_iter_dense_block() {
let mut block = Vec::new();
block.push(3u32);
block.extend((0..BLOCK_SIZE).map(|i| i + BLOCK_SIZE + 1));
test_optional_index_iter_aux(&block, 3 * BLOCK_SIZE);
block.extend((0..ELEMENTS_PER_BLOCK).map(|i| i + ELEMENTS_PER_BLOCK + 1));
test_optional_index_iter_aux(&block, 3 * ELEMENTS_PER_BLOCK);
}
#[test]
@@ -215,12 +238,12 @@ mod bench {
let vals: Vec<RowId> = (0..TOTAL_NUM_VALUES)
.map(|_| rng.gen_bool(fill_ratio))
.enumerate()
.filter(|(pos, val)| *val)
.filter(|(_pos, val)| *val)
.map(|(pos, _)| pos as RowId)
.collect();
serialize_optional_index(&&vals[..], TOTAL_NUM_VALUES, &mut out).unwrap();
let codec = open_optional_index(OwnedBytes::new(out)).unwrap();
codec
open_optional_index(OwnedBytes::new(out)).unwrap()
}
fn random_range_iterator(
@@ -242,7 +265,7 @@ mod bench {
}
fn n_percent_step_iterator(percent: f32, num_values: u32) -> impl Iterator<Item = u32> {
let ratio = percent as f32 / 100.0;
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)

2
columnar/src/column_index/serialize.rs
View File

@@ -30,6 +30,7 @@ impl<'a> SerializableColumnIndex<'a> {
}
}
/// Serialize a column index.
pub fn serialize_column_index(
column_index: SerializableColumnIndex,
output: &mut impl Write,
@@ -51,6 +52,7 @@ pub fn serialize_column_index(
Ok(column_index_num_bytes)
}
/// Open a serialized column index.
pub fn open_column_index(mut bytes: OwnedBytes) -> io::Result<ColumnIndex> {
if bytes.is_empty() {
return Err(io::Error::new(

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

@@ -5,12 +5,12 @@ use crate::iterable::Iterable;
use crate::{ColumnIndex, ColumnValues, MergeRowOrder};
pub(crate) struct MergedColumnValues<'a, T> {
pub(crate) column_indexes: &'a [Option<ColumnIndex>],
pub(crate) column_indexes: &'a [ColumnIndex],
pub(crate) column_values: &'a [Option<Arc<dyn ColumnValues<T>>>],
pub(crate) merge_row_order: &'a MergeRowOrder,
}
impl<'a, T: Copy + PartialOrd + Debug> Iterable<T> for MergedColumnValues<'a, T> {
impl<'a, T: Copy + PartialOrd + Debug + 'static> Iterable<T> for MergedColumnValues<'a, T> {
fn boxed_iter(&self) -> Box<dyn Iterator<Item = T> + '_> {
match self.merge_row_order {
MergeRowOrder::Stack(_) => Box::new(
@@ -23,8 +23,7 @@ impl<'a, T: Copy + PartialOrd + Debug> Iterable<T> for MergedColumnValues<'a, T>
shuffle_merge_order
.iter_new_to_old_row_addrs()
.flat_map(|row_addr| {
let column_index =
self.column_indexes[row_addr.segment_ord as usize].as_ref()?;
let column_index = &self.column_indexes[row_addr.segment_ord as usize];
let column_values =
self.column_values[row_addr.segment_ord as usize].as_ref()?;
let value_range = column_index.value_row_ids(row_addr.row_id);

165
columnar/src/column_values/mod.rs
View File

@@ -2,7 +2,7 @@
//! # `fastfield_codecs`
//!
//! - Columnar storage of data for tantivy [`Column`].
//! - Columnar storage of data for tantivy [`crate::Column`].
//! - Encode data in different codecs.
//! - Monotonically map values to u64/u128
@@ -10,6 +10,7 @@ use std::fmt::Debug;
use std::ops::{Range, RangeInclusive};
use std::sync::Arc;
use downcast_rs::DowncastSync;
pub use monotonic_mapping::{MonotonicallyMappableToU64, StrictlyMonotonicFn};
pub use monotonic_mapping_u128::MonotonicallyMappableToU128;
@@ -25,7 +26,10 @@ mod monotonic_column;
pub(crate) use merge::MergedColumnValues;
pub use stats::ColumnStats;
pub use u128_based::{open_u128_mapped, serialize_column_values_u128};
pub use u128_based::{
open_u128_as_compact_u64, open_u128_mapped, serialize_column_values_u128,
CompactSpaceU64Accessor,
};
pub use u64_based::{
load_u64_based_column_values, serialize_and_load_u64_based_column_values,
serialize_u64_based_column_values, CodecType, ALL_U64_CODEC_TYPES,
@@ -41,7 +45,7 @@ use crate::RowId;
///
/// Any methods with a default and specialized implementation need to be called in the
/// wrappers that implement the trait: Arc and MonotonicMappingColumn
pub trait ColumnValues<T: PartialOrd = u64>: Send + Sync {
pub trait ColumnValues<T: PartialOrd = u64>: Send + Sync + DowncastSync {
/// Return the value associated with the given idx.
///
/// This accessor should return as fast as possible.
@@ -58,10 +62,50 @@ pub trait ColumnValues<T: PartialOrd = u64>: Send + Sync {
/// # Panics
///
/// May panic if `idx` is greater than the column length.
fn get_vals(&self, idx: &[u32], output: &mut [T]) {
assert!(idx.len() == output.len());
for (out, idx) in output.iter_mut().zip(idx.iter()) {
*out = self.get_val(*idx as u32);
fn get_vals(&self, indexes: &[u32], output: &mut [T]) {
assert!(indexes.len() == output.len());
let out_and_idx_chunks = output.chunks_exact_mut(4).zip(indexes.chunks_exact(4));
for (out_x4, idx_x4) in out_and_idx_chunks {
out_x4[0] = self.get_val(idx_x4[0]);
out_x4[1] = self.get_val(idx_x4[1]);
out_x4[2] = self.get_val(idx_x4[2]);
out_x4[3] = self.get_val(idx_x4[3]);
}
let out_and_idx_chunks = output
.chunks_exact_mut(4)
.into_remainder()
.iter_mut()
.zip(indexes.chunks_exact(4).remainder());
for (out, idx) in out_and_idx_chunks {
*out = self.get_val(*idx);
}
}
/// Allows to push down multiple fetch calls, to avoid dynamic dispatch overhead.
/// The slightly weird `Option<T>` in output allows pushdown to full columns.
///
/// idx and output should have the same length
///
/// # Panics
///
/// May panic if `idx` is greater than the column length.
fn get_vals_opt(&self, indexes: &[u32], output: &mut [Option<T>]) {
assert!(indexes.len() == output.len());
let out_and_idx_chunks = output.chunks_exact_mut(4).zip(indexes.chunks_exact(4));
for (out_x4, idx_x4) in out_and_idx_chunks {
out_x4[0] = Some(self.get_val(idx_x4[0]));
out_x4[1] = Some(self.get_val(idx_x4[1]));
out_x4[2] = Some(self.get_val(idx_x4[2]));
out_x4[3] = Some(self.get_val(idx_x4[3]));
}
let out_and_idx_chunks = output
.chunks_exact_mut(4)
.into_remainder()
.iter_mut()
.zip(indexes.chunks_exact(4).remainder());
for (out, idx) in out_and_idx_chunks {
*out = Some(self.get_val(*idx));
}
}
@@ -83,7 +127,6 @@ pub trait ColumnValues<T: PartialOrd = u64>: Send + Sync {
/// Get the row ids of values which are in the provided value range.
///
/// Note that position == docid for single value fast fields
#[inline(always)]
fn get_row_ids_for_value_range(
&self,
value_range: RangeInclusive<T>,
@@ -91,7 +134,7 @@ pub trait ColumnValues<T: PartialOrd = u64>: Send + Sync {
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.start..row_id_range.end {
for idx in row_id_range {
let val = self.get_val(idx);
if value_range.contains(&val) {
row_id_hits.push(idx);
@@ -99,20 +142,26 @@ pub trait ColumnValues<T: PartialOrd = u64>: Send + Sync {
}
}
/// Returns the minimum value for this fast field.
/// Returns a lower bound for this column of values.
///
/// This min_value may not be exact.
/// For instance, the min value does not take in account of possible
/// deleted document. All values are however guaranteed to be higher than
/// `.min_value()`.
/// All values are guaranteed to be higher than `.min_value()`
/// but this value is not necessary the best boundary value.
///
/// We have
/// ∀i < self.num_vals(), self.get_val(i) >= self.min_value()
/// But we don't have necessarily
/// ∃i < self.num_vals(), self.get_val(i) == self.min_value()
fn min_value(&self) -> T;
/// Returns the maximum value for this fast field.
/// Returns an upper bound for this column of values.
///
/// This max_value may not be exact.
/// For instance, the max value does not take in account of possible
/// deleted document. All values are however guaranteed to be higher than
/// `.max_value()`.
/// All values are guaranteed to be lower than `.max_value()`
/// but this value is not necessary the best boundary value.
///
/// We have
/// ∀i < self.num_vals(), self.get_val(i) <= self.max_value()
/// But we don't have necessarily
/// ∃i < self.num_vals(), self.get_val(i) == self.max_value()
fn max_value(&self) -> T;
/// The number of values in the column.
@@ -123,13 +172,40 @@ pub trait ColumnValues<T: PartialOrd = u64>: Send + Sync {
Box::new((0..self.num_vals()).map(|idx| self.get_val(idx)))
}
}
downcast_rs::impl_downcast!(sync ColumnValues<T> where T: PartialOrd);
impl<T: Copy + PartialOrd + Debug> ColumnValues<T> for Arc<dyn ColumnValues<T>> {
/// Empty column of values.
pub struct EmptyColumnValues;
impl<T: PartialOrd + Default> ColumnValues<T> for EmptyColumnValues {
fn get_val(&self, _idx: u32) -> T {
panic!("Internal Error: Called get_val of empty column.")
}
fn min_value(&self) -> T {
T::default()
}
fn max_value(&self) -> T {
T::default()
}
fn num_vals(&self) -> u32 {
0
}
}
impl<T: Copy + PartialOrd + Debug + 'static> ColumnValues<T> for Arc<dyn ColumnValues<T>> {
#[inline(always)]
fn get_val(&self, idx: u32) -> T {
self.as_ref().get_val(idx)
}
#[inline(always)]
fn get_vals_opt(&self, indexes: &[u32], output: &mut [Option<T>]) {
self.as_ref().get_vals_opt(indexes, output)
}
#[inline(always)]
fn min_value(&self) -> T {
self.as_ref().min_value()
@@ -167,54 +243,5 @@ impl<T: Copy + PartialOrd + Debug> ColumnValues<T> for Arc<dyn ColumnValues<T>>
}
}
/// Wraps an cloneable iterator into a `Column`.
pub struct IterColumn<T>(T);
impl<T> From<T> for IterColumn<T>
where T: Iterator + Clone + ExactSizeIterator
{
fn from(iter: T) -> Self {
IterColumn(iter)
}
}
impl<T> ColumnValues<T::Item> for IterColumn<T>
where
T: Iterator + Clone + ExactSizeIterator + Send + Sync,
T::Item: PartialOrd + Debug,
{
fn get_val(&self, idx: u32) -> T::Item {
self.0.clone().nth(idx as usize).unwrap()
}
fn min_value(&self) -> T::Item {
self.0.clone().next().unwrap()
}
fn max_value(&self) -> T::Item {
self.0.clone().last().unwrap()
}
fn num_vals(&self) -> u32 {
self.0.len() as u32
}
fn iter(&self) -> Box<dyn Iterator<Item = T::Item> + '_> {
Box::new(self.0.clone())
}
}
#[cfg(all(test, feature = "unstable"))]
mod bench;
#[cfg(test)]
mod tests {
use super::*;
#[test]
fn test_range_as_col() {
let col = IterColumn::from(10..100);
assert_eq!(col.num_vals(), 90);
assert_eq!(col.max_value(), 99);
}
}

20
columnar/src/column_values/monotonic_column.rs
View File

@@ -31,10 +31,10 @@ pub fn monotonic_map_column<C, T, Input, Output>(
monotonic_mapping: T,
) -> impl ColumnValues<Output>
where
C: ColumnValues<Input>,
T: StrictlyMonotonicFn<Input, Output> + Send + Sync,
Input: PartialOrd + Debug + Send + Sync + Clone,
Output: PartialOrd + Debug + Send + Sync + Clone,
C: ColumnValues<Input> + 'static,
T: StrictlyMonotonicFn<Input, Output> + Send + Sync + 'static,
Input: PartialOrd + Debug + Send + Sync + Clone + 'static,
Output: PartialOrd + Debug + Send + Sync + Clone + 'static,
{
MonotonicMappingColumn {
from_column,
@@ -45,12 +45,12 @@ where
impl<C, T, Input, Output> ColumnValues<Output> for MonotonicMappingColumn<C, T, Input>
where
C: ColumnValues<Input>,
T: StrictlyMonotonicFn<Input, Output> + Send + Sync,
Input: PartialOrd + Send + Debug + Sync + Clone,
Output: PartialOrd + Send + Debug + Sync + Clone,
C: ColumnValues<Input> + 'static,
T: StrictlyMonotonicFn<Input, Output> + Send + Sync + 'static,
Input: PartialOrd + Send + Debug + Sync + Clone + 'static,
Output: PartialOrd + Send + Debug + Sync + Clone + 'static,
{
#[inline]
#[inline(always)]
fn get_val(&self, idx: u32) -> Output {
let from_val = self.from_column.get_val(idx);
self.monotonic_mapping.mapping(from_val)
@@ -107,7 +107,7 @@ mod tests {
#[test]
fn test_monotonic_mapping_iter() {
let vals: Vec<u64> = (0..100u64).map(|el| el * 10).collect();
let col = VecColumn::from(&vals);
let col = VecColumn::from(vals);
let mapped = monotonic_map_column(
col,
StrictlyMonotonicMappingInverter::from(StrictlyMonotonicMappingToInternal::<i64>::new()),

4
columnar/src/column_values/monotonic_mapping.rs
View File

@@ -139,12 +139,12 @@ impl MonotonicallyMappableToU64 for i64 {
impl MonotonicallyMappableToU64 for DateTime {
#[inline(always)]
fn to_u64(self) -> u64 {
common::i64_to_u64(self.into_timestamp_micros())
common::i64_to_u64(self.into_timestamp_nanos())
}
#[inline(always)]
fn from_u64(val: u64) -> Self {
DateTime::from_timestamp_micros(common::u64_to_i64(val))
DateTime::from_timestamp_nanos(common::u64_to_i64(val))
}
}

2
columnar/src/column_values/u128_based/compact_space/blank_range.rs
View File

@@ -38,6 +38,6 @@ impl Ord for BlankRange {
}
impl PartialOrd for BlankRange {
fn partial_cmp(&self, other: &Self) -> Option<std::cmp::Ordering> {
Some(self.blank_size().cmp(&other.blank_size()))
Some(self.cmp(other))
}
}

37
columnar/src/column_values/u128_based/compact_space/build_compact_space.rs
View File

@@ -10,7 +10,7 @@ use super::{CompactSpace, RangeMapping};
/// Put the blanks for the sorted values into a binary heap
fn get_blanks(values_sorted: &BTreeSet<u128>) -> BinaryHeap<BlankRange> {
let mut blanks: BinaryHeap<BlankRange> = BinaryHeap::new();
for (first, second) in values_sorted.iter().tuple_windows() {
for (first, second) in values_sorted.iter().copied().tuple_windows() {
// Correctness Overflow: the values are deduped and sorted (BTreeSet property), that means
// there's always space between two values.
let blank_range = first + 1..=second - 1;
@@ -65,12 +65,12 @@ pub fn get_compact_space(
return compact_space_builder.finish();
}
let mut blanks: BinaryHeap<BlankRange> = get_blanks(values_deduped_sorted);
// Replace after stabilization of https://github.com/rust-lang/rust/issues/62924
// We start by space that's limited to min_value..=max_value
let min_value = *values_deduped_sorted.iter().next().unwrap_or(&0);
let max_value = *values_deduped_sorted.iter().last().unwrap_or(&0);
// Replace after stabilization of https://github.com/rust-lang/rust/issues/62924
let min_value = values_deduped_sorted.iter().next().copied().unwrap_or(0);
let max_value = values_deduped_sorted.iter().last().copied().unwrap_or(0);
let mut blanks: BinaryHeap<BlankRange> = get_blanks(values_deduped_sorted);
// +1 for null, in case min and max covers the whole space, we are off by one.
let mut amplitude_compact_space = (max_value - min_value).saturating_add(1);
@@ -84,6 +84,7 @@ pub fn get_compact_space(
let mut amplitude_bits: u8 = num_bits(amplitude_compact_space);
let mut blank_collector = BlankCollector::new();
// We will stage blanks until they reduce the compact space by at least 1 bit and then flush
// them if the metadata cost is lower than the total number of saved bits.
// Binary heap to process the gaps by their size
@@ -93,6 +94,7 @@ pub fn get_compact_space(
let staged_spaces_sum: u128 = blank_collector.staged_blanks_sum();
let amplitude_new_compact_space = amplitude_compact_space - staged_spaces_sum;
let amplitude_new_bits = num_bits(amplitude_new_compact_space);
if amplitude_bits == amplitude_new_bits {
continue;
}
@@ -100,7 +102,16 @@ pub fn get_compact_space(
// TODO: Maybe calculate exact cost of blanks and run this more expensive computation only,
// when amplitude_new_bits changes
let cost = blank_collector.num_staged_blanks() * cost_per_blank;
if cost >= saved_bits {
// We want to end up with a compact space that fits into 32 bits.
// In order to deal with pathological cases, we force the algorithm to keep
// refining the compact space the amplitude bits is lower than 32.
//
// The worst case scenario happens for a large number of u128s regularly
// spread over the full u128 space.
//
// This change will force the algorithm to degenerate into dictionary encoding.
if amplitude_bits <= 32 && cost >= saved_bits {
// Continue here, since although we walk over the blanks by size,
// we can potentially save a lot at the last bits, which are smaller blanks
//
@@ -115,6 +126,8 @@ pub fn get_compact_space(
compact_space_builder.add_blanks(blank_collector.drain().map(|blank| blank.blank_range()));
}
assert!(amplitude_bits <= 32);
// special case, when we don't collected any blanks because:
// * the data is empty (early exit)
// * the algorithm did decide it's not worth the cost, which can be the case for single values
@@ -199,7 +212,7 @@ impl CompactSpaceBuilder {
covered_space.push(0..=0); // empty data case
};
let mut compact_start: u64 = 1; // 0 is reserved for `null`
let mut compact_start: u32 = 1; // 0 is reserved for `null`
let mut ranges_mapping: Vec<RangeMapping> = Vec::with_capacity(covered_space.len());
for cov in covered_space {
let range_mapping = super::RangeMapping {
@@ -218,6 +231,7 @@ impl CompactSpaceBuilder {
#[cfg(test)]
mod tests {
use super::*;
use crate::column_values::u128_based::compact_space::COST_PER_BLANK_IN_BITS;
#[test]
fn test_binary_heap_pop_order() {
@@ -228,4 +242,11 @@ mod tests {
assert_eq!(blanks.pop().unwrap().blank_size(), 101);
assert_eq!(blanks.pop().unwrap().blank_size(), 11);
}
#[test]
fn test_worst_case_scenario() {
let vals: BTreeSet<u128> = (0..8).map(|i| i * ((1u128 << 34) / 8)).collect();
let compact_space = get_compact_space(&vals, vals.len() as u32, COST_PER_BLANK_IN_BITS);
assert!(compact_space.amplitude_compact_space() < u32::MAX as u128);
}
}

227
columnar/src/column_values/u128_based/compact_space/mod.rs
View File

@@ -22,7 +22,7 @@ mod build_compact_space;
use build_compact_space::get_compact_space;
use common::{BinarySerializable, CountingWriter, OwnedBytes, VInt, VIntU128};
use tantivy_bitpacker::{self, BitPacker, BitUnpacker};
use tantivy_bitpacker::{BitPacker, BitUnpacker};
use crate::column_values::ColumnValues;
use crate::RowId;
@@ -42,15 +42,15 @@ pub struct CompactSpace {
#[derive(Debug, Clone, Eq, PartialEq)]
struct RangeMapping {
value_range: RangeInclusive<u128>,
compact_start: u64,
compact_start: u32,
}
impl RangeMapping {
fn range_length(&self) -> u64 {
(self.value_range.end() - self.value_range.start()) as u64 + 1
fn range_length(&self) -> u32 {
(self.value_range.end() - self.value_range.start()) as u32 + 1
}
// The last value of the compact space in this range
fn compact_end(&self) -> u64 {
fn compact_end(&self) -> u32 {
self.compact_start + self.range_length() - 1
}
}
@@ -81,7 +81,7 @@ impl BinarySerializable for CompactSpace {
let num_ranges = VInt::deserialize(reader)?.0;
let mut ranges_mapping: Vec<RangeMapping> = vec![];
let mut value = 0u128;
let mut compact_start = 1u64; // 0 is reserved for `null`
let mut compact_start = 1u32; // 0 is reserved for `null`
for _ in 0..num_ranges {
let blank_delta_start = VIntU128::deserialize(reader)?.0;
value += blank_delta_start;
@@ -122,10 +122,10 @@ impl CompactSpace {
/// Returns either Ok(the value in the compact space) or if it is outside the compact space the
/// Err(position where it would be inserted)
fn u128_to_compact(&self, value: u128) -> Result<u64, usize> {
fn u128_to_compact(&self, value: u128) -> Result<u32, usize> {
self.ranges_mapping
.binary_search_by(|probe| {
let value_range = &probe.value_range;
let value_range: &RangeInclusive<u128> = &probe.value_range;
if value < *value_range.start() {
Ordering::Greater
} else if value > *value_range.end() {
@@ -136,19 +136,19 @@ impl CompactSpace {
})
.map(|pos| {
let range_mapping = &self.ranges_mapping[pos];
let pos_in_range = (value - range_mapping.value_range.start()) as u64;
let pos_in_range: u32 = (value - range_mapping.value_range.start()) as u32;
range_mapping.compact_start + pos_in_range
})
}
/// Unpacks a value from compact space u64 to u128 space
fn compact_to_u128(&self, compact: u64) -> u128 {
/// Unpacks a value from compact space u32 to u128 space
fn compact_to_u128(&self, compact: u32) -> u128 {
let pos = self
.ranges_mapping
.binary_search_by_key(&compact, |range_mapping| range_mapping.compact_start)
// Correctness: Overflow. The first range starts at compact space 0, the error from
// binary search can never be 0
.map_or_else(|e| e - 1, |v| v);
.unwrap_or_else(|e| e - 1);
let range_mapping = &self.ranges_mapping[pos];
let diff = compact - range_mapping.compact_start;
@@ -178,11 +178,15 @@ impl CompactSpaceCompressor {
/// Taking the vals as Vec may cost a lot of memory. It is used to sort the vals.
pub fn train_from(iter: impl Iterator<Item = u128>) -> Self {
let mut values_sorted = BTreeSet::new();
// Total number of values, with their redundancy.
let mut total_num_values = 0u32;
for val in iter {
total_num_values += 1u32;
values_sorted.insert(val);
}
let min_value = *values_sorted.iter().next().unwrap_or(&0);
let max_value = *values_sorted.iter().last().unwrap_or(&0);
let compact_space =
get_compact_space(&values_sorted, total_num_values, COST_PER_BLANK_IN_BITS);
let amplitude_compact_space = compact_space.amplitude_compact_space();
@@ -193,13 +197,12 @@ impl CompactSpaceCompressor {
);
let num_bits = tantivy_bitpacker::compute_num_bits(amplitude_compact_space as u64);
let min_value = *values_sorted.iter().next().unwrap_or(&0);
let max_value = *values_sorted.iter().last().unwrap_or(&0);
assert_eq!(
compact_space
.u128_to_compact(max_value)
.expect("could not convert max value to compact space"),
amplitude_compact_space as u64
amplitude_compact_space as u32
);
CompactSpaceCompressor {
params: IPCodecParams {
@@ -240,7 +243,7 @@ impl CompactSpaceCompressor {
"Could not convert value to compact_space. This is a bug.",
)
})?;
bitpacker.write(compact, self.params.num_bits, write)?;
bitpacker.write(compact as u64, self.params.num_bits, write)?;
}
bitpacker.close(write)?;
self.write_footer(write)?;
@@ -289,6 +292,63 @@ impl BinarySerializable for IPCodecParams {
}
}
/// Exposes the compact space compressed values as u64.
///
/// This allows faster access to the values, as u64 is faster to work with than u128.
/// It also allows to handle u128 values like u64, via the `open_u64_lenient` as a uniform
/// access interface.
///
/// When converting from the internal u64 to u128 `compact_to_u128` can be used.
pub struct CompactSpaceU64Accessor(CompactSpaceDecompressor);
impl CompactSpaceU64Accessor {
pub(crate) fn open(data: OwnedBytes) -> io::Result<CompactSpaceU64Accessor> {
let decompressor = CompactSpaceU64Accessor(CompactSpaceDecompressor::open(data)?);
Ok(decompressor)
}
/// Convert a compact space value to u128
pub fn compact_to_u128(&self, compact: u32) -> u128 {
self.0.compact_to_u128(compact)
}
}
impl ColumnValues<u64> for CompactSpaceU64Accessor {
#[inline]
fn get_val(&self, doc: u32) -> u64 {
let compact = self.0.get_compact(doc);
compact as u64
}
fn min_value(&self) -> u64 {
self.0.u128_to_compact(self.0.min_value()).unwrap() as u64
}
fn max_value(&self) -> u64 {
self.0.u128_to_compact(self.0.max_value()).unwrap() as u64
}
fn num_vals(&self) -> u32 {
self.0.params.num_vals
}
#[inline]
fn iter(&self) -> Box<dyn Iterator<Item = u64> + '_> {
Box::new(self.0.iter_compact().map(|el| el as u64))
}
#[inline]
fn get_row_ids_for_value_range(
&self,
value_range: RangeInclusive<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)
}
}
impl ColumnValues<u128> for CompactSpaceDecompressor {
#[inline]
fn get_val(&self, doc: u32) -> u128 {
@@ -314,48 +374,6 @@ impl ColumnValues<u128> for CompactSpaceDecompressor {
#[inline]
fn get_row_ids_for_value_range(
&self,
value_range: RangeInclusive<u128>,
positions_range: Range<u32>,
positions: &mut Vec<u32>,
) {
self.get_positions_for_value_range(value_range, positions_range, positions)
}
}
impl CompactSpaceDecompressor {
pub fn open(data: OwnedBytes) -> io::Result<CompactSpaceDecompressor> {
let (data_slice, footer_len_bytes) = data.split_at(data.len() - 4);
let footer_len = u32::deserialize(&mut &footer_len_bytes[..])?;
let data_footer = &data_slice[data_slice.len() - footer_len as usize..];
let params = IPCodecParams::deserialize(&mut &data_footer[..])?;
let decompressor = CompactSpaceDecompressor { data, params };
Ok(decompressor)
}
/// Converting to compact space for the decompressor is more complex, since we may get values
/// which are outside the compact space. e.g. if we map
/// 1000 => 5
/// 2000 => 6
///
/// and we want a mapping for 1005, there is no equivalent compact space. We instead return an
/// error with the index of the next range.
fn u128_to_compact(&self, value: u128) -> Result<u64, usize> {
self.params.compact_space.u128_to_compact(value)
}
fn compact_to_u128(&self, compact: u64) -> u128 {
self.params.compact_space.compact_to_u128(compact)
}
/// Comparing on compact space: Random dataset 0,24 (50% random hit) - 1.05 GElements/s
/// Comparing on compact space: Real dataset 1.08 GElements/s
///
/// Comparing on original space: Real dataset .06 GElements/s (not completely optimized)
#[inline]
pub fn get_positions_for_value_range(
&self,
value_range: RangeInclusive<u128>,
position_range: Range<u32>,
@@ -395,44 +413,42 @@ impl CompactSpaceDecompressor {
range_mapping.compact_end()
});
let range = compact_from..=compact_to;
let value_range = compact_from..=compact_to;
self.get_positions_for_compact_value_range(value_range, position_range, positions);
}
}
let scan_num_docs = position_range.end - position_range.start;
impl CompactSpaceDecompressor {
pub fn open(data: OwnedBytes) -> io::Result<CompactSpaceDecompressor> {
let (data_slice, footer_len_bytes) = data.split_at(data.len() - 4);
let footer_len = u32::deserialize(&mut &footer_len_bytes[..])?;
let step_size = 4;
let cutoff = position_range.start + scan_num_docs - scan_num_docs % step_size;
let data_footer = &data_slice[data_slice.len() - footer_len as usize..];
let params = IPCodecParams::deserialize(&mut &data_footer[..])?;
let decompressor = CompactSpaceDecompressor { data, params };
let mut push_if_in_range = |idx, val| {
if range.contains(&val) {
positions.push(idx);
}
};
let get_val = |idx| self.params.bit_unpacker.get(idx, &self.data);
// unrolled loop
for idx in (position_range.start..cutoff).step_by(step_size as usize) {
let idx1 = idx;
let idx2 = idx + 1;
let idx3 = idx + 2;
let idx4 = idx + 3;
let val1 = get_val(idx1);
let val2 = get_val(idx2);
let val3 = get_val(idx3);
let val4 = get_val(idx4);
push_if_in_range(idx1, val1);
push_if_in_range(idx2, val2);
push_if_in_range(idx3, val3);
push_if_in_range(idx4, val4);
}
Ok(decompressor)
}
// handle rest
for idx in cutoff..position_range.end {
push_if_in_range(idx, get_val(idx));
}
/// Converting to compact space for the decompressor is more complex, since we may get values
/// which are outside the compact space. e.g. if we map
/// 1000 => 5
/// 2000 => 6
///
/// and we want a mapping for 1005, there is no equivalent compact space. We instead return an
/// error with the index of the next range.
fn u128_to_compact(&self, value: u128) -> Result<u32, usize> {
self.params.compact_space.u128_to_compact(value)
}
fn compact_to_u128(&self, compact: u32) -> u128 {
self.params.compact_space.compact_to_u128(compact)
}
#[inline]
fn iter_compact(&self) -> impl Iterator<Item = u64> + '_ {
(0..self.params.num_vals).map(move |idx| self.params.bit_unpacker.get(idx, &self.data))
fn iter_compact(&self) -> impl Iterator<Item = u32> + '_ {
(0..self.params.num_vals)
.map(move |idx| self.params.bit_unpacker.get(idx, &self.data) as u32)
}
#[inline]
@@ -443,9 +459,14 @@ impl CompactSpaceDecompressor {
.map(|compact| self.compact_to_u128(compact))
}
#[inline]
pub fn get_compact(&self, idx: u32) -> u32 {
self.params.bit_unpacker.get(idx, &self.data) as u32
}
#[inline]
pub fn get(&self, idx: u32) -> u128 {
let compact = self.params.bit_unpacker.get(idx, &self.data);
let compact = self.get_compact(idx);
self.compact_to_u128(compact)
}
@@ -456,6 +477,20 @@ impl CompactSpaceDecompressor {
pub fn max_value(&self) -> u128 {
self.params.max_value
}
fn get_positions_for_compact_value_range(
&self,
value_range: RangeInclusive<u32>,
position_range: Range<u32>,
positions: &mut Vec<u32>,
) {
self.params.bit_unpacker.get_ids_for_value_range(
*value_range.start() as u64..=*value_range.end() as u64,
position_range,
&self.data,
positions,
);
}
}
#[cfg(test)]
@@ -469,12 +504,12 @@ mod tests {
#[test]
fn compact_space_test() {
let ips = &[
let ips: BTreeSet<u128> = [
2u128, 4u128, 1000, 1001, 1002, 1003, 1004, 1005, 1008, 1010, 1012, 1260,
]
.into_iter()
.collect();
let compact_space = get_compact_space(ips, ips.len() as u32, 11);
let compact_space = get_compact_space(&ips, ips.len() as u32, 11);
let amplitude = compact_space.amplitude_compact_space();
assert_eq!(amplitude, 17);
assert_eq!(1, compact_space.u128_to_compact(2).unwrap());
@@ -497,8 +532,8 @@ mod tests {
);
for ip in ips {
let compact = compact_space.u128_to_compact(*ip).unwrap();
assert_eq!(compact_space.compact_to_u128(compact), *ip);
let compact = compact_space.u128_to_compact(ip).unwrap();
assert_eq!(compact_space.compact_to_u128(compact), ip);
}
}
@@ -524,7 +559,7 @@ mod tests {
.map(|pos| pos as u32)
.collect::<Vec<_>>();
let mut positions = Vec::new();
decompressor.get_positions_for_value_range(
decompressor.get_row_ids_for_value_range(
range,
0..decompressor.num_vals(),
&mut positions,
@@ -569,7 +604,7 @@ mod tests {
let val = *val;
let pos = pos as u32;
let mut positions = Vec::new();
decomp.get_positions_for_value_range(val..=val, pos..pos + 1, &mut positions);
decomp.get_row_ids_for_value_range(val..=val, pos..pos + 1, &mut positions);
assert_eq!(positions, vec![pos]);
}

21
columnar/src/column_values/u128_based/mod.rs
View File

@@ -6,7 +6,9 @@ use std::sync::Arc;
mod compact_space;
use common::{BinarySerializable, OwnedBytes, VInt};
use compact_space::{CompactSpaceCompressor, CompactSpaceDecompressor};
pub use compact_space::{
CompactSpaceCompressor, CompactSpaceDecompressor, CompactSpaceU64Accessor,
};
use crate::column_values::monotonic_map_column;
use crate::column_values::monotonic_mapping::{
@@ -108,6 +110,23 @@ pub fn open_u128_mapped<T: MonotonicallyMappableToU128 + Debug>(
StrictlyMonotonicMappingToInternal::<T>::new().into();
Ok(Arc::new(monotonic_map_column(reader, inverted)))
}
/// Returns the u64 representation of the u128 data.
/// The internal representation of the data as u64 is useful for faster processing.
///
/// In order to convert to u128 back cast to `CompactSpaceU64Accessor` and call
/// `compact_to_u128`.
///
/// # 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>>> {
let header = U128Header::deserialize(&mut bytes)?;
assert_eq!(header.codec_type, U128FastFieldCodecType::CompactSpace);
let reader = CompactSpaceU64Accessor::open(bytes)?;
Ok(Arc::new(reader))
}
#[cfg(test)]
pub mod tests {
use super::*;

63
columnar/src/column_values/u64_based/bitpacked.rs
View File

@@ -1,4 +1,6 @@
use std::io::{self, Write};
use std::num::NonZeroU64;
use std::ops::{Range, RangeInclusive};
use common::{BinarySerializable, OwnedBytes};
use fastdivide::DividerU64;
@@ -16,12 +18,51 @@ pub struct BitpackedReader {
stats: ColumnStats,
}
#[inline(always)]
const fn div_ceil(n: u64, q: NonZeroU64) -> u64 {
// copied from unstable rust standard library.
let d = n / q.get();
let r = n % q.get();
if r > 0 {
d + 1
} else {
d
}
}
// The bitpacked codec applies a linear transformation `f` over data that are bitpacked.
// f is defined by:
// f: bitpacked -> stats.min_value + stats.gcd * bitpacked
//
// In order to run range queries, we invert the transformation.
// `transform_range_before_linear_transformation` returns the range of values
// [min_bipacked_value..max_bitpacked_value] such that
// f(bitpacked) ∈ [min_value, max_value] <=> bitpacked ∈ [min_bitpacked_value, max_bitpacked_value]
fn transform_range_before_linear_transformation(
stats: &ColumnStats,
range: RangeInclusive<u64>,
) -> Option<RangeInclusive<u64>> {
if range.is_empty() {
return None;
}
if stats.min_value > *range.end() {
return None;
}
if stats.max_value < *range.start() {
return None;
}
let shifted_range =
range.start().saturating_sub(stats.min_value)..=range.end().saturating_sub(stats.min_value);
let start_before_gcd_multiplication: u64 = div_ceil(*shifted_range.start(), stats.gcd);
let end_before_gcd_multiplication: u64 = *shifted_range.end() / stats.gcd;
Some(start_before_gcd_multiplication..=end_before_gcd_multiplication)
}
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)
}
#[inline]
fn min_value(&self) -> u64 {
self.stats.min_value
@@ -34,6 +75,26 @@ impl ColumnValues for BitpackedReader {
fn num_vals(&self) -> RowId {
self.stats.num_rows
}
fn get_row_ids_for_value_range(
&self,
range: RangeInclusive<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,
);
}
}
fn num_bits(stats: &ColumnStats) -> u8 {

7
columnar/src/column_values/u64_based/blockwise_linear.rs
View File

@@ -63,7 +63,10 @@ impl BlockwiseLinearEstimator {
if self.block.is_empty() {
return;
}
let line = Line::train(&VecColumn::from(&self.block));
let column = VecColumn::from(std::mem::take(&mut self.block));
let line = Line::train(&column);
self.block = column.into();
let mut max_value = 0u64;
for (i, buffer_val) in self.block.iter().enumerate() {
let interpolated_val = line.eval(i as u32);
@@ -125,7 +128,7 @@ impl ColumnCodecEstimator for BlockwiseLinearEstimator {
*buffer_val = gcd_divider.divide(*buffer_val - stats.min_value);
}
let line = Line::train(&VecColumn::from(&buffer));
let line = Line::train(&VecColumn::from(buffer.to_vec()));
assert!(!buffer.is_empty());

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

@@ -184,7 +184,7 @@ mod tests {
}
fn test_eval_max_err(ys: &[u64]) -> Option<u64> {
let line = Line::train(&VecColumn::from(&ys));
let line = Line::train(&VecColumn::from(ys.to_vec()));
ys.iter()
.enumerate()
.map(|(x, y)| y.wrapping_sub(line.eval(x as u32)))

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

@@ -173,7 +173,9 @@ impl LinearCodecEstimator {
fn collect_before_line_estimation(&mut self, value: u64) {
self.block.push(value);
if self.block.len() == LINE_ESTIMATION_BLOCK_LEN {
let line = Line::train(&VecColumn::from(&self.block));
let column = VecColumn::from(std::mem::take(&mut self.block));
let line = Line::train(&column);
self.block = column.into();
let block = std::mem::take(&mut self.block);
for val in block {
self.collect_after_line_estimation(&line, val);

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

@@ -27,7 +27,7 @@ pub struct StatsCollector {
// This is the same as computing the difference between the values and the first value.
//
// This way, we can compress i64-converted-to-u64 (e.g. timestamp that were supplied in
// seconds, only to be converted in microseconds).
// seconds, only to be converted in nanoseconds).
increment_gcd_opt: Option<(NonZeroU64, DividerU64)>,
first_value_opt: Option<u64>,
}

17
columnar/src/column_values/u64_based/tests.rs
View File

@@ -1,6 +1,5 @@
use proptest::prelude::*;
use proptest::strategy::Strategy;
use proptest::{num, prop_oneof, proptest};
use proptest::{prop_oneof, proptest};
#[test]
fn test_serialize_and_load_simple() {
@@ -99,14 +98,28 @@ pub(crate) fn create_and_validate<TColumnCodec: ColumnCodec>(
let reader = TColumnCodec::load(OwnedBytes::new(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() {
let val = reader.get_val(doc as u32);
assert_eq!(
val, orig_val,
"val `{val}` does not match orig_val {orig_val:?}, in data set {name}, data `{vals:?}`",
);
buffer.resize(1, 0);
reader.get_vals(&[doc as u32], &mut buffer);
let val = buffer[0];
assert_eq!(
val, orig_val,
"val `{val}` does not match orig_val {orig_val:?}, in data set {name}, data `{vals:?}`",
);
}
let all_docs: Vec<u32> = (0..vals.len() as u32).collect();
buffer.resize(all_docs.len(), 0);
reader.get_vals(&all_docs, &mut buffer);
assert_eq!(vals, buffer);
if !vals.is_empty() {
let test_rand_idx = rand::thread_rng().gen_range(0..=vals.len() - 1);
let expected_positions: Vec<u32> = vals

20
columnar/src/column_values/vec_column.rs
View File

@@ -4,14 +4,14 @@ use tantivy_bitpacker::minmax;
use crate::ColumnValues;
/// VecColumn provides `Column` over a slice.
pub struct VecColumn<'a, T = u64> {
pub(crate) values: &'a [T],
/// VecColumn provides `Column` over a `Vec<T>`.
pub struct VecColumn<T = u64> {
pub(crate) values: Vec<T>,
pub(crate) min_value: T,
pub(crate) max_value: T,
}
impl<'a, T: Copy + PartialOrd + Send + Sync + Debug> ColumnValues<T> for VecColumn<'a, T> {
impl<T: Copy + PartialOrd + Send + Sync + Debug + 'static> ColumnValues<T> for VecColumn<T> {
fn get_val(&self, position: u32) -> T {
self.values[position as usize]
}
@@ -37,11 +37,8 @@ impl<'a, T: Copy + PartialOrd + Send + Sync + Debug> ColumnValues<T> for VecColu
}
}
impl<'a, T: Copy + PartialOrd + Default, V> From<&'a V> for VecColumn<'a, T>
where V: AsRef<[T]> + ?Sized
{
fn from(values: &'a V) -> Self {
let values = values.as_ref();
impl<T: Copy + PartialOrd + Default> From<Vec<T>> for VecColumn<T> {
fn from(values: Vec<T>) -> Self {
let (min_value, max_value) = minmax(values.iter().copied()).unwrap_or_default();
Self {
values,
@@ -50,3 +47,8 @@ where V: AsRef<[T]> + ?Sized
}
}
}
impl From<VecColumn> for Vec<u64> {
fn from(column: VecColumn) -> Self {
column.values
}
}

20
columnar/src/columnar/column_type.rs
View File

@@ -1,3 +1,4 @@
use std::fmt;
use std::fmt::Debug;
use std::net::Ipv6Addr;
@@ -21,6 +22,22 @@ pub enum ColumnType {
DateTime = 7u8,
}
impl fmt::Display for ColumnType {
fn fmt(&self, f: &mut fmt::Formatter) -> fmt::Result {
let short_str = match self {
ColumnType::I64 => "i64",
ColumnType::U64 => "u64",
ColumnType::F64 => "f64",
ColumnType::Bytes => "bytes",
ColumnType::Str => "str",
ColumnType::Bool => "bool",
ColumnType::IpAddr => "ip",
ColumnType::DateTime => "datetime",
};
write!(f, "{short_str}")
}
}
// The order needs to match _exactly_ the order in the enum
const COLUMN_TYPES: [ColumnType; 8] = [
ColumnType::I64,
@@ -37,6 +54,9 @@ impl ColumnType {
pub fn to_code(self) -> u8 {
self as u8
}
pub fn is_date_time(&self) -> bool {
self == &ColumnType::DateTime
}
pub(crate) fn try_from_code(code: u8) -> Result<ColumnType, InvalidData> {
COLUMN_TYPES.get(code as usize).copied().ok_or(InvalidData)

38
columnar/src/columnar/merge/merge_dict_column.rs
View File

@@ -1,7 +1,7 @@
use std::io::{self, Write};
use common::{BitSet, CountingWriter, ReadOnlyBitSet};
use sstable::{SSTable, TermOrdinal};
use sstable::{SSTable, Streamer, TermOrdinal, VoidSSTable};
use super::term_merger::TermMerger;
use crate::column::serialize_column_mappable_to_u64;
@@ -56,17 +56,19 @@ impl<'a> RemappedTermOrdinalsValues<'a> {
.bytes_columns
.iter()
.enumerate()
.flat_map(|(segment_ord, byte_column)| {
let segment_ord = self.term_ord_mapping.get_segment(segment_ord as u32);
byte_column.iter().flat_map(move |bytes_column| {
bytes_column
.ords()
.values
.iter()
.map(move |term_ord| segment_ord[term_ord as usize])
})
.flat_map(|(seg_ord, bytes_column_opt)| {
let bytes_column = bytes_column_opt.as_ref()?;
Some((seg_ord, bytes_column))
})
.flat_map(move |(seg_ord, bytes_column)| {
let term_ord_after_merge_mapping =
self.term_ord_mapping.get_segment(seg_ord as u32);
bytes_column
.ords()
.values
.iter()
.map(move |term_ord| term_ord_after_merge_mapping[term_ord as usize])
});
// TODO see if we can better decompose the mapping / and the stacking
Box::new(iter)
}
@@ -124,16 +126,20 @@ fn serialize_merged_dict(
let mut term_ord_mapping = TermOrdinalMapping::default();
let mut field_term_streams = Vec::new();
for column in bytes_columns.iter().flatten() {
term_ord_mapping.add_segment(column.dictionary.num_terms());
let terms = column.dictionary.stream()?;
field_term_streams.push(terms);
for column_opt in bytes_columns.iter() {
if let Some(column) = column_opt {
term_ord_mapping.add_segment(column.dictionary.num_terms());
let terms: Streamer<VoidSSTable> = column.dictionary.stream()?;
field_term_streams.push(terms);
} else {
term_ord_mapping.add_segment(0);
field_term_streams.push(Streamer::empty());
}
}
let mut merged_terms = TermMerger::new(field_term_streams);
let mut sstable_builder = sstable::VoidSSTable::writer(output);
// TODO support complex `merge_row_order`.
match merge_row_order {
MergeRowOrder::Stack(_) => {
let mut current_term_ord = 0;

15
columnar/src/columnar/merge/merge_mapping.rs
View File

@@ -11,6 +11,17 @@ pub struct StackMergeOrder {
}
impl StackMergeOrder {
#[cfg(test)]
pub fn stack_for_test(num_rows_per_columnar: &[u32]) -> StackMergeOrder {
let mut cumulated_row_ids: Vec<RowId> = Vec::with_capacity(num_rows_per_columnar.len());
let mut cumulated_row_id = 0;
for &num_rows in num_rows_per_columnar {
cumulated_row_id += num_rows;
cumulated_row_ids.push(cumulated_row_id);
}
StackMergeOrder { cumulated_row_ids }
}
pub fn stack(columnars: &[&ColumnarReader]) -> StackMergeOrder {
let mut cumulated_row_ids: Vec<RowId> = Vec::with_capacity(columnars.len());
let mut cumulated_row_id = 0;
@@ -41,8 +52,8 @@ pub enum MergeRowOrder {
/// Columnar tables are simply stacked one above the other.
/// If the i-th columnar_readers has n_rows_i rows, then
/// in the resulting columnar,
/// rows [r0..n_row_0) contains the row of columnar_readers[0], in ordder
/// rows [n_row_0..n_row_0 + n_row_1 contains the row of columnar_readers[1], in order.
/// rows [r0..n_row_0) contains the row of `columnar_readers[0]`, in ordder
/// rows [n_row_0..n_row_0 + n_row_1 contains the row of `columnar_readers[1]`, in order.
/// ..
/// No documents is deleted.
Stack(StackMergeOrder),

291
columnar/src/columnar/merge/mod.rs
View File

@@ -2,11 +2,12 @@ mod merge_dict_column;
mod merge_mapping;
mod term_merger;
use std::collections::{BTreeMap, HashMap, HashSet};
use std::collections::{BTreeMap, HashSet};
use std::io;
use std::net::Ipv6Addr;
use std::sync::Arc;
use itertools::Itertools;
pub use merge_mapping::{MergeRowOrder, ShuffleMergeOrder, StackMergeOrder};
use super::writer::ColumnarSerializer;
@@ -17,7 +18,8 @@ use crate::columnar::writer::CompatibleNumericalTypes;
use crate::columnar::ColumnarReader;
use crate::dynamic_column::DynamicColumn;
use crate::{
BytesColumn, Column, ColumnIndex, ColumnType, ColumnValues, NumericalType, NumericalValue,
BytesColumn, Column, ColumnIndex, ColumnType, ColumnValues, DynamicColumnHandle, NumericalType,
NumericalValue,
};
/// Column types are grouped into different categories.
@@ -27,14 +29,16 @@ use crate::{
/// In practise, today, only Numerical colummns are coerced into one type today.
///
/// See also [README.md].
#[derive(Copy, Clone, Eq, PartialEq, Hash, Debug)]
enum ColumnTypeCategory {
Bool,
Str,
///
/// The ordering has to match the ordering of the variants in [ColumnType].
#[derive(Copy, Clone, Eq, PartialOrd, Ord, PartialEq, Hash, Debug)]
pub(crate) enum ColumnTypeCategory {
Numerical,
DateTime,
Bytes,
Str,
Bool,
IpAddr,
DateTime,
}
impl From<ColumnType> for ColumnTypeCategory {
@@ -78,20 +82,37 @@ pub fn merge_columnar(
output: &mut impl io::Write,
) -> io::Result<()> {
let mut serializer = ColumnarSerializer::new(output);
let num_rows_per_columnar = columnar_readers
.iter()
.map(|reader| reader.num_rows())
.collect::<Vec<u32>>();
let columns_to_merge =
group_columns_for_merge(columnar_readers, required_columns, &merge_row_order)?;
for res in columns_to_merge {
let ((column_name, _column_type_category), grouped_columns) = res;
let grouped_columns = grouped_columns.open(&merge_row_order)?;
if grouped_columns.is_empty() {
continue;
}
let column_type = grouped_columns.column_type_after_merge();
let mut columns = grouped_columns.columns;
coerce_columns(column_type, &mut columns)?;
let columns_to_merge = group_columns_for_merge(columnar_readers, required_columns)?;
for ((column_name, column_type), columns) in columns_to_merge {
let mut column_serializer =
serializer.serialize_column(column_name.as_bytes(), column_type);
serializer.start_serialize_column(column_name.as_bytes(), column_type);
merge_column(
column_type,
&num_rows_per_columnar,
columns,
&merge_row_order,
&mut column_serializer,
)?;
column_serializer.finalize()?;
}
serializer.finalize(merge_row_order.num_rows())?;
serializer.finalize(merge_row_order.num_rows())?;
Ok(())
}
@@ -108,6 +129,7 @@ fn dynamic_column_to_u64_monotonic(dynamic_column: DynamicColumn) -> Option<Colu
fn merge_column(
column_type: ColumnType,
num_docs_per_column: &[u32],
columns: Vec<Option<DynamicColumn>>,
merge_row_order: &MergeRowOrder,
wrt: &mut impl io::Write,
@@ -118,17 +140,19 @@ fn merge_column(
| ColumnType::F64
| ColumnType::DateTime
| ColumnType::Bool => {
let mut column_indexes: Vec<Option<ColumnIndex>> = Vec::with_capacity(columns.len());
let mut column_indexes: Vec<ColumnIndex> = Vec::with_capacity(columns.len());
let mut column_values: Vec<Option<Arc<dyn ColumnValues>>> =
Vec::with_capacity(columns.len());
for dynamic_column_opt in columns {
if let Some(Column { idx, values }) =
for (i, dynamic_column_opt) in columns.into_iter().enumerate() {
if let Some(Column { index: idx, values }) =
dynamic_column_opt.and_then(dynamic_column_to_u64_monotonic)
{
column_indexes.push(Some(idx));
column_indexes.push(idx);
column_values.push(Some(values));
} else {
column_indexes.push(None);
column_indexes.push(ColumnIndex::Empty {
num_docs: num_docs_per_column[i],
});
column_values.push(None);
}
}
@@ -142,15 +166,19 @@ fn merge_column(
serialize_column_mappable_to_u64(merged_column_index, &merge_column_values, wrt)?;
}
ColumnType::IpAddr => {
let mut column_indexes: Vec<Option<ColumnIndex>> = Vec::with_capacity(columns.len());
let mut column_indexes: Vec<ColumnIndex> = Vec::with_capacity(columns.len());
let mut column_values: Vec<Option<Arc<dyn ColumnValues<Ipv6Addr>>>> =
Vec::with_capacity(columns.len());
for dynamic_column_opt in columns {
if let Some(DynamicColumn::IpAddr(Column { idx, values })) = dynamic_column_opt {
column_indexes.push(Some(idx));
for (i, dynamic_column_opt) in columns.into_iter().enumerate() {
if let Some(DynamicColumn::IpAddr(Column { index: idx, values })) =
dynamic_column_opt
{
column_indexes.push(idx);
column_values.push(Some(values));
} else {
column_indexes.push(None);
column_indexes.push(ColumnIndex::Empty {
num_docs: num_docs_per_column[i],
});
column_values.push(None);
}
}
@@ -166,20 +194,22 @@ fn merge_column(
serialize_column_mappable_to_u128(merged_column_index, &merge_column_values, wrt)?;
}
ColumnType::Bytes | ColumnType::Str => {
let mut column_indexes: Vec<Option<ColumnIndex>> = Vec::with_capacity(columns.len());
let mut column_indexes: Vec<ColumnIndex> = Vec::with_capacity(columns.len());
let mut bytes_columns: Vec<Option<BytesColumn>> = Vec::with_capacity(columns.len());
for dynamic_column_opt in columns {
for (i, dynamic_column_opt) in columns.into_iter().enumerate() {
match dynamic_column_opt {
Some(DynamicColumn::Str(str_column)) => {
column_indexes.push(Some(str_column.term_ord_column.idx.clone()));
column_indexes.push(str_column.term_ord_column.index.clone());
bytes_columns.push(Some(str_column.into()));
}
Some(DynamicColumn::Bytes(bytes_column)) => {
column_indexes.push(Some(bytes_column.term_ord_column.idx.clone()));
column_indexes.push(bytes_column.term_ord_column.index.clone());
bytes_columns.push(Some(bytes_column));
}
_ => {
column_indexes.push(None);
column_indexes.push(ColumnIndex::Empty {
num_docs: num_docs_per_column[i],
});
bytes_columns.push(None);
}
}
@@ -195,40 +225,12 @@ fn merge_column(
struct GroupedColumns {
required_column_type: Option<ColumnType>,
columns: Vec<Option<DynamicColumn>>,
column_category: ColumnTypeCategory,
}
impl GroupedColumns {
fn for_category(column_category: ColumnTypeCategory, num_columnars: usize) -> Self {
GroupedColumns {
required_column_type: None,
columns: vec![None; num_columnars],
column_category,
}
}
/// Set the dynamic column for a given columnar.
fn set_column(&mut self, columnar_id: usize, column: DynamicColumn) {
self.columns[columnar_id] = Some(column);
}
/// Force the existence of a column, as well as its type.
fn require_type(&mut self, required_type: ColumnType) -> io::Result<()> {
if let Some(existing_required_type) = self.required_column_type {
if existing_required_type == required_type {
// This was just a duplicate in the `required_columns`.
// Nothing to do.
return Ok(());
} else {
return Err(io::Error::new(
io::ErrorKind::InvalidInput,
"Required column conflicts with another required column of the same type \
category.",
));
}
}
self.required_column_type = Some(required_type);
Ok(())
/// Check is column group can be skipped during serialization.
fn is_empty(&self) -> bool {
self.required_column_type.is_none() && self.columns.iter().all(Option::is_none)
}
/// Returns the column type after merge.
@@ -250,11 +252,76 @@ impl GroupedColumns {
}
// At the moment, only the numerical categorical column type has more than one possible
// column type.
assert_eq!(self.column_category, ColumnTypeCategory::Numerical);
assert!(self
.columns
.iter()
.flatten()
.all(|el| ColumnTypeCategory::from(el.column_type()) == ColumnTypeCategory::Numerical));
merged_numerical_columns_type(self.columns.iter().flatten()).into()
}
}
struct GroupedColumnsHandle {
required_column_type: Option<ColumnType>,
columns: Vec<Option<DynamicColumnHandle>>,
}
impl GroupedColumnsHandle {
fn new(num_columnars: usize) -> Self {
GroupedColumnsHandle {
required_column_type: None,
columns: vec![None; num_columnars],
}
}
fn open(self, merge_row_order: &MergeRowOrder) -> io::Result<GroupedColumns> {
let mut columns: Vec<Option<DynamicColumn>> = Vec::new();
for (columnar_id, column) in self.columns.iter().enumerate() {
if let Some(column) = column {
let column = column.open()?;
// We skip columns that end up with 0 documents.
// That way, we make sure they don't end up influencing the merge type or
// creating empty columns.
if is_empty_after_merge(merge_row_order, &column, columnar_id) {
columns.push(None);
} else {
columns.push(Some(column));
}
} else {
columns.push(None);
}
}
Ok(GroupedColumns {
required_column_type: self.required_column_type,
columns,
})
}
/// Set the dynamic column for a given columnar.
fn set_column(&mut self, columnar_id: usize, column: DynamicColumnHandle) {
self.columns[columnar_id] = Some(column);
}
/// Force the existence of a column, as well as its type.
fn require_type(&mut self, required_type: ColumnType) -> io::Result<()> {
if let Some(existing_required_type) = self.required_column_type {
if existing_required_type == required_type {
// This was just a duplicate in the `required_columns`.
// Nothing to do.
return Ok(());
} else {
return Err(io::Error::new(
io::ErrorKind::InvalidInput,
"Required column conflicts with another required column of the same type \
category.",
));
}
}
self.required_column_type = Some(required_type);
Ok(())
}
}
/// Returns the type of the merged numerical column.
///
/// This function picks the first numerical type out of i64, u64, f64 (order matters
@@ -275,48 +342,92 @@ fn merged_numerical_columns_type<'a>(
compatible_numerical_types.to_numerical_type()
}
#[allow(clippy::type_complexity)]
fn group_columns_for_merge(
columnar_readers: &[&ColumnarReader],
required_columns: &[(String, ColumnType)],
) -> io::Result<BTreeMap<(String, ColumnType), Vec<Option<DynamicColumn>>>> {
// Each column name may have multiple types of column associated.
// For merging we are interested in the same column type category since they can be merged.
let mut columns_grouped: HashMap<(String, ColumnTypeCategory), GroupedColumns> = HashMap::new();
fn is_empty_after_merge(
merge_row_order: &MergeRowOrder,
column: &DynamicColumn,
columnar_ord: usize,
) -> bool {
if column.num_values() == 0u32 {
// It was empty before the merge.
return true;
}
match merge_row_order {
MergeRowOrder::Stack(_) => {
// If we are stacking the columnar, no rows are being deleted.
false
}
MergeRowOrder::Shuffled(shuffled) => {
if let Some(alive_bitset) = &shuffled.alive_bitsets[columnar_ord] {
let column_index = column.column_index();
match column_index {
ColumnIndex::Empty { .. } => true,
ColumnIndex::Full => alive_bitset.len() == 0,
ColumnIndex::Optional(optional_index) => {
for doc in optional_index.iter_rows() {
if alive_bitset.contains(doc) {
return false;
}
}
true
}
ColumnIndex::Multivalued(multivalued_index) => {
for (doc_id, (start_index, end_index)) in multivalued_index
.start_index_column
.iter()
.tuple_windows()
.enumerate()
{
let doc_id = doc_id as u32;
if start_index == end_index {
// There are no values in this document
continue;
}
// The document contains values and is present in the alive bitset.
// The column is therefore not empty.
if alive_bitset.contains(doc_id) {
return false;
}
}
true
}
}
} else {
// No document is being deleted.
// The shuffle is applying a permutation.
false
}
}
}
}
/// Iterates over the columns of the columnar readers, grouped by column name.
/// Key functionality is that `open` of the Columns is done lazy per group.
fn group_columns_for_merge<'a>(
columnar_readers: &'a [&'a ColumnarReader],
required_columns: &'a [(String, ColumnType)],
_merge_row_order: &'a MergeRowOrder,
) -> io::Result<BTreeMap<(String, ColumnTypeCategory), GroupedColumnsHandle>> {
let mut columns: BTreeMap<(String, ColumnTypeCategory), GroupedColumnsHandle> = BTreeMap::new();
for &(ref column_name, column_type) in required_columns {
columns_grouped
columns
.entry((column_name.clone(), column_type.into()))
.or_insert_with(|| {
GroupedColumns::for_category(column_type.into(), columnar_readers.len())
})
.or_insert_with(|| GroupedColumnsHandle::new(columnar_readers.len()))
.require_type(column_type)?;
}
for (columnar_id, columnar_reader) in columnar_readers.iter().enumerate() {
let column_name_and_handle = columnar_reader.list_columns()?;
let column_name_and_handle = columnar_reader.iter_columns()?;
for (column_name, handle) in column_name_and_handle {
let column_category: ColumnTypeCategory = handle.column_type().into();
let column = handle.open()?;
columns_grouped
columns
.entry((column_name, column_category))
.or_insert_with(|| {
GroupedColumns::for_category(column_category, columnar_readers.len())
})
.set_column(columnar_id, column);
.or_insert_with(|| GroupedColumnsHandle::new(columnar_readers.len()))
.set_column(columnar_id, handle);
}
}
let mut merge_columns: BTreeMap<(String, ColumnType), Vec<Option<DynamicColumn>>> =
Default::default();
for ((column_name, _), mut grouped_columns) in columns_grouped {
let column_type = grouped_columns.column_type_after_merge();
coerce_columns(column_type, &mut grouped_columns.columns)?;
merge_columns.insert((column_name, column_type), grouped_columns.columns);
}
Ok(merge_columns)
Ok(columns)
}
fn coerce_columns(
@@ -361,8 +472,8 @@ fn coerce_column(column_type: ColumnType, column: DynamicColumn) -> io::Result<D
fn min_max_if_numerical(column: &DynamicColumn) -> Option<(NumericalValue, NumericalValue)> {
match column {
DynamicColumn::I64(column) => Some((column.min_value().into(), column.max_value().into())),
DynamicColumn::U64(column) => Some((column.min_value().into(), column.min_value().into())),
DynamicColumn::F64(column) => Some((column.min_value().into(), column.min_value().into())),
DynamicColumn::U64(column) => Some((column.min_value().into(), column.max_value().into())),
DynamicColumn::F64(column) => Some((column.min_value().into(), column.max_value().into())),
DynamicColumn::Bool(_)
| DynamicColumn::IpAddr(_)
| DynamicColumn::DateTime(_)

292
columnar/src/columnar/merge/tests.rs
View File

@@ -23,70 +23,73 @@ fn test_column_coercion_to_u64() {
let columnar1 = make_columnar("numbers", &[1i64]);
// u64 type
let columnar2 = make_columnar("numbers", &[u64::MAX]);
let column_map: BTreeMap<(String, ColumnType), Vec<Option<DynamicColumn>>> =
group_columns_for_merge(&[&columnar1, &columnar2], &[]).unwrap();
let columnars = &[&columnar1, &columnar2];
let merge_order = StackMergeOrder::stack(columnars).into();
let column_map: BTreeMap<(String, ColumnTypeCategory), GroupedColumnsHandle> =
group_columns_for_merge(columnars, &[], &merge_order).unwrap();
assert_eq!(column_map.len(), 1);
assert!(column_map.contains_key(&("numbers".to_string(), ColumnType::U64)));
}
#[test]
fn test_column_no_coercion_if_all_the_same() {
let columnar1 = make_columnar("numbers", &[1u64]);
let columnar2 = make_columnar("numbers", &[2u64]);
let column_map: BTreeMap<(String, ColumnType), Vec<Option<DynamicColumn>>> =
group_columns_for_merge(&[&columnar1, &columnar2], &[]).unwrap();
assert_eq!(column_map.len(), 1);
assert!(column_map.contains_key(&("numbers".to_string(), ColumnType::U64)));
assert!(column_map.contains_key(&("numbers".to_string(), ColumnTypeCategory::Numerical)));
}
#[test]
fn test_column_coercion_to_i64() {
let columnar1 = make_columnar("numbers", &[-1i64]);
let columnar2 = make_columnar("numbers", &[2u64]);
let column_map: BTreeMap<(String, ColumnType), Vec<Option<DynamicColumn>>> =
group_columns_for_merge(&[&columnar1, &columnar2], &[]).unwrap();
let columnars = &[&columnar1, &columnar2];
let merge_order = StackMergeOrder::stack(columnars).into();
let column_map: BTreeMap<(String, ColumnTypeCategory), GroupedColumnsHandle> =
group_columns_for_merge(columnars, &[], &merge_order).unwrap();
assert_eq!(column_map.len(), 1);
assert!(column_map.contains_key(&("numbers".to_string(), ColumnType::I64)));
assert!(column_map.contains_key(&("numbers".to_string(), ColumnTypeCategory::Numerical)));
}
#[test]
fn test_impossible_coercion_returns_an_error() {
let columnar1 = make_columnar("numbers", &[u64::MAX]);
let group_error =
group_columns_for_merge(&[&columnar1], &[("numbers".to_string(), ColumnType::I64)])
.map(|_| ())
.unwrap_err();
assert_eq!(group_error.kind(), io::ErrorKind::InvalidInput);
}
//#[test]
// fn test_impossible_coercion_returns_an_error() {
// let columnar1 = make_columnar("numbers", &[u64::MAX]);
// let merge_order = StackMergeOrder::stack(&[&columnar1]).into();
// let group_error = group_columns_for_merge_iter(
//&[&columnar1],
//&[("numbers".to_string(), ColumnType::I64)],
//&merge_order,
//)
//.unwrap_err();
// assert_eq!(group_error.kind(), io::ErrorKind::InvalidInput);
//}
#[test]
fn test_group_columns_with_required_column() {
let columnar1 = make_columnar("numbers", &[1i64]);
let columnar2 = make_columnar("numbers", &[2u64]);
let column_map: BTreeMap<(String, ColumnType), Vec<Option<DynamicColumn>>> =
let columnars = &[&columnar1, &columnar2];
let merge_order = StackMergeOrder::stack(columnars).into();
let column_map: BTreeMap<(String, ColumnTypeCategory), GroupedColumnsHandle> =
group_columns_for_merge(
&[&columnar1, &columnar2],
&[("numbers".to_string(), ColumnType::U64)],
&merge_order,
)
.unwrap();
assert_eq!(column_map.len(), 1);
assert!(column_map.contains_key(&("numbers".to_string(), ColumnType::U64)));
assert!(column_map.contains_key(&("numbers".to_string(), ColumnTypeCategory::Numerical)));
}
#[test]
fn test_group_columns_required_column_with_no_existing_columns() {
let columnar1 = make_columnar("numbers", &[2u64]);
let columnar2 = make_columnar("numbers", &[2u64]);
let column_map: BTreeMap<(String, ColumnType), Vec<Option<DynamicColumn>>> =
group_columns_for_merge(
&[&columnar1, &columnar2],
&[("required_col".to_string(), ColumnType::Str)],
)
.unwrap();
let columnars = &[&columnar1, &columnar2];
let merge_order = StackMergeOrder::stack(columnars).into();
let column_map: BTreeMap<_, _> = group_columns_for_merge(
columnars,
&[("required_col".to_string(), ColumnType::Str)],
&merge_order,
)
.unwrap();
assert_eq!(column_map.len(), 2);
let columns = column_map
.get(&("required_col".to_string(), ColumnType::Str))
.unwrap();
let columns = &column_map
.get(&("required_col".to_string(), ColumnTypeCategory::Str))
.unwrap()
.columns;
assert_eq!(columns.len(), 2);
assert!(columns[0].is_none());
assert!(columns[1].is_none());
@@ -96,35 +99,42 @@ fn test_group_columns_required_column_with_no_existing_columns() {
fn test_group_columns_required_column_is_above_all_columns_have_the_same_type_rule() {
let columnar1 = make_columnar("numbers", &[2i64]);
let columnar2 = make_columnar("numbers", &[2i64]);
let column_map: BTreeMap<(String, ColumnType), Vec<Option<DynamicColumn>>> =
let columnars = &[&columnar1, &columnar2];
let merge_order = StackMergeOrder::stack(columnars).into();
let column_map: BTreeMap<(String, ColumnTypeCategory), GroupedColumnsHandle> =
group_columns_for_merge(
&[&columnar1, &columnar2],
columnars,
&[("numbers".to_string(), ColumnType::U64)],
&merge_order,
)
.unwrap();
assert_eq!(column_map.len(), 1);
assert!(column_map.contains_key(&("numbers".to_string(), ColumnType::U64)));
assert!(column_map.contains_key(&("numbers".to_string(), ColumnTypeCategory::Numerical)));
}
#[test]
fn test_missing_column() {
let columnar1 = make_columnar("numbers", &[-1i64]);
let columnar2 = make_columnar("numbers2", &[2u64]);
let column_map: BTreeMap<(String, ColumnType), Vec<Option<DynamicColumn>>> =
group_columns_for_merge(&[&columnar1, &columnar2], &[]).unwrap();
let columnars = &[&columnar1, &columnar2];
let merge_order = StackMergeOrder::stack(columnars).into();
let column_map: BTreeMap<(String, ColumnTypeCategory), GroupedColumnsHandle> =
group_columns_for_merge(columnars, &[], &merge_order).unwrap();
assert_eq!(column_map.len(), 2);
assert!(column_map.contains_key(&("numbers".to_string(), ColumnType::I64)));
assert!(column_map.contains_key(&("numbers".to_string(), ColumnTypeCategory::Numerical)));
{
let columns = column_map
.get(&("numbers".to_string(), ColumnType::I64))
.unwrap();
let columns = &column_map
.get(&("numbers".to_string(), ColumnTypeCategory::Numerical))
.unwrap()
.columns;
assert!(columns[0].is_some());
assert!(columns[1].is_none());
}
{
let columns = column_map
.get(&("numbers2".to_string(), ColumnType::U64))
.unwrap();
let columns = &column_map
.get(&("numbers2".to_string(), ColumnTypeCategory::Numerical))
.unwrap()
.columns;
assert!(columns[0].is_none());
assert!(columns[1].is_some());
}
@@ -153,20 +163,24 @@ fn make_numerical_columnar_multiple_columns(
ColumnarReader::open(buffer).unwrap()
}
fn make_byte_columnar_multiple_columns(columns: &[(&str, &[&[&[u8]]])]) -> ColumnarReader {
#[track_caller]
fn make_byte_columnar_multiple_columns(
columns: &[(&str, &[&[&[u8]]])],
num_rows: u32,
) -> ColumnarReader {
let mut dataframe_writer = ColumnarWriter::default();
for (column_name, column_values) in columns {
assert_eq!(
column_values.len(),
num_rows as usize,
"All columns must have `{num_rows}` rows"
);
for (row_id, vals) in column_values.iter().enumerate() {
for val in vals.iter() {
dataframe_writer.record_bytes(row_id as u32, column_name, val);
}
}
}
let num_rows = columns
.iter()
.map(|(_, val_rows)| val_rows.len() as RowId)
.max()
.unwrap_or(0u32);
let mut buffer: Vec<u8> = Vec::new();
dataframe_writer
.serialize(num_rows, None, &mut buffer)
@@ -218,7 +232,9 @@ fn test_merge_columnar_numbers() {
assert_eq!(columnar_reader.num_columns(), 1);
let cols = columnar_reader.read_columns("numbers").unwrap();
let dynamic_column = cols[0].open().unwrap();
let DynamicColumn::F64(vals) = dynamic_column else { panic!() };
let DynamicColumn::F64(vals) = dynamic_column else {
panic!()
};
assert_eq!(vals.get_cardinality(), Cardinality::Optional);
assert_eq!(vals.first(0u32), Some(-1f64));
assert_eq!(vals.first(1u32), None);
@@ -244,7 +260,11 @@ fn test_merge_columnar_texts() {
assert_eq!(columnar_reader.num_columns(), 1);
let cols = columnar_reader.read_columns("texts").unwrap();
let dynamic_column = cols[0].open().unwrap();
let DynamicColumn::Str(vals) = dynamic_column else { panic!() };
let DynamicColumn::Str(vals) = dynamic_column else {
panic!()
};
assert_eq!(vals.ords().get_cardinality(), Cardinality::Optional);
let get_str_for_ord = |ord| {
let mut out = String::new();
vals.ord_to_str(ord, &mut out).unwrap();
@@ -272,8 +292,8 @@ fn test_merge_columnar_texts() {
#[test]
fn test_merge_columnar_byte() {
let columnar1 = make_byte_columnar_multiple_columns(&[("bytes", &[&[b"bbbb"], &[b"baaa"]])]);
let columnar2 = make_byte_columnar_multiple_columns(&[("bytes", &[&[], &[b"a"]])]);
let columnar1 = make_byte_columnar_multiple_columns(&[("bytes", &[&[b"bbbb"], &[b"baaa"]])], 2);
let columnar2 = make_byte_columnar_multiple_columns(&[("bytes", &[&[], &[b"a"]])], 2);
let mut buffer = Vec::new();
let columnars = &[&columnar1, &columnar2];
let stack_merge_order = StackMergeOrder::stack(columnars);
@@ -289,7 +309,9 @@ fn test_merge_columnar_byte() {
assert_eq!(columnar_reader.num_columns(), 1);
let cols = columnar_reader.read_columns("bytes").unwrap();
let dynamic_column = cols[0].open().unwrap();
let DynamicColumn::Bytes(vals) = dynamic_column else { panic!() };
let DynamicColumn::Bytes(vals) = dynamic_column else {
panic!()
};
let get_bytes_for_ord = |ord| {
let mut out = Vec::new();
vals.ord_to_bytes(ord, &mut out).unwrap();
@@ -316,3 +338,155 @@ fn test_merge_columnar_byte() {
assert_eq!(get_bytes_for_row(2), b"");
assert_eq!(get_bytes_for_row(3), b"a");
}
#[test]
fn test_merge_columnar_byte_with_missing() {
let columnar1 = make_byte_columnar_multiple_columns(&[], 3);
let columnar2 = make_byte_columnar_multiple_columns(&[("col", &[&[b"b"], &[]])], 2);
let columnar3 = make_byte_columnar_multiple_columns(
&[
("col", &[&[], &[b"b"], &[b"a", b"b"]]),
("col2", &[&[b"hello"], &[], &[b"a", b"b"]]),
],
3,
);
let mut buffer = Vec::new();
let columnars = &[&columnar1, &columnar2, &columnar3];
let stack_merge_order = StackMergeOrder::stack(columnars);
crate::columnar::merge_columnar(
columnars,
&[],
MergeRowOrder::Stack(stack_merge_order),
&mut buffer,
)
.unwrap();
let columnar_reader = ColumnarReader::open(buffer).unwrap();
assert_eq!(columnar_reader.num_rows(), 3 + 2 + 3);
assert_eq!(columnar_reader.num_columns(), 2);
let cols = columnar_reader.read_columns("col").unwrap();
let dynamic_column = cols[0].open().unwrap();
let DynamicColumn::Bytes(vals) = dynamic_column else {
panic!()
};
let get_bytes_for_ord = |ord| {
let mut out = Vec::new();
vals.ord_to_bytes(ord, &mut out).unwrap();
out
};
assert_eq!(vals.dictionary.num_terms(), 2);
assert_eq!(get_bytes_for_ord(0), b"a");
assert_eq!(get_bytes_for_ord(1), b"b");
let get_bytes_for_row = |row_id| {
let terms: Vec<Vec<u8>> = vals
.term_ords(row_id)
.map(|term_ord| {
let mut out = Vec::new();
vals.ord_to_bytes(term_ord, &mut out).unwrap();
out
})
.collect();
terms
};
assert!(get_bytes_for_row(0).is_empty());
assert!(get_bytes_for_row(1).is_empty());
assert!(get_bytes_for_row(2).is_empty());
assert_eq!(get_bytes_for_row(3), vec![b"b".to_vec()]);
assert!(get_bytes_for_row(4).is_empty());
assert!(get_bytes_for_row(5).is_empty());
assert_eq!(get_bytes_for_row(6), vec![b"b".to_vec()]);
assert_eq!(get_bytes_for_row(7), vec![b"a".to_vec(), b"b".to_vec()]);
}
#[test]
fn test_merge_columnar_different_types() {
let columnar1 = make_text_columnar_multiple_columns(&[("mixed", &[&["a"]])]);
let columnar2 = make_text_columnar_multiple_columns(&[("mixed", &[&[], &["b"]])]);
let columnar3 = make_columnar("mixed", &[1i64]);
let mut buffer = Vec::new();
let columnars = &[&columnar1, &columnar2, &columnar3];
let stack_merge_order = StackMergeOrder::stack(columnars);
crate::columnar::merge_columnar(
columnars,
&[],
MergeRowOrder::Stack(stack_merge_order),
&mut buffer,
)
.unwrap();
let columnar_reader = ColumnarReader::open(buffer).unwrap();
assert_eq!(columnar_reader.num_rows(), 4);
assert_eq!(columnar_reader.num_columns(), 2);
let cols = columnar_reader.read_columns("mixed").unwrap();
// numeric column
let dynamic_column = cols[0].open().unwrap();
let DynamicColumn::I64(vals) = dynamic_column else {
panic!()
};
assert_eq!(vals.get_cardinality(), Cardinality::Optional);
assert_eq!(vals.values_for_doc(0).collect_vec(), vec![]);
assert_eq!(vals.values_for_doc(1).collect_vec(), vec![]);
assert_eq!(vals.values_for_doc(2).collect_vec(), vec![]);
assert_eq!(vals.values_for_doc(3).collect_vec(), vec![1]);
assert_eq!(vals.values_for_doc(4).collect_vec(), vec![]);
// text column
let dynamic_column = cols[1].open().unwrap();
let DynamicColumn::Str(vals) = dynamic_column else {
panic!()
};
assert_eq!(vals.ords().get_cardinality(), Cardinality::Optional);
let get_str_for_ord = |ord| {
let mut out = String::new();
vals.ord_to_str(ord, &mut out).unwrap();
out
};
assert_eq!(vals.dictionary.num_terms(), 2);
assert_eq!(get_str_for_ord(0), "a");
assert_eq!(get_str_for_ord(1), "b");
let get_str_for_row = |row_id| {
let term_ords: Vec<String> = vals
.term_ords(row_id)
.map(|el| {
let mut out = String::new();
vals.ord_to_str(el, &mut out).unwrap();
out
})
.collect();
term_ords
};
assert_eq!(get_str_for_row(0), vec!["a".to_string()]);
assert_eq!(get_str_for_row(1), Vec::<String>::new());
assert_eq!(get_str_for_row(2), vec!["b".to_string()]);
assert_eq!(get_str_for_row(3), Vec::<String>::new());
}
#[test]
fn test_merge_columnar_different_empty_cardinality() {
let columnar1 = make_text_columnar_multiple_columns(&[("mixed", &[&["a"]])]);
let columnar2 = make_columnar("mixed", &[1i64]);
let mut buffer = Vec::new();
let columnars = &[&columnar1, &columnar2];
let stack_merge_order = StackMergeOrder::stack(columnars);
crate::columnar::merge_columnar(
columnars,
&[],
MergeRowOrder::Stack(stack_merge_order),
&mut buffer,
)
.unwrap();
let columnar_reader = ColumnarReader::open(buffer).unwrap();
assert_eq!(columnar_reader.num_rows(), 2);
assert_eq!(columnar_reader.num_columns(), 2);
let cols = columnar_reader.read_columns("mixed").unwrap();
// numeric column
let dynamic_column = cols[0].open().unwrap();
assert_eq!(dynamic_column.get_cardinality(), Cardinality::Optional);
// text column
let dynamic_column = cols[1].open().unwrap();
assert_eq!(dynamic_column.get_cardinality(), Cardinality::Optional);
}

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

@@ -5,6 +5,8 @@ mod reader;
mod writer;
pub use column_type::{ColumnType, HasAssociatedColumnType};
#[cfg(test)]
pub(crate) use merge::ColumnTypeCategory;
pub use merge::{merge_columnar, MergeRowOrder, ShuffleMergeOrder, StackMergeOrder};
pub use reader::ColumnarReader;
pub use writer::ColumnarWriter;

81
columnar/src/columnar/reader/mod.rs
View File

@@ -1,4 +1,4 @@
use std::{io, mem};
use std::{fmt, io, mem};
use common::file_slice::FileSlice;
use common::BinarySerializable;
@@ -21,6 +21,32 @@ pub struct ColumnarReader {
num_rows: RowId,
}
impl fmt::Debug for ColumnarReader {
fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
let num_rows = self.num_rows();
let columns = self.list_columns().unwrap();
let num_cols = columns.len();
let mut debug_struct = f.debug_struct("Columnar");
debug_struct
.field("num_rows", &num_rows)
.field("num_cols", &num_cols);
for (col_name, dynamic_column_handle) in columns.into_iter().take(5) {
let col = dynamic_column_handle.open().unwrap();
if col.num_values() > 10 {
debug_struct.field(&col_name, &"..");
} else {
debug_struct.field(&col_name, &col);
}
}
if num_cols > 5 {
debug_struct.finish_non_exhaustive()?;
} else {
debug_struct.finish()?;
}
Ok(())
}
}
/// Functions by both the async/sync code listing columns.
/// It takes a stream from the column sstable and return the list of
/// `DynamicColumn` available in it.
@@ -76,30 +102,41 @@ impl ColumnarReader {
pub fn num_rows(&self) -> RowId {
self.num_rows
}
// Iterate over the columns in a sorted way
pub fn iter_columns(
&self,
) -> io::Result<impl Iterator<Item = (String, DynamicColumnHandle)> + '_> {
let mut stream = self.column_dictionary.stream()?;
Ok(std::iter::from_fn(move || {
if stream.advance() {
let key_bytes: &[u8] = stream.key();
let column_code: u8 = key_bytes.last().cloned().unwrap();
// TODO Error Handling. The API gets quite ugly when returning the error here, so
// instead we could just check the first N columns upfront.
let column_type: ColumnType = ColumnType::try_from_code(column_code)
.map_err(|_| io_invalid_data(format!("Unknown column code `{column_code}`")))
.unwrap();
let range = stream.value().clone();
let column_name =
// The last two bytes are respectively the 0u8 separator and the column_type.
String::from_utf8_lossy(&key_bytes[..key_bytes.len() - 2]).to_string();
let file_slice = self
.column_data
.slice(range.start as usize..range.end as usize);
let column_handle = DynamicColumnHandle {
file_slice,
column_type,
};
Some((column_name, column_handle))
} else {
None
}
}))
}
// TODO Add unit tests
pub fn list_columns(&self) -> io::Result<Vec<(String, DynamicColumnHandle)>> {
let mut stream = self.column_dictionary.stream()?;
let mut results = Vec::new();
while stream.advance() {
let key_bytes: &[u8] = stream.key();
let column_code: u8 = key_bytes.last().cloned().unwrap();
let column_type: ColumnType = ColumnType::try_from_code(column_code)
.map_err(|_| io_invalid_data(format!("Unknown column code `{column_code}`")))?;
let range = stream.value().clone();
let column_name =
// The last two bytes are respectively the 0u8 separator and the column_type.
String::from_utf8_lossy(&key_bytes[..key_bytes.len() - 2]).to_string();
let file_slice = self
.column_data
.slice(range.start as usize..range.end as usize);
let column_handle = DynamicColumnHandle {
file_slice,
column_type,
};
results.push((column_name, column_handle));
}
Ok(results)
Ok(self.iter_columns()?.collect())
}
fn stream_for_column_range(&self, column_name: &str) -> sstable::StreamerBuilder<RangeSSTable> {

3
columnar/src/columnar/writer/column_writers.rs
View File

@@ -269,7 +269,8 @@ impl StrOrBytesColumnWriter {
dictionaries: &mut [DictionaryBuilder],
arena: &mut MemoryArena,
) {
let unordered_id = dictionaries[self.dictionary_id as usize].get_or_allocate_id(bytes);
let unordered_id =
dictionaries[self.dictionary_id as usize].get_or_allocate_id(bytes, arena);
self.column_writer.record(doc, unordered_id, arena);
}

88
columnar/src/columnar/writer/mod.rs
View File

@@ -13,9 +13,7 @@ pub(crate) use serializer::ColumnarSerializer;
use stacker::{Addr, ArenaHashMap, MemoryArena};
use crate::column_index::SerializableColumnIndex;
use crate::column_values::{
ColumnValues, MonotonicallyMappableToU128, MonotonicallyMappableToU64, VecColumn,
};
use crate::column_values::{MonotonicallyMappableToU128, MonotonicallyMappableToU64};
use crate::columnar::column_type::ColumnType;
use crate::columnar::writer::column_writers::{
ColumnWriter, NumericalColumnWriter, StrOrBytesColumnWriter,
@@ -79,7 +77,6 @@ fn mutate_or_create_column<V, TMutator>(
impl ColumnarWriter {
pub fn mem_usage(&self) -> usize {
// TODO add dictionary builders.
self.arena.mem_usage()
+ self.numerical_field_hash_map.mem_usage()
+ self.bool_field_hash_map.mem_usage()
@@ -87,6 +84,11 @@ impl ColumnarWriter {
+ self.str_field_hash_map.mem_usage()
+ self.ip_addr_field_hash_map.mem_usage()
+ self.datetime_field_hash_map.mem_usage()
+ self
.dictionaries
.iter()
.map(|dict| dict.mem_usage())
.sum::<usize>()
}
/// Returns the list of doc ids from 0..num_docs sorted by the `sort_field`
@@ -98,22 +100,37 @@ impl ColumnarWriter {
///
/// The sort applied is stable.
pub fn sort_order(&self, sort_field: &str, num_docs: RowId, reversed: bool) -> Vec<u32> {
let Some(numerical_col_writer) =
self.numerical_field_hash_map.get::<NumericalColumnWriter>(sort_field.as_bytes()) else {
return Vec::new();
let Some(numerical_col_writer) = self
.numerical_field_hash_map
.get::<NumericalColumnWriter>(sort_field.as_bytes())
.or_else(|| {
self.datetime_field_hash_map
.get::<NumericalColumnWriter>(sort_field.as_bytes())
})
else {
return Vec::new();
};
let mut symbols_buffer = Vec::new();
let mut values = Vec::new();
let mut last_doc_opt: Option<RowId> = None;
let mut start_doc_check_fill = 0;
let mut current_doc_opt: Option<RowId> = None;
// Assumption: NewDoc will never call the same doc twice and is strictly increasing between
// calls
for op in numerical_col_writer.operation_iterator(&self.arena, None, &mut symbols_buffer) {
match op {
ColumnOperation::NewDoc(doc) => {
last_doc_opt = Some(doc);
current_doc_opt = Some(doc);
}
ColumnOperation::Value(numerical_value) => {
if let Some(last_doc) = last_doc_opt {
if let Some(current_doc) = current_doc_opt {
// Fill up with 0.0 since last doc
values.extend((start_doc_check_fill..current_doc).map(|doc| (0.0, doc)));
start_doc_check_fill = current_doc + 1;
// handle multi values
current_doc_opt = None;
let score: f32 = f64::coerce(numerical_value) as f32;
values.push((score, last_doc));
values.push((score, current_doc));
}
}
}
@@ -123,9 +140,9 @@ impl ColumnarWriter {
}
values.sort_by(|(left_score, _), (right_score, _)| {
if reversed {
right_score.partial_cmp(left_score).unwrap()
right_score.total_cmp(left_score)
} else {
left_score.partial_cmp(right_score).unwrap()
left_score.total_cmp(right_score)
}
});
values.into_iter().map(|(_score, doc)| doc).collect()
@@ -257,7 +274,7 @@ impl ColumnarWriter {
let mut column: ColumnWriter = column_opt.unwrap_or_default();
column.record(
doc,
NumericalValue::I64(datetime.into_timestamp_micros()),
NumericalValue::I64(datetime.into_timestamp_nanos()),
arena,
);
column
@@ -319,7 +336,7 @@ impl ColumnarWriter {
let mut columns: Vec<(&[u8], ColumnType, Addr)> = self
.numerical_field_hash_map
.iter()
.map(|(column_name, addr, _)| {
.map(|(column_name, addr)| {
let numerical_column_writer: NumericalColumnWriter =
self.numerical_field_hash_map.read(addr);
let column_type = numerical_column_writer.numerical_type().into();
@@ -329,27 +346,27 @@ impl ColumnarWriter {
columns.extend(
self.bytes_field_hash_map
.iter()
.map(|(term, addr, _)| (term, ColumnType::Bytes, addr)),
.map(|(term, addr)| (term, ColumnType::Bytes, addr)),
);
columns.extend(
self.str_field_hash_map
.iter()
.map(|(column_name, addr, _)| (column_name, ColumnType::Str, addr)),
.map(|(column_name, addr)| (column_name, ColumnType::Str, addr)),
);
columns.extend(
self.bool_field_hash_map
.iter()
.map(|(column_name, addr, _)| (column_name, ColumnType::Bool, addr)),
.map(|(column_name, addr)| (column_name, ColumnType::Bool, addr)),
);
columns.extend(
self.ip_addr_field_hash_map
.iter()
.map(|(column_name, addr, _)| (column_name, ColumnType::IpAddr, addr)),
.map(|(column_name, addr)| (column_name, ColumnType::IpAddr, addr)),
);
columns.extend(
self.datetime_field_hash_map
.iter()
.map(|(column_name, addr, _)| (column_name, ColumnType::DateTime, addr)),
.map(|(column_name, addr)| (column_name, ColumnType::DateTime, addr)),
);
columns.sort_unstable_by_key(|(column_name, col_type, _)| (*column_name, *col_type));
@@ -361,7 +378,7 @@ impl ColumnarWriter {
let column_writer: ColumnWriter = self.bool_field_hash_map.read(addr);
let cardinality = column_writer.get_cardinality(num_docs);
let mut column_serializer =
serializer.serialize_column(column_name, column_type);
serializer.start_serialize_column(column_name, column_type);
serialize_bool_column(
cardinality,
num_docs,
@@ -373,12 +390,13 @@ impl ColumnarWriter {
buffers,
&mut column_serializer,
)?;
column_serializer.finalize()?;
}
ColumnType::IpAddr => {
let column_writer: ColumnWriter = self.ip_addr_field_hash_map.read(addr);
let cardinality = column_writer.get_cardinality(num_docs);
let mut column_serializer =
serializer.serialize_column(column_name, ColumnType::IpAddr);
serializer.start_serialize_column(column_name, ColumnType::IpAddr);
serialize_ip_addr_column(
cardinality,
num_docs,
@@ -390,6 +408,7 @@ impl ColumnarWriter {
buffers,
&mut column_serializer,
)?;
column_serializer.finalize()?;
}
ColumnType::Bytes | ColumnType::Str => {
let str_or_bytes_column_writer: StrOrBytesColumnWriter =
@@ -404,7 +423,7 @@ impl ColumnarWriter {
.column_writer
.get_cardinality(num_docs);
let mut column_serializer =
serializer.serialize_column(column_name, column_type);
serializer.start_serialize_column(column_name, column_type);
serialize_bytes_or_str_column(
cardinality,
num_docs,
@@ -416,15 +435,17 @@ impl ColumnarWriter {
&mut symbol_byte_buffer,
),
buffers,
&self.arena,
&mut column_serializer,
)?;
column_serializer.finalize()?;
}
ColumnType::F64 | ColumnType::I64 | ColumnType::U64 => {
let numerical_column_writer: NumericalColumnWriter =
self.numerical_field_hash_map.read(addr);
let cardinality = numerical_column_writer.cardinality(num_docs);
let mut column_serializer =
serializer.serialize_column(column_name, column_type);
serializer.start_serialize_column(column_name, column_type);
let numerical_type = column_type.numerical_type().unwrap();
serialize_numerical_column(
cardinality,
@@ -438,12 +459,13 @@ impl ColumnarWriter {
buffers,
&mut column_serializer,
)?;
column_serializer.finalize()?;
}
ColumnType::DateTime => {
let column_writer: ColumnWriter = self.datetime_field_hash_map.read(addr);
let cardinality = column_writer.get_cardinality(num_docs);
let mut column_serializer =
serializer.serialize_column(column_name, ColumnType::DateTime);
serializer.start_serialize_column(column_name, ColumnType::DateTime);
serialize_numerical_column(
cardinality,
num_docs,
@@ -456,6 +478,7 @@ impl ColumnarWriter {
buffers,
&mut column_serializer,
)?;
column_serializer.finalize()?;
}
};
}
@@ -466,6 +489,7 @@ impl ColumnarWriter {
// Serialize [Dictionary, Column, dictionary num bytes U32::LE]
// Column: [Column Index, Column Values, column index num bytes U32::LE]
#[allow(clippy::too_many_arguments)]
fn serialize_bytes_or_str_column(
cardinality: Cardinality,
num_docs: RowId,
@@ -473,6 +497,7 @@ fn serialize_bytes_or_str_column(
dictionary_builder: &DictionaryBuilder,
operation_it: impl Iterator<Item = ColumnOperation<UnorderedId>>,
buffers: &mut SpareBuffers,
arena: &MemoryArena,
wrt: impl io::Write,
) -> io::Result<()> {
let SpareBuffers {
@@ -481,7 +506,8 @@ fn serialize_bytes_or_str_column(
..
} = buffers;
let mut counting_writer = CountingWriter::wrap(wrt);
let term_id_mapping: TermIdMapping = dictionary_builder.serialize(&mut counting_writer)?;
let term_id_mapping: TermIdMapping =
dictionary_builder.serialize(arena, &mut counting_writer)?;
let dictionary_num_bytes: u32 = counting_writer.written_bytes() as u32;
let mut wrt = counting_writer.finish();
let operation_iterator = operation_it.map(|symbol: ColumnOperation<UnorderedId>| {
@@ -617,10 +643,7 @@ fn send_to_serialize_column_mappable_to_u128<
value_index_builders: &mut PreallocatedIndexBuilders,
values: &mut Vec<T>,
mut wrt: impl io::Write,
) -> io::Result<()>
where
for<'a> VecColumn<'a, T>: ColumnValues<T>,
{
) -> io::Result<()> {
values.clear();
// TODO: split index and values
let serializable_column_index = match cardinality {
@@ -673,10 +696,7 @@ fn send_to_serialize_column_mappable_to_u64(
value_index_builders: &mut PreallocatedIndexBuilders,
values: &mut Vec<u64>,
mut wrt: impl io::Write,
) -> io::Result<()>
where
for<'a> VecColumn<'a, u64>: ColumnValues<u64>,
{
) -> io::Result<()> {
values.clear();
let serializable_column_index = match cardinality {
Cardinality::Full => {

26
columnar/src/columnar/writer/serializer.rs
View File

@@ -18,7 +18,12 @@ pub struct ColumnarSerializer<W: io::Write> {
/// code.
fn prepare_key(key: &[u8], column_type: ColumnType, buffer: &mut Vec<u8>) {
buffer.clear();
buffer.extend_from_slice(key);
// Convert 0 bytes to '0' string, as 0 bytes are reserved for the end of the path.
if key.contains(&0u8) {
buffer.extend(key.iter().map(|&b| if b == 0 { b'0' } else { b }));
} else {
buffer.extend_from_slice(key);
}
buffer.push(0u8);
buffer.push(column_type.to_code());
}
@@ -34,11 +39,12 @@ impl<W: io::Write> ColumnarSerializer<W> {
}
}
pub fn serialize_column<'a>(
/// Creates a ColumnSerializer.
pub fn start_serialize_column<'a>(
&'a mut self,
column_name: &[u8],
column_type: ColumnType,
) -> impl io::Write + 'a {
) -> ColumnSerializer<'a, W> {
let start_offset = self.wrt.written_bytes();
prepare_key(column_name, column_type, &mut self.prepare_key_buffer);
ColumnSerializer {
@@ -60,20 +66,21 @@ impl<W: io::Write> ColumnarSerializer<W> {
}
}
struct ColumnSerializer<'a, W: io::Write> {
pub struct ColumnSerializer<'a, W: io::Write> {
columnar_serializer: &'a mut ColumnarSerializer<W>,
start_offset: u64,
}
impl<'a, W: io::Write> Drop for ColumnSerializer<'a, W> {
fn drop(&mut self) {
impl<'a, W: io::Write> ColumnSerializer<'a, W> {
pub fn finalize(self) -> io::Result<()> {
let end_offset: u64 = self.columnar_serializer.wrt.written_bytes();
let byte_range = self.start_offset..end_offset;
self.columnar_serializer.sstable_range.insert_cannot_fail(
self.columnar_serializer.sstable_range.insert(
&self.columnar_serializer.prepare_key_buffer[..],
&byte_range,
);
)?;
self.columnar_serializer.prepare_key_buffer.clear();
Ok(())
}
}
@@ -94,14 +101,13 @@ impl<'a, W: io::Write> io::Write for ColumnSerializer<'a, W> {
#[cfg(test)]
mod tests {
use super::*;
use crate::columnar::column_type::ColumnType;
#[test]
fn test_prepare_key_bytes() {
let mut buffer: Vec<u8> = b"somegarbage".to_vec();
prepare_key(b"root\0child", ColumnType::Str, &mut buffer);
assert_eq!(buffer.len(), 12);
assert_eq!(&buffer[..10], b"root\0child");
assert_eq!(&buffer[..10], b"root0child");
assert_eq!(buffer[10], 0u8);
assert_eq!(buffer[11], ColumnType::Str.to_code());
}

49
columnar/src/dictionary.rs
View File

@@ -1,7 +1,7 @@
use std::io;
use fnv::FnvHashMap;
use sstable::SSTable;
use stacker::{MemoryArena, SharedArenaHashMap};
pub(crate) struct TermIdMapping {
unordered_to_ord: Vec<OrderedId>,
@@ -31,26 +31,38 @@ pub struct OrderedId(pub u32);
/// mapping.
#[derive(Default)]
pub(crate) struct DictionaryBuilder {
dict: FnvHashMap<Vec<u8>, UnorderedId>,
dict: SharedArenaHashMap,
}
impl DictionaryBuilder {
/// Get or allocate an unordered id.
/// (This ID is simply an auto-incremented id.)
pub fn get_or_allocate_id(&mut self, term: &[u8]) -> UnorderedId {
if let Some(term_id) = self.dict.get(term) {
return *term_id;
}
let new_id = UnorderedId(self.dict.len() as u32);
self.dict.insert(term.to_vec(), new_id);
new_id
pub fn get_or_allocate_id(&mut self, term: &[u8], arena: &mut MemoryArena) -> UnorderedId {
let next_id = self.dict.len() as u32;
let unordered_id = self
.dict
.mutate_or_create(term, arena, |unordered_id: Option<u32>| {
if let Some(unordered_id) = unordered_id {
unordered_id
} else {
next_id
}
});
UnorderedId(unordered_id)
}
/// Serialize the dictionary into an fst, and returns the
/// `UnorderedId -> TermOrdinal` map.
pub fn serialize<'a, W: io::Write + 'a>(&self, wrt: &mut W) -> io::Result<TermIdMapping> {
let mut terms: Vec<(&[u8], UnorderedId)> =
self.dict.iter().map(|(k, v)| (k.as_slice(), *v)).collect();
pub fn serialize<'a, W: io::Write + 'a>(
&self,
arena: &MemoryArena,
wrt: &mut W,
) -> io::Result<TermIdMapping> {
let mut terms: Vec<(&[u8], UnorderedId)> = self
.dict
.iter(arena)
.map(|(k, v)| (k, arena.read(v)))
.collect();
terms.sort_unstable_by_key(|(key, _)| *key);
// TODO Remove the allocation.
let mut unordered_to_ord: Vec<OrderedId> = vec![OrderedId(0u32); terms.len()];
@@ -63,6 +75,10 @@ impl DictionaryBuilder {
sstable_builder.finish()?;
Ok(TermIdMapping { unordered_to_ord })
}
pub(crate) fn mem_usage(&self) -> usize {
self.dict.mem_usage()
}
}
#[cfg(test)]
@@ -71,12 +87,13 @@ mod tests {
#[test]
fn test_dictionary_builder() {
let mut arena = MemoryArena::default();
let mut dictionary_builder = DictionaryBuilder::default();
let hello_uid = dictionary_builder.get_or_allocate_id(b"hello");
let happy_uid = dictionary_builder.get_or_allocate_id(b"happy");
let tax_uid = dictionary_builder.get_or_allocate_id(b"tax");
let hello_uid = dictionary_builder.get_or_allocate_id(b"hello", &mut arena);
let happy_uid = dictionary_builder.get_or_allocate_id(b"happy", &mut arena);
let tax_uid = dictionary_builder.get_or_allocate_id(b"tax", &mut arena);
let mut buffer = Vec::new();
let id_mapping = dictionary_builder.serialize(&mut buffer).unwrap();
let id_mapping = dictionary_builder.serialize(&arena, &mut buffer).unwrap();
assert_eq!(id_mapping.to_ord(hello_uid), OrderedId(1));
assert_eq!(id_mapping.to_ord(happy_uid), OrderedId(0));
assert_eq!(id_mapping.to_ord(tax_uid), OrderedId(2));

94
columnar/src/dynamic_column.rs
View File

@@ -1,14 +1,14 @@
use std::io;
use std::net::Ipv6Addr;
use std::sync::Arc;
use std::{fmt, io};
use common::file_slice::FileSlice;
use common::{DateTime, HasLen, OwnedBytes};
use common::{ByteCount, DateTime, HasLen, OwnedBytes};
use crate::column::{BytesColumn, Column, StrColumn};
use crate::column_values::{monotonic_map_column, StrictlyMonotonicFn};
use crate::columnar::ColumnType;
use crate::{Cardinality, NumericalType};
use crate::{Cardinality, ColumnIndex, ColumnValues, NumericalType};
#[derive(Clone)]
pub enum DynamicColumn {
@@ -22,19 +22,54 @@ pub enum DynamicColumn {
Str(StrColumn),
}
impl DynamicColumn {
pub fn get_cardinality(&self) -> Cardinality {
impl fmt::Debug for DynamicColumn {
fn fmt(&self, f: &mut fmt::Formatter) -> fmt::Result {
write!(f, "[{} {} |", self.get_cardinality(), self.column_type())?;
match self {
DynamicColumn::Bool(c) => c.get_cardinality(),
DynamicColumn::I64(c) => c.get_cardinality(),
DynamicColumn::U64(c) => c.get_cardinality(),
DynamicColumn::F64(c) => c.get_cardinality(),
DynamicColumn::IpAddr(c) => c.get_cardinality(),
DynamicColumn::DateTime(c) => c.get_cardinality(),
DynamicColumn::Bytes(c) => c.ords().get_cardinality(),
DynamicColumn::Str(c) => c.ords().get_cardinality(),
DynamicColumn::Bool(col) => write!(f, " {col:?}")?,
DynamicColumn::I64(col) => write!(f, " {col:?}")?,
DynamicColumn::U64(col) => write!(f, " {col:?}")?,
DynamicColumn::F64(col) => write!(f, "{col:?}")?,
DynamicColumn::IpAddr(col) => write!(f, "{col:?}")?,
DynamicColumn::DateTime(col) => write!(f, "{col:?}")?,
DynamicColumn::Bytes(col) => write!(f, "{col:?}")?,
DynamicColumn::Str(col) => write!(f, "{col:?}")?,
}
write!(f, "]")
}
}
impl DynamicColumn {
pub fn column_index(&self) -> &ColumnIndex {
match self {
DynamicColumn::Bool(c) => &c.index,
DynamicColumn::I64(c) => &c.index,
DynamicColumn::U64(c) => &c.index,
DynamicColumn::F64(c) => &c.index,
DynamicColumn::IpAddr(c) => &c.index,
DynamicColumn::DateTime(c) => &c.index,
DynamicColumn::Bytes(c) => &c.ords().index,
DynamicColumn::Str(c) => &c.ords().index,
}
}
pub fn get_cardinality(&self) -> Cardinality {
self.column_index().get_cardinality()
}
pub fn num_values(&self) -> u32 {
match self {
DynamicColumn::Bool(c) => c.values.num_vals(),
DynamicColumn::I64(c) => c.values.num_vals(),
DynamicColumn::U64(c) => c.values.num_vals(),
DynamicColumn::F64(c) => c.values.num_vals(),
DynamicColumn::IpAddr(c) => c.values.num_vals(),
DynamicColumn::DateTime(c) => c.values.num_vals(),
DynamicColumn::Bytes(c) => c.ords().values.num_vals(),
DynamicColumn::Str(c) => c.ords().values.num_vals(),
}
}
pub fn column_type(&self) -> ColumnType {
match self {
DynamicColumn::Bool(_) => ColumnType::Bool,
@@ -73,11 +108,11 @@ impl DynamicColumn {
fn coerce_to_f64(self) -> Option<DynamicColumn> {
match self {
DynamicColumn::I64(column) => Some(DynamicColumn::F64(Column {
idx: column.idx,
index: column.index,
values: Arc::new(monotonic_map_column(column.values, MapI64ToF64)),
})),
DynamicColumn::U64(column) => Some(DynamicColumn::F64(Column {
idx: column.idx,
index: column.index,
values: Arc::new(monotonic_map_column(column.values, MapU64ToF64)),
})),
DynamicColumn::F64(_) => Some(self),
@@ -91,7 +126,7 @@ impl DynamicColumn {
return None;
}
Some(DynamicColumn::I64(Column {
idx: column.idx,
index: column.index,
values: Arc::new(monotonic_map_column(column.values, MapU64ToI64)),
}))
}
@@ -106,7 +141,7 @@ impl DynamicColumn {
return None;
}
Some(DynamicColumn::U64(Column {
idx: column.idx,
index: column.index,
values: Arc::new(monotonic_map_column(column.values, MapI64ToU64)),
}))
}
@@ -193,7 +228,7 @@ static_dynamic_conversions!(StrColumn, Str);
static_dynamic_conversions!(BytesColumn, Bytes);
static_dynamic_conversions!(Column<Ipv6Addr>, IpAddr);
#[derive(Clone)]
#[derive(Clone, Debug)]
pub struct DynamicColumnHandle {
pub(crate) file_slice: FileSlice,
pub(crate) column_type: ColumnType,
@@ -212,7 +247,12 @@ impl DynamicColumnHandle {
}
/// Returns the `u64` fast field reader reader associated with `fields` of types
/// Str, u64, i64, f64, or datetime.
/// Str, u64, i64, f64, bool, ip, or datetime.
///
/// Notice that for IpAddr, the fastfield reader will return the u64 representation of the
/// IpAddr.
/// In order to convert to u128 back cast to `CompactSpaceU64Accessor` and call
/// `compact_to_u128`.
///
/// If not, the fastfield reader will returns the u64-value associated with the original
/// FastValue.
@@ -223,9 +263,15 @@ impl DynamicColumnHandle {
let column: BytesColumn = crate::column::open_column_bytes(column_bytes)?;
Ok(Some(column.term_ord_column))
}
ColumnType::Bool => Ok(None),
ColumnType::IpAddr => Ok(None),
ColumnType::I64 | ColumnType::U64 | ColumnType::F64 | ColumnType::DateTime => {
ColumnType::IpAddr => {
let column = crate::column::open_column_u128_as_compact_u64(column_bytes)?;
Ok(Some(column))
}
ColumnType::Bool
| ColumnType::I64
| ColumnType::U64
| ColumnType::F64
| ColumnType::DateTime => {
let column = crate::column::open_column_u64::<u64>(column_bytes)?;
Ok(Some(column))
}
@@ -248,8 +294,8 @@ impl DynamicColumnHandle {
Ok(dynamic_column)
}
pub fn num_bytes(&self) -> usize {
self.file_slice.len()
pub fn num_bytes(&self) -> ByteCount {
self.file_slice.len().into()
}
pub fn column_type(&self) -> ColumnType {

45
columnar/src/lib.rs
View File

@@ -1,3 +1,22 @@
//! # Tantivy-Columnar
//!
//! `tantivy-columnar`provides a columnar storage for tantivy.
//! The crate allows for efficient read operations on specific columns rather than entire records.
//!
//! ## Overview
//!
//! - **columnar**: Reading, writing, and merging multiple columns:
//! - **[ColumnarWriter]**: Makes it possible to create a new columnar.
//! - **[ColumnarReader]**: The ColumnarReader makes it possible to access a set of columns
//! associated to field names.
//! - **[merge_columnar]**: Contains the functionalities to merge multiple ColumnarReader or
//! segments into a single one.
//!
//! - **column**: A single column, which contains
//! - [column_index]: Resolves the rows for a document id. Manages the cardinality of the
//! column.
//! - [column_values]: Stores the values of a column in a dense format.
#![cfg_attr(all(feature = "unstable", test), feature(test))]
#[cfg(test)]
@@ -7,10 +26,12 @@ extern crate more_asserts;
#[cfg(all(test, feature = "unstable"))]
extern crate test;
use std::fmt::Display;
use std::io;
mod block_accessor;
mod column;
mod column_index;
pub mod column_index;
pub mod column_values;
mod columnar;
mod dictionary;
@@ -19,9 +40,12 @@ mod iterable;
pub(crate) mod utils;
mod value;
pub use block_accessor::ColumnBlockAccessor;
pub use column::{BytesColumn, Column, StrColumn};
pub use column_index::ColumnIndex;
pub use column_values::{ColumnValues, MonotonicallyMappableToU128, MonotonicallyMappableToU64};
pub use column_values::{
ColumnValues, EmptyColumnValues, MonotonicallyMappableToU128, MonotonicallyMappableToU64,
};
pub use columnar::{
merge_columnar, ColumnType, ColumnarReader, ColumnarWriter, HasAssociatedColumnType,
MergeRowOrder, ShuffleMergeOrder, StackMergeOrder,
@@ -34,7 +58,7 @@ pub use self::dynamic_column::{DynamicColumn, DynamicColumnHandle};
pub type RowId = u32;
pub type DocId = u32;
#[derive(Clone, Copy)]
#[derive(Clone, Copy, Debug)]
pub struct RowAddr {
pub segment_ord: u32,
pub row_id: RowId,
@@ -71,6 +95,17 @@ pub enum Cardinality {
Multivalued = 2,
}
impl Display for Cardinality {
fn fmt(&self, f: &mut std::fmt::Formatter) -> std::fmt::Result {
let short_str = match self {
Cardinality::Full => "full",
Cardinality::Optional => "opt",
Cardinality::Multivalued => "mult",
};
write!(f, "{short_str}")
}
}
impl Cardinality {
pub fn is_optional(&self) -> bool {
matches!(self, Cardinality::Optional)
@@ -78,10 +113,12 @@ impl Cardinality {
pub fn is_multivalue(&self) -> bool {
matches!(self, Cardinality::Multivalued)
}
pub fn is_full(&self) -> bool {
matches!(self, Cardinality::Full)
}
pub(crate) fn to_code(self) -> u8 {
self as u8
}
pub(crate) fn try_from_code(code: u8) -> Result<Cardinality, InvalidData> {
match code {
0 => Ok(Cardinality::Full),

757
columnar/src/tests.rs
View File

@@ -1,10 +1,19 @@
use std::collections::HashMap;
use std::fmt::Debug;
use std::net::Ipv6Addr;
use common::DateTime;
use proptest::prelude::*;
use proptest::sample::subsequence;
use crate::column_values::MonotonicallyMappableToU128;
use crate::columnar::ColumnType;
use crate::columnar::{ColumnType, ColumnTypeCategory};
use crate::dynamic_column::{DynamicColumn, DynamicColumnHandle};
use crate::value::NumericalValue;
use crate::{Cardinality, ColumnarReader, ColumnarWriter};
use crate::value::{Coerce, NumericalValue};
use crate::{
BytesColumn, Cardinality, Column, ColumnarReader, ColumnarWriter, RowAddr, RowId,
ShuffleMergeOrder, StackMergeOrder,
};
#[test]
fn test_dataframe_writer_str() {
@@ -17,7 +26,7 @@ fn test_dataframe_writer_str() {
assert_eq!(columnar.num_columns(), 1);
let cols: Vec<DynamicColumnHandle> = columnar.read_columns("my_string").unwrap();
assert_eq!(cols.len(), 1);
assert_eq!(cols[0].num_bytes(), 158);
assert_eq!(cols[0].num_bytes(), 73);
}
#[test]
@@ -31,7 +40,7 @@ fn test_dataframe_writer_bytes() {
assert_eq!(columnar.num_columns(), 1);
let cols: Vec<DynamicColumnHandle> = columnar.read_columns("my_string").unwrap();
assert_eq!(cols.len(), 1);
assert_eq!(cols[0].num_bytes(), 158);
assert_eq!(cols[0].num_bytes(), 73);
}
#[test]
@@ -48,7 +57,9 @@ fn test_dataframe_writer_bool() {
assert_eq!(cols[0].num_bytes(), 22);
assert_eq!(cols[0].column_type(), ColumnType::Bool);
let dyn_bool_col = cols[0].open().unwrap();
let DynamicColumn::Bool(bool_col) = dyn_bool_col else { panic!(); };
let DynamicColumn::Bool(bool_col) = dyn_bool_col else {
panic!();
};
let vals: Vec<Option<bool>> = (0..5).map(|row_id| bool_col.first(row_id)).collect();
assert_eq!(&vals, &[None, Some(false), None, Some(true), None,]);
}
@@ -70,7 +81,9 @@ fn test_dataframe_writer_u64_multivalued() {
assert_eq!(cols.len(), 1);
assert_eq!(cols[0].num_bytes(), 29);
let dyn_i64_col = cols[0].open().unwrap();
let DynamicColumn::I64(divisor_col) = dyn_i64_col else { panic!(); };
let DynamicColumn::I64(divisor_col) = dyn_i64_col else {
panic!();
};
assert_eq!(
divisor_col.get_cardinality(),
crate::Cardinality::Multivalued
@@ -92,7 +105,9 @@ fn test_dataframe_writer_ip_addr() {
assert_eq!(cols[0].num_bytes(), 42);
assert_eq!(cols[0].column_type(), ColumnType::IpAddr);
let dyn_bool_col = cols[0].open().unwrap();
let DynamicColumn::IpAddr(ip_col) = dyn_bool_col else { panic!(); };
let DynamicColumn::IpAddr(ip_col) = dyn_bool_col else {
panic!();
};
let vals: Vec<Option<Ipv6Addr>> = (0..5).map(|row_id| ip_col.first(row_id)).collect();
assert_eq!(
&vals,
@@ -125,8 +140,10 @@ fn test_dataframe_writer_numerical() {
// - null footer 6 bytes
assert_eq!(cols[0].num_bytes(), 33);
let column = cols[0].open().unwrap();
let DynamicColumn::I64(column_i64) = column else { panic!(); };
assert_eq!(column_i64.idx.get_cardinality(), Cardinality::Optional);
let DynamicColumn::I64(column_i64) = column else {
panic!();
};
assert_eq!(column_i64.index.get_cardinality(), Cardinality::Optional);
assert_eq!(column_i64.first(0), None);
assert_eq!(column_i64.first(1), Some(12i64));
assert_eq!(column_i64.first(2), Some(13i64));
@@ -136,6 +153,46 @@ fn test_dataframe_writer_numerical() {
assert_eq!(column_i64.first(6), None); //< we can change the spec for that one.
}
#[test]
fn test_dataframe_sort_by_full() {
let mut dataframe_writer = ColumnarWriter::default();
dataframe_writer.record_numerical(0u32, "value", NumericalValue::U64(1));
dataframe_writer.record_numerical(1u32, "value", NumericalValue::U64(2));
let data = dataframe_writer.sort_order("value", 2, false);
assert_eq!(data, vec![0, 1]);
}
#[test]
fn test_dataframe_sort_by_opt() {
let mut dataframe_writer = ColumnarWriter::default();
dataframe_writer.record_numerical(1u32, "value", NumericalValue::U64(3));
dataframe_writer.record_numerical(3u32, "value", NumericalValue::U64(2));
let data = dataframe_writer.sort_order("value", 5, false);
// 0, 2, 4 is 0.0
assert_eq!(data, vec![0, 2, 4, 3, 1]);
let data = dataframe_writer.sort_order("value", 5, true);
assert_eq!(
data,
vec![4, 2, 0, 3, 1].into_iter().rev().collect::<Vec<_>>()
);
}
#[test]
fn test_dataframe_sort_by_multi() {
let mut dataframe_writer = ColumnarWriter::default();
// valid for sort
dataframe_writer.record_numerical(1u32, "value", NumericalValue::U64(2));
// those are ignored for sort
dataframe_writer.record_numerical(1u32, "value", NumericalValue::U64(4));
dataframe_writer.record_numerical(1u32, "value", NumericalValue::U64(4));
// valid for sort
dataframe_writer.record_numerical(3u32, "value", NumericalValue::U64(3));
// ignored, would change sort order
dataframe_writer.record_numerical(3u32, "value", NumericalValue::U64(1));
let data = dataframe_writer.sort_order("value", 4, false);
assert_eq!(data, vec![0, 2, 1, 3]);
}
#[test]
fn test_dictionary_encoded_str() {
let mut buffer = Vec::new();
@@ -149,7 +206,9 @@ fn test_dictionary_encoded_str() {
assert_eq!(columnar_reader.num_columns(), 2);
let col_handles = columnar_reader.read_columns("my.column").unwrap();
assert_eq!(col_handles.len(), 1);
let DynamicColumn::Str(str_col) = col_handles[0].open().unwrap() else { panic!(); };
let DynamicColumn::Str(str_col) = col_handles[0].open().unwrap() else {
panic!();
};
let index: Vec<Option<u64>> = (0..5).map(|row_id| str_col.ords().first(row_id)).collect();
assert_eq!(index, &[None, Some(0), None, Some(2), Some(1)]);
assert_eq!(str_col.num_rows(), 5);
@@ -181,7 +240,9 @@ fn test_dictionary_encoded_bytes() {
assert_eq!(columnar_reader.num_columns(), 2);
let col_handles = columnar_reader.read_columns("my.column").unwrap();
assert_eq!(col_handles.len(), 1);
let DynamicColumn::Bytes(bytes_col) = col_handles[0].open().unwrap() else { panic!(); };
let DynamicColumn::Bytes(bytes_col) = col_handles[0].open().unwrap() else {
panic!();
};
let index: Vec<Option<u64>> = (0..5)
.map(|row_id| bytes_col.ords().first(row_id))
.collect();
@@ -210,3 +271,675 @@ fn test_dictionary_encoded_bytes() {
.unwrap();
assert_eq!(term_buffer, b"b");
}
fn num_strategy() -> impl Strategy<Value = NumericalValue> {
prop_oneof![
3 => Just(NumericalValue::U64(0u64)),
3 => Just(NumericalValue::U64(u64::MAX)),
3 => Just(NumericalValue::I64(0i64)),
3 => Just(NumericalValue::I64(i64::MIN)),
3 => Just(NumericalValue::I64(i64::MAX)),
3 => Just(NumericalValue::F64(1.2f64)),
1 => any::<f64>().prop_map(NumericalValue::from),
1 => any::<u64>().prop_map(NumericalValue::from),
1 => any::<i64>().prop_map(NumericalValue::from),
]
}
#[derive(Debug, Clone, Copy)]
enum ColumnValue {
Str(&'static str),
Bytes(&'static [u8]),
Numerical(NumericalValue),
IpAddr(Ipv6Addr),
Bool(bool),
DateTime(DateTime),
}
impl<T: Into<NumericalValue>> From<T> for ColumnValue {
fn from(val: T) -> ColumnValue {
ColumnValue::Numerical(val.into())
}
}
impl ColumnValue {
pub(crate) fn column_type_category(&self) -> ColumnTypeCategory {
match self {
ColumnValue::Str(_) => ColumnTypeCategory::Str,
ColumnValue::Bytes(_) => ColumnTypeCategory::Bytes,
ColumnValue::Numerical(_) => ColumnTypeCategory::Numerical,
ColumnValue::IpAddr(_) => ColumnTypeCategory::IpAddr,
ColumnValue::Bool(_) => ColumnTypeCategory::Bool,
ColumnValue::DateTime(_) => ColumnTypeCategory::DateTime,
}
}
}
fn column_name_strategy() -> impl Strategy<Value = &'static str> {
prop_oneof![Just("c1"), Just("c2")]
}
fn string_strategy() -> impl Strategy<Value = &'static str> {
prop_oneof![Just("a"), Just("b")]
}
fn bytes_strategy() -> impl Strategy<Value = &'static [u8]> {
prop_oneof![Just(&[0u8][..]), Just(&[1u8][..])]
}
// A random column value
fn column_value_strategy() -> impl Strategy<Value = ColumnValue> {
prop_oneof![
10 => string_strategy().prop_map(ColumnValue::Str),
1 => bytes_strategy().prop_map(ColumnValue::Bytes),
40 => num_strategy().prop_map(ColumnValue::Numerical),
1 => (1u16..3u16).prop_map(|ip_addr_byte| ColumnValue::IpAddr(Ipv6Addr::new(
127,
0,
0,
0,
0,
0,
0,
ip_addr_byte
))),
1 => any::<bool>().prop_map(ColumnValue::Bool),
1 => (0_679_723_993i64..1_679_723_995i64)
.prop_map(|val| { ColumnValue::DateTime(DateTime::from_timestamp_secs(val)) })
]
}
// A document contains up to 4 values.
fn doc_strategy() -> impl Strategy<Value = Vec<(&'static str, ColumnValue)>> {
proptest::collection::vec((column_name_strategy(), column_value_strategy()), 0..=4)
}
fn num_docs_strategy() -> impl Strategy<Value = usize> {
prop_oneof!(
// We focus heavily on the 0..2 case as we assume it is sufficient to cover all edge cases.
0usize..=3usize,
// We leave 50% of the effort exploring more defensively.
3usize..=12usize
)
}
// A columnar contains up to 2 docs.
fn columnar_docs_strategy() -> impl Strategy<Value = Vec<Vec<(&'static str, ColumnValue)>>> {
num_docs_strategy()
.prop_flat_map(|num_docs| proptest::collection::vec(doc_strategy(), num_docs))
}
fn columnar_docs_and_mapping_strategy(
) -> impl Strategy<Value = (Vec<Vec<(&'static str, ColumnValue)>>, Vec<RowId>)> {
columnar_docs_strategy().prop_flat_map(|docs| {
permutation_strategy(docs.len()).prop_map(move |permutation| (docs.clone(), permutation))
})
}
fn permutation_strategy(n: usize) -> impl Strategy<Value = Vec<RowId>> {
Just((0u32..n as RowId).collect()).prop_shuffle()
}
fn permutation_and_subset_strategy(n: usize) -> impl Strategy<Value = Vec<usize>> {
let vals: Vec<usize> = (0..n).collect();
subsequence(vals, 0..=n).prop_shuffle()
}
fn build_columnar_with_mapping(
docs: &[Vec<(&'static str, ColumnValue)>],
old_to_new_row_ids_opt: Option<&[RowId]>,
) -> ColumnarReader {
let num_docs = docs.len() as u32;
let mut buffer = Vec::new();
let mut columnar_writer = ColumnarWriter::default();
for (doc_id, vals) in docs.iter().enumerate() {
for (column_name, col_val) in vals {
match *col_val {
ColumnValue::Str(str_val) => {
columnar_writer.record_str(doc_id as u32, column_name, str_val);
}
ColumnValue::Bytes(bytes) => {
columnar_writer.record_bytes(doc_id as u32, column_name, bytes)
}
ColumnValue::Numerical(num) => {
columnar_writer.record_numerical(doc_id as u32, column_name, num);
}
ColumnValue::IpAddr(ip_addr) => {
columnar_writer.record_ip_addr(doc_id as u32, column_name, ip_addr);
}
ColumnValue::Bool(bool_val) => {
columnar_writer.record_bool(doc_id as u32, column_name, bool_val);
}
ColumnValue::DateTime(date_time) => {
columnar_writer.record_datetime(doc_id as u32, column_name, date_time);
}
}
}
}
columnar_writer
.serialize(num_docs, old_to_new_row_ids_opt, &mut buffer)
.unwrap();
ColumnarReader::open(buffer).unwrap()
}
fn build_columnar(docs: &[Vec<(&'static str, ColumnValue)>]) -> ColumnarReader {
build_columnar_with_mapping(docs, None)
}
fn assert_columnar_eq_strict(left: &ColumnarReader, right: &ColumnarReader) {
assert_columnar_eq(left, right, false);
}
fn assert_columnar_eq(
left: &ColumnarReader,
right: &ColumnarReader,
lenient_on_numerical_value: bool,
) {
assert_eq!(left.num_rows(), right.num_rows());
let left_columns = left.list_columns().unwrap();
let right_columns = right.list_columns().unwrap();
assert_eq!(left_columns.len(), right_columns.len());
for i in 0..left_columns.len() {
assert_eq!(left_columns[i].0, right_columns[i].0);
let left_column = left_columns[i].1.open().unwrap();
let right_column = right_columns[i].1.open().unwrap();
assert_dyn_column_eq(&left_column, &right_column, lenient_on_numerical_value);
}
}
fn assert_column_eq<T: Copy + PartialOrd + Debug + Send + Sync + 'static>(
left: &Column<T>,
right: &Column<T>,
) {
assert_eq!(left.get_cardinality(), right.get_cardinality());
assert_eq!(left.num_docs(), right.num_docs());
let num_docs = left.num_docs();
for doc in 0..num_docs {
assert_eq!(
left.index.value_row_ids(doc),
right.index.value_row_ids(doc)
);
}
assert_eq!(left.values.num_vals(), right.values.num_vals());
let num_vals = left.values.num_vals();
for i in 0..num_vals {
assert_eq!(left.values.get_val(i), right.values.get_val(i));
}
}
fn assert_bytes_column_eq(left: &BytesColumn, right: &BytesColumn) {
assert_eq!(
left.term_ord_column.get_cardinality(),
right.term_ord_column.get_cardinality()
);
assert_eq!(left.num_rows(), right.num_rows());
assert_column_eq(&left.term_ord_column, &right.term_ord_column);
assert_eq!(left.dictionary.num_terms(), right.dictionary.num_terms());
let num_terms = left.dictionary.num_terms();
let mut left_terms = left.dictionary.stream().unwrap();
let mut right_terms = right.dictionary.stream().unwrap();
for _ in 0..num_terms {
assert!(left_terms.advance());
assert!(right_terms.advance());
assert_eq!(left_terms.key(), right_terms.key());
}
assert!(!left_terms.advance());
assert!(!right_terms.advance());
}
fn assert_dyn_column_eq(
left_dyn_column: &DynamicColumn,
right_dyn_column: &DynamicColumn,
lenient_on_numerical_value: bool,
) {
assert_eq!(
&left_dyn_column.get_cardinality(),
&right_dyn_column.get_cardinality()
);
match &(left_dyn_column, right_dyn_column) {
(DynamicColumn::Bool(left_col), DynamicColumn::Bool(right_col)) => {
assert_column_eq(left_col, right_col);
}
(DynamicColumn::I64(left_col), DynamicColumn::I64(right_col)) => {
assert_column_eq(left_col, right_col);
}
(DynamicColumn::U64(left_col), DynamicColumn::U64(right_col)) => {
assert_column_eq(left_col, right_col);
}
(DynamicColumn::F64(left_col), DynamicColumn::F64(right_col)) => {
assert_column_eq(left_col, right_col);
}
(DynamicColumn::DateTime(left_col), DynamicColumn::DateTime(right_col)) => {
assert_column_eq(left_col, right_col);
}
(DynamicColumn::IpAddr(left_col), DynamicColumn::IpAddr(right_col)) => {
assert_column_eq(left_col, right_col);
}
(DynamicColumn::Bytes(left_col), DynamicColumn::Bytes(right_col)) => {
assert_bytes_column_eq(left_col, right_col);
}
(DynamicColumn::Str(left_col), DynamicColumn::Str(right_col)) => {
assert_bytes_column_eq(left_col, right_col);
}
(left, right) => {
if lenient_on_numerical_value {
assert_eq!(
ColumnTypeCategory::from(left.column_type()),
ColumnTypeCategory::from(right.column_type())
);
} else {
panic!(
"Column type are not the same: {:?} vs {:?}",
left.column_type(),
right.column_type()
);
}
}
}
}
trait AssertEqualToColumnValue {
fn assert_equal_to_column_value(&self, column_value: &ColumnValue);
}
impl AssertEqualToColumnValue for bool {
fn assert_equal_to_column_value(&self, column_value: &ColumnValue) {
let ColumnValue::Bool(val) = column_value else {
panic!()
};
assert_eq!(self, val);
}
}
impl AssertEqualToColumnValue for Ipv6Addr {
fn assert_equal_to_column_value(&self, column_value: &ColumnValue) {
let ColumnValue::IpAddr(val) = column_value else {
panic!()
};
assert_eq!(self, val);
}
}
impl<T: Coerce + PartialEq + Debug + Into<NumericalValue>> AssertEqualToColumnValue for T {
fn assert_equal_to_column_value(&self, column_value: &ColumnValue) {
let ColumnValue::Numerical(num) = column_value else {
panic!()
};
assert_eq!(self, &T::coerce(*num));
}
}
impl AssertEqualToColumnValue for DateTime {
fn assert_equal_to_column_value(&self, column_value: &ColumnValue) {
let ColumnValue::DateTime(dt) = column_value else {
panic!()
};
assert_eq!(self, dt);
}
}
fn assert_column_values<
T: AssertEqualToColumnValue + PartialEq + Copy + PartialOrd + Debug + Send + Sync + 'static,
>(
col: &Column<T>,
expected: &HashMap<u32, Vec<&ColumnValue>>,
) {
let mut num_non_empty_rows = 0;
for doc in 0..col.num_docs() {
let doc_vals: Vec<T> = col.values_for_doc(doc).collect();
if doc_vals.is_empty() {
continue;
}
num_non_empty_rows += 1;
let expected_vals = expected.get(&doc).unwrap();
assert_eq!(doc_vals.len(), expected_vals.len());
for (val, &expected) in doc_vals.iter().zip(expected_vals.iter()) {
val.assert_equal_to_column_value(expected)
}
}
assert_eq!(num_non_empty_rows, expected.len());
}
fn assert_bytes_column_values(
col: &BytesColumn,
expected: &HashMap<u32, Vec<&ColumnValue>>,
is_str: bool,
) {
let mut num_non_empty_rows = 0;
let mut buffer = Vec::new();
for doc in 0..col.term_ord_column.num_docs() {
let doc_vals: Vec<u64> = col.term_ords(doc).collect();
if doc_vals.is_empty() {
continue;
}
let expected_vals = expected.get(&doc).unwrap();
assert_eq!(doc_vals.len(), expected_vals.len());
for (&expected_col_val, &ord) in expected_vals.iter().zip(&doc_vals) {
col.ord_to_bytes(ord, &mut buffer).unwrap();
match expected_col_val {
ColumnValue::Str(str_val) => {
assert!(is_str);
assert_eq!(str_val.as_bytes(), &buffer);
}
ColumnValue::Bytes(bytes_val) => {
assert!(!is_str);
assert_eq!(bytes_val, &buffer);
}
_ => {
panic!();
}
}
}
num_non_empty_rows += 1;
}
assert_eq!(num_non_empty_rows, expected.len());
}
// This proptest attempts to create a tiny columnar based of up to 3 rows, and checks that the
// resulting columnar matches the row data.
proptest! {
#![proptest_config(ProptestConfig::with_cases(500))]
#[test]
fn test_single_columnar_builder_proptest(docs in columnar_docs_strategy()) {
let columnar = build_columnar(&docs[..]);
assert_eq!(columnar.num_rows() as usize, docs.len());
let mut expected_columns: HashMap<(&str, ColumnTypeCategory), HashMap<u32, Vec<&ColumnValue>> > = Default::default();
for (doc_id, doc_vals) in docs.iter().enumerate() {
for (col_name, col_val) in doc_vals {
expected_columns
.entry((col_name, col_val.column_type_category()))
.or_default()
.entry(doc_id as u32)
.or_default()
.push(col_val);
}
}
let column_list = columnar.list_columns().unwrap();
assert_eq!(expected_columns.len(), column_list.len());
for (column_name, column) in column_list {
let dynamic_column = column.open().unwrap();
let col_category: ColumnTypeCategory = dynamic_column.column_type().into();
let expected_col_values: &HashMap<u32, Vec<&ColumnValue>> = expected_columns.get(&(column_name.as_str(), col_category)).unwrap();
match &dynamic_column {
DynamicColumn::Bool(col) =>
assert_column_values(col, expected_col_values),
DynamicColumn::I64(col) =>
assert_column_values(col, expected_col_values),
DynamicColumn::U64(col) =>
assert_column_values(col, expected_col_values),
DynamicColumn::F64(col) =>
assert_column_values(col, expected_col_values),
DynamicColumn::IpAddr(col) =>
assert_column_values(col, expected_col_values),
DynamicColumn::DateTime(col) =>
assert_column_values(col, expected_col_values),
DynamicColumn::Bytes(col) =>
assert_bytes_column_values(col, expected_col_values, false),
DynamicColumn::Str(col) =>
assert_bytes_column_values(col, expected_col_values, true),
}
}
}
}
// Same as `test_single_columnar_builder_proptest` but with a shuffling mapping.
proptest! {
#![proptest_config(ProptestConfig::with_cases(500))]
#[test]
fn test_single_columnar_builder_with_shuffle_proptest((docs, mapping) in columnar_docs_and_mapping_strategy()) {
let columnar = build_columnar_with_mapping(&docs[..], Some(&mapping));
assert_eq!(columnar.num_rows() as usize, docs.len());
let mut expected_columns: HashMap<(&str, ColumnTypeCategory), HashMap<u32, Vec<&ColumnValue>> > = Default::default();
for (doc_id, doc_vals) in docs.iter().enumerate() {
for (col_name, col_val) in doc_vals {
expected_columns
.entry((col_name, col_val.column_type_category()))
.or_default()
.entry(mapping[doc_id])
.or_default()
.push(col_val);
}
}
let column_list = columnar.list_columns().unwrap();
assert_eq!(expected_columns.len(), column_list.len());
for (column_name, column) in column_list {
let dynamic_column = column.open().unwrap();
let col_category: ColumnTypeCategory = dynamic_column.column_type().into();
let expected_col_values: &HashMap<u32, Vec<&ColumnValue>> = expected_columns.get(&(column_name.as_str(), col_category)).unwrap();
for _doc_id in 0..columnar.num_rows() {
match &dynamic_column {
DynamicColumn::Bool(col) =>
assert_column_values(col, expected_col_values),
DynamicColumn::I64(col) =>
assert_column_values(col, expected_col_values),
DynamicColumn::U64(col) =>
assert_column_values(col, expected_col_values),
DynamicColumn::F64(col) =>
assert_column_values(col, expected_col_values),
DynamicColumn::IpAddr(col) =>
assert_column_values(col, expected_col_values),
DynamicColumn::DateTime(col) =>
assert_column_values(col, expected_col_values),
DynamicColumn::Bytes(col) =>
assert_bytes_column_values(col, expected_col_values, false),
DynamicColumn::Str(col) =>
assert_bytes_column_values(col, expected_col_values, true),
}
}
}
}
}
// This tests create 2 or 3 random small columnar and attempts to merge them.
// It compares the resulting merged dataframe with what would have been obtained by building the
// dataframe from the concatenated rows to begin with.
proptest! {
#![proptest_config(ProptestConfig::with_cases(1000))]
#[test]
fn test_columnar_merge_proptest(columnar_docs in proptest::collection::vec(columnar_docs_strategy(), 2..=3)) {
let columnar_readers: Vec<ColumnarReader> = columnar_docs.iter()
.map(|docs| build_columnar(&docs[..]))
.collect::<Vec<_>>();
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();
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[..]);
assert_columnar_eq_strict(&merged_columnar, &expected_merged_columnar);
}
}
#[test]
fn test_columnar_merging_empty_columnar() {
let columnar_docs: Vec<Vec<Vec<(&str, ColumnValue)>>> =
vec![vec![], vec![vec![("c1", ColumnValue::Str("a"))]]];
let columnar_readers: Vec<ColumnarReader> = columnar_docs
.iter()
.map(|docs| build_columnar(&docs[..]))
.collect::<Vec<_>>();
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[..]);
crate::merge_columnar(
&columnar_readers_arr[..],
&[],
crate::MergeRowOrder::Stack(stack_merge_order),
&mut output,
)
.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[..]);
assert_columnar_eq_strict(&merged_columnar, &expected_merged_columnar);
}
#[test]
fn test_columnar_merging_number_columns() {
let columnar_docs: Vec<Vec<Vec<(&str, ColumnValue)>>> = vec![
// columnar 1
vec![
// doc 1.1
vec![("c2", ColumnValue::Numerical(0i64.into()))],
],
// columnar2
vec![
// doc 2.1
vec![("c2", ColumnValue::Numerical(0u64.into()))],
// doc 2.2
vec![("c2", ColumnValue::Numerical(u64::MAX.into()))],
],
];
let columnar_readers: Vec<ColumnarReader> = columnar_docs
.iter()
.map(|docs| build_columnar(&docs[..]))
.collect::<Vec<_>>();
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[..]);
crate::merge_columnar(
&columnar_readers_arr[..],
&[],
crate::MergeRowOrder::Stack(stack_merge_order),
&mut output,
)
.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[..]);
assert_columnar_eq_strict(&merged_columnar, &expected_merged_columnar);
}
// TODO add non trivial remap and merge
// TODO test required_columns
// TODO document edge case: required_columns incompatible with values.
fn columnar_docs_and_remap(
) -> impl Strategy<Value = (Vec<Vec<Vec<(&'static str, ColumnValue)>>>, Vec<RowAddr>)> {
proptest::collection::vec(columnar_docs_strategy(), 2..=3).prop_flat_map(
|columnars_docs: Vec<Vec<Vec<(&str, ColumnValue)>>>| {
let row_addrs: Vec<RowAddr> = columnars_docs
.iter()
.enumerate()
.flat_map(|(segment_ord, columnar_docs)| {
(0u32..columnar_docs.len() as u32).map(move |row_id| RowAddr {
segment_ord: segment_ord as u32,
row_id,
})
})
.collect();
permutation_and_subset_strategy(row_addrs.len()).prop_map(move |shuffled_subset| {
let shuffled_row_addr_subset: Vec<RowAddr> =
shuffled_subset.iter().map(|ord| row_addrs[*ord]).collect();
(columnars_docs.clone(), shuffled_row_addr_subset)
})
},
)
}
proptest! {
#![proptest_config(ProptestConfig::with_cases(1000))]
#[test]
fn test_columnar_merge_and_remap_proptest((columnar_docs, shuffle_merge_order) in columnar_docs_and_remap()) {
let shuffled_rows: Vec<Vec<(&'static str, ColumnValue)>> = shuffle_merge_order.iter()
.map(|row_addr| columnar_docs[row_addr.segment_ord as usize][row_addr.row_id as usize].clone())
.collect();
let expected_merged_columnar = build_columnar(&shuffled_rows[..]);
let columnar_readers: Vec<ColumnarReader> = columnar_docs.iter()
.map(|docs| build_columnar(&docs[..]))
.collect::<Vec<_>>();
let columnar_readers_arr: Vec<&ColumnarReader> = columnar_readers.iter().collect();
let mut output: Vec<u8> = Vec::new();
let segment_num_rows: Vec<RowId> = columnar_docs.iter().map(|docs| docs.len() as RowId).collect();
let shuffle_merge_order = ShuffleMergeOrder::for_test(&segment_num_rows, shuffle_merge_order);
crate::merge_columnar(&columnar_readers_arr[..], &[], shuffle_merge_order.into(), &mut output).unwrap();
let merged_columnar = ColumnarReader::open(output).unwrap();
assert_columnar_eq(&merged_columnar, &expected_merged_columnar, true);
}
}
#[test]
fn test_columnar_merge_empty() {
let columnar_reader_1 = build_columnar(&[]);
let rows: &[Vec<_>] = &[vec![("c1", ColumnValue::Str("a"))]][..];
let columnar_reader_2 = build_columnar(rows);
let mut output: Vec<u8> = Vec::new();
let segment_num_rows: Vec<RowId> = vec![0, 0];
let shuffle_merge_order = ShuffleMergeOrder::for_test(&segment_num_rows, vec![]);
crate::merge_columnar(
&[&columnar_reader_1, &columnar_reader_2],
&[],
shuffle_merge_order.into(),
&mut output,
)
.unwrap();
let merged_columnar = ColumnarReader::open(output).unwrap();
assert_eq!(merged_columnar.num_rows(), 0);
assert_eq!(merged_columnar.num_columns(), 0);
}
#[test]
fn test_columnar_merge_single_str_column() {
let columnar_reader_1 = build_columnar(&[]);
let rows: &[Vec<_>] = &[vec![("c1", ColumnValue::Str("a"))]][..];
let columnar_reader_2 = build_columnar(rows);
let mut output: Vec<u8> = Vec::new();
let segment_num_rows: Vec<RowId> = vec![0, 1];
let shuffle_merge_order = ShuffleMergeOrder::for_test(
&segment_num_rows,
vec![RowAddr {
segment_ord: 1u32,
row_id: 0u32,
}],
);
crate::merge_columnar(
&[&columnar_reader_1, &columnar_reader_2],
&[],
shuffle_merge_order.into(),
&mut output,
)
.unwrap();
let merged_columnar = ColumnarReader::open(output).unwrap();
assert_eq!(merged_columnar.num_rows(), 1);
assert_eq!(merged_columnar.num_columns(), 1);
}
#[test]
fn test_delete_decrease_cardinality() {
let columnar_reader_1 = build_columnar(&[]);
let rows: &[Vec<_>] = &[
vec![
("c", ColumnValue::from(0i64)),
("c", ColumnValue::from(0i64)),
],
vec![("c", ColumnValue::from(0i64))],
][..];
// c is multivalued here
let columnar_reader_2 = build_columnar(rows);
let mut output: Vec<u8> = Vec::new();
let shuffle_merge_order = ShuffleMergeOrder::for_test(
&[0, 2],
vec![RowAddr {
segment_ord: 1u32,
row_id: 1u32,
}],
);
crate::merge_columnar(
&[&columnar_reader_1, &columnar_reader_2],
&[],
shuffle_merge_order.into(),
&mut output,
)
.unwrap();
let merged_columnar = ColumnarReader::open(output).unwrap();
assert_eq!(merged_columnar.num_rows(), 1);
assert_eq!(merged_columnar.num_columns(), 1);
let cols = merged_columnar.read_columns("c").unwrap();
assert_eq!(cols.len(), 1);
assert_eq!(cols[0].column_type(), ColumnType::I64);
assert_eq!(cols[0].open().unwrap().get_cardinality(), Cardinality::Full);
}

2
columnar/src/value.rs
View File

@@ -109,7 +109,7 @@ impl Coerce for f64 {
impl Coerce for DateTime {
fn coerce(value: NumericalValue) -> Self {
let timestamp_micros = i64::coerce(value);
DateTime::from_timestamp_micros(timestamp_micros)
DateTime::from_timestamp_nanos(timestamp_micros)
}
}

4
common/Cargo.toml
View File

@@ -1,6 +1,6 @@
[package]
name = "tantivy-common"
version = "0.5.0"
version = "0.7.0"
authors = ["Paul Masurel <paul@quickwit.io>", "Pascal Seitz <pascal@quickwit.io>"]
license = "MIT"
edition = "2021"
@@ -14,7 +14,7 @@ repository = "https://github.com/quickwit-oss/tantivy"
[dependencies]
byteorder = "1.4.3"
ownedbytes = { version= "0.5", path="../ownedbytes" }
ownedbytes = { version= "0.7", path="../ownedbytes" }
async-trait = "0.1"
time = { version = "0.3.10", features = ["serde-well-known"] }
serde = { version = "1.0.136", features = ["derive"] }

39
common/benches/bench.rs Normal file
View File

@@ -0,0 +1,39 @@
#![feature(test)]
extern crate test;
#[cfg(test)]
mod tests {
use rand::seq::IteratorRandom;
use rand::thread_rng;
use tantivy_common::serialize_vint_u32;
use test::Bencher;
#[bench]
fn bench_vint(b: &mut Bencher) {
let vals: Vec<u32> = (0..20_000).collect();
b.iter(|| {
let mut out = 0u64;
for val in vals.iter().cloned() {
let mut buf = [0u8; 8];
serialize_vint_u32(val, &mut buf);
out += u64::from(buf[0]);
}
out
});
}
#[bench]
fn bench_vint_rand(b: &mut Bencher) {
let vals: Vec<u32> = (0..20_000).choose_multiple(&mut thread_rng(), 100_000);
b.iter(|| {
let mut out = 0u64;
for val in vals.iter().cloned() {
let mut buf = [0u8; 8];
serialize_vint_u32(val, &mut buf);
out += u64::from(buf[0]);
}
out
});
}
}

7
common/src/bitset.rs
View File

@@ -1,9 +1,10 @@
use std::convert::TryInto;
use std::io::Write;
use std::{fmt, io, u64};
use ownedbytes::OwnedBytes;
use crate::ByteCount;
#[derive(Clone, Copy, Eq, PartialEq)]
pub struct TinySet(u64);
@@ -386,8 +387,8 @@ impl ReadOnlyBitSet {
}
/// Number of bytes used in the bitset representation.
pub fn num_bytes(&self) -> usize {
self.data.len()
pub fn num_bytes(&self) -> ByteCount {
self.data.len().into()
}
}

114
common/src/byte_count.rs Normal file
View File

@@ -0,0 +1,114 @@
use std::iter::Sum;
use std::ops::{Add, AddAssign};
use serde::{Deserialize, Serialize};
/// Indicates space usage in bytes
#[derive(Copy, Clone, Default, PartialEq, Eq, PartialOrd, Ord, Serialize, Deserialize)]
pub struct ByteCount(u64);
impl std::fmt::Debug for ByteCount {
fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
f.write_str(&self.human_readable())
}
}
impl std::fmt::Display for ByteCount {
fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
f.write_str(&self.human_readable())
}
}
const SUFFIX_AND_THRESHOLD: [(&str, u64); 5] = [
("KB", 1_000),
("MB", 1_000_000),
("GB", 1_000_000_000),
("TB", 1_000_000_000_000),
("PB", 1_000_000_000_000_000),
];
impl ByteCount {
#[inline]
pub fn get_bytes(&self) -> u64 {
self.0
}
pub fn human_readable(&self) -> String {
for (suffix, threshold) in SUFFIX_AND_THRESHOLD.iter().rev() {
if self.get_bytes() >= *threshold {
let unit_num = self.get_bytes() as f64 / *threshold as f64;
return format!("{unit_num:.2} {suffix}");
}
}
format!("{:.2} B", self.get_bytes())
}
}
impl From<u64> for ByteCount {
fn from(value: u64) -> Self {
ByteCount(value)
}
}
impl From<usize> for ByteCount {
fn from(value: usize) -> Self {
ByteCount(value as u64)
}
}
impl Sum for ByteCount {
#[inline]
fn sum<I: Iterator<Item = Self>>(iter: I) -> Self {
iter.fold(ByteCount::default(), |acc, x| acc + x)
}
}
impl PartialEq<u64> for ByteCount {
#[inline]
fn eq(&self, other: &u64) -> bool {
self.get_bytes() == *other
}
}
impl PartialOrd<u64> for ByteCount {
#[inline]
fn partial_cmp(&self, other: &u64) -> Option<std::cmp::Ordering> {
self.get_bytes().partial_cmp(other)
}
}
impl Add for ByteCount {
type Output = Self;
#[inline]
fn add(self, other: Self) -> Self {
Self(self.get_bytes() + other.get_bytes())
}
}
impl AddAssign for ByteCount {
#[inline]
fn add_assign(&mut self, other: Self) {
*self = Self(self.get_bytes() + other.get_bytes());
}
}
#[cfg(test)]
mod test {
use crate::ByteCount;
#[test]
fn test_bytes() {
assert_eq!(ByteCount::from(0u64).human_readable(), "0 B");
assert_eq!(ByteCount::from(300u64).human_readable(), "300 B");
assert_eq!(ByteCount::from(1_000_000u64).human_readable(), "1.00 MB");
assert_eq!(ByteCount::from(1_500_000u64).human_readable(), "1.50 MB");
assert_eq!(
ByteCount::from(1_500_000_000u64).human_readable(),
"1.50 GB"
);
assert_eq!(
ByteCount::from(3_213_000_000_000u64).human_readable(),
"3.21 TB"
);
}
}

92
common/src/datetime.rs
View File

@@ -1,25 +1,31 @@
use std::fmt;
use std::io::{Read, Write};
use serde::{Deserialize, Serialize};
use time::format_description::well_known::Rfc3339;
use time::{OffsetDateTime, PrimitiveDateTime, UtcOffset};
/// DateTime Precision
use crate::BinarySerializable;
/// Precision with which datetimes are truncated when stored in fast fields. This setting is only
/// relevant for fast fields. In the docstore, datetimes are always saved with nanosecond precision.
#[derive(
Clone, Copy, Debug, Hash, PartialEq, Eq, PartialOrd, Ord, Serialize, Deserialize, Default,
)]
#[serde(rename_all = "lowercase")]
pub enum DatePrecision {
/// Seconds precision
pub enum DateTimePrecision {
/// Second precision.
#[default]
Seconds,
/// Milli-seconds precision.
/// Millisecond precision.
Milliseconds,
/// Micro-seconds precision.
/// Microsecond precision.
Microseconds,
/// Nanosecond precision.
Nanoseconds,
}
/// A date/time value with microsecond precision.
/// A date/time value with nanoseconds precision.
///
/// This timestamp does not carry any explicit time zone information.
/// Users are responsible for applying the provided conversion
@@ -29,31 +35,48 @@ pub enum DatePrecision {
/// All constructors and conversions are provided as explicit
/// functions and not by implementing any `From`/`Into` traits
/// to prevent unintended usage.
#[derive(Clone, Default, Copy, PartialEq, Eq, PartialOrd, Ord, Hash)]
#[derive(Clone, Default, Copy, PartialEq, Eq, PartialOrd, Ord, Hash, Serialize, Deserialize)]
pub struct DateTime {
// Timestamp in microseconds.
pub(crate) timestamp_micros: i64,
// Timestamp in nanoseconds.
pub(crate) timestamp_nanos: i64,
}
impl DateTime {
/// Minimum possible `DateTime` value.
pub const MIN: DateTime = DateTime {
timestamp_nanos: i64::MIN,
};
/// Maximum possible `DateTime` value.
pub const MAX: DateTime = DateTime {
timestamp_nanos: i64::MAX,
};
/// Create new from UNIX timestamp in seconds
pub const fn from_timestamp_secs(seconds: i64) -> Self {
Self {
timestamp_micros: seconds * 1_000_000,
timestamp_nanos: seconds * 1_000_000_000,
}
}
/// Create new from UNIX timestamp in milliseconds
pub const fn from_timestamp_millis(milliseconds: i64) -> Self {
Self {
timestamp_micros: milliseconds * 1_000,
timestamp_nanos: milliseconds * 1_000_000,
}
}
/// Create new from UNIX timestamp in microseconds.
pub const fn from_timestamp_micros(microseconds: i64) -> Self {
Self {
timestamp_micros: microseconds,
timestamp_nanos: microseconds * 1_000,
}
}
/// Create new from UNIX timestamp in nanoseconds.
pub const fn from_timestamp_nanos(nanoseconds: i64) -> Self {
Self {
timestamp_nanos: nanoseconds,
}
}
@@ -61,9 +84,9 @@ impl DateTime {
///
/// The given date/time is converted to UTC and the actual
/// time zone is discarded.
pub const fn from_utc(dt: OffsetDateTime) -> Self {
let timestamp_micros = dt.unix_timestamp() * 1_000_000 + dt.microsecond() as i64;
Self { timestamp_micros }
pub fn from_utc(dt: OffsetDateTime) -> Self {
let timestamp_nanos = dt.unix_timestamp_nanos() as i64;
Self { timestamp_nanos }
}
/// Create new from `PrimitiveDateTime`
@@ -77,23 +100,27 @@ impl DateTime {
/// Convert to UNIX timestamp in seconds.
pub const fn into_timestamp_secs(self) -> i64 {
self.timestamp_micros / 1_000_000
self.timestamp_nanos / 1_000_000_000
}
/// Convert to UNIX timestamp in milliseconds.
pub const fn into_timestamp_millis(self) -> i64 {
self.timestamp_micros / 1_000
self.timestamp_nanos / 1_000_000
}
/// Convert to UNIX timestamp in microseconds.
pub const fn into_timestamp_micros(self) -> i64 {
self.timestamp_micros
self.timestamp_nanos / 1_000
}
/// Convert to UNIX timestamp in nanoseconds.
pub const fn into_timestamp_nanos(self) -> i64 {
self.timestamp_nanos
}
/// Convert to UTC `OffsetDateTime`
pub fn into_utc(self) -> OffsetDateTime {
let timestamp_nanos = self.timestamp_micros as i128 * 1000;
let utc_datetime = OffsetDateTime::from_unix_timestamp_nanos(timestamp_nanos)
let utc_datetime = OffsetDateTime::from_unix_timestamp_nanos(self.timestamp_nanos as i128)
.expect("valid UNIX timestamp");
debug_assert_eq!(UtcOffset::UTC, utc_datetime.offset());
utc_datetime
@@ -116,21 +143,34 @@ impl DateTime {
}
/// Truncates the microseconds value to the corresponding precision.
pub fn truncate(self, precision: DatePrecision) -> Self {
pub fn truncate(self, precision: DateTimePrecision) -> Self {
let truncated_timestamp_micros = match precision {
DatePrecision::Seconds => (self.timestamp_micros / 1_000_000) * 1_000_000,
DatePrecision::Milliseconds => (self.timestamp_micros / 1_000) * 1_000,
DatePrecision::Microseconds => self.timestamp_micros,
DateTimePrecision::Seconds => (self.timestamp_nanos / 1_000_000_000) * 1_000_000_000,
DateTimePrecision::Milliseconds => (self.timestamp_nanos / 1_000_000) * 1_000_000,
DateTimePrecision::Microseconds => (self.timestamp_nanos / 1_000) * 1_000,
DateTimePrecision::Nanoseconds => self.timestamp_nanos,
};
Self {
timestamp_micros: truncated_timestamp_micros,
timestamp_nanos: truncated_timestamp_micros,
}
}
}
impl fmt::Debug for DateTime {
fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
fn fmt(&self, f: &mut fmt::Formatter) -> fmt::Result {
let utc_rfc3339 = self.into_utc().format(&Rfc3339).map_err(|_| fmt::Error)?;
f.write_str(&utc_rfc3339)
}
}
impl BinarySerializable for DateTime {
fn serialize<W: Write + ?Sized>(&self, writer: &mut W) -> std::io::Result<()> {
let timestamp_micros = self.into_timestamp_micros();
<i64 as BinarySerializable>::serialize(&timestamp_micros, writer)
}
fn deserialize<R: Read>(reader: &mut R) -> std::io::Result<Self> {
let timestamp_micros = <i64 as BinarySerializable>::deserialize(reader)?;
Ok(Self::from_timestamp_micros(timestamp_micros))
}
}

88
common/src/file_slice.rs
View File

@@ -1,3 +1,4 @@
use std::fs::File;
use std::ops::{Deref, Range, RangeBounds};
use std::sync::Arc;
use std::{fmt, io};
@@ -5,7 +6,7 @@ use std::{fmt, io};
use async_trait::async_trait;
use ownedbytes::{OwnedBytes, StableDeref};
use crate::HasLen;
use crate::{ByteCount, HasLen};
/// Objects that represents files sections in tantivy.
///
@@ -32,6 +33,62 @@ pub trait FileHandle: 'static + Send + Sync + HasLen + fmt::Debug {
}
}
#[derive(Debug)]
/// A File with it's length included.
pub struct WrapFile {
file: File,
len: usize,
}
impl WrapFile {
/// Creates a new WrapFile and stores its length.
pub fn new(file: File) -> io::Result<Self> {
let len = file.metadata()?.len() as usize;
Ok(WrapFile { file, len })
}
}
#[async_trait]
impl FileHandle for WrapFile {
fn read_bytes(&self, range: Range<usize>) -> io::Result<OwnedBytes> {
let file_len = self.len();
// Calculate the actual range to read, ensuring it stays within file boundaries
let start = range.start;
let end = range.end.min(file_len);
// Ensure the start is before the end of the range
if start >= end {
return Err(io::Error::new(io::ErrorKind::InvalidInput, "Invalid range"));
}
let mut buffer = vec![0; end - start];
#[cfg(unix)]
{
use std::os::unix::prelude::FileExt;
self.file.read_exact_at(&mut buffer, start as u64)?;
}
#[cfg(not(unix))]
{
use std::io::{Read, Seek};
let mut file = self.file.try_clone()?; // Clone the file to read from it separately
// Seek to the start position in the file
file.seek(io::SeekFrom::Start(start as u64))?;
// Read the data into the buffer
file.read_exact(&mut buffer)?;
}
Ok(OwnedBytes::new(buffer))
}
// todo implement async
}
impl HasLen for WrapFile {
fn len(&self) -> usize {
self.len
}
}
#[async_trait]
impl FileHandle for &'static [u8] {
fn read_bytes(&self, range: Range<usize>) -> io::Result<OwnedBytes> {
@@ -67,6 +124,30 @@ impl fmt::Debug for FileSlice {
}
}
impl FileSlice {
pub fn stream_file_chunks(&self) -> impl Iterator<Item = io::Result<OwnedBytes>> + '_ {
let len = self.range.end;
let mut start = self.range.start;
std::iter::from_fn(move || {
/// Returns chunks of 1MB of data from the FileHandle.
const CHUNK_SIZE: usize = 1024 * 1024; // 1MB
if start < len {
let end = (start + CHUNK_SIZE).min(len);
let range = start..end;
let chunk = self.data.read_bytes(range);
start += CHUNK_SIZE;
match chunk {
Ok(chunk) => Some(Ok(chunk)),
Err(e) => Some(Err(e)),
}
} else {
None
}
})
}
}
/// Takes a range, a `RangeBounds` object, and returns
/// a `Range` that corresponds to the relative application of the
/// `RangeBounds` object to the original `Range`.
@@ -216,6 +297,11 @@ impl FileSlice {
pub fn slice_to(&self, to_offset: usize) -> FileSlice {
self.slice(0..to_offset)
}
/// Returns the byte count of the FileSlice.
pub fn num_bytes(&self) -> ByteCount {
self.range.len().into()
}
}
#[async_trait]

26
common/src/group_by.rs
View File

@@ -27,15 +27,15 @@ pub trait GroupByIteratorExtended: Iterator {
where
Self: Sized,
F: FnMut(&Self::Item) -> K,
K: PartialEq + Copy,
Self::Item: Copy,
K: PartialEq + Clone,
Self::Item: Clone,
{
GroupByIterator::new(self, key)
}
}
impl<I: Iterator> GroupByIteratorExtended for I {}
pub struct GroupByIterator<I, F, K: Copy>
pub struct GroupByIterator<I, F, K: Clone>
where
I: Iterator,
F: FnMut(&I::Item) -> K,
@@ -50,7 +50,7 @@ where
inner: Rc<RefCell<GroupByShared<I, F, K>>>,
}
struct GroupByShared<I, F, K: Copy>
struct GroupByShared<I, F, K: Clone>
where
I: Iterator,
F: FnMut(&I::Item) -> K,
@@ -63,7 +63,7 @@ impl<I, F, K> GroupByIterator<I, F, K>
where
I: Iterator,
F: FnMut(&I::Item) -> K,
K: Copy,
K: Clone,
{
fn new(inner: I, group_by_fn: F) -> Self {
let inner = GroupByShared {
@@ -80,28 +80,28 @@ where
impl<I, F, K> Iterator for GroupByIterator<I, F, K>
where
I: Iterator,
I::Item: Copy,
I::Item: Clone,
F: FnMut(&I::Item) -> K,
K: Copy,
K: Clone,
{
type Item = (K, GroupIterator<I, F, K>);
fn next(&mut self) -> Option<Self::Item> {
let mut inner = self.inner.borrow_mut();
let value = *inner.iter.peek()?;
let value = inner.iter.peek()?.clone();
let key = (inner.group_by_fn)(&value);
let inner = self.inner.clone();
let group_iter = GroupIterator {
inner,
group_key: key,
group_key: key.clone(),
};
Some((key, group_iter))
}
}
pub struct GroupIterator<I, F, K: Copy>
pub struct GroupIterator<I, F, K: Clone>
where
I: Iterator,
F: FnMut(&I::Item) -> K,
@@ -110,10 +110,10 @@ where
group_key: K,
}
impl<I, F, K: PartialEq + Copy> Iterator for GroupIterator<I, F, K>
impl<I, F, K: PartialEq + Clone> Iterator for GroupIterator<I, F, K>
where
I: Iterator,
I::Item: Copy,
I::Item: Clone,
F: FnMut(&I::Item) -> K,
{
type Item = I::Item;
@@ -121,7 +121,7 @@ where
fn next(&mut self) -> Option<Self::Item> {
let mut inner = self.inner.borrow_mut();
// peek if next value is in group
let peek_val = *inner.iter.peek()?;
let peek_val = inner.iter.peek()?.clone();
if (inner.group_by_fn)(&peek_val) == self.group_key {
inner.iter.next()
} else {

112
common/src/json_path_writer.rs Normal file
View File

@@ -0,0 +1,112 @@
use crate::replace_in_place;
/// Separates the different segments of a json path.
pub const JSON_PATH_SEGMENT_SEP: u8 = 1u8;
pub const JSON_PATH_SEGMENT_SEP_STR: &str =
unsafe { std::str::from_utf8_unchecked(&[JSON_PATH_SEGMENT_SEP]) };
/// Create a new JsonPathWriter, that creates flattened json paths for tantivy.
#[derive(Clone, Debug, Default)]
pub struct JsonPathWriter {
path: String,
indices: Vec<usize>,
expand_dots: bool,
}
impl JsonPathWriter {
pub fn new() -> Self {
JsonPathWriter {
path: String::new(),
indices: Vec::new(),
expand_dots: false,
}
}
/// When expand_dots is enabled, json object like
/// `{"k8s.node.id": 5}` is processed as if it was
/// `{"k8s": {"node": {"id": 5}}}`.
/// This option has the merit of allowing users to
/// write queries like `k8s.node.id:5`.
/// On the other, enabling that feature can lead to
/// ambiguity.
#[inline]
pub fn set_expand_dots(&mut self, expand_dots: bool) {
self.expand_dots = expand_dots;
}
/// Push a new segment to the path.
#[inline]
pub fn push(&mut self, segment: &str) {
let len_path = self.path.len();
self.indices.push(len_path);
if !self.path.is_empty() {
self.path.push_str(JSON_PATH_SEGMENT_SEP_STR);
}
self.path.push_str(segment);
if self.expand_dots {
// This might include the separation byte, which is ok because it is not a dot.
let appended_segment = &mut self.path[len_path..];
// The unsafe below is safe as long as b'.' and JSON_PATH_SEGMENT_SEP are
// valid single byte ut8 strings.
// By utf-8 design, they cannot be part of another codepoint.
unsafe {
replace_in_place(b'.', JSON_PATH_SEGMENT_SEP, appended_segment.as_bytes_mut())
};
}
}
/// Remove the last segment. Does nothing if the path is empty.
#[inline]
pub fn pop(&mut self) {
if let Some(last_idx) = self.indices.pop() {
self.path.truncate(last_idx);
}
}
/// Clear the path.
#[inline]
pub fn clear(&mut self) {
self.path.clear();
self.indices.clear();
}
/// Get the current path.
#[inline]
pub fn as_str(&self) -> &str {
&self.path
}
}
impl From<JsonPathWriter> for String {
#[inline]
fn from(value: JsonPathWriter) -> Self {
value.path
}
}
#[cfg(test)]
mod tests {
use super::*;
#[test]
fn json_path_writer_test() {
let mut writer = JsonPathWriter::new();
writer.push("root");
assert_eq!(writer.as_str(), "root");
writer.push("child");
assert_eq!(writer.as_str(), "root\u{1}child");
writer.pop();
assert_eq!(writer.as_str(), "root");
writer.push("k8s.node.id");
assert_eq!(writer.as_str(), "root\u{1}k8s.node.id");
writer.set_expand_dots(true);
writer.pop();
writer.push("k8s.node.id");
assert_eq!(writer.as_str(), "root\u{1}k8s\u{1}node\u{1}id");
}
}

10
common/src/lib.rs
View File

@@ -5,20 +5,23 @@ use std::ops::Deref;
pub use byteorder::LittleEndian as Endianness;
mod bitset;
mod byte_count;
mod datetime;
pub mod file_slice;
mod group_by;
mod json_path_writer;
mod serialize;
mod vint;
mod writer;
pub use bitset::*;
pub use datetime::{DatePrecision, DateTime};
pub use byte_count::ByteCount;
pub use datetime::{DateTime, DateTimePrecision};
pub use group_by::GroupByIteratorExtended;
pub use json_path_writer::JsonPathWriter;
pub use ownedbytes::{OwnedBytes, StableDeref};
pub use serialize::{BinarySerializable, DeserializeFrom, FixedSize};
pub use vint::{
deserialize_vint_u128, read_u32_vint, read_u32_vint_no_advance, serialize_vint_u128,
serialize_vint_u32, write_u32_vint, VInt, VIntU128,
read_u32_vint, read_u32_vint_no_advance, serialize_vint_u32, write_u32_vint, VInt, VIntU128,
};
pub use writer::{AntiCallToken, CountingWriter, TerminatingWrite};
@@ -113,6 +116,7 @@ pub fn u64_to_f64(val: u64) -> f64 {
///
/// This function assumes that the needle is rarely contained in the bytes string
/// and offers a fast path if the needle is not present.
#[inline]
pub fn replace_in_place(needle: u8, replacement: u8, bytes: &mut [u8]) {
if !bytes.contains(&needle) {
return;

41
common/src/serialize.rs
View File

@@ -1,3 +1,4 @@
use std::borrow::Cow;
use std::io::{Read, Write};
use std::{fmt, io};
@@ -249,11 +250,47 @@ impl BinarySerializable for String {
}
}
impl<'a> BinarySerializable for Cow<'a, str> {
fn serialize<W: Write + ?Sized>(&self, writer: &mut W) -> io::Result<()> {
let data: &[u8] = self.as_bytes();
VInt(data.len() as u64).serialize(writer)?;
writer.write_all(data)
}
fn deserialize<R: Read>(reader: &mut R) -> io::Result<Cow<'a, str>> {
let string_length = VInt::deserialize(reader)?.val() as usize;
let mut result = String::with_capacity(string_length);
reader
.take(string_length as u64)
.read_to_string(&mut result)?;
Ok(Cow::Owned(result))
}
}
impl<'a> BinarySerializable for Cow<'a, [u8]> {
fn serialize<W: Write + ?Sized>(&self, writer: &mut W) -> io::Result<()> {
VInt(self.len() as u64).serialize(writer)?;
for it in self.iter() {
it.serialize(writer)?;
}
Ok(())
}
fn deserialize<R: Read>(reader: &mut R) -> io::Result<Cow<'a, [u8]>> {
let num_items = VInt::deserialize(reader)?.val();
let mut items: Vec<u8> = Vec::with_capacity(num_items as usize);
for _ in 0..num_items {
let item = u8::deserialize(reader)?;
items.push(item);
}
Ok(Cow::Owned(items))
}
}
#[cfg(test)]
pub mod test {
use super::{VInt, *};
use crate::serialize::BinarySerializable;
use super::*;
pub fn fixed_size_test<O: BinarySerializable + FixedSize + Default>() {
let mut buffer = Vec::new();
O::default().serialize(&mut buffer).unwrap();

77
common/src/vint.rs
View File

@@ -1,8 +1,6 @@
use std::io;
use std::io::{Read, Write};
use byteorder::{ByteOrder, LittleEndian};
use super::BinarySerializable;
/// Variable int serializes a u128 number
@@ -19,26 +17,6 @@ pub fn serialize_vint_u128(mut val: u128, output: &mut Vec<u8>) {
}
}
/// Deserializes a u128 number
///
/// Returns the number and the slice after the vint
pub fn deserialize_vint_u128(data: &[u8]) -> io::Result<(u128, &[u8])> {
let mut result = 0u128;
let mut shift = 0u64;
for i in 0..19 {
let b = data[i];
result |= u128::from(b % 128u8) << shift;
if b >= STOP_BIT {
return Ok((result, &data[i + 1..]));
}
shift += 7;
}
Err(io::Error::new(
io::ErrorKind::InvalidData,
"Failed to deserialize u128 vint",
))
}
/// Wrapper over a `u128` that serializes as a variable int.
#[derive(Clone, Copy, Debug, Eq, PartialEq)]
pub struct VIntU128(pub u128);
@@ -80,17 +58,13 @@ pub struct VInt(pub u64);
const STOP_BIT: u8 = 128;
#[inline]
pub fn serialize_vint_u32(val: u32, buf: &mut [u8; 8]) -> &[u8] {
const START_2: u64 = 1 << 7;
const START_3: u64 = 1 << 14;
const START_4: u64 = 1 << 21;
const START_5: u64 = 1 << 28;
const STOP_1: u64 = START_2 - 1;
const STOP_2: u64 = START_3 - 1;
const STOP_3: u64 = START_4 - 1;
const STOP_4: u64 = START_5 - 1;
const MASK_1: u64 = 127;
const MASK_2: u64 = MASK_1 << 7;
const MASK_3: u64 = MASK_2 << 7;
@@ -99,25 +73,29 @@ pub fn serialize_vint_u32(val: u32, buf: &mut [u8; 8]) -> &[u8] {
let val = u64::from(val);
const STOP_BIT: u64 = 128u64;
let (res, num_bytes) = match val {
0..=STOP_1 => (val | STOP_BIT, 1),
START_2..=STOP_2 => (
let (res, num_bytes) = if val < START_2 {
(val | STOP_BIT, 1)
} else if val < START_3 {
(
(val & MASK_1) | ((val & MASK_2) << 1) | (STOP_BIT << (8)),
2,
),
START_3..=STOP_3 => (
)
} else if val < START_4 {
(
(val & MASK_1) | ((val & MASK_2) << 1) | ((val & MASK_3) << 2) | (STOP_BIT << (8 * 2)),
3,
),
START_4..=STOP_4 => (
)
} else if val < START_5 {
(
(val & MASK_1)
| ((val & MASK_2) << 1)
| ((val & MASK_3) << 2)
| ((val & MASK_4) << 3)
| (STOP_BIT << (8 * 3)),
4,
),
_ => (
)
} else {
(
(val & MASK_1)
| ((val & MASK_2) << 1)
| ((val & MASK_3) << 2)
@@ -125,9 +103,9 @@ pub fn serialize_vint_u32(val: u32, buf: &mut [u8; 8]) -> &[u8] {
| ((val & MASK_5) << 4)
| (STOP_BIT << (8 * 4)),
5,
),
)
};
LittleEndian::write_u64(&mut buf[..], res);
*buf = res.to_le_bytes();
&buf[0..num_bytes]
}
@@ -245,7 +223,6 @@ impl BinarySerializable for VInt {
mod tests {
use super::{serialize_vint_u32, BinarySerializable, VInt};
use crate::vint::{deserialize_vint_u128, serialize_vint_u128, VIntU128};
fn aux_test_vint(val: u64) {
let mut v = [14u8; 10];
@@ -284,27 +261,7 @@ mod tests {
let mut buffer2 = [0u8; 8];
let len_vint = VInt(val as u64).serialize_into(&mut buffer);
let res2 = serialize_vint_u32(val, &mut buffer2);
assert_eq!(&buffer[..len_vint], res2, "array wrong for {}", val);
}
fn aux_test_vint_u128(val: u128) {
let mut data = vec![];
serialize_vint_u128(val, &mut data);
let (deser_val, _data) = deserialize_vint_u128(&data).unwrap();
assert_eq!(val, deser_val);
let mut out = vec![];
VIntU128(val).serialize(&mut out).unwrap();
let deser_val = VIntU128::deserialize(&mut &out[..]).unwrap();
assert_eq!(val, deser_val.0);
}
#[test]
fn test_vint_u128() {
aux_test_vint_u128(0);
aux_test_vint_u128(1);
aux_test_vint_u128(u128::MAX / 3);
aux_test_vint_u128(u128::MAX);
assert_eq!(&buffer[..len_vint], res2, "array wrong for {val}");
}
#[test]

71
examples/aggregation.rs
View File

@@ -7,17 +7,12 @@
// ---
use serde_json::{Deserializer, Value};
use tantivy::aggregation::agg_req::{
Aggregation, Aggregations, BucketAggregation, BucketAggregationType, MetricAggregation,
RangeAggregation,
};
use tantivy::aggregation::agg_req::Aggregations;
use tantivy::aggregation::agg_result::AggregationResults;
use tantivy::aggregation::bucket::RangeAggregationRange;
use tantivy::aggregation::metric::AverageAggregation;
use tantivy::aggregation::AggregationCollector;
use tantivy::query::AllQuery;
use tantivy::schema::{self, IndexRecordOption, Schema, TextFieldIndexing, FAST};
use tantivy::Index;
use tantivy::{Index, IndexWriter, TantivyDocument};
fn main() -> tantivy::Result<()> {
// # Create Schema
@@ -42,7 +37,7 @@ fn main() -> tantivy::Result<()> {
.set_index_option(IndexRecordOption::WithFreqs)
.set_tokenizer("raw"),
)
.set_fast()
.set_fast(None)
.set_stored();
schema_builder.add_text_field("category", text_fieldtype);
schema_builder.add_f64_field("stock", FAST);
@@ -137,10 +132,10 @@ fn main() -> tantivy::Result<()> {
let stream = Deserializer::from_str(data).into_iter::<Value>();
let mut index_writer = index.writer(50_000_000)?;
let mut index_writer: IndexWriter = index.writer(50_000_000)?;
let mut num_indexed = 0;
for value in stream {
let doc = schema.parse_document(&serde_json::to_string(&value.unwrap())?)?;
let doc = TantivyDocument::parse_json(&schema, &serde_json::to_string(&value.unwrap())?)?;
index_writer.add_document(doc)?;
num_indexed += 1;
if num_indexed > 4 {
@@ -192,58 +187,11 @@ fn main() -> tantivy::Result<()> {
//
let agg_req: Aggregations = serde_json::from_str(agg_req_str)?;
let collector = AggregationCollector::from_aggs(agg_req, None);
let collector = AggregationCollector::from_aggs(agg_req, Default::default());
let agg_res: AggregationResults = searcher.search(&AllQuery, &collector).unwrap();
let res2: Value = serde_json::to_value(agg_res)?;
// ### Request Rust API
//
// This is exactly the same request as above, but via the rust structures.
//
let agg_req: Aggregations = vec![(
"group_by_stock".to_string(),
Aggregation::Bucket(BucketAggregation {
bucket_agg: BucketAggregationType::Range(RangeAggregation {
field: "stock".to_string(),
ranges: vec![
RangeAggregationRange {
key: Some("few".into()),
from: None,
to: Some(1f64),
},
RangeAggregationRange {
key: Some("some".into()),
from: Some(1f64),
to: Some(10f64),
},
RangeAggregationRange {
key: Some("many".into()),
from: Some(10f64),
to: None,
},
],
..Default::default()
}),
sub_aggregation: vec![(
"average_price".to_string(),
Aggregation::Metric(MetricAggregation::Average(
AverageAggregation::from_field_name("price".to_string()),
)),
)]
.into_iter()
.collect(),
}),
)]
.into_iter()
.collect();
let collector = AggregationCollector::from_aggs(agg_req, None);
// We use the `AllQuery` which will pass all documents to the AggregationCollector.
let agg_res: AggregationResults = searcher.search(&AllQuery, &collector).unwrap();
let res1: Value = serde_json::to_value(agg_res)?;
let res: Value = serde_json::to_value(agg_res)?;
// ### Aggregation Result
//
@@ -261,8 +209,7 @@ fn main() -> tantivy::Result<()> {
}
"#;
let expected_json: Value = serde_json::from_str(expected_res)?;
assert_eq!(expected_json, res1);
assert_eq!(expected_json, res2);
assert_eq!(expected_json, res);
// ### Request 2
//
@@ -287,7 +234,7 @@ fn main() -> tantivy::Result<()> {
let agg_req: Aggregations = serde_json::from_str(agg_req_str)?;
let collector = AggregationCollector::from_aggs(agg_req, None);
let collector = AggregationCollector::from_aggs(agg_req, Default::default());
let agg_res: AggregationResults = searcher.search(&AllQuery, &collector).unwrap();
let res: Value = serde_json::to_value(agg_res)?;

26
examples/basic_search.rs
View File

@@ -15,7 +15,7 @@
use tantivy::collector::TopDocs;
use tantivy::query::QueryParser;
use tantivy::schema::*;
use tantivy::{doc, Index, ReloadPolicy};
use tantivy::{doc, Index, IndexWriter, ReloadPolicy};
use tempfile::TempDir;
fn main() -> tantivy::Result<()> {
@@ -75,7 +75,7 @@ fn main() -> tantivy::Result<()> {
// Here we give tantivy a budget of `50MB`.
// Using a bigger memory_arena for the indexer may increase
// throughput, but 50 MB is already plenty.
let mut index_writer = index.writer(50_000_000)?;
let mut index_writer: IndexWriter = index.writer(50_000_000)?;
// Let's index our documents!
// We first need a handle on the title and the body field.
@@ -87,7 +87,7 @@ fn main() -> tantivy::Result<()> {
let title = schema.get_field("title").unwrap();
let body = schema.get_field("body").unwrap();
let mut old_man_doc = Document::default();
let mut old_man_doc = TantivyDocument::default();
old_man_doc.add_text(title, "The Old Man and the Sea");
old_man_doc.add_text(
body,
@@ -164,7 +164,7 @@ fn main() -> tantivy::Result<()> {
// will reload the index automatically after each commit.
let reader = index
.reader_builder()
.reload_policy(ReloadPolicy::OnCommit)
.reload_policy(ReloadPolicy::OnCommitWithDelay)
.try_into()?;
// We now need to acquire a searcher.
@@ -217,9 +217,23 @@ fn main() -> tantivy::Result<()> {
// the document returned will only contain
// a title.
for (_score, doc_address) in top_docs {
let retrieved_doc = searcher.doc(doc_address)?;
println!("{}", schema.to_json(&retrieved_doc));
let retrieved_doc: TantivyDocument = searcher.doc(doc_address)?;
println!("{}", retrieved_doc.to_json(&schema));
}
// We can also get an explanation to understand
// how a found document got its score.
let query = query_parser.parse_query("title:sea^20 body:whale^70")?;
let (_score, doc_address) = searcher
.search(&query, &TopDocs::with_limit(1))?
.into_iter()
.next()
.unwrap();
let explanation = query.explain(&searcher, doc_address)?;
println!("{}", explanation.to_pretty_json());
Ok(())
}

4
examples/custom_collector.rs
View File

@@ -13,7 +13,7 @@ use columnar::Column;
use tantivy::collector::{Collector, SegmentCollector};
use tantivy::query::QueryParser;
use tantivy::schema::{Schema, FAST, INDEXED, TEXT};
use tantivy::{doc, Index, Score, SegmentReader};
use tantivy::{doc, Index, IndexWriter, Score, SegmentReader};
#[derive(Default)]
struct Stats {
@@ -142,7 +142,7 @@ fn main() -> tantivy::Result<()> {
// this example.
let index = Index::create_in_ram(schema);
let mut index_writer = index.writer(50_000_000)?;
let mut index_writer: IndexWriter = index.writer(50_000_000)?;
index_writer.add_document(doc!(
product_name => "Super Broom 2000",
product_description => "While it is ok for short distance travel, this broom \

10
examples/custom_tokenizer.rs
View File

@@ -6,7 +6,7 @@ use tantivy::collector::TopDocs;
use tantivy::query::QueryParser;
use tantivy::schema::*;
use tantivy::tokenizer::NgramTokenizer;
use tantivy::{doc, Index};
use tantivy::{doc, Index, IndexWriter};
fn main() -> tantivy::Result<()> {
// # Defining the schema
@@ -53,7 +53,7 @@ fn main() -> tantivy::Result<()> {
// this will store tokens of 3 characters each
index
.tokenizers()
.register("ngram3", NgramTokenizer::new(3, 3, false));
.register("ngram3", NgramTokenizer::new(3, 3, false).unwrap());
// To insert document we need an index writer.
// There must be only one writer at a time.
@@ -62,7 +62,7 @@ fn main() -> tantivy::Result<()> {
//
// Here we use a buffer of 50MB per thread. Using a bigger
// memory arena for the indexer can increase its throughput.
let mut index_writer = index.writer(50_000_000)?;
let mut index_writer: IndexWriter = index.writer(50_000_000)?;
index_writer.add_document(doc!(
title => "The Old Man and the Sea",
body => "He was an old man who fished alone in a skiff in the Gulf Stream and \
@@ -103,8 +103,8 @@ fn main() -> tantivy::Result<()> {
let top_docs = searcher.search(&query, &TopDocs::with_limit(10))?;
for (_, doc_address) in top_docs {
let retrieved_doc = searcher.doc(doc_address)?;
println!("{}", schema.to_json(&retrieved_doc));
let retrieved_doc: TantivyDocument = searcher.doc(doc_address)?;
println!("{}", retrieved_doc.to_json(&schema));
}
Ok(())

20
examples/date_time_field.rs
View File

@@ -4,8 +4,8 @@
use tantivy::collector::TopDocs;
use tantivy::query::QueryParser;
use tantivy::schema::{DateOptions, Schema, Value, INDEXED, STORED, STRING};
use tantivy::Index;
use tantivy::schema::{DateOptions, Document, OwnedValue, Schema, INDEXED, STORED, STRING};
use tantivy::{Index, IndexWriter, TantivyDocument};
fn main() -> tantivy::Result<()> {
// # Defining the schema
@@ -13,7 +13,7 @@ fn main() -> tantivy::Result<()> {
let opts = DateOptions::from(INDEXED)
.set_stored()
.set_fast()
.set_precision(tantivy::DatePrecision::Seconds);
.set_precision(tantivy::DateTimePrecision::Seconds);
// Add `occurred_at` date field type
let occurred_at = schema_builder.add_date_field("occurred_at", opts);
let event_type = schema_builder.add_text_field("event", STRING | STORED);
@@ -22,16 +22,18 @@ fn main() -> tantivy::Result<()> {
// # Indexing documents
let index = Index::create_in_ram(schema.clone());
let mut index_writer = index.writer(50_000_000)?;
let mut index_writer: IndexWriter = index.writer(50_000_000)?;
// The dates are passed as string in the RFC3339 format
let doc = schema.parse_document(
let doc = TantivyDocument::parse_json(
&schema,
r#"{
"occurred_at": "2022-06-22T12:53:50.53Z",
"event": "pull-request"
}"#,
)?;
index_writer.add_document(doc)?;
let doc = schema.parse_document(
let doc = TantivyDocument::parse_json(
&schema,
r#"{
"occurred_at": "2022-06-22T13:00:00.22Z",
"event": "comment"
@@ -58,13 +60,13 @@ fn main() -> tantivy::Result<()> {
let count_docs = searcher.search(&*query, &TopDocs::with_limit(4))?;
assert_eq!(count_docs.len(), 1);
for (_score, doc_address) in count_docs {
let retrieved_doc = searcher.doc(doc_address)?;
let retrieved_doc = searcher.doc::<TantivyDocument>(doc_address)?;
assert!(matches!(
retrieved_doc.get_first(occurred_at),
Some(Value::Date(_))
Some(OwnedValue::Date(_))
));
assert_eq!(
schema.to_json(&retrieved_doc),
retrieved_doc.to_json(&schema),
r#"{"event":["comment"],"occurred_at":["2022-06-22T13:00:00.22Z"]}"#
);
}

12
examples/deleting_updating_documents.rs
View File

@@ -11,7 +11,7 @@
use tantivy::collector::TopDocs;
use tantivy::query::TermQuery;
use tantivy::schema::*;
use tantivy::{doc, Index, IndexReader};
use tantivy::{doc, Index, IndexReader, IndexWriter};
// A simple helper function to fetch a single document
// given its id from our index.
@@ -19,7 +19,7 @@ use tantivy::{doc, Index, IndexReader};
fn extract_doc_given_isbn(
reader: &IndexReader,
isbn_term: &Term,
) -> tantivy::Result<Option<Document>> {
) -> tantivy::Result<Option<TantivyDocument>> {
let searcher = reader.searcher();
// This is the simplest query you can think of.
@@ -69,10 +69,10 @@ fn main() -> tantivy::Result<()> {
let index = Index::create_in_ram(schema.clone());
let mut index_writer = index.writer(50_000_000)?;
let mut index_writer: IndexWriter = index.writer(50_000_000)?;
// Let's add a couple of documents, for the sake of the example.
let mut old_man_doc = Document::default();
let mut old_man_doc = TantivyDocument::default();
old_man_doc.add_text(title, "The Old Man and the Sea");
index_writer.add_document(doc!(
isbn => "978-0099908401",
@@ -94,7 +94,7 @@ fn main() -> tantivy::Result<()> {
// Oops our frankenstein doc seems misspelled
let frankenstein_doc_misspelled = extract_doc_given_isbn(&reader, &frankenstein_isbn)?.unwrap();
assert_eq!(
schema.to_json(&frankenstein_doc_misspelled),
frankenstein_doc_misspelled.to_json(&schema),
r#"{"isbn":["978-9176370711"],"title":["Frankentein"]}"#,
);
@@ -136,7 +136,7 @@ fn main() -> tantivy::Result<()> {
// No more typo!
let frankenstein_new_doc = extract_doc_given_isbn(&reader, &frankenstein_isbn)?.unwrap();
assert_eq!(
schema.to_json(&frankenstein_new_doc),
frankenstein_new_doc.to_json(&schema),
r#"{"isbn":["978-9176370711"],"title":["Frankenstein"]}"#,
);

4
examples/faceted_search.rs
View File

@@ -17,7 +17,7 @@
use tantivy::collector::FacetCollector;
use tantivy::query::{AllQuery, TermQuery};
use tantivy::schema::*;
use tantivy::{doc, Index};
use tantivy::{doc, Index, IndexWriter};
fn main() -> tantivy::Result<()> {
// Let's create a temporary directory for the sake of this example
@@ -30,7 +30,7 @@ fn main() -> tantivy::Result<()> {
let schema = schema_builder.build();
let index = Index::create_in_ram(schema);
let mut index_writer = index.writer(30_000_000)?;
let mut index_writer: IndexWriter = index.writer(30_000_000)?;
// For convenience, tantivy also comes with a macro to
// reduce the boilerplate above.

9
examples/faceted_search_with_tweaked_score.rs
View File

@@ -12,7 +12,7 @@ use std::collections::HashSet;
use tantivy::collector::TopDocs;
use tantivy::query::BooleanQuery;
use tantivy::schema::*;
use tantivy::{doc, DocId, Index, Score, SegmentReader};
use tantivy::{doc, DocId, Index, IndexWriter, Score, SegmentReader};
fn main() -> tantivy::Result<()> {
let mut schema_builder = Schema::builder();
@@ -23,7 +23,7 @@ fn main() -> tantivy::Result<()> {
let schema = schema_builder.build();
let index = Index::create_in_ram(schema);
let mut index_writer = index.writer(30_000_000)?;
let mut index_writer: IndexWriter = index.writer(30_000_000)?;
index_writer.add_document(doc!(
title => "Fried egg",
@@ -91,11 +91,10 @@ fn main() -> tantivy::Result<()> {
.iter()
.map(|(_, doc_id)| {
searcher
.doc(*doc_id)
.doc::<TantivyDocument>(*doc_id)
.unwrap()
.get_first(title)
.unwrap()
.as_text()
.and_then(|v| v.as_str())
.unwrap()
.to_owned()
})

10
examples/fuzzy_search.rs
View File

@@ -14,7 +14,7 @@
use tantivy::collector::{Count, TopDocs};
use tantivy::query::FuzzyTermQuery;
use tantivy::schema::*;
use tantivy::{doc, Index, ReloadPolicy};
use tantivy::{doc, Index, IndexWriter, ReloadPolicy};
use tempfile::TempDir;
fn main() -> tantivy::Result<()> {
@@ -66,7 +66,7 @@ fn main() -> tantivy::Result<()> {
// Here we give tantivy a budget of `50MB`.
// Using a bigger memory_arena for the indexer may increase
// throughput, but 50 MB is already plenty.
let mut index_writer = index.writer(50_000_000)?;
let mut index_writer: IndexWriter = index.writer(50_000_000)?;
// Let's index our documents!
// We first need a handle on the title and the body field.
@@ -123,7 +123,7 @@ fn main() -> tantivy::Result<()> {
// will reload the index automatically after each commit.
let reader = index
.reader_builder()
.reload_policy(ReloadPolicy::OnCommit)
.reload_policy(ReloadPolicy::OnCommitWithDelay)
.try_into()?;
// We now need to acquire a searcher.
@@ -151,10 +151,10 @@ fn main() -> tantivy::Result<()> {
assert_eq!(count, 3);
assert_eq!(top_docs.len(), 3);
for (score, doc_address) in top_docs {
let retrieved_doc = searcher.doc(doc_address)?;
// Note that the score is not lower for the fuzzy hit.
// There's an issue open for that: https://github.com/quickwit-oss/tantivy/issues/563
println!("score {score:?} doc {}", schema.to_json(&retrieved_doc));
let retrieved_doc: TantivyDocument = searcher.doc(doc_address)?;
println!("score {score:?} doc {}", retrieved_doc.to_json(&schema));
// score 1.0 doc {"title":["The Diary of Muadib"]}
//
// score 1.0 doc {"title":["The Diary of a Young Girl"]}

2
examples/index_from_multiple_threads.rs
View File

@@ -96,7 +96,7 @@ fn main() -> tantivy::Result<()> {
let mut index_writer_wlock = index_writer.write().unwrap();
index_writer_wlock.commit()?
};
println!("committed with opstamp {}", opstamp);
println!("committed with opstamp {opstamp}");
thread::sleep(Duration::from_millis(500));
}

4
examples/index_with_json.rs
View File

@@ -21,7 +21,7 @@ fn main() -> tantivy::Result<()> {
}"#;
// We can parse our document
let _mice_and_men_doc = schema.parse_document(mice_and_men_doc_json)?;
let _mice_and_men_doc = TantivyDocument::parse_json(&schema, mice_and_men_doc_json)?;
// Multi-valued field are allowed, they are
// expressed in JSON by an array.
@@ -30,7 +30,7 @@ fn main() -> tantivy::Result<()> {
"title": ["Frankenstein", "The Modern Prometheus"],
"year": 1818
}"#;
let _frankenstein_doc = schema.parse_document(frankenstein_json)?;
let _frankenstein_doc = TantivyDocument::parse_json(&schema, frankenstein_json)?;
// Note that the schema is saved in your index directory.
//

4
examples/integer_range_search.rs
View File

@@ -5,7 +5,7 @@
use tantivy::collector::Count;
use tantivy::query::RangeQuery;
use tantivy::schema::{Schema, INDEXED};
use tantivy::{doc, Index, Result};
use tantivy::{doc, Index, IndexWriter, Result};
fn main() -> Result<()> {
// For the sake of simplicity, this schema will only have 1 field
@@ -17,7 +17,7 @@ fn main() -> Result<()> {
let index = Index::create_in_ram(schema);
let reader = index.reader()?;
{
let mut index_writer = index.writer_with_num_threads(1, 6_000_000)?;
let mut index_writer: IndexWriter = index.writer_with_num_threads(1, 6_000_000)?;
for year in 1950u64..2019u64 {
index_writer.add_document(doc!(year_field => year))?;
}

13
examples/ip_field.rs
View File

@@ -6,7 +6,7 @@
use tantivy::collector::{Count, TopDocs};
use tantivy::query::QueryParser;
use tantivy::schema::{Schema, FAST, INDEXED, STORED, STRING};
use tantivy::Index;
use tantivy::{Index, IndexWriter, TantivyDocument};
fn main() -> tantivy::Result<()> {
// # Defining the schema
@@ -22,20 +22,22 @@ fn main() -> tantivy::Result<()> {
// # Indexing documents
let index = Index::create_in_ram(schema.clone());
let mut index_writer = index.writer(50_000_000)?;
let mut index_writer: IndexWriter = index.writer(50_000_000)?;
// ### IPv4
// Adding documents that contain an IPv4 address. Notice that the IP addresses are passed as
// `String`. Since the field is of type ip, we parse the IP address from the string and store it
// internally as IPv6.
let doc = schema.parse_document(
let doc = TantivyDocument::parse_json(
&schema,
r#"{
"ip": "192.168.0.33",
"event_type": "login"
}"#,
)?;
index_writer.add_document(doc)?;
let doc = schema.parse_document(
let doc = TantivyDocument::parse_json(
&schema,
r#"{
"ip": "192.168.0.80",
"event_type": "checkout"
@@ -44,7 +46,8 @@ fn main() -> tantivy::Result<()> {
index_writer.add_document(doc)?;
// ### IPv6
// Adding a document that contains an IPv6 address.
let doc = schema.parse_document(
let doc = TantivyDocument::parse_json(
&schema,
r#"{
"ip": "2001:0db8:85a3:0000:0000:8a2e:0370:7334",
"event_type": "checkout"

8
examples/iterating_docs_and_positions.rs
View File

@@ -10,7 +10,7 @@
// ---
// Importing tantivy...
use tantivy::schema::*;
use tantivy::{doc, DocSet, Index, Postings, TERMINATED};
use tantivy::{doc, DocSet, Index, IndexWriter, Postings, TERMINATED};
fn main() -> tantivy::Result<()> {
// We first create a schema for the sake of the
@@ -24,7 +24,7 @@ fn main() -> tantivy::Result<()> {
let index = Index::create_in_ram(schema);
let mut index_writer = index.writer_with_num_threads(1, 50_000_000)?;
let mut index_writer: IndexWriter = index.writer_with_num_threads(1, 50_000_000)?;
index_writer.add_document(doc!(title => "The Old Man and the Sea"))?;
index_writer.add_document(doc!(title => "Of Mice and Men"))?;
index_writer.add_document(doc!(title => "The modern Promotheus"))?;
@@ -84,7 +84,7 @@ fn main() -> tantivy::Result<()> {
// Doc 0: TermFreq 2: [0, 4]
// Doc 2: TermFreq 1: [0]
// ```
println!("Doc {}: TermFreq {}: {:?}", doc_id, term_freq, positions);
println!("Doc {doc_id}: TermFreq {term_freq}: {positions:?}");
doc_id = segment_postings.advance();
}
}
@@ -125,7 +125,7 @@ fn main() -> tantivy::Result<()> {
// Once again these docs MAY contains deleted documents as well.
let docs = block_segment_postings.docs();
// Prints `Docs [0, 2].`
println!("Docs {:?}", docs);
println!("Docs {docs:?}");
block_segment_postings.advance();
}
}

10
examples/json_field.rs
View File

@@ -7,7 +7,7 @@
use tantivy::collector::{Count, TopDocs};
use tantivy::query::QueryParser;
use tantivy::schema::{Schema, FAST, STORED, STRING, TEXT};
use tantivy::Index;
use tantivy::{Index, IndexWriter, TantivyDocument};
fn main() -> tantivy::Result<()> {
// # Defining the schema
@@ -20,8 +20,9 @@ fn main() -> tantivy::Result<()> {
// # Indexing documents
let index = Index::create_in_ram(schema.clone());
let mut index_writer = index.writer(50_000_000)?;
let doc = schema.parse_document(
let mut index_writer: IndexWriter = index.writer(50_000_000)?;
let doc = TantivyDocument::parse_json(
&schema,
r#"{
"timestamp": "2022-02-22T23:20:50.53Z",
"event_type": "click",
@@ -33,7 +34,8 @@ fn main() -> tantivy::Result<()> {
}"#,
)?;
index_writer.add_document(doc)?;
let doc = schema.parse_document(
let doc = TantivyDocument::parse_json(
&schema,
r#"{
"timestamp": "2022-02-22T23:20:51.53Z",
"event_type": "click",

83
examples/phrase_prefix_search.rs Normal file
View File

@@ -0,0 +1,83 @@
use tantivy::collector::TopDocs;
use tantivy::query::QueryParser;
use tantivy::schema::*;
use tantivy::{doc, Index, IndexWriter, ReloadPolicy, Result};
use tempfile::TempDir;
fn main() -> Result<()> {
let index_path = TempDir::new()?;
let mut schema_builder = Schema::builder();
schema_builder.add_text_field("title", TEXT | STORED);
schema_builder.add_text_field("body", TEXT);
let schema = schema_builder.build();
let title = schema.get_field("title").unwrap();
let body = schema.get_field("body").unwrap();
let index = Index::create_in_dir(&index_path, schema)?;
let mut index_writer: IndexWriter = index.writer(50_000_000)?;
index_writer.add_document(doc!(
title => "The Old Man and the Sea",
body => "He was an old man who fished alone in a skiff in the Gulf Stream and he had gone \
eighty-four days now without taking a fish.",
))?;
index_writer.add_document(doc!(
title => "Of Mice and Men",
body => "A few miles south of Soledad, the Salinas River drops in close to the hillside \
bank and runs deep and green. The water is warm too, for it has slipped twinkling \
over the yellow sands in the sunlight before reaching the narrow pool. On one \
side of the river the golden foothill slopes curve up to the strong and rocky \
Gabilan Mountains, but on the valley side the water is lined with trees—willows \
fresh and green with every spring, carrying in their lower leaf junctures the \
debris of the winter’s flooding; and sycamores with mottled, white, recumbent \
limbs and branches that arch over the pool"
))?;
// Multivalued field just need to be repeated.
index_writer.add_document(doc!(
title => "Frankenstein",
title => "The Modern Prometheus",
body => "You will rejoice to hear that no disaster has accompanied the commencement of an \
enterprise which you have regarded with such evil forebodings. I arrived here \
yesterday, and my first task is to assure my dear sister of my welfare and \
increasing confidence in the success of my undertaking."
))?;
index_writer.commit()?;
let reader = index
.reader_builder()
.reload_policy(ReloadPolicy::OnCommitWithDelay)
.try_into()?;
let searcher = reader.searcher();
let query_parser = QueryParser::for_index(&index, vec![title, body]);
// This will match documents containing the phrase "in the"
// followed by some word starting with "su",
// i.e. it will match "in the sunlight" and "in the success",
// 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 mut titles = top_docs
.into_iter()
.map(|(_score, doc_address)| {
let doc = searcher.doc::<TantivyDocument>(doc_address)?;
let title = doc
.get_first(title)
.and_then(|v| v.as_str())
.unwrap()
.to_owned();
Ok(title)
})
.collect::<Result<Vec<_>>>()?;
titles.sort_unstable();
assert_eq!(titles, ["Frankenstein", "Of Mice and Men"]);
Ok(())
}

17
examples/pre_tokenized_text.rs
View File

@@ -12,12 +12,13 @@
use tantivy::collector::{Count, TopDocs};
use tantivy::query::TermQuery;
use tantivy::schema::*;
use tantivy::tokenizer::{PreTokenizedString, SimpleTokenizer, Token, Tokenizer};
use tantivy::{doc, Index, ReloadPolicy};
use tantivy::tokenizer::{PreTokenizedString, SimpleTokenizer, Token, TokenStream, Tokenizer};
use tantivy::{doc, Index, IndexWriter, ReloadPolicy};
use tempfile::TempDir;
fn pre_tokenize_text(text: &str) -> Vec<Token> {
let mut token_stream = SimpleTokenizer.token_stream(text);
let mut tokenizer = SimpleTokenizer::default();
let mut token_stream = tokenizer.token_stream(text);
let mut tokens = vec![];
while token_stream.advance() {
tokens.push(token_stream.token().clone());
@@ -37,7 +38,7 @@ fn main() -> tantivy::Result<()> {
let index = Index::create_in_dir(&index_path, schema.clone())?;
let mut index_writer = index.writer(50_000_000)?;
let mut index_writer: IndexWriter = index.writer(50_000_000)?;
// We can create a document manually, by setting the fields
// one by one in a Document object.
@@ -82,7 +83,7 @@ fn main() -> tantivy::Result<()> {
}]
}"#;
let short_man_doc = schema.parse_document(short_man_json)?;
let short_man_doc = TantivyDocument::parse_json(&schema, short_man_json)?;
index_writer.add_document(short_man_doc)?;
@@ -93,7 +94,7 @@ fn main() -> tantivy::Result<()> {
let reader = index
.reader_builder()
.reload_policy(ReloadPolicy::OnCommit)
.reload_policy(ReloadPolicy::OnCommitWithDelay)
.try_into()?;
let searcher = reader.searcher();
@@ -114,8 +115,8 @@ fn main() -> tantivy::Result<()> {
// Note that the tokens are not stored along with the original text
// in the document store
for (_score, doc_address) in top_docs {
let retrieved_doc = searcher.doc(doc_address)?;
println!("Document: {}", schema.to_json(&retrieved_doc));
let retrieved_doc: TantivyDocument = searcher.doc(doc_address)?;
println!("{}", retrieved_doc.to_json(&schema));
}
// In contrary to the previous query, when we search for the "man" term we

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