rust/tantivy
1
0
Fork 0
mirror of https://github.com/quickwit-oss/tantivy.git synced 2025-12-28 04:52:55 +00:00
Code Issues Packages Projects Releases Wiki Activity

Compare commits

base: rust:agg_format
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:remove_dyn
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

1 Commits
agg_format ... remove_dyn

Author SHA1 Message Date
Pascal Seitz
6037cdfe7e remove dynamic dispatch in collect_segment 2023-03-02 20:16:45 +08:00
375 changed files with 13743 additions and 37082 deletions
Expand all files Collapse all files

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

@@ -3,23 +3,20 @@ name: Coverage
on:
push:
branches: [main]
# Ensures that we cancel running jobs for the same PR / same workflow.
concurrency:
group: ${{ github.workflow }}-${{ github.event.pull_request.number || github.ref }}
cancel-in-progress: true
pull_request:
branches: [main]
jobs:
coverage:
runs-on: ubuntu-latest
steps:
- uses: actions/checkout@v4
- uses: actions/checkout@v3
- name: Install Rust
run: rustup toolchain install nightly-2024-07-01 --profile minimal --component llvm-tools-preview
run: rustup toolchain install nightly --profile minimal --component llvm-tools-preview
- uses: Swatinem/rust-cache@v2
- uses: taiki-e/install-action@cargo-llvm-cov
- name: Generate code coverage
run: cargo +nightly-2024-07-01 llvm-cov --all-features --workspace --doctests --lcov --output-path lcov.info
run: cargo +nightly 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

@@ -8,18 +8,13 @@ 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@v4
- uses: actions/checkout@v3
- name: Install stable
uses: actions-rs/toolchain@v1
with:

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

@@ -9,18 +9,13 @@ 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@v4
- uses: actions/checkout@v3
- name: Install nightly
uses: actions-rs/toolchain@v1
@@ -39,13 +34,6 @@ 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:
@@ -60,14 +48,14 @@ jobs:
strategy:
matrix:
features: [
{ label: "all", flags: "mmap,stopwords,lz4-compression,zstd-compression,failpoints" },
{ label: "all", flags: "mmap,stopwords,brotli-compression,lz4-compression,snappy-compression,zstd-compression,failpoints" },
{ label: "quickwit", flags: "mmap,quickwit,failpoints" }
]
name: test-${{ matrix.features.label}}
steps:
- uses: actions/checkout@v4
- uses: actions/checkout@v3
- name: Install stable
uses: actions-rs/toolchain@v1

2
.gitignore vendored
View File

@@ -13,5 +13,3 @@ 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.
## [fieldnorm/](src/fieldnorm): Here is my doc, how many tokens in this field?
## [fieldnorms/](src/fieldnorms): 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

@@ -1,187 +1,3 @@
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

112
Cargo.toml
View File

@@ -1,6 +1,6 @@
[package]
name = "tantivy"
version = "0.23.0"
version = "0.19.0"
authors = ["Paul Masurel <paul.masurel@gmail.com>"]
license = "MIT"
categories = ["database-implementations", "data-structures"]
@@ -11,85 +11,73 @@ repository = "https://github.com/quickwit-oss/tantivy"
readme = "README.md"
keywords = ["search", "information", "retrieval"]
edition = "2021"
rust-version = "1.66"
exclude = ["benches/*.json", "benches/*.txt"]
rust-version = "1.62"
[dependencies]
oneshot = "0.1.7"
base64 = "0.22.0"
oneshot = "0.1.5"
base64 = "0.21.0"
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 = "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 }
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 }
tempfile = { version = "3.3.0", optional = true }
log = "0.4.16"
serde = { version = "1.0.136", features = ["derive"] }
serde_json = "1.0.79"
fs4 = { version = "0.8.0", optional = true }
num_cpus = "1.13.1"
fs2 = { version = "0.4.3", 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 = "1.2.1"
bitpacking = { version = "0.9.2", default-features = false, features = [
"bitpacker4x",
] }
census = "0.4.2"
downcast-rs = "1.2.0"
bitpacking = { version = "0.8.4", default-features = false, features = ["bitpacker4x"] }
census = "0.4.0"
rustc-hash = "1.1.0"
thiserror = "1.0.30"
htmlescape = "0.3.1"
fail = { version = "0.5.0", optional = true }
fail = "0.5.0"
murmurhash32 = "0.2.0"
time = { version = "0.3.10", features = ["serde-well-known"] }
smallvec = "1.8.0"
rayon = "1.5.2"
lru = "0.12.0"
lru = "0.9.0"
fastdivide = "0.4.0"
itertools = "0.13.0"
itertools = "0.10.3"
measure_time = "0.8.2"
async-trait = "0.1.53"
arc-swap = "1.5.0"
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 = "0.3.0", features = ["use_serde"] }
hyperloglogplus = { version = "0.4.1", features = ["const-loop"] }
futures-util = { version = "0.3.28", optional = true }
fnv = "1.0.7"
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" }
[target.'cfg(windows)'.dependencies]
winapi = "0.3.9"
[dev-dependencies]
binggan = "0.8.0"
rand = "0.8.5"
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"
@@ -100,44 +88,27 @@ 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 = ["fs4", "tempfile", "memmap2"]
mmap = ["fs2", "tempfile", "memmap2"]
stopwords = []
brotli-compression = ["brotli"]
lz4-compression = ["lz4_flex"]
snappy-compression = ["snap"]
zstd-compression = ["zstd"]
failpoints = ["fail", "fail/failpoints"]
unstable = [] # useful for benches.
failpoints = ["fail/failpoints"]
unstable = [] # useful for benches.
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"]
quickwit = ["sstable"]
[workspace]
members = [
"query-grammar",
"bitpacker",
"common",
"ownedbytes",
"stacker",
"sstable",
"tokenizer-api",
"columnar",
]
members = ["query-grammar", "bitpacker", "common", "ownedbytes", "stacker", "sstable", "tokenizer-api", "columnar"]
# Following the "fail" crate best practises, we isolate
# tests that define specific behavior in fail check points
@@ -149,7 +120,7 @@ members = [
[[test]]
name = "failpoints"
path = "tests/failpoints/mod.rs"
required-features = ["failpoints"]
required-features = ["fail/failpoints"]
[[bench]]
name = "analyzer"
@@ -159,6 +130,3 @@ harness = false
name = "index-bench"
harness = false
[[bench]]
name = "agg_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:

42
README.md
View File

@@ -5,29 +5,28 @@
[![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)
<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](https://tantivy-search.github.io/logo/tantivy-logo.png)
## Fast full-text search engine library written in Rust
**Tantivy** is a **full-text search engine library** written in Rust.
**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.
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.
Tantivy is, in fact, strongly inspired by Lucene's design.
## Benchmark
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.
The following [benchmark](https://tantivy-search.github.io/bench/) breaks down the
# Benchmark
The following [benchmark](https://tantivy-search.github.io/bench/) breakdowns
performance for different types of queries/collections.
Your mileage WILL vary depending on the nature of queries and their load.
<img src="doc/assets/images/searchbenchmark.png">
Details about the benchmark can be found at this [repository](https://github.com/quickwit-oss/search-benchmark-game).
## Features
# 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))
@@ -43,7 +42,7 @@ Details about the benchmark can be found at this [repository](https://github.com
- 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)
- Compressed document store (LZ4, Zstd, None, Brotli, Snap)
- Range queries
- Faceted search
- Configurable indexing (optional term frequency and position indexing)
@@ -53,11 +52,11 @@ Details about the benchmark can be found at this [repository](https://github.com
- 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.
@@ -67,7 +66,7 @@ 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.
@@ -78,16 +77,16 @@ 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:
@@ -98,11 +97,10 @@ 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;
<img align="center" src="doc/assets/images/paradedb.png" alt="ParadeDB" height="25" width="auto" /> &nbsp;
<img align="center" src="doc/assets/images/etsy.png" alt="Etsy" height="25" width="auto" />&nbsp;
<img align="center" src="doc/assets/images/Nuclia.png#gh-light-mode-only" alt="Nuclia" height="25" width="auto" /> &nbsp;
<img align="center" src="doc/assets/images/humanfirst.png#gh-light-mode-only" alt="Humanfirst.ai" height="30" width="auto" />
<img align="center" src="doc/assets/images/element.io.svg#gh-light-mode-only" alt="Element.io" height="25" width="auto" />
@@ -111,7 +109,7 @@ 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
View File

@@ -1,21 +0,0 @@
# 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 Normal file
View File

@@ -0,0 +1,23 @@
# 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

458
benches/agg_bench.rs
View File

@@ -1,458 +0,0 @@
use binggan::{black_box, InputGroup, PeakMemAlloc, INSTRUMENTED_SYSTEM};
use rand::prelude::SliceRandom;
use rand::rngs::StdRng;
use rand::{Rng, SeedableRng};
use rand_distr::Distribution;
use serde_json::json;
use tantivy::aggregation::agg_req::Aggregations;
use tantivy::aggregation::AggregationCollector;
use tantivy::query::{AllQuery, TermQuery};
use tantivy::schema::{IndexRecordOption, Schema, TextFieldIndexing, FAST, STRING};
use tantivy::{doc, Index, Term};
#[global_allocator]
pub static GLOBAL: &PeakMemAlloc<std::alloc::System> = &INSTRUMENTED_SYSTEM;
/// Mini macro to register a function via its name
/// runner.register("average_u64", move |index| average_u64(index));
macro_rules! register {
($runner:expr, $func:ident) => {
$runner.register(stringify!($func), move |index| $func(index))
};
}
fn main() {
let inputs = vec![
("full", get_test_index_bench(Cardinality::Full).unwrap()),
(
"dense",
get_test_index_bench(Cardinality::OptionalDense).unwrap(),
),
(
"sparse",
get_test_index_bench(Cardinality::OptionalSparse).unwrap(),
),
(
"multivalue",
get_test_index_bench(Cardinality::Multivalued).unwrap(),
),
];
bench_agg(InputGroup::new_with_inputs(inputs));
}
fn bench_agg(mut group: InputGroup<Index>) {
group.set_alloc(GLOBAL); // Set the peak mem allocator. This will enable peak memory reporting.
register!(group, average_u64);
register!(group, average_f64);
register!(group, average_f64_u64);
register!(group, stats_f64);
register!(group, extendedstats_f64);
register!(group, percentiles_f64);
register!(group, terms_few);
register!(group, terms_many);
register!(group, terms_many_top_1000);
register!(group, terms_many_order_by_term);
register!(group, terms_many_with_top_hits);
register!(group, terms_many_with_avg_sub_agg);
register!(group, terms_many_json_mixed_type_with_avg_sub_agg);
register!(group, cardinality_agg);
register!(group, terms_few_with_cardinality_agg);
register!(group, range_agg);
register!(group, range_agg_with_avg_sub_agg);
register!(group, range_agg_with_term_agg_few);
register!(group, range_agg_with_term_agg_many);
register!(group, histogram);
register!(group, histogram_hard_bounds);
register!(group, histogram_with_avg_sub_agg);
register!(group, avg_and_range_with_avg_sub_agg);
group.run();
}
fn exec_term_with_agg(index: &Index, agg_req: serde_json::Value) {
let agg_req: Aggregations = serde_json::from_value(agg_req).unwrap();
let reader = index.reader().unwrap();
let text_field = reader.searcher().schema().get_field("text").unwrap();
let term_query = TermQuery::new(
Term::from_field_text(text_field, "cool"),
IndexRecordOption::Basic,
);
let collector = get_collector(agg_req);
let searcher = reader.searcher();
black_box(searcher.search(&term_query, &collector).unwrap());
}
fn average_u64(index: &Index) {
let agg_req = json!({
"average": { "avg": { "field": "score", } }
});
exec_term_with_agg(index, agg_req)
}
fn average_f64(index: &Index) {
let agg_req = json!({
"average": { "avg": { "field": "score_f64", } }
});
exec_term_with_agg(index, agg_req)
}
fn average_f64_u64(index: &Index) {
let agg_req = json!({
"average_f64": { "avg": { "field": "score_f64" } },
"average": { "avg": { "field": "score" } },
});
exec_term_with_agg(index, agg_req)
}
fn stats_f64(index: &Index) {
let agg_req = json!({
"average_f64": { "stats": { "field": "score_f64", } }
});
exec_term_with_agg(index, agg_req)
}
fn extendedstats_f64(index: &Index) {
let agg_req = json!({
"extendedstats_f64": { "extended_stats": { "field": "score_f64", } }
});
exec_term_with_agg(index, agg_req)
}
fn percentiles_f64(index: &Index) {
let agg_req = json!({
"mypercentiles": {
"percentiles": {
"field": "score_f64",
"percents": [ 95, 99, 99.9 ]
}
}
});
execute_agg(index, agg_req);
}
fn cardinality_agg(index: &Index) {
let agg_req = json!({
"cardinality": {
"cardinality": {
"field": "text_many_terms"
},
}
});
execute_agg(index, agg_req);
}
fn terms_few_with_cardinality_agg(index: &Index) {
let agg_req = json!({
"my_texts": {
"terms": { "field": "text_few_terms" },
"aggs": {
"cardinality": {
"cardinality": {
"field": "text_many_terms"
},
}
}
},
});
execute_agg(index, agg_req);
}
fn terms_few(index: &Index) {
let agg_req = json!({
"my_texts": { "terms": { "field": "text_few_terms" } },
});
execute_agg(index, agg_req);
}
fn terms_many(index: &Index) {
let agg_req = json!({
"my_texts": { "terms": { "field": "text_many_terms" } },
});
execute_agg(index, agg_req);
}
fn terms_many_top_1000(index: &Index) {
let agg_req = json!({
"my_texts": { "terms": { "field": "text_many_terms", "size": 1000 } },
});
execute_agg(index, agg_req);
}
fn terms_many_order_by_term(index: &Index) {
let agg_req = json!({
"my_texts": { "terms": { "field": "text_many_terms", "order": { "_key": "desc" } } },
});
execute_agg(index, agg_req);
}
fn terms_many_with_top_hits(index: &Index) {
let agg_req = json!({
"my_texts": {
"terms": { "field": "text_many_terms" },
"aggs": {
"top_hits": { "top_hits":
{
"sort": [
{ "score": "desc" }
],
"size": 2,
"doc_value_fields": ["score_f64"]
}
}
}
},
});
execute_agg(index, agg_req);
}
fn terms_many_with_avg_sub_agg(index: &Index) {
let agg_req = json!({
"my_texts": {
"terms": { "field": "text_many_terms" },
"aggs": {
"average_f64": { "avg": { "field": "score_f64" } }
}
},
});
execute_agg(index, agg_req);
}
fn terms_many_json_mixed_type_with_avg_sub_agg(index: &Index) {
let agg_req = json!({
"my_texts": {
"terms": { "field": "json.mixed_type" },
"aggs": {
"average_f64": { "avg": { "field": "score_f64" } }
}
},
});
execute_agg(index, agg_req);
}
fn execute_agg(index: &Index, agg_req: serde_json::Value) {
let agg_req: Aggregations = serde_json::from_value(agg_req).unwrap();
let collector = get_collector(agg_req);
let reader = index.reader().unwrap();
let searcher = reader.searcher();
black_box(searcher.search(&AllQuery, &collector).unwrap());
}
fn range_agg(index: &Index) {
let agg_req = json!({
"range_f64": { "range": { "field": "score_f64", "ranges": [
{ "from": 3, "to": 7000 },
{ "from": 7000, "to": 20000 },
{ "from": 20000, "to": 30000 },
{ "from": 30000, "to": 40000 },
{ "from": 40000, "to": 50000 },
{ "from": 50000, "to": 60000 }
] } },
});
execute_agg(index, agg_req);
}
fn range_agg_with_avg_sub_agg(index: &Index) {
let agg_req = json!({
"rangef64": {
"range": {
"field": "score_f64",
"ranges": [
{ "from": 3, "to": 7000 },
{ "from": 7000, "to": 20000 },
{ "from": 20000, "to": 30000 },
{ "from": 30000, "to": 40000 },
{ "from": 40000, "to": 50000 },
{ "from": 50000, "to": 60000 }
]
},
"aggs": {
"average_f64": { "avg": { "field": "score_f64" } }
}
},
});
execute_agg(index, agg_req);
}
fn range_agg_with_term_agg_few(index: &Index) {
let agg_req = json!({
"rangef64": {
"range": {
"field": "score_f64",
"ranges": [
{ "from": 3, "to": 7000 },
{ "from": 7000, "to": 20000 },
{ "from": 20000, "to": 30000 },
{ "from": 30000, "to": 40000 },
{ "from": 40000, "to": 50000 },
{ "from": 50000, "to": 60000 }
]
},
"aggs": {
"my_texts": { "terms": { "field": "text_few_terms" } },
}
},
});
execute_agg(index, agg_req);
}
fn range_agg_with_term_agg_many(index: &Index) {
let agg_req = json!({
"rangef64": {
"range": {
"field": "score_f64",
"ranges": [
{ "from": 3, "to": 7000 },
{ "from": 7000, "to": 20000 },
{ "from": 20000, "to": 30000 },
{ "from": 30000, "to": 40000 },
{ "from": 40000, "to": 50000 },
{ "from": 50000, "to": 60000 }
]
},
"aggs": {
"my_texts": { "terms": { "field": "text_many_terms" } },
}
},
});
execute_agg(index, agg_req);
}
fn histogram(index: &Index) {
let agg_req = json!({
"rangef64": {
"histogram": {
"field": "score_f64",
"interval": 100 // 1000 buckets
},
}
});
execute_agg(index, agg_req);
}
fn histogram_hard_bounds(index: &Index) {
let agg_req = json!({
"rangef64": { "histogram": { "field": "score_f64", "interval": 100, "hard_bounds": { "min": 1000, "max": 300000 } } },
});
execute_agg(index, agg_req);
}
fn histogram_with_avg_sub_agg(index: &Index) {
let agg_req = json!({
"rangef64": {
"histogram": { "field": "score_f64", "interval": 100 },
"aggs": {
"average_f64": { "avg": { "field": "score_f64" } }
}
}
});
execute_agg(index, agg_req);
}
fn avg_and_range_with_avg_sub_agg(index: &Index) {
let agg_req = json!({
"rangef64": {
"range": {
"field": "score_f64",
"ranges": [
{ "from": 3, "to": 7000 },
{ "from": 7000, "to": 20000 },
{ "from": 20000, "to": 60000 }
]
},
"aggs": {
"average_in_range": { "avg": { "field": "score" } }
}
},
"average": { "avg": { "field": "score" } }
});
execute_agg(index, agg_req);
}
#[derive(Clone, Copy, Hash, Default, Debug, PartialEq, Eq, PartialOrd, Ord)]
enum Cardinality {
/// All documents contain exactly one value.
/// `Full` is the default for auto-detecting the Cardinality, since it is the most strict.
#[default]
Full = 0,
/// All documents contain at most one value.
OptionalDense = 1,
/// All documents may contain any number of values.
Multivalued = 2,
/// 1 / 20 documents has a value
OptionalSparse = 3,
}
fn get_collector(agg_req: Aggregations) -> AggregationCollector {
AggregationCollector::from_aggs(agg_req, Default::default())
}
fn get_test_index_bench(cardinality: Cardinality) -> tantivy::Result<Index> {
let mut schema_builder = Schema::builder();
let text_fieldtype = tantivy::schema::TextOptions::default()
.set_indexing_options(
TextFieldIndexing::default().set_index_option(IndexRecordOption::WithFreqs),
)
.set_stored();
let text_field = schema_builder.add_text_field("text", text_fieldtype);
let json_field = schema_builder.add_json_field("json", FAST);
let text_field_many_terms = schema_builder.add_text_field("text_many_terms", STRING | FAST);
let text_field_few_terms = schema_builder.add_text_field("text_few_terms", STRING | FAST);
let score_fieldtype = tantivy::schema::NumericOptions::default().set_fast();
let score_field = schema_builder.add_u64_field("score", score_fieldtype.clone());
let score_field_f64 = schema_builder.add_f64_field("score_f64", score_fieldtype.clone());
let score_field_i64 = schema_builder.add_i64_field("score_i64", score_fieldtype);
let index = Index::create_from_tempdir(schema_builder.build())?;
let few_terms_data = ["INFO", "ERROR", "WARN", "DEBUG"];
let lg_norm = rand_distr::LogNormal::new(2.996f64, 0.979f64).unwrap();
let many_terms_data = (0..150_000)
.map(|num| format!("author{num}"))
.collect::<Vec<_>>();
{
let mut rng = StdRng::from_seed([1u8; 32]);
let mut index_writer = index.writer_with_num_threads(1, 200_000_000)?;
// To make the different test cases comparable we just change one doc to force the
// cardinality
if cardinality == Cardinality::OptionalDense {
index_writer.add_document(doc!())?;
}
if cardinality == Cardinality::Multivalued {
index_writer.add_document(doc!(
json_field => json!({"mixed_type": 10.0}),
json_field => json!({"mixed_type": 10.0}),
text_field => "cool",
text_field => "cool",
text_field_many_terms => "cool",
text_field_many_terms => "cool",
text_field_few_terms => "cool",
text_field_few_terms => "cool",
score_field => 1u64,
score_field => 1u64,
score_field_f64 => lg_norm.sample(&mut rng),
score_field_f64 => lg_norm.sample(&mut rng),
score_field_i64 => 1i64,
score_field_i64 => 1i64,
))?;
}
let mut doc_with_value = 1_000_000;
if cardinality == Cardinality::OptionalSparse {
doc_with_value /= 20;
}
let _val_max = 1_000_000.0;
for _ in 0..doc_with_value {
let val: f64 = rng.gen_range(0.0..1_000_000.0);
let json = if rng.gen_bool(0.1) {
// 10% are numeric values
json!({ "mixed_type": val })
} else {
json!({"mixed_type": many_terms_data.choose(&mut rng).unwrap().to_string()})
};
index_writer.add_document(doc!(
text_field => "cool",
json_field => json,
text_field_many_terms => many_terms_data.choose(&mut rng).unwrap().to_string(),
text_field_few_terms => few_terms_data.choose(&mut rng).unwrap().to_string(),
score_field => val as u64,
score_field_f64 => lg_norm.sample(&mut rng),
score_field_i64 => val as i64,
))?;
if cardinality == Cardinality::OptionalSparse {
for _ in 0..20 {
index_writer.add_document(doc!(text_field => "cool"))?;
}
}
}
// writing the segment
index_writer.commit()?;
}
Ok(index)
}

27
benches/analyzer.rs
View File

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

1000
benches/gh.json
View File

File diff suppressed because one or more lines are too long

267
benches/index-bench.rs
View File

@@ -1,99 +1,10 @@
use criterion::{criterion_group, criterion_main, BatchSize, Bencher, Criterion, Throughput};
use tantivy::schema::{TantivyDocument, FAST, INDEXED, STORED, STRING, TEXT};
use tantivy::{tokenizer, Index, IndexWriter};
use criterion::{criterion_group, criterion_main, Criterion};
use pprof::criterion::{Output, PProfProfiler};
use tantivy::schema::{INDEXED, STORED, STRING, TEXT};
use tantivy::Index;
const HDFS_LOGS: &str = include_str!("hdfs.json");
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::Value = serde_json::from_str(doc_json).unwrap();
tantivy::doc!(json_field=>json_val)
})
}
const NUM_REPEATS: usize = 2;
pub fn hdfs_index_benchmark(c: &mut Criterion) {
let schema = {
@@ -103,14 +14,7 @@ pub fn hdfs_index_benchmark(c: &mut Criterion) {
schema_builder.add_text_field("severity", STRING);
schema_builder.build()
};
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 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);
@@ -119,100 +23,99 @@ 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 | FAST);
schema_builder.add_json_field("json", TEXT);
schema_builder.build()
};
let mut group = c.benchmark_group("index-hdfs");
group.throughput(Throughput::Bytes(HDFS_LOGS.len() as u64));
group.sample_size(20);
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" };
{
let parse_json = false;
// for parse_json in [false, true] {
let suffix = if parse_json {
format!("{suffix}-with-json-parsing")
} else {
suffix.to_string()
};
let bench_name = format!("{prefix}{suffix}");
group.bench_function(bench_name, |b| {
benchmark(b, HDFS_LOGS, schema.clone(), commit, parse_json, is_dynamic)
});
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();
}
}
}
}
}
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-gh-fast", |b| {
benchmark_dynamic_json(b, GH_LOGS, dynamic_schema_fast.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-with-commit", |b| {
benchmark_dynamic_json(b, GH_LOGS, dynamic_schema_fast.clone(), true, 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();
}
}
})
});
}
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-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();
})
});
group.bench_function("index-wiki-with-commit", |b| {
benchmark_dynamic_json(b, WIKI, dynamic_schema.clone(), true, 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();
})
});
}
criterion_group! {
name = benches;
config = Criterion::default();
config = Criterion::default().with_profiler(PProfProfiler::new(100, Output::Flamegraph(None)));
targets = hdfs_index_benchmark
}
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);
criterion_main!(benches);

1000
benches/wiki.json
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 = "0.6.0"
version = "0.3.0"
edition = "2021"
authors = ["Paul Masurel <paul.masurel@gmail.com>"]
license = "MIT"
@@ -15,7 +15,6 @@ 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

@@ -1,13 +1,10 @@
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()
@@ -121,125 +118,6 @@ 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)]
@@ -322,58 +200,4 @@ 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

@@ -64,8 +64,10 @@ 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: vec![0; 8],
compressed_blocks,
buffer: vec![],
offset_and_bits: vec![],
}

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

@@ -1,365 +0,0 @@
//! 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
View File

@@ -1,165 +0,0 @@
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
View File

@@ -1,13 +0,0 @@
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

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

93
cliff.toml
View File

@@ -1,93 +0,0 @@
# 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

34
columnar/Cargo.toml
View File

@@ -1,38 +1,28 @@
[package]
name = "tantivy-columnar"
version = "0.3.0"
version = "0.1.0"
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 = "0.13.0"
itertools = "0.10.5"
log = "0.4.17"
fnv = "1.0.7"
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 = { 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/" }
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/" }
serde = "1.0.152"
downcast-rs = "1.2.0"
[dev-dependencies]
proptest = "1"
more-asserts = "0.3.1"
rand = "0.8"
binggan = "0.8.1"
[[bench]]
name = "bench_merge"
harness = false
[[bench]]
name = "bench_access"
harness = false
rand = "0.8.5"
[features]
unstable = []

67
columnar/benches/bench_access.rs
View File

@@ -1,67 +0,0 @@
use binggan::{black_box, InputGroup};
use common::*;
use tantivy_columnar::Column;
pub mod common;
const NUM_DOCS: u32 = 2_000_000;
pub fn generate_columnar_and_open(card: Card, num_docs: u32) -> Column {
let reader = generate_columnar_with_name(card, num_docs, "price");
reader.read_columns("price").unwrap()[0]
.open_u64_lenient()
.unwrap()
.unwrap()
}
fn main() {
let mut inputs = Vec::new();
let mut add_card = |card1: Card| {
inputs.push((
format!("{card1}"),
generate_columnar_and_open(card1, NUM_DOCS),
));
};
add_card(Card::MultiSparse);
add_card(Card::Multi);
add_card(Card::Sparse);
add_card(Card::Dense);
add_card(Card::Full);
bench_group(InputGroup::new_with_inputs(inputs));
}
fn bench_group(mut runner: InputGroup<Column>) {
runner.register("access_values_for_doc", |column| {
let mut sum = 0;
for i in 0..NUM_DOCS {
for value in column.values_for_doc(i) {
sum += value;
}
}
black_box(sum);
});
runner.register("access_first_vals", |column| {
let mut sum = 0;
const BLOCK_SIZE: usize = 32;
let mut docs = vec![0; BLOCK_SIZE];
let mut buffer = vec![None; BLOCK_SIZE];
for i in (0..NUM_DOCS).step_by(BLOCK_SIZE) {
// fill docs
for idx in 0..BLOCK_SIZE {
docs[idx] = idx as u32 + i;
}
column.first_vals(&docs, &mut buffer);
for val in buffer.iter() {
let Some(val) = val else { continue };
sum += *val;
}
}
black_box(sum);
});
runner.run();
}

155
columnar/benches/bench_first_vals.rs
View File

@@ -1,155 +0,0 @@
#![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, &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);
}

49
columnar/benches/bench_merge.rs
View File

@@ -1,49 +0,0 @@
pub mod common;
use binggan::{black_box, BenchRunner};
use common::{generate_columnar_with_name, Card};
use tantivy_columnar::*;
const NUM_DOCS: u32 = 100_000;
fn main() {
let mut inputs = Vec::new();
let mut add_combo = |card1: Card, card2: Card| {
inputs.push((
format!("merge_{card1}_and_{card2}"),
vec![
generate_columnar_with_name(card1, NUM_DOCS, "price"),
generate_columnar_with_name(card2, NUM_DOCS, "price"),
],
));
};
add_combo(Card::Multi, Card::Multi);
add_combo(Card::MultiSparse, Card::MultiSparse);
add_combo(Card::Dense, Card::Dense);
add_combo(Card::Sparse, Card::Sparse);
add_combo(Card::Sparse, Card::Dense);
add_combo(Card::MultiSparse, Card::Dense);
add_combo(Card::MultiSparse, Card::Sparse);
add_combo(Card::Multi, Card::Dense);
add_combo(Card::Multi, Card::Sparse);
let runner: BenchRunner = BenchRunner::new();
let mut group = runner.new_group();
for (input_name, columnar_readers) in inputs.iter() {
group.register_with_input(
input_name,
columnar_readers,
move |columnar_readers: &Vec<ColumnarReader>| {
let mut out = Vec::new();
let columnar_readers = columnar_readers.iter().collect::<Vec<_>>();
let merge_row_order = StackMergeOrder::stack(&columnar_readers[..]);
merge_columnar(&columnar_readers, &[], merge_row_order.into(), &mut out).unwrap();
black_box(out);
},
);
}
group.run();
}

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

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

@@ -16,6 +16,14 @@ 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();

59
columnar/benches/common.rs
View File

@@ -1,59 +0,0 @@
extern crate tantivy_columnar;
use core::fmt;
use std::fmt::{Display, Formatter};
use tantivy_columnar::{ColumnarReader, ColumnarWriter};
pub enum Card {
MultiSparse,
Multi,
Sparse,
Dense,
Full,
}
impl Display for Card {
fn fmt(&self, f: &mut Formatter) -> fmt::Result {
match self {
Card::MultiSparse => write!(f, "multi sparse 1/13"),
Card::Multi => write!(f, "multi 2x"),
Card::Sparse => write!(f, "sparse 1/13"),
Card::Dense => write!(f, "dense 1/12"),
Card::Full => write!(f, "full"),
}
}
}
pub fn generate_columnar_with_name(card: Card, num_docs: u32, column_name: &str) -> ColumnarReader {
let mut columnar_writer = ColumnarWriter::default();
if let Card::MultiSparse = card {
columnar_writer.record_numerical(0, column_name, 10u64);
columnar_writer.record_numerical(0, column_name, 10u64);
}
for i in 0..num_docs {
match card {
Card::MultiSparse | Card::Sparse => {
if i % 13 == 0 {
columnar_writer.record_numerical(i, column_name, i as u64);
}
}
Card::Dense => {
if i % 12 == 0 {
columnar_writer.record_numerical(i, column_name, i as u64);
}
}
Card::Full => {
columnar_writer.record_numerical(i, column_name, i as u64);
}
Card::Multi => {
columnar_writer.record_numerical(i, column_name, i as u64);
columnar_writer.record_numerical(i, column_name, i as u64);
}
}
}
let mut wrt: Vec<u8> = Vec::new();
columnar_writer.serialize(num_docs, &mut wrt).unwrap();
ColumnarReader::open(wrt).unwrap()
}

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

@@ -8,6 +8,7 @@ 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 = []

BIN
columnar/compat_tests_data/v1.columnar
View File

Binary file not shown.

BIN
columnar/compat_tests_data/v2.columnar
View File

Binary file not shown.

158
columnar/src/block_accessor.rs
View File

@@ -1,158 +0,0 @@
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};
@@ -21,22 +21,7 @@ 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
@@ -71,12 +56,6 @@ 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
@@ -84,7 +63,7 @@ impl From<StrColumn> for BytesColumn {
}
impl StrColumn {
pub fn wrap(bytes_column: BytesColumn) -> StrColumn {
pub(crate) fn wrap(bytes_column: BytesColumn) -> StrColumn {
StrColumn(bytes_column)
}

108
columnar/src/column/mod.rs
View File

@@ -1,48 +1,29 @@
mod dictionary_encoded;
mod serialize;
use std::fmt::{self, Debug};
use std::fmt::Debug;
use std::io::Write;
use std::ops::{Range, RangeInclusive};
use std::ops::{Deref, 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_u128_as_compact_u64,
open_column_u64, serialize_column_mappable_to_u128, serialize_column_mappable_to_u64,
open_column_bytes, open_column_str, open_column_u128, open_column_u64,
serialize_column_mappable_to_u128, serialize_column_mappable_to_u64,
};
use crate::column_index::{ColumnIndex, Set};
use crate::column_index::ColumnIndex;
use crate::column_values::monotonic_mapping::StrictlyMonotonicMappingToInternal;
use crate::column_values::{monotonic_map_column, ColumnValues};
use crate::{Cardinality, DocId, EmptyColumnValues, MonotonicallyMappableToU64, RowId};
use crate::{Cardinality, MonotonicallyMappableToU64, RowId};
#[derive(Clone)]
pub struct Column<T = u64> {
pub index: ColumnIndex,
pub idx: 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(
@@ -50,7 +31,7 @@ impl<T: MonotonicallyMappableToU64> Column<T> {
StrictlyMonotonicMappingToInternal::<T>::new(),
));
Column {
index: self.index,
idx: self.idx,
values,
}
}
@@ -59,11 +40,11 @@ impl<T: MonotonicallyMappableToU64> Column<T> {
impl<T: PartialOrd + Copy + Debug + Send + Sync + 'static> Column<T> {
#[inline]
pub fn get_cardinality(&self) -> Cardinality {
self.index.get_cardinality()
self.idx.get_cardinality()
}
pub fn num_docs(&self) -> RowId {
match &self.index {
match &self.idx {
ColumnIndex::Empty { num_docs } => *num_docs,
ColumnIndex::Full => self.values.num_vals(),
ColumnIndex::Optional(optional_index) => optional_index.num_docs(),
@@ -83,60 +64,16 @@ 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()
}
/// 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)
pub fn values_for_doc(&self, row_id: RowId) -> impl Iterator<Item = T> + '_ {
self.value_row_ids(row_id)
.map(|value_row_id: RowId| self.values.get_val(value_row_id))
}
/// Get the docids of values which are in the provided value and docid range.
/// Get the docids of values which are in the provided value range.
#[inline]
pub fn get_docids_for_value_range(
&self,
@@ -145,19 +82,17 @@ 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
.index
.docid_range_to_rowids(selected_docid_range.clone());
let rowid_range = self.idx.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.index
self.idx
.select_batch_in_place(selected_docid_range.start, doc_ids);
}
/// Fills the output vector with the (possibly multiple values that are associated_with
/// Fils the output vector with the (possibly multiple values that are associated_with
/// `row_id`.
///
/// This method clears the `output` vector.
@@ -174,6 +109,14 @@ 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)
@@ -195,7 +138,6 @@ 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)
}
@@ -209,7 +151,7 @@ impl<T: PartialOrd + Debug + Send + Sync + Copy + 'static> ColumnValues<T>
}
fn num_vals(&self) -> u32 {
match &self.column.index {
match &self.column.idx {
ColumnIndex::Empty { .. } => 0u32,
ColumnIndex::Full => self.column.values.num_vals(),
ColumnIndex::Optional(optional_idx) => optional_idx.num_docs(),

47
columnar/src/column/serialize.rs
View File

@@ -12,7 +12,7 @@ use crate::column_values::{
CodecType, MonotonicallyMappableToU128, MonotonicallyMappableToU64,
};
use crate::iterable::Iterable;
use crate::{StrColumn, Version};
use crate::StrColumn;
pub fn serialize_column_mappable_to_u128<T: MonotonicallyMappableToU128>(
column_index: SerializableColumnIndex<'_>,
@@ -40,10 +40,7 @@ pub fn serialize_column_mappable_to_u64<T: MonotonicallyMappableToU64>(
Ok(())
}
pub fn open_column_u64<T: MonotonicallyMappableToU64>(
bytes: OwnedBytes,
format_version: Version,
) -> io::Result<Column<T>> {
pub fn open_column_u64<T: MonotonicallyMappableToU64>(bytes: OwnedBytes) -> io::Result<Column<T>> {
let (body, column_index_num_bytes_payload) = bytes.rsplit(4);
let column_index_num_bytes = u32::from_le_bytes(
column_index_num_bytes_payload
@@ -52,17 +49,16 @@ pub fn open_column_u64<T: MonotonicallyMappableToU64>(
.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, format_version)?;
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 {
index: column_index,
idx: column_index,
values: column_values,
})
}
pub fn open_column_u128<T: MonotonicallyMappableToU128>(
bytes: OwnedBytes,
format_version: Version,
) -> io::Result<Column<T>> {
let (body, column_index_num_bytes_payload) = bytes.rsplit(4);
let column_index_num_bytes = u32::from_le_bytes(
@@ -72,50 +68,27 @@ pub fn open_column_u128<T: MonotonicallyMappableToU128>(
.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, format_version)?;
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 {
index: column_index,
idx: 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,
format_version: Version,
) -> 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, format_version)?;
let column_values = crate::column_values::open_u128_as_compact_u64(column_values_data)?;
Ok(Column {
index: column_index,
values: column_values,
})
}
pub fn open_column_bytes(data: OwnedBytes, format_version: Version) -> io::Result<BytesColumn> {
pub fn open_column_bytes(data: OwnedBytes) -> io::Result<BytesColumn> {
let (body, dictionary_len_bytes) = data.rsplit(4);
let dictionary_len = u32::from_le_bytes(dictionary_len_bytes.as_slice().try_into().unwrap());
let (dictionary_bytes, column_bytes) = body.split(dictionary_len as usize);
let dictionary = Arc::new(Dictionary::from_bytes(dictionary_bytes)?);
let term_ord_column = crate::column::open_column_u64::<u64>(column_bytes, format_version)?;
let term_ord_column = crate::column::open_column_u64::<u64>(column_bytes)?;
Ok(BytesColumn {
dictionary,
term_ord_column,
})
}
pub fn open_column_str(data: OwnedBytes, format_version: Version) -> io::Result<StrColumn> {
let bytes_column = open_column_bytes(data, format_version)?;
pub fn open_column_str(data: OwnedBytes) -> io::Result<StrColumn> {
let bytes_column = open_column_bytes(data)?;
Ok(StrColumn::wrap(bytes_column))
}

153
columnar/src/column_index/merge/mod.rs
View File

@@ -1,82 +1,29 @@
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};
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
}
}
// 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)
}
pub fn merge_column_index<'a>(
columns: &'a [ColumnIndex],
columns: &'a [Option<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, merge_row_order);
let cardinality_after_merge = detect_cardinality(columns);
match merge_row_order {
MergeRowOrder::Stack(stack_merge_order) => {
merge_column_index_stacked(columns, cardinality_after_merge, stack_merge_order)
@@ -95,68 +42,45 @@ pub fn merge_column_index<'a>(
#[cfg(test)]
mod tests {
use common::OwnedBytes;
use crate::column_index::merge::detect_cardinality;
use crate::column_index::multivalued_index::{
open_multivalued_index, serialize_multivalued_index, MultiValueIndex,
};
use crate::column_index::multivalued_index::MultiValueIndex;
use crate::column_index::{merge_column_index, OptionalIndex, SerializableColumnIndex};
use crate::{
Cardinality, ColumnIndex, MergeRowOrder, RowAddr, RowId, ShuffleMergeOrder, StackMergeOrder,
};
use crate::{Cardinality, ColumnIndex, MergeRowOrder, RowAddr, RowId, ShuffleMergeOrder};
#[test]
fn test_detect_cardinality() {
assert_eq!(
detect_cardinality(&[], &StackMergeOrder::stack_for_test(&[]).into()),
Cardinality::Full
);
assert_eq!(detect_cardinality(&[]), 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(
&[optional_index.clone(), ColumnIndex::Empty { num_docs: 0 }],
&StackMergeOrder::stack_for_test(&[1, 0]).into()
),
detect_cardinality(&[Some(optional_index.clone()), None]),
Cardinality::Optional
);
assert_eq!(
detect_cardinality(
&[optional_index.clone(), ColumnIndex::Full],
&StackMergeOrder::stack_for_test(&[1, 1]).into()
),
detect_cardinality(&[Some(optional_index.clone()), Some(ColumnIndex::Full)]),
Cardinality::Optional
);
assert_eq!(
detect_cardinality(
&[
multivalued_index.clone(),
ColumnIndex::Empty { num_docs: 0 }
],
&StackMergeOrder::stack_for_test(&[1, 0]).into()
),
detect_cardinality(&[Some(multivalued_index.clone()), None]),
Cardinality::Multivalued
);
assert_eq!(
detect_cardinality(
&[multivalued_index.clone(), optional_index.clone()],
&StackMergeOrder::stack_for_test(&[1, 1]).into()
),
detect_cardinality(&[
Some(multivalued_index.clone()),
Some(optional_index.clone())
]),
Cardinality::Multivalued
);
assert_eq!(
detect_cardinality(
&[optional_index, multivalued_index],
&StackMergeOrder::stack_for_test(&[1, 1]).into()
),
detect_cardinality(&[Some(optional_index), Some(multivalued_index)]),
Cardinality::Multivalued
);
}
#[test]
fn test_merge_index_multivalued_sorted() {
let column_indexes: Vec<ColumnIndex> = vec![MultiValueIndex::for_test(&[0, 2, 5]).into()];
let column_indexes: Vec<Option<ColumnIndex>> =
vec![Some(MultiValueIndex::for_test(&[0, 2, 5]).into())];
let merge_row_order: MergeRowOrder = ShuffleMergeOrder::for_test(
&[2],
vec![
@@ -172,23 +96,18 @@ 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 mut output = Vec::new();
serialize_multivalued_index(&start_index_iterable, &mut output).unwrap();
let multivalue =
open_multivalued_index(OwnedBytes::new(output), crate::Version::V2).unwrap();
let start_indexes: Vec<RowId> = multivalue.get_start_index_column().iter().collect();
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<ColumnIndex> = vec![
MultiValueIndex::for_test(&[0, 2, 5]).into(),
ColumnIndex::Empty { num_docs: 0 },
MultiValueIndex::for_test(&[0, 1, 4]).into(),
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 merge_row_order: MergeRowOrder = ShuffleMergeOrder::for_test(
&[2, 0, 2],
@@ -208,16 +127,10 @@ 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 mut output = Vec::new();
serialize_multivalued_index(&start_index_iterable, &mut output).unwrap();
let multivalue =
open_multivalued_index(OwnedBytes::new(output), crate::Version::V2).unwrap();
let start_indexes: Vec<RowId> = multivalue.get_start_index_column().iter().collect();
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]);
}
}

101
columnar/src/column_index/merge/shuffled.rs
View File

@@ -1,13 +1,11 @@
use std::iter;
use crate::column_index::{
SerializableColumnIndex, SerializableMultivalueIndex, SerializableOptionalIndex, Set,
};
use crate::column_index::{SerializableColumnIndex, Set};
use crate::iterable::Iterable;
use crate::{Cardinality, ColumnIndex, RowId, ShuffleMergeOrder};
pub fn merge_column_index_shuffled<'a>(
column_indexes: &'a [ColumnIndex],
column_indexes: &'a [Option<ColumnIndex>],
cardinality_after_merge: Cardinality,
shuffle_merge_order: &'a ShuffleMergeOrder,
) -> SerializableColumnIndex<'a> {
@@ -16,24 +14,15 @@ pub fn merge_column_index_shuffled<'a>(
Cardinality::Optional => {
let non_null_row_ids =
merge_column_index_shuffled_optional(column_indexes, shuffle_merge_order);
SerializableColumnIndex::Optional(SerializableOptionalIndex {
SerializableColumnIndex::Optional {
non_null_row_ids,
num_rows: shuffle_merge_order.num_rows(),
})
}
}
Cardinality::Multivalued => {
let non_null_row_ids =
merge_column_index_shuffled_optional(column_indexes, shuffle_merge_order);
SerializableColumnIndex::Multivalued(SerializableMultivalueIndex {
doc_ids_with_values: SerializableOptionalIndex {
non_null_row_ids,
num_rows: shuffle_merge_order.num_rows(),
},
start_offsets: merge_column_index_shuffled_multivalued(
column_indexes,
shuffle_merge_order,
),
})
let multivalue_start_index =
merge_column_index_shuffled_multivalued(column_indexes, shuffle_merge_order);
SerializableColumnIndex::Multivalued(multivalue_start_index)
}
}
}
@@ -44,41 +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 [ColumnIndex],
column_indexes: &'a [Option<ColumnIndex>],
merge_order: &'a ShuffleMergeOrder,
) -> Box<dyn Iterable<RowId> + 'a> {
Box::new(ShuffledIndex {
Box::new(ShuffledOptionalIndex {
column_indexes,
merge_order,
})
}
struct ShuffledIndex<'a> {
column_indexes: &'a [ColumnIndex],
struct ShuffledOptionalIndex<'a> {
column_indexes: &'a [Option<ColumnIndex>],
merge_order: &'a ShuffleMergeOrder,
}
impl<'a> Iterable<u32> for ShuffledIndex<'a> {
impl<'a> Iterable<u32> for ShuffledOptionalIndex<'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 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
}
}),
)
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
}
}))
}
}
fn merge_column_index_shuffled_multivalued<'a>(
column_indexes: &'a [ColumnIndex],
column_indexes: &'a [Option<ColumnIndex>],
merge_order: &'a ShuffleMergeOrder,
) -> Box<dyn Iterable<RowId> + 'a> {
Box::new(ShuffledMultivaluedIndex {
@@ -88,16 +77,19 @@ fn merge_column_index_shuffled_multivalued<'a>(
}
struct ShuffledMultivaluedIndex<'a> {
column_indexes: &'a [ColumnIndex],
column_indexes: &'a [Option<ColumnIndex>],
merge_order: &'a ShuffleMergeOrder,
}
fn iter_num_values<'a>(
column_indexes: &'a [ColumnIndex],
column_indexes: &'a [Option<ColumnIndex>],
merge_order: &'a ShuffleMergeOrder,
) -> impl Iterator<Item = u32> + 'a {
merge_order.iter_new_to_old_row_addrs().map(|row_addr| {
let column_index = &column_indexes[row_addr.segment_ord as usize];
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;
};
match column_index {
ColumnIndex::Empty { .. } => 0u32,
ColumnIndex::Full => 1,
@@ -113,18 +105,11 @@ fn iter_num_values<'a>(
/// Transforms an iterator containing the number of vals per row (with `num_rows` elements)
/// into a `start_offset` iterator starting at 0 and (with `num_rows + 1` element)
///
/// This will filter values with 0 values as these are covered by the optional index in the
/// multivalue index.
fn integrate_num_vals(num_vals: impl Iterator<Item = u32>) -> impl Iterator<Item = RowId> {
iter::once(0u32).chain(
num_vals
.filter(|num_vals| *num_vals != 0)
.scan(0, |state, num_vals| {
*state += num_vals;
Some(*state)
}),
)
iter::once(0u32).chain(num_vals.scan(0, |state, num_vals| {
*state += num_vals;
Some(*state)
}))
}
impl<'a> Iterable<u32> for ShuffledMultivaluedIndex<'a> {
@@ -152,13 +137,13 @@ mod tests {
#[test]
fn test_integrate_num_vals_several() {
assert!(integrate_num_vals([3, 0, 10, 20].into_iter()).eq([0, 3, 13, 33].into_iter()));
assert!(integrate_num_vals([3, 0, 10, 20].into_iter()).eq([0, 3, 3, 13, 33].into_iter()));
}
#[test]
fn test_merge_column_index_optional_shuffle() {
let optional_index: ColumnIndex = OptionalIndex::for_test(2, &[0]).into();
let column_indexes = [optional_index, ColumnIndex::Full];
let column_indexes = vec![Some(optional_index), Some(ColumnIndex::Full)];
let row_addrs = vec![
RowAddr {
segment_ord: 0u32,
@@ -175,13 +160,7 @@ mod tests {
Cardinality::Optional,
&shuffle_merge_order,
);
let SerializableColumnIndex::Optional(SerializableOptionalIndex {
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]);

250
columnar/src/column_index/merge/stacked.rs
View File

@@ -1,8 +1,6 @@
use std::ops::Range;
use std::iter;
use crate::column_index::multivalued_index::{MultiValueIndex, SerializableMultivalueIndex};
use crate::column_index::serialize::SerializableOptionalIndex;
use crate::column_index::SerializableColumnIndex;
use crate::column_index::{SerializableColumnIndex, Set};
use crate::iterable::Iterable;
use crate::{Cardinality, ColumnIndex, RowId, StackMergeOrder};
@@ -11,157 +9,31 @@ use crate::{Cardinality, ColumnIndex, RowId, StackMergeOrder};
///
/// There are no sort nor deletes involved.
pub fn merge_column_index_stacked<'a>(
columns: &'a [ColumnIndex],
columns: &'a [Option<ColumnIndex>],
cardinality_after_merge: Cardinality,
stack_merge_order: &'a StackMergeOrder,
) -> SerializableColumnIndex<'a> {
match cardinality_after_merge {
Cardinality::Full => SerializableColumnIndex::Full,
Cardinality::Optional => SerializableColumnIndex::Optional(SerializableOptionalIndex {
Cardinality::Optional => SerializableColumnIndex::Optional {
non_null_row_ids: Box::new(StackedOptionalIndex {
columns,
stack_merge_order,
}),
num_rows: stack_merge_order.num_rows(),
}),
Cardinality::Multivalued => {
let serializable_multivalue_index =
make_serializable_multivalued_index(columns, stack_merge_order);
SerializableColumnIndex::Multivalued(serializable_multivalue_index)
}
}
}
struct StackedDocIdsWithValues<'a> {
column_indexes: &'a [ColumnIndex],
stack_merge_order: &'a StackMergeOrder,
}
impl Iterable<u32> for StackedDocIdsWithValues<'_> {
fn boxed_iter(&self) -> Box<dyn Iterator<Item = u32> + '_> {
Box::new((0..self.column_indexes.len()).flat_map(|i| {
let column_index = &self.column_indexes[i];
let doc_range = self.stack_merge_order.columnar_range(i);
get_doc_ids_with_values(column_index, doc_range)
}))
}
}
fn get_doc_ids_with_values<'a>(
column_index: &'a ColumnIndex,
doc_range: Range<u32>,
) -> Box<dyn Iterator<Item = u32> + 'a> {
match column_index {
ColumnIndex::Empty { .. } => Box::new(0..0),
ColumnIndex::Full => Box::new(doc_range),
ColumnIndex::Optional(optional_index) => Box::new(
optional_index
.iter_rows()
.map(move |row| row + doc_range.start),
),
ColumnIndex::Multivalued(multivalued_index) => match multivalued_index {
MultiValueIndex::MultiValueIndexV1(multivalued_index) => {
Box::new((0..multivalued_index.num_docs()).filter_map(move |docid| {
let range = multivalued_index.range(docid);
if range.is_empty() {
None
} else {
Some(docid + doc_range.start)
}
}))
}
MultiValueIndex::MultiValueIndexV2(multivalued_index) => Box::new(
multivalued_index
.optional_index
.iter_rows()
.map(move |row| row + doc_range.start),
),
},
}
}
fn stack_doc_ids_with_values<'a>(
column_indexes: &'a [ColumnIndex],
stack_merge_order: &'a StackMergeOrder,
) -> SerializableOptionalIndex<'a> {
let num_rows = stack_merge_order.num_rows();
SerializableOptionalIndex {
non_null_row_ids: Box::new(StackedDocIdsWithValues {
column_indexes,
stack_merge_order,
}),
num_rows,
}
}
struct StackedStartOffsets<'a> {
column_indexes: &'a [ColumnIndex],
stack_merge_order: &'a StackMergeOrder,
}
fn get_num_values_iterator<'a>(
column_index: &'a ColumnIndex,
num_docs: u32,
) -> Box<dyn Iterator<Item = u32> + 'a> {
match column_index {
ColumnIndex::Empty { .. } => Box::new(std::iter::empty()),
ColumnIndex::Full => Box::new(std::iter::repeat(1u32).take(num_docs as usize)),
ColumnIndex::Optional(optional_index) => {
Box::new(std::iter::repeat(1u32).take(optional_index.num_non_nulls() as usize))
Cardinality::Multivalued => {
let stacked_multivalued_index = StackedMultivaluedIndex {
columns,
stack_merge_order,
};
SerializableColumnIndex::Multivalued(Box::new(stacked_multivalued_index))
}
ColumnIndex::Multivalued(multivalued_index) => Box::new(
multivalued_index
.get_start_index_column()
.iter()
.scan(0u32, |previous_start_offset, current_start_offset| {
let num_vals = current_start_offset - *previous_start_offset;
*previous_start_offset = current_start_offset;
Some(num_vals)
})
.skip(1),
),
}
}
impl<'a> Iterable<u32> for StackedStartOffsets<'a> {
fn boxed_iter(&self) -> Box<dyn Iterator<Item = u32> + '_> {
let num_values_it = (0..self.column_indexes.len()).flat_map(|columnar_id| {
let num_docs = self.stack_merge_order.columnar_range(columnar_id).len() as u32;
let column_index = &self.column_indexes[columnar_id];
get_num_values_iterator(column_index, num_docs)
});
Box::new(std::iter::once(0u32).chain(num_values_it.into_iter().scan(
0u32,
|cumulated, el| {
*cumulated += el;
Some(*cumulated)
},
)))
}
}
fn stack_start_offsets<'a>(
column_indexes: &'a [ColumnIndex],
stack_merge_order: &'a StackMergeOrder,
) -> Box<dyn Iterable<u32> + 'a> {
Box::new(StackedStartOffsets {
column_indexes,
stack_merge_order,
})
}
fn make_serializable_multivalued_index<'a>(
columns: &'a [ColumnIndex],
stack_merge_order: &'a StackMergeOrder,
) -> SerializableMultivalueIndex<'a> {
SerializableMultivalueIndex {
doc_ids_with_values: stack_doc_ids_with_values(columns, stack_merge_order),
start_offsets: stack_start_offsets(columns, stack_merge_order),
}
}
struct StackedOptionalIndex<'a> {
columns: &'a [ColumnIndex],
columns: &'a [Option<ColumnIndex>],
stack_merge_order: &'a StackMergeOrder,
}
@@ -174,19 +46,111 @@ 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 {
ColumnIndex::Full => Box::new(columnar_row_range),
ColumnIndex::Optional(optional_index) => Box::new(
Some(ColumnIndex::Full) => Box::new(columnar_row_range),
Some(ColumnIndex::Optional(optional_index)) => Box::new(
optional_index
.iter_rows()
.map(move |row_id: RowId| columnar_row_range.start + row_id),
),
ColumnIndex::Multivalued(_) => {
Some(ColumnIndex::Multivalued(_)) => {
panic!("No multivalued index is allowed when stacking column index");
}
ColumnIndex::Empty { .. } => Box::new(std::iter::empty()),
None | Some(ColumnIndex::Empty { .. }) => Box::new(std::iter::empty()),
};
rows_it
}),
)
}
}
#[derive(Clone, Copy)]
struct StackedMultivaluedIndex<'a> {
columns: &'a [Option<ColumnIndex>],
stack_merge_order: &'a StackMergeOrder,
}
fn convert_column_opt_to_multivalued_index<'a>(
column_index_opt: Option<&'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)) => {
Box::new(
(0..num_rows)
// TODO optimize
.map(|row_id| optional_index.rank(row_id))
.chain(std::iter::once(optional_index.num_non_nulls())),
)
}
Some(ColumnIndex::Multivalued(multivalued_index)) => {
multivalued_index.start_index_column.iter()
}
}
}
impl<'a> Iterable<RowId> for StackedMultivaluedIndex<'a> {
fn boxed_iter(&self) -> Box<dyn Iterator<Item = RowId> + '_> {
let multivalued_indexes =
self.columns
.iter()
.map(Option::as_ref)
.enumerate()
.map(|(columnar_id, column_opt)| {
let num_rows =
self.stack_merge_order.columnar_range(columnar_id).len() as RowId;
convert_column_opt_to_multivalued_index(column_opt, num_rows)
});
stack_multivalued_indexes(multivalued_indexes)
}
}
// Refactor me
fn stack_multivalued_indexes<'a>(
mut multivalued_indexes: impl Iterator<Item = Box<dyn Iterator<Item = RowId> + 'a>> + 'a,
) -> Box<dyn Iterator<Item = RowId> + 'a> {
let mut offset = 0;
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() {
last_row_id = offset + row_id;
return Some(last_row_id);
}
offset = last_row_id;
loop {
current_it = multivalued_indexes.next();
if current_it.as_mut()?.next().is_some() {
break;
}
}
}))
}
#[cfg(test)]
mod tests {
use crate::RowId;
fn it<'a>(row_ids: &'a [RowId]) -> Box<dyn Iterator<Item = RowId> + 'a> {
Box::new(row_ids.iter().copied())
}
#[test]
fn test_stack() {
let columns = [
it(&[0u32, 0u32]),
it(&[0u32, 1u32, 1u32, 4u32]),
it(&[0u32, 3u32, 5u32]),
it(&[0u32, 4u32]),
]
.into_iter();
let start_offsets: Vec<RowId> = super::stack_multivalued_indexes(columns).collect();
assert_eq!(start_offsets, &[0, 0, 1, 1, 4, 7, 9, 13]);
}
}

125
columnar/src/column_index/mod.rs
View File

@@ -1,8 +1,3 @@
//! # `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;
@@ -11,16 +6,13 @@ mod serialize;
use std::ops::Range;
pub use merge::merge_column_index;
pub(crate) use multivalued_index::SerializableMultivalueIndex;
pub use optional_index::{OptionalIndex, Set};
pub use serialize::{
open_column_index, serialize_column_index, SerializableColumnIndex, SerializableOptionalIndex,
};
pub use serialize::{open_column_index, serialize_column_index, SerializableColumnIndex};
use crate::column_index::multivalued_index::MultiValueIndex;
use crate::{Cardinality, DocId, RowId};
#[derive(Clone, Debug)]
#[derive(Clone)]
pub enum ColumnIndex {
Empty {
num_docs: u32,
@@ -45,15 +37,11 @@ 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,
}
@@ -86,89 +74,24 @@ impl ColumnIndex {
}
}
/// 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> {
pub fn docid_range_to_rowids(&self, doc_id: Range<DocId>) -> Range<RowId> {
match self {
ColumnIndex::Empty { .. } => 0..0,
ColumnIndex::Full => doc_id_range,
ColumnIndex::Full => doc_id,
ColumnIndex::Optional(optional_index) => {
let row_start = optional_index.rank(doc_id_range.start);
let row_end = optional_index.rank(doc_id_range.end);
let row_start = optional_index.rank(doc_id.start);
let row_end = optional_index.rank(doc_id.end);
row_start..row_end
}
ColumnIndex::Multivalued(multivalued_index) => match multivalued_index {
MultiValueIndex::MultiValueIndexV1(index) => {
let row_start = index.start_index_column.get_val(doc_id_range.start);
let row_end = index.start_index_column.get_val(doc_id_range.end);
row_start..row_end
}
MultiValueIndex::MultiValueIndexV2(index) => {
// In this case we will use the optional_index select the next values
// that are valid. There are different cases to consider:
// Not exists below means does not exist in the optional
// index, because it has no values.
// * doc_id_range may cover a range of docids which are non existent
// => rank
// will give us the next document outside the range with a value. They both
// get the same rank and therefore return a zero range
//
// * doc_id_range.start and doc_id_range.end may not exist, but docids in
// between may have values
// => rank will give us the next document outside the range with a value.
//
// * doc_id_range.start may be not existent but doc_id_range.end may exist
// * doc_id_range.start may exist but doc_id_range.end may not exist
// * doc_id_range.start and doc_id_range.end may exist
// => rank on doc_id_range.end will give use the next value, which matches
// how the `start_index_column` works, so we get the value start of the next
// docid which we use to create the exclusive range.
//
let rank_start = index.optional_index.rank(doc_id_range.start);
let row_start = index.start_index_column.get_val(rank_start);
let rank_end = index.optional_index.rank(doc_id_range.end);
let row_end = index.start_index_column.get_val(rank_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);
row_start..row_end
}
},
let row_start = multivalued_index.start_index_column.get_val(start_docid);
let row_end = multivalued_index.start_index_column.get_val(end_docid);
row_start..row_end
}
}
}
@@ -190,21 +113,3 @@ 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
);
}
}

317
columnar/src/column_index/multivalued_index.rs
View File

@@ -3,98 +3,56 @@ use std::io::Write;
use std::ops::Range;
use std::sync::Arc;
use common::{CountingWriter, OwnedBytes};
use common::OwnedBytes;
use super::optional_index::{open_optional_index, serialize_optional_index};
use super::{OptionalIndex, SerializableOptionalIndex, Set};
use crate::column_values::{
load_u64_based_column_values, serialize_u64_based_column_values, CodecType, ColumnValues,
};
use crate::iterable::Iterable;
use crate::{DocId, RowId, Version};
pub struct SerializableMultivalueIndex<'a> {
pub doc_ids_with_values: SerializableOptionalIndex<'a>,
pub start_offsets: Box<dyn Iterable<u32> + 'a>,
}
use crate::{DocId, RowId};
pub fn serialize_multivalued_index(
multivalued_index: &SerializableMultivalueIndex,
multivalued_index: &dyn Iterable<RowId>,
output: &mut impl Write,
) -> io::Result<()> {
let SerializableMultivalueIndex {
doc_ids_with_values,
start_offsets,
} = multivalued_index;
let mut count_writer = CountingWriter::wrap(output);
let SerializableOptionalIndex {
non_null_row_ids,
num_rows,
} = doc_ids_with_values;
serialize_optional_index(&**non_null_row_ids, *num_rows, &mut count_writer)?;
let optional_len = count_writer.written_bytes() as u32;
let output = count_writer.finish();
serialize_u64_based_column_values(
&**start_offsets,
multivalued_index,
&[CodecType::Bitpacked, CodecType::Linear],
output,
)?;
output.write_all(&optional_len.to_le_bytes())?;
Ok(())
}
pub fn open_multivalued_index(
bytes: OwnedBytes,
format_version: Version,
) -> io::Result<MultiValueIndex> {
match format_version {
Version::V1 => {
let start_index_column: Arc<dyn ColumnValues<RowId>> =
load_u64_based_column_values(bytes)?;
Ok(MultiValueIndex::MultiValueIndexV1(MultiValueIndexV1 {
start_index_column,
}))
}
Version::V2 => {
let (body_bytes, optional_index_len) = bytes.rsplit(4);
let optional_index_len =
u32::from_le_bytes(optional_index_len.as_slice().try_into().unwrap());
let (optional_index_bytes, start_index_bytes) =
body_bytes.split(optional_index_len as usize);
let optional_index = open_optional_index(optional_index_bytes)?;
let start_index_column: Arc<dyn ColumnValues<RowId>> =
load_u64_based_column_values(start_index_bytes)?;
Ok(MultiValueIndex::MultiValueIndexV2(MultiValueIndexV2 {
optional_index,
start_index_column,
}))
}
}
pub fn open_multivalued_index(bytes: OwnedBytes) -> io::Result<MultiValueIndex> {
let start_index_column: Arc<dyn ColumnValues<RowId>> = load_u64_based_column_values(bytes)?;
Ok(MultiValueIndex { start_index_column })
}
#[derive(Clone)]
/// Index to resolve value range for given doc_id.
/// Starts at 0.
pub enum MultiValueIndex {
MultiValueIndexV1(MultiValueIndexV1),
MultiValueIndexV2(MultiValueIndexV2),
}
#[derive(Clone)]
/// Index to resolve value range for given doc_id.
/// Starts at 0.
pub struct MultiValueIndexV1 {
pub struct MultiValueIndex {
pub start_index_column: Arc<dyn crate::ColumnValues<RowId>>,
}
impl MultiValueIndexV1 {
impl From<Arc<dyn ColumnValues<RowId>>> for MultiValueIndex {
fn from(start_index_column: Arc<dyn ColumnValues<RowId>>) -> Self {
MultiValueIndex { start_index_column }
}
}
impl MultiValueIndex {
pub fn for_test(start_offsets: &[RowId]) -> MultiValueIndex {
let mut buffer = Vec::new();
serialize_multivalued_index(&start_offsets, &mut buffer).unwrap();
let bytes = OwnedBytes::new(buffer);
open_multivalued_index(bytes).unwrap()
}
/// Returns `[start, end)`, such that the values associated with
/// the given document are `start..end`.
#[inline]
pub(crate) fn range(&self, doc_id: DocId) -> Range<RowId> {
if doc_id >= self.num_docs() {
return 0..0;
}
let start = self.start_index_column.get_val(doc_id);
let end = self.start_index_column.get_val(doc_id + 1);
start..end
@@ -117,6 +75,7 @@ impl MultiValueIndexV1 {
///
/// TODO: Instead of a linear scan we can employ a exponential search into binary search to
/// match a docid to its value position.
#[allow(clippy::bool_to_int_with_if)]
pub(crate) fn select_batch_in_place(&self, docid_start: DocId, ranks: &mut Vec<u32>) {
if ranks.is_empty() {
return;
@@ -144,170 +103,14 @@ impl MultiValueIndexV1 {
}
}
#[derive(Clone)]
/// Index to resolve value range for given doc_id.
/// Starts at 0.
pub struct MultiValueIndexV2 {
pub optional_index: OptionalIndex,
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 {
let index = match self {
MultiValueIndex::MultiValueIndexV1(idx) => &idx.start_index_column,
MultiValueIndex::MultiValueIndexV2(idx) => &idx.start_index_column,
};
f.debug_struct("MultiValuedIndex")
.field("num_rows", &index.num_vals())
.finish_non_exhaustive()
}
}
impl MultiValueIndex {
pub fn for_test(start_offsets: &[RowId]) -> MultiValueIndex {
assert!(!start_offsets.is_empty());
assert_eq!(start_offsets[0], 0);
let mut doc_with_values = Vec::new();
let mut compact_start_offsets: Vec<u32> = vec![0];
for doc in 0..start_offsets.len() - 1 {
if start_offsets[doc] < start_offsets[doc + 1] {
doc_with_values.push(doc as RowId);
compact_start_offsets.push(start_offsets[doc + 1]);
}
}
let serializable_multivalued_index = SerializableMultivalueIndex {
doc_ids_with_values: SerializableOptionalIndex {
non_null_row_ids: Box::new(&doc_with_values[..]),
num_rows: start_offsets.len() as u32 - 1,
},
start_offsets: Box::new(&compact_start_offsets[..]),
};
let mut buffer = Vec::new();
serialize_multivalued_index(&serializable_multivalued_index, &mut buffer).unwrap();
let bytes = OwnedBytes::new(buffer);
open_multivalued_index(bytes, Version::V2).unwrap()
}
pub fn get_start_index_column(&self) -> &Arc<dyn crate::ColumnValues<RowId>> {
match self {
MultiValueIndex::MultiValueIndexV1(idx) => &idx.start_index_column,
MultiValueIndex::MultiValueIndexV2(idx) => &idx.start_index_column,
}
}
/// Returns `[start, end)` values range, such that the values associated with
/// the given document are `start..end`.
#[inline]
pub(crate) fn range(&self, doc_id: DocId) -> Range<RowId> {
match self {
MultiValueIndex::MultiValueIndexV1(idx) => idx.range(doc_id),
MultiValueIndex::MultiValueIndexV2(idx) => idx.range(doc_id),
}
}
/// Returns the number of documents in the index.
#[inline]
pub fn num_docs(&self) -> u32 {
match self {
MultiValueIndex::MultiValueIndexV1(idx) => idx.start_index_column.num_vals() - 1,
MultiValueIndex::MultiValueIndexV2(idx) => idx.optional_index.num_docs(),
}
}
/// Converts a list of ranks (row ids of values) in a 1:n index to the corresponding list of
/// docids. Positions are converted inplace to docids.
///
/// Since there is no index for value pos -> docid, but docid -> value pos range, we scan the
/// index.
///
/// Correctness: positions needs to be sorted. idx_reader needs to contain monotonically
/// increasing positions.
///
/// TODO: Instead of a linear scan we can employ a exponential search into binary search to
/// match a docid to its value position.
pub(crate) fn select_batch_in_place(&self, docid_start: DocId, ranks: &mut Vec<u32>) {
match self {
MultiValueIndex::MultiValueIndexV1(idx) => {
idx.select_batch_in_place(docid_start, ranks)
}
MultiValueIndex::MultiValueIndexV2(idx) => {
idx.select_batch_in_place(docid_start, ranks)
}
}
}
}
impl MultiValueIndexV2 {
/// Returns `[start, end)`, such that the values associated with
/// the given document are `start..end`.
#[inline]
pub(crate) fn range(&self, doc_id: DocId) -> Range<RowId> {
let Some(rank) = self.optional_index.rank_if_exists(doc_id) else {
return 0..0;
};
let start = self.start_index_column.get_val(rank);
let end = self.start_index_column.get_val(rank + 1);
start..end
}
/// Returns the number of documents in the index.
#[inline]
pub fn num_docs(&self) -> u32 {
self.optional_index.num_docs()
}
/// Converts a list of ranks (row ids of values) in a 1:n index to the corresponding list of
/// docids. Positions are converted inplace to docids.
///
/// Since there is no index for value pos -> docid, but docid -> value pos range, we scan the
/// index.
///
/// Correctness: positions needs to be sorted. idx_reader needs to contain monotonically
/// increasing positions.
///
/// TODO: Instead of a linear scan we can employ a exponential search into binary search to
/// match a docid to its value position.
pub(crate) fn select_batch_in_place(&self, docid_start: DocId, ranks: &mut Vec<u32>) {
if ranks.is_empty() {
return;
}
let mut cur_pos_in_idx = self.optional_index.rank(docid_start);
let mut last_doc = None;
assert!(cur_pos_in_idx <= ranks[0]);
let mut write_doc_pos = 0;
for i in 0..ranks.len() {
let pos = ranks[i];
loop {
let end = self.start_index_column.get_val(cur_pos_in_idx + 1);
if end > pos {
ranks[write_doc_pos] = cur_pos_in_idx;
write_doc_pos += if last_doc == Some(cur_pos_in_idx) {
0
} else {
1
};
last_doc = Some(cur_pos_in_idx);
break;
}
cur_pos_in_idx += 1;
}
}
ranks.truncate(write_doc_pos);
for rank in ranks.iter_mut() {
*rank = self.optional_index.select(*rank);
}
}
}
#[cfg(test)]
mod tests {
use std::ops::Range;
use std::sync::Arc;
use super::MultiValueIndex;
use crate::{ColumnarReader, DynamicColumn};
use crate::column_values::IterColumn;
use crate::{ColumnValues, RowId};
fn index_to_pos_helper(
index: &MultiValueIndex,
@@ -321,12 +124,13 @@ mod tests {
#[test]
fn test_positions_to_docid() {
let index = MultiValueIndex::for_test(&[0, 10, 12, 15, 22, 23]);
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);
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]);
assert_eq!(index_to_pos_helper(&index, 1..5, positions), vec![1, 3, 4]);
assert_eq!(index_to_pos_helper(&index, 0..5, &[9]), vec![0]);
assert_eq!(index_to_pos_helper(&index, 1..5, &[10]), vec![1]);
assert_eq!(index_to_pos_helper(&index, 1..5, &[11]), vec![1]);
@@ -334,67 +138,4 @@ mod tests {
assert_eq!(index_to_pos_helper(&index, 2..5, &[12, 14]), vec![2]);
assert_eq!(index_to_pos_helper(&index, 2..5, &[12, 14, 15]), vec![2, 3]);
}
#[test]
fn test_range_to_rowids() {
use crate::ColumnarWriter;
let mut columnar_writer = ColumnarWriter::default();
// This column gets coerced to u64
columnar_writer.record_numerical(1, "full", u64::MAX);
columnar_writer.record_numerical(1, "full", u64::MAX);
columnar_writer.record_numerical(5, "full", u64::MAX);
columnar_writer.record_numerical(5, "full", u64::MAX);
let mut wrt: Vec<u8> = Vec::new();
columnar_writer.serialize(7, &mut wrt).unwrap();
let reader = ColumnarReader::open(wrt).unwrap();
// Open the column as u64
let column = reader.read_columns("full").unwrap()[0]
.open()
.unwrap()
.coerce_numerical(crate::NumericalType::U64)
.unwrap();
let DynamicColumn::U64(column) = column else {
panic!();
};
let row_id_range = column.index.docid_range_to_rowids(1..2);
assert_eq!(row_id_range, 0..2);
let row_id_range = column.index.docid_range_to_rowids(0..2);
assert_eq!(row_id_range, 0..2);
let row_id_range = column.index.docid_range_to_rowids(0..4);
assert_eq!(row_id_range, 0..2);
let row_id_range = column.index.docid_range_to_rowids(3..4);
assert_eq!(row_id_range, 2..2);
let row_id_range = column.index.docid_range_to_rowids(1..6);
assert_eq!(row_id_range, 0..4);
let row_id_range = column.index.docid_range_to_rowids(3..6);
assert_eq!(row_id_range, 2..4);
let row_id_range = column.index.docid_range_to_rowids(0..6);
assert_eq!(row_id_range, 0..4);
let row_id_range = column.index.docid_range_to_rowids(0..6);
assert_eq!(row_id_range, 0..4);
let check = |range, expected| {
let full_range = 0..=u64::MAX;
let mut docids = Vec::new();
column.get_docids_for_value_range(full_range, range, &mut docids);
assert_eq!(docids, expected);
};
// check(0..1, vec![]);
// check(0..2, vec![1]);
check(1..2, vec![1]);
}
}

33
columnar/src/column_index/optional_index/mod.rs
View File

@@ -21,6 +21,8 @@ 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,
@@ -86,21 +88,6 @@ pub struct OptionalIndex {
block_metas: Arc<[BlockMeta]>,
}
impl<'a> Iterable<u32> for &'a OptionalIndex {
fn boxed_iter(&self) -> Box<dyn Iterator<Item = u32> + '_> {
Box::new(self.iter_rows())
}
}
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
@@ -113,8 +100,8 @@ struct RowAddr {
#[inline(always)]
fn row_addr_from_row_id(row_id: RowId) -> RowAddr {
RowAddr {
block_id: (row_id / ELEMENTS_PER_BLOCK) as u16,
in_block_row_id: (row_id % ELEMENTS_PER_BLOCK) as u16,
block_id: (row_id / BLOCK_SIZE) as u16,
in_block_row_id: (row_id % BLOCK_SIZE) as u16,
}
}
@@ -189,20 +176,14 @@ 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,
} = row_addr_from_row_id(doc_id);
let block_meta = self.block_metas[block_id as usize];
let block = self.block(block_meta);
let block_offset_row_id = match block {
Block::Dense(dense_block) => dense_block.rank(in_block_row_id),
Block::Sparse(sparse_block) => sparse_block.rank(in_block_row_id),
@@ -210,15 +191,13 @@ 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.get(block_id as usize)?;
let block_meta = self.block_metas[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),
@@ -503,7 +482,7 @@ fn deserialize_optional_index_block_metadatas(
non_null_rows_before_block += num_non_null_rows;
}
block_metas.resize(
((num_rows + ELEMENTS_PER_BLOCK - 1) / ELEMENTS_PER_BLOCK) as usize,
((num_rows + BLOCK_SIZE - 1) / BLOCK_SIZE) as usize,
BlockMeta {
non_null_rows_before_block,
start_byte_offset,

8
columnar/src/column_index/optional_index/set.rs
View File

@@ -28,11 +28,10 @@ pub trait Set<T> {
/// Returns true if the elements is contained in the Set
fn contains(&self, el: T) -> bool;
/// Returns the element's rank (its position in the set).
/// If the set does not contain the element, it will return the next existing elements rank.
/// Returns the number of rows in the set that are < `el`
fn rank(&self, el: T) -> T;
/// If the set contains `el`, returns the element's rank (its position in the set).
/// If the set contains `el` returns the element rank.
/// If the set does not contain the element, it returns `None`.
fn rank_if_exists(&self, el: T) -> Option<T>;
@@ -40,8 +39,7 @@ pub trait Set<T> {
///
/// # Panics
///
/// May panic if rank is greater or equal to the number of
/// elements in the Set.
/// May panic if rank is greater than 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

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

42
columnar/src/column_index/optional_index/set_block/tests.rs
View File

@@ -22,8 +22,8 @@ fn test_set_helper<C: SetCodec<Item = u16>>(vals: &[u16]) -> usize {
vals.iter().cloned().take_while(|v| *v < val).count() as u16
);
}
for (rank, val) in vals.iter().enumerate() {
assert_eq!(tested_set.select(rank as u16), *val);
for rank in 0..vals.len() {
assert_eq!(tested_set.select(rank as u16), vals[rank]);
}
buffer.len()
}
@@ -107,41 +107,3 @@ fn test_simple_translate_codec_idx_to_original_idx_dense() {
assert_eq!(i, select_cursor.select(i));
}
}
#[test]
fn test_simple_translate_idx_to_value_idx_dense() {
let mut buffer = Vec::new();
DenseBlockCodec::serialize([1, 10].iter().copied(), &mut buffer).unwrap();
let tested_set = DenseBlockCodec::open(buffer.as_slice());
assert!(tested_set.contains(1));
assert!(!tested_set.contains(2));
assert_eq!(tested_set.rank(0), 0);
assert_eq!(tested_set.rank(1), 0);
for rank in 2..10 {
// ranks that don't exist select the next highest one
assert_eq!(tested_set.rank_if_exists(rank), None);
assert_eq!(tested_set.rank(rank), 1);
}
assert_eq!(tested_set.rank(10), 1);
}
#[test]
fn test_simple_translate_idx_to_value_idx_sparse() {
let mut buffer = Vec::new();
SparseBlockCodec::serialize([1, 10].iter().copied(), &mut buffer).unwrap();
let tested_set = SparseBlockCodec::open(buffer.as_slice());
assert!(tested_set.contains(1));
assert!(!tested_set.contains(2));
assert_eq!(tested_set.rank(0), 0);
assert_eq!(tested_set.select(tested_set.rank(0)), 1);
assert_eq!(tested_set.rank(1), 0);
assert_eq!(tested_set.select(tested_set.rank(1)), 1);
for rank in 2..10 {
// ranks that don't exist select the next highest one
assert_eq!(tested_set.rank_if_exists(rank), None);
assert_eq!(tested_set.rank(rank), 1);
assert_eq!(tested_set.select(tested_set.rank(rank)), 10);
}
assert_eq!(tested_set.rank(10), 1);
assert_eq!(tested_set.select(tested_set.rank(10)), 10);
}

41
columnar/src/column_index/optional_index/tests.rs
View File

@@ -1,29 +1,8 @@
use proptest::prelude::*;
use proptest::prelude::{any, prop, *};
use proptest::strategy::Strategy;
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, &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() {
@@ -56,7 +35,7 @@ proptest! {
#[test]
fn test_with_random_sets_simple() {
let vals = 10..ELEMENTS_PER_BLOCK * 2;
let vals = 10..BLOCK_SIZE * 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();
@@ -192,7 +171,7 @@ fn test_optional_index_rank() {
test_optional_index_rank_aux(&[0u32, 1u32]);
let mut block = Vec::new();
block.push(3u32);
block.extend((0..ELEMENTS_PER_BLOCK).map(|i| i + ELEMENTS_PER_BLOCK + 1));
block.extend((0..BLOCK_SIZE).map(|i| i + BLOCK_SIZE + 1));
test_optional_index_rank_aux(&block);
}
@@ -206,8 +185,8 @@ 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..ELEMENTS_PER_BLOCK).map(|i| i + ELEMENTS_PER_BLOCK + 1));
test_optional_index_iter_aux(&block, 3 * ELEMENTS_PER_BLOCK);
block.extend((0..BLOCK_SIZE).map(|i| i + BLOCK_SIZE + 1));
test_optional_index_iter_aux(&block, 3 * BLOCK_SIZE);
}
#[test]
@@ -236,12 +215,12 @@ 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();
open_optional_index(OwnedBytes::new(out)).unwrap()
let codec = open_optional_index(OwnedBytes::new(out)).unwrap();
codec
}
fn random_range_iterator(
@@ -263,7 +242,7 @@ mod bench {
}
fn n_percent_step_iterator(percent: f32, num_values: u32) -> impl Iterator<Item = u32> {
let ratio = percent / 100.0;
let ratio = percent as f32 / 100.0;
let step_size = (1f32 / ratio) as u32;
let deviation = step_size - 1;
random_range_iterator(0, num_values, step_size, deviation)

45
columnar/src/column_index/serialize.rs
View File

@@ -3,45 +3,33 @@ use std::io::Write;
use common::{CountingWriter, OwnedBytes};
use super::multivalued_index::SerializableMultivalueIndex;
use super::OptionalIndex;
use crate::column_index::multivalued_index::serialize_multivalued_index;
use crate::column_index::optional_index::serialize_optional_index;
use crate::column_index::ColumnIndex;
use crate::iterable::Iterable;
use crate::{Cardinality, RowId, Version};
pub struct SerializableOptionalIndex<'a> {
pub non_null_row_ids: Box<dyn Iterable<RowId> + 'a>,
pub num_rows: RowId,
}
impl<'a> From<&'a OptionalIndex> for SerializableOptionalIndex<'a> {
fn from(optional_index: &'a OptionalIndex) -> Self {
SerializableOptionalIndex {
non_null_row_ids: Box::new(optional_index),
num_rows: optional_index.num_docs(),
}
}
}
use crate::{Cardinality, RowId};
pub enum SerializableColumnIndex<'a> {
Full,
Optional(SerializableOptionalIndex<'a>),
Multivalued(SerializableMultivalueIndex<'a>),
Optional {
non_null_row_ids: Box<dyn Iterable<RowId> + 'a>,
num_rows: RowId,
},
// TODO remove the Arc<dyn> apart from serialization this is not
// dynamic at all.
Multivalued(Box<dyn Iterable<RowId> + 'a>),
}
impl<'a> SerializableColumnIndex<'a> {
pub fn get_cardinality(&self) -> Cardinality {
match self {
SerializableColumnIndex::Full => Cardinality::Full,
SerializableColumnIndex::Optional(_) => Cardinality::Optional,
SerializableColumnIndex::Optional { .. } => Cardinality::Optional,
SerializableColumnIndex::Multivalued(_) => Cardinality::Multivalued,
}
}
}
/// Serialize a column index.
pub fn serialize_column_index(
column_index: SerializableColumnIndex,
output: &mut impl Write,
@@ -51,23 +39,19 @@ pub fn serialize_column_index(
output.write_all(&[cardinality])?;
match column_index {
SerializableColumnIndex::Full => {}
SerializableColumnIndex::Optional(SerializableOptionalIndex {
SerializableColumnIndex::Optional {
non_null_row_ids,
num_rows,
}) => serialize_optional_index(non_null_row_ids.as_ref(), num_rows, &mut output)?,
} => serialize_optional_index(non_null_row_ids.as_ref(), num_rows, &mut output)?,
SerializableColumnIndex::Multivalued(multivalued_index) => {
serialize_multivalued_index(&multivalued_index, &mut output)?
serialize_multivalued_index(&*multivalued_index, &mut output)?
}
}
let column_index_num_bytes = output.written_bytes() as u32;
Ok(column_index_num_bytes)
}
/// Open a serialized column index.
pub fn open_column_index(
mut bytes: OwnedBytes,
format_version: Version,
) -> io::Result<ColumnIndex> {
pub fn open_column_index(mut bytes: OwnedBytes) -> io::Result<ColumnIndex> {
if bytes.is_empty() {
return Err(io::Error::new(
io::ErrorKind::UnexpectedEof,
@@ -84,8 +68,7 @@ pub fn open_column_index(
Ok(ColumnIndex::Optional(optional_index))
}
Cardinality::Multivalued => {
let multivalue_index =
super::multivalued_index::open_multivalued_index(bytes, format_version)?;
let multivalue_index = super::multivalued_index::open_multivalued_index(bytes)?;
Ok(ColumnIndex::Multivalued(multivalue_index))
}
}

6
columnar/src/column_values/bench.rs
View File

@@ -34,7 +34,6 @@ fn compute_stats(vals: impl Iterator<Item = u64>) -> ColumnStats {
fn value_iter() -> impl Iterator<Item = u64> {
0..20_000
}
fn get_reader_for_bench<Codec: ColumnCodec>(data: &[u64]) -> Codec::ColumnValues {
let mut bytes = Vec::new();
let stats = compute_stats(data.iter().cloned());
@@ -42,13 +41,10 @@ fn get_reader_for_bench<Codec: ColumnCodec>(data: &[u64]) -> Codec::ColumnValues
for val in data {
codec_serializer.collect(*val);
}
codec_serializer
.serialize(&stats, Box::new(data.iter().copied()).as_mut(), &mut bytes)
.unwrap();
codec_serializer.serialize(&stats, Box::new(data.iter().copied()).as_mut(), &mut bytes);
Codec::load(OwnedBytes::new(bytes)).unwrap()
}
fn bench_get<Codec: ColumnCodec>(b: &mut Bencher, data: &[u64]) {
let col = get_reader_for_bench::<Codec>(data);
b.iter(|| {

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 [ColumnIndex],
pub(crate) column_indexes: &'a [Option<ColumnIndex>],
pub(crate) column_values: &'a [Option<Arc<dyn ColumnValues<T>>>],
pub(crate) merge_row_order: &'a MergeRowOrder,
}
impl<'a, T: Copy + PartialOrd + Debug + 'static> Iterable<T> for MergedColumnValues<'a, T> {
impl<'a, T: Copy + PartialOrd + Debug> Iterable<T> for MergedColumnValues<'a, T> {
fn boxed_iter(&self) -> Box<dyn Iterator<Item = T> + '_> {
match self.merge_row_order {
MergeRowOrder::Stack(_) => Box::new(
@@ -23,7 +23,8 @@ impl<'a, T: Copy + PartialOrd + Debug + 'static> Iterable<T> for MergedColumnVal
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];
let column_index =
self.column_indexes[row_addr.segment_ord as usize].as_ref()?;
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 [`crate::Column`].
//! - Columnar storage of data for tantivy [`Column`].
//! - Encode data in different codecs.
//! - Monotonically map values to u64/u128
@@ -10,7 +10,6 @@ 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;
@@ -26,10 +25,7 @@ mod monotonic_column;
pub(crate) use merge::MergedColumnValues;
pub use stats::ColumnStats;
pub use u128_based::{
open_u128_as_compact_u64, open_u128_mapped, serialize_column_values_u128,
CompactSpaceU64Accessor,
};
pub use u128_based::{open_u128_mapped, serialize_column_values_u128};
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,
@@ -45,7 +41,7 @@ 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 + DowncastSync {
pub trait ColumnValues<T: PartialOrd = u64>: Send + Sync {
/// Return the value associated with the given idx.
///
/// This accessor should return as fast as possible.
@@ -62,50 +58,10 @@ pub trait ColumnValues<T: PartialOrd = u64>: Send + Sync + DowncastSync {
/// # Panics
///
/// May panic if `idx` is greater than the column length.
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));
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);
}
}
@@ -127,6 +83,7 @@ pub trait ColumnValues<T: PartialOrd = u64>: Send + Sync + DowncastSync {
/// 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>,
@@ -134,7 +91,7 @@ pub trait ColumnValues<T: PartialOrd = u64>: Send + Sync + DowncastSync {
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 {
for idx in row_id_range.start..row_id_range.end {
let val = self.get_val(idx);
if value_range.contains(&val) {
row_id_hits.push(idx);
@@ -142,26 +99,20 @@ pub trait ColumnValues<T: PartialOrd = u64>: Send + Sync + DowncastSync {
}
}
/// Returns a lower bound for this column of values.
/// Returns the minimum value for this fast field.
///
/// 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()
/// 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()`.
fn min_value(&self) -> T;
/// Returns an upper bound for this column of values.
/// Returns the maximum value for this fast field.
///
/// 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()
/// 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()`.
fn max_value(&self) -> T;
/// The number of values in the column.
@@ -172,40 +123,13 @@ pub trait ColumnValues<T: PartialOrd = u64>: Send + Sync + DowncastSync {
Box::new((0..self.num_vals()).map(|idx| self.get_val(idx)))
}
}
downcast_rs::impl_downcast!(sync ColumnValues<T> where T: PartialOrd);
/// 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>> {
impl<T: Copy + PartialOrd + Debug> 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()
@@ -243,5 +167,54 @@ impl<T: Copy + PartialOrd + Debug + 'static> ColumnValues<T> for Arc<dyn ColumnV
}
}
/// 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> + 'static,
T: StrictlyMonotonicFn<Input, Output> + Send + Sync + 'static,
Input: PartialOrd + Debug + Send + Sync + Clone + 'static,
Output: PartialOrd + Debug + Send + Sync + Clone + 'static,
C: ColumnValues<Input>,
T: StrictlyMonotonicFn<Input, Output> + Send + Sync,
Input: PartialOrd + Debug + Send + Sync + Clone,
Output: PartialOrd + Debug + Send + Sync + Clone,
{
MonotonicMappingColumn {
from_column,
@@ -45,12 +45,12 @@ where
impl<C, T, Input, Output> ColumnValues<Output> for MonotonicMappingColumn<C, T, Input>
where
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,
C: ColumnValues<Input>,
T: StrictlyMonotonicFn<Input, Output> + Send + Sync,
Input: PartialOrd + Send + Debug + Sync + Clone,
Output: PartialOrd + Send + Debug + Sync + Clone,
{
#[inline(always)]
#[inline]
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_nanos())
common::i64_to_u64(self.into_timestamp_micros())
}
#[inline(always)]
fn from_u64(val: u64) -> Self {
DateTime::from_timestamp_nanos(common::u64_to_i64(val))
DateTime::from_timestamp_micros(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.cmp(other))
Some(self.blank_size().cmp(&other.blank_size()))
}
}

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().copied().tuple_windows() {
for (first, second) in values_sorted.iter().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();
}
// We start by space that's limited to min_value..=max_value
// 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);
// 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);
// +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);
@@ -84,7 +84,6 @@ 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
@@ -94,7 +93,6 @@ 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;
}
@@ -102,16 +100,7 @@ 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;
// 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 {
if 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
//
@@ -126,8 +115,6 @@ 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
@@ -212,7 +199,7 @@ impl CompactSpaceBuilder {
covered_space.push(0..=0); // empty data case
};
let mut compact_start: u32 = 1; // 0 is reserved for `null`
let mut compact_start: u64 = 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 {
@@ -231,7 +218,6 @@ 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() {
@@ -242,11 +228,4 @@ 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::{BitPacker, BitUnpacker};
use tantivy_bitpacker::{self, 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: u32,
compact_start: u64,
}
impl RangeMapping {
fn range_length(&self) -> u32 {
(self.value_range.end() - self.value_range.start()) as u32 + 1
fn range_length(&self) -> u64 {
(self.value_range.end() - self.value_range.start()) as u64 + 1
}
// The last value of the compact space in this range
fn compact_end(&self) -> u32 {
fn compact_end(&self) -> u64 {
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 = 1u32; // 0 is reserved for `null`
let mut compact_start = 1u64; // 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<u32, usize> {
fn u128_to_compact(&self, value: u128) -> Result<u64, usize> {
self.ranges_mapping
.binary_search_by(|probe| {
let value_range: &RangeInclusive<u128> = &probe.value_range;
let value_range = &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: u32 = (value - range_mapping.value_range.start()) as u32;
let pos_in_range = (value - range_mapping.value_range.start()) as u64;
range_mapping.compact_start + pos_in_range
})
}
/// Unpacks a value from compact space u32 to u128 space
fn compact_to_u128(&self, compact: u32) -> u128 {
/// Unpacks a value from compact space u64 to u128 space
fn compact_to_u128(&self, compact: u64) -> 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
.unwrap_or_else(|e| e - 1);
.map_or_else(|e| e - 1, |v| v);
let range_mapping = &self.ranges_mapping[pos];
let diff = compact - range_mapping.compact_start;
@@ -178,15 +178,11 @@ 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();
@@ -197,12 +193,13 @@ 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 u32
amplitude_compact_space as u64
);
CompactSpaceCompressor {
params: IPCodecParams {
@@ -243,7 +240,7 @@ impl CompactSpaceCompressor {
"Could not convert value to compact_space. This is a bug.",
)
})?;
bitpacker.write(compact as u64, self.params.num_bits, write)?;
bitpacker.write(compact, self.params.num_bits, write)?;
}
bitpacker.close(write)?;
self.write_footer(write)?;
@@ -292,63 +289,6 @@ 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 {
@@ -374,6 +314,48 @@ 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>,
@@ -413,42 +395,44 @@ impl ColumnValues<u128> for CompactSpaceDecompressor {
range_mapping.compact_end()
});
let value_range = compact_from..=compact_to;
self.get_positions_for_compact_value_range(value_range, position_range, positions);
}
}
let range = compact_from..=compact_to;
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 scan_num_docs = position_range.end - position_range.start;
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 step_size = 4;
let cutoff = position_range.start + scan_num_docs - scan_num_docs % step_size;
Ok(decompressor)
}
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);
}
/// 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)
// handle rest
for idx in cutoff..position_range.end {
push_if_in_range(idx, get_val(idx));
}
}
#[inline]
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)
fn iter_compact(&self) -> impl Iterator<Item = u64> + '_ {
(0..self.params.num_vals).map(move |idx| self.params.bit_unpacker.get(idx, &self.data))
}
#[inline]
@@ -459,14 +443,9 @@ 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.get_compact(idx);
let compact = self.params.bit_unpacker.get(idx, &self.data);
self.compact_to_u128(compact)
}
@@ -477,20 +456,6 @@ 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)]
@@ -504,12 +469,12 @@ mod tests {
#[test]
fn compact_space_test() {
let ips: BTreeSet<u128> = [
let ips = &[
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());
@@ -532,8 +497,8 @@ 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);
}
}
@@ -559,7 +524,7 @@ mod tests {
.map(|pos| pos as u32)
.collect::<Vec<_>>();
let mut positions = Vec::new();
decompressor.get_row_ids_for_value_range(
decompressor.get_positions_for_value_range(
range,
0..decompressor.num_vals(),
&mut positions,
@@ -604,7 +569,7 @@ mod tests {
let val = *val;
let pos = pos as u32;
let mut positions = Vec::new();
decomp.get_row_ids_for_value_range(val..=val, pos..pos + 1, &mut positions);
decomp.get_positions_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

@@ -6,9 +6,7 @@ use std::sync::Arc;
mod compact_space;
use common::{BinarySerializable, OwnedBytes, VInt};
pub use compact_space::{
CompactSpaceCompressor, CompactSpaceDecompressor, CompactSpaceU64Accessor,
};
use compact_space::{CompactSpaceCompressor, CompactSpaceDecompressor};
use crate::column_values::monotonic_map_column;
use crate::column_values::monotonic_mapping::{
@@ -110,23 +108,6 @@ 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

@@ -1,6 +1,4 @@
use std::io::{self, Write};
use std::num::NonZeroU64;
use std::ops::{Range, RangeInclusive};
use common::{BinarySerializable, OwnedBytes};
use fastdivide::DividerU64;
@@ -18,51 +16,12 @@ 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
@@ -75,26 +34,6 @@ 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

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

@@ -173,9 +173,7 @@ impl LinearCodecEstimator {
fn collect_before_line_estimation(&mut self, value: u64) {
self.block.push(value);
if self.block.len() == LINE_ESTIMATION_BLOCK_LEN {
let column = VecColumn::from(std::mem::take(&mut self.block));
let line = Line::train(&column);
self.block = column.into();
let line = Line::train(&VecColumn::from(&self.block));
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 nanoseconds).
// seconds, only to be converted in microseconds).
increment_gcd_opt: Option<(NonZeroU64, DividerU64)>,
first_value_opt: Option<u64>,
}

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

@@ -1,5 +1,6 @@
use proptest::prelude::*;
use proptest::{prop_oneof, proptest};
use proptest::strategy::Strategy;
use proptest::{num, prop_oneof, proptest};
#[test]
fn test_serialize_and_load_simple() {
@@ -98,28 +99,14 @@ 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 `Vec<T>`.
pub struct VecColumn<T = u64> {
pub(crate) values: Vec<T>,
/// VecColumn provides `Column` over a slice.
pub struct VecColumn<'a, T = u64> {
pub(crate) values: &'a [T],
pub(crate) min_value: T,
pub(crate) max_value: T,
}
impl<T: Copy + PartialOrd + Send + Sync + Debug + 'static> ColumnValues<T> for VecColumn<T> {
impl<'a, T: Copy + PartialOrd + Send + Sync + Debug> ColumnValues<T> for VecColumn<'a, T> {
fn get_val(&self, position: u32) -> T {
self.values[position as usize]
}
@@ -37,8 +37,11 @@ impl<T: Copy + PartialOrd + Send + Sync + Debug + 'static> ColumnValues<T> for V
}
}
impl<T: Copy + PartialOrd + Default> From<Vec<T>> for VecColumn<T> {
fn from(values: Vec<T>) -> Self {
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();
let (min_value, max_value) = minmax(values.iter().copied()).unwrap_or_default();
Self {
values,
@@ -47,8 +50,3 @@ impl<T: Copy + PartialOrd + Default> From<Vec<T>> for VecColumn<T> {
}
}
}
impl From<VecColumn> for Vec<u64> {
fn from(column: VecColumn) -> Self {
column.values
}
}

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

@@ -1,4 +1,3 @@
use std::fmt;
use std::fmt::Debug;
use std::net::Ipv6Addr;
@@ -22,22 +21,6 @@ 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,
@@ -54,9 +37,6 @@ 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)

25
columnar/src/columnar/format_version.rs
View File

@@ -1,6 +1,3 @@
use core::fmt;
use std::fmt::{Display, Formatter};
use crate::InvalidData;
pub const VERSION_FOOTER_NUM_BYTES: usize = MAGIC_BYTES.len() + std::mem::size_of::<u32>();
@@ -11,7 +8,7 @@ const MAGIC_BYTES: [u8; 4] = [2, 113, 119, 66];
pub fn footer() -> [u8; VERSION_FOOTER_NUM_BYTES] {
let mut footer_bytes = [0u8; VERSION_FOOTER_NUM_BYTES];
footer_bytes[0..4].copy_from_slice(&CURRENT_VERSION.to_bytes());
footer_bytes[0..4].copy_from_slice(&Version::V1.to_bytes());
footer_bytes[4..8].copy_from_slice(&MAGIC_BYTES[..]);
footer_bytes
}
@@ -23,22 +20,10 @@ pub fn parse_footer(footer_bytes: [u8; VERSION_FOOTER_NUM_BYTES]) -> Result<Vers
Version::try_from_bytes(footer_bytes[0..4].try_into().unwrap())
}
pub const CURRENT_VERSION: Version = Version::V2;
#[derive(Debug, Copy, Clone, Eq, PartialEq)]
#[repr(u32)]
pub enum Version {
V1 = 1u32,
V2 = 2u32,
}
impl Display for Version {
fn fmt(&self, f: &mut Formatter) -> fmt::Result {
match self {
Version::V1 => write!(f, "v1"),
Version::V2 => write!(f, "v2"),
}
}
}
impl Version {
@@ -50,7 +35,6 @@ impl Version {
let code = u32::from_le_bytes(bytes);
match code {
1u32 => Ok(Version::V1),
2u32 => Ok(Version::V2),
_ => Err(InvalidData),
}
}
@@ -63,9 +47,9 @@ mod tests {
use super::*;
#[test]
fn test_footer_deserialization() {
fn test_footer_dserialization() {
let parsed_version: Version = parse_footer(footer()).unwrap();
assert_eq!(Version::V2, parsed_version);
assert_eq!(Version::V1, parsed_version);
}
#[test]
@@ -79,10 +63,11 @@ mod tests {
for &i in &version_to_tests {
let version_res = Version::try_from_bytes(i.to_le_bytes());
if let Ok(version) = version_res {
assert_eq!(version, Version::V1);
assert_eq!(version.to_bytes(), i.to_le_bytes());
valid_versions.insert(i);
}
}
assert_eq!(valid_versions.len(), 2);
assert_eq!(valid_versions.len(), 1);
}
}

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, Streamer, TermOrdinal, VoidSSTable};
use sstable::{SSTable, TermOrdinal};
use super::term_merger::TermMerger;
use crate::column::serialize_column_mappable_to_u64;
@@ -56,19 +56,17 @@ impl<'a> RemappedTermOrdinalsValues<'a> {
.bytes_columns
.iter()
.enumerate()
.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])
.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])
})
});
// TODO see if we can better decompose the mapping / and the stacking
Box::new(iter)
}
@@ -126,20 +124,16 @@ fn serialize_merged_dict(
let mut term_ord_mapping = TermOrdinalMapping::default();
let mut field_term_streams = Vec::new();
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());
}
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);
}
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

@@ -11,17 +11,6 @@ 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;
@@ -52,8 +41,8 @@ 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),

280
columnar/src/columnar/merge/mod.rs
View File

@@ -2,7 +2,7 @@ mod merge_dict_column;
mod merge_mapping;
mod term_merger;
use std::collections::{BTreeMap, HashSet};
use std::collections::{BTreeMap, HashMap, HashSet};
use std::io;
use std::net::Ipv6Addr;
use std::sync::Arc;
@@ -17,8 +17,7 @@ use crate::columnar::writer::CompatibleNumericalTypes;
use crate::columnar::ColumnarReader;
use crate::dynamic_column::DynamicColumn;
use crate::{
BytesColumn, Column, ColumnIndex, ColumnType, ColumnValues, DynamicColumnHandle, NumericalType,
NumericalValue,
BytesColumn, Column, ColumnIndex, ColumnType, ColumnValues, NumericalType, NumericalValue,
};
/// Column types are grouped into different categories.
@@ -28,16 +27,14 @@ use crate::{
/// In practise, today, only Numerical colummns are coerced into one type today.
///
/// See also [README.md].
///
/// 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,
Bytes,
Str,
#[derive(Copy, Clone, Eq, PartialEq, Hash, Debug)]
enum ColumnTypeCategory {
Bool,
IpAddr,
Str,
Numerical,
DateTime,
Bytes,
IpAddr,
}
impl From<ColumnType> for ColumnTypeCategory {
@@ -81,37 +78,20 @@ 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.start_serialize_column(column_name.as_bytes(), column_type);
serializer.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())?;
Ok(())
}
@@ -128,7 +108,6 @@ 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,
@@ -139,19 +118,17 @@ fn merge_column(
| ColumnType::F64
| ColumnType::DateTime
| ColumnType::Bool => {
let mut column_indexes: Vec<ColumnIndex> = Vec::with_capacity(columns.len());
let mut column_indexes: Vec<Option<ColumnIndex>> = Vec::with_capacity(columns.len());
let mut column_values: Vec<Option<Arc<dyn ColumnValues>>> =
Vec::with_capacity(columns.len());
for (i, dynamic_column_opt) in columns.into_iter().enumerate() {
if let Some(Column { index: idx, values }) =
for dynamic_column_opt in columns {
if let Some(Column { idx, values }) =
dynamic_column_opt.and_then(dynamic_column_to_u64_monotonic)
{
column_indexes.push(idx);
column_indexes.push(Some(idx));
column_values.push(Some(values));
} else {
column_indexes.push(ColumnIndex::Empty {
num_docs: num_docs_per_column[i],
});
column_indexes.push(None);
column_values.push(None);
}
}
@@ -165,19 +142,15 @@ fn merge_column(
serialize_column_mappable_to_u64(merged_column_index, &merge_column_values, wrt)?;
}
ColumnType::IpAddr => {
let mut column_indexes: Vec<ColumnIndex> = Vec::with_capacity(columns.len());
let mut column_indexes: Vec<Option<ColumnIndex>> = Vec::with_capacity(columns.len());
let mut column_values: Vec<Option<Arc<dyn ColumnValues<Ipv6Addr>>>> =
Vec::with_capacity(columns.len());
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);
for dynamic_column_opt in columns {
if let Some(DynamicColumn::IpAddr(Column { idx, values })) = dynamic_column_opt {
column_indexes.push(Some(idx));
column_values.push(Some(values));
} else {
column_indexes.push(ColumnIndex::Empty {
num_docs: num_docs_per_column[i],
});
column_indexes.push(None);
column_values.push(None);
}
}
@@ -193,22 +166,20 @@ fn merge_column(
serialize_column_mappable_to_u128(merged_column_index, &merge_column_values, wrt)?;
}
ColumnType::Bytes | ColumnType::Str => {
let mut column_indexes: Vec<ColumnIndex> = Vec::with_capacity(columns.len());
let mut column_indexes: Vec<Option<ColumnIndex>> = Vec::with_capacity(columns.len());
let mut bytes_columns: Vec<Option<BytesColumn>> = Vec::with_capacity(columns.len());
for (i, dynamic_column_opt) in columns.into_iter().enumerate() {
for dynamic_column_opt in columns {
match dynamic_column_opt {
Some(DynamicColumn::Str(str_column)) => {
column_indexes.push(str_column.term_ord_column.index.clone());
column_indexes.push(Some(str_column.term_ord_column.idx.clone()));
bytes_columns.push(Some(str_column.into()));
}
Some(DynamicColumn::Bytes(bytes_column)) => {
column_indexes.push(bytes_column.term_ord_column.index.clone());
column_indexes.push(Some(bytes_column.term_ord_column.idx.clone()));
bytes_columns.push(Some(bytes_column));
}
_ => {
column_indexes.push(ColumnIndex::Empty {
num_docs: num_docs_per_column[i],
});
column_indexes.push(None);
bytes_columns.push(None);
}
}
@@ -224,12 +195,40 @@ fn merge_column(
struct GroupedColumns {
required_column_type: Option<ColumnType>,
columns: Vec<Option<DynamicColumn>>,
column_category: ColumnTypeCategory,
}
impl GroupedColumns {
/// 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)
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(())
}
/// Returns the column type after merge.
@@ -251,76 +250,11 @@ impl GroupedColumns {
}
// At the moment, only the numerical categorical column type has more than one possible
// column type.
assert!(self
.columns
.iter()
.flatten()
.all(|el| ColumnTypeCategory::from(el.column_type()) == ColumnTypeCategory::Numerical));
assert_eq!(self.column_category, 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
@@ -341,80 +275,48 @@ fn merged_numerical_columns_type<'a>(
compatible_numerical_types.to_numerical_type()
}
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 alive_docid in alive_bitset.iter() {
if !multivalued_index.range(alive_docid).is_empty() {
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();
#[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();
for &(ref column_name, column_type) in required_columns {
columns
columns_grouped
.entry((column_name.clone(), column_type.into()))
.or_insert_with(|| GroupedColumnsHandle::new(columnar_readers.len()))
.or_insert_with(|| {
GroupedColumns::for_category(column_type.into(), columnar_readers.len())
})
.require_type(column_type)?;
}
for (columnar_id, columnar_reader) in columnar_readers.iter().enumerate() {
let column_name_and_handle = columnar_reader.iter_columns()?;
let column_name_and_handle = columnar_reader.list_columns()?;
for (column_name, handle) in column_name_and_handle {
let column_category: ColumnTypeCategory = handle.column_type().into();
columns
let column = handle.open()?;
columns_grouped
.entry((column_name, column_category))
.or_insert_with(|| GroupedColumnsHandle::new(columnar_readers.len()))
.set_column(columnar_id, handle);
.or_insert_with(|| {
GroupedColumns::for_category(column_category, columnar_readers.len())
})
.set_column(columnar_id, column);
}
}
Ok(columns)
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)
}
fn coerce_columns(
@@ -459,8 +361,8 @@ 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.max_value().into())),
DynamicColumn::F64(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::Bool(_)
| DynamicColumn::IpAddr(_)
| DynamicColumn::DateTime(_)

308
columnar/src/columnar/merge/tests.rs
View File

@@ -1,5 +1,3 @@
use itertools::Itertools;
use super::*;
use crate::{Cardinality, ColumnarWriter, HasAssociatedColumnType, RowId};
@@ -14,7 +12,7 @@ fn make_columnar<T: Into<NumericalValue> + HasAssociatedColumnType + Copy>(
}
let mut buffer: Vec<u8> = Vec::new();
dataframe_writer
.serialize(vals.len() as RowId, &mut buffer)
.serialize(vals.len() as RowId, None, &mut buffer)
.unwrap();
ColumnarReader::open(buffer).unwrap()
}
@@ -25,73 +23,70 @@ fn test_column_coercion_to_u64() {
let columnar1 = make_columnar("numbers", &[1i64]);
// u64 type
let columnar2 = make_columnar("numbers", &[u64::MAX]);
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();
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(), ColumnTypeCategory::Numerical)));
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)));
}
#[test]
fn test_column_coercion_to_i64() {
let columnar1 = make_columnar("numbers", &[-1i64]);
let columnar2 = make_columnar("numbers", &[2u64]);
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();
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(), ColumnTypeCategory::Numerical)));
assert!(column_map.contains_key(&("numbers".to_string(), ColumnType::I64)));
}
//#[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_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_group_columns_with_required_column() {
let columnar1 = make_columnar("numbers", &[1i64]);
let columnar2 = make_columnar("numbers", &[2u64]);
let columnars = &[&columnar1, &columnar2];
let merge_order = StackMergeOrder::stack(columnars).into();
let column_map: BTreeMap<(String, ColumnTypeCategory), GroupedColumnsHandle> =
let column_map: BTreeMap<(String, ColumnType), Vec<Option<DynamicColumn>>> =
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(), ColumnTypeCategory::Numerical)));
assert!(column_map.contains_key(&("numbers".to_string(), ColumnType::U64)));
}
#[test]
fn test_group_columns_required_column_with_no_existing_columns() {
let columnar1 = make_columnar("numbers", &[2u64]);
let columnar2 = make_columnar("numbers", &[2u64]);
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();
let column_map: BTreeMap<(String, ColumnType), Vec<Option<DynamicColumn>>> =
group_columns_for_merge(
&[&columnar1, &columnar2],
&[("required_col".to_string(), ColumnType::Str)],
)
.unwrap();
assert_eq!(column_map.len(), 2);
let columns = &column_map
.get(&("required_col".to_string(), ColumnTypeCategory::Str))
.unwrap()
.columns;
let columns = column_map
.get(&("required_col".to_string(), ColumnType::Str))
.unwrap();
assert_eq!(columns.len(), 2);
assert!(columns[0].is_none());
assert!(columns[1].is_none());
@@ -101,42 +96,35 @@ 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 columnars = &[&columnar1, &columnar2];
let merge_order = StackMergeOrder::stack(columnars).into();
let column_map: BTreeMap<(String, ColumnTypeCategory), GroupedColumnsHandle> =
let column_map: BTreeMap<(String, ColumnType), Vec<Option<DynamicColumn>>> =
group_columns_for_merge(
columnars,
&[&columnar1, &columnar2],
&[("numbers".to_string(), ColumnType::U64)],
&merge_order,
)
.unwrap();
assert_eq!(column_map.len(), 1);
assert!(column_map.contains_key(&("numbers".to_string(), ColumnTypeCategory::Numerical)));
assert!(column_map.contains_key(&("numbers".to_string(), ColumnType::U64)));
}
#[test]
fn test_missing_column() {
let columnar1 = make_columnar("numbers", &[-1i64]);
let columnar2 = make_columnar("numbers2", &[2u64]);
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();
let column_map: BTreeMap<(String, ColumnType), Vec<Option<DynamicColumn>>> =
group_columns_for_merge(&[&columnar1, &columnar2], &[]).unwrap();
assert_eq!(column_map.len(), 2);
assert!(column_map.contains_key(&("numbers".to_string(), ColumnTypeCategory::Numerical)));
assert!(column_map.contains_key(&("numbers".to_string(), ColumnType::I64)));
{
let columns = &column_map
.get(&("numbers".to_string(), ColumnTypeCategory::Numerical))
.unwrap()
.columns;
let columns = column_map
.get(&("numbers".to_string(), ColumnType::I64))
.unwrap();
assert!(columns[0].is_some());
assert!(columns[1].is_none());
}
{
let columns = &column_map
.get(&("numbers2".to_string(), ColumnTypeCategory::Numerical))
.unwrap()
.columns;
let columns = column_map
.get(&("numbers2".to_string(), ColumnType::U64))
.unwrap();
assert!(columns[0].is_none());
assert!(columns[1].is_some());
}
@@ -159,30 +147,30 @@ fn make_numerical_columnar_multiple_columns(
.max()
.unwrap_or(0u32);
let mut buffer: Vec<u8> = Vec::new();
dataframe_writer.serialize(num_rows, &mut buffer).unwrap();
dataframe_writer
.serialize(num_rows, None, &mut buffer)
.unwrap();
ColumnarReader::open(buffer).unwrap()
}
#[track_caller]
fn make_byte_columnar_multiple_columns(
columns: &[(&str, &[&[&[u8]]])],
num_rows: u32,
) -> ColumnarReader {
fn make_byte_columnar_multiple_columns(columns: &[(&str, &[&[&[u8]]])]) -> 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, &mut buffer).unwrap();
dataframe_writer
.serialize(num_rows, None, &mut buffer)
.unwrap();
ColumnarReader::open(buffer).unwrap()
}
@@ -201,7 +189,9 @@ fn make_text_columnar_multiple_columns(columns: &[(&str, &[&[&str]])]) -> Column
.max()
.unwrap_or(0u32);
let mut buffer: Vec<u8> = Vec::new();
dataframe_writer.serialize(num_rows, &mut buffer).unwrap();
dataframe_writer
.serialize(num_rows, None, &mut buffer)
.unwrap();
ColumnarReader::open(buffer).unwrap()
}
@@ -228,9 +218,7 @@ 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);
@@ -256,11 +244,7 @@ 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!()
};
assert_eq!(vals.ords().get_cardinality(), Cardinality::Optional);
let DynamicColumn::Str(vals) = dynamic_column else { panic!() };
let get_str_for_ord = |ord| {
let mut out = String::new();
vals.ord_to_str(ord, &mut out).unwrap();
@@ -288,8 +272,8 @@ fn test_merge_columnar_texts() {
#[test]
fn test_merge_columnar_byte() {
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 columnar1 = make_byte_columnar_multiple_columns(&[("bytes", &[&[b"bbbb"], &[b"baaa"]])]);
let columnar2 = make_byte_columnar_multiple_columns(&[("bytes", &[&[], &[b"a"]])]);
let mut buffer = Vec::new();
let columnars = &[&columnar1, &columnar2];
let stack_merge_order = StackMergeOrder::stack(columnars);
@@ -305,9 +289,7 @@ 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();
@@ -334,155 +316,3 @@ 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);
}

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

@@ -5,9 +5,6 @@ mod reader;
mod writer;
pub use column_type::{ColumnType, HasAssociatedColumnType};
pub use format_version::{Version, CURRENT_VERSION};
#[cfg(test)]
pub(crate) use merge::ColumnTypeCategory;
pub use merge::{merge_columnar, MergeRowOrder, ShuffleMergeOrder, StackMergeOrder};
pub use reader::ColumnarReader;
pub use writer::ColumnarWriter;

98
columnar/src/columnar/reader/mod.rs
View File

@@ -1,4 +1,4 @@
use std::{fmt, io, mem};
use std::{io, mem};
use common::file_slice::FileSlice;
use common::BinarySerializable;
@@ -6,7 +6,7 @@ use sstable::{Dictionary, RangeSSTable};
use crate::columnar::{format_version, ColumnType};
use crate::dynamic_column::DynamicColumnHandle;
use crate::{RowId, Version};
use crate::RowId;
fn io_invalid_data(msg: String) -> io::Error {
io::Error::new(io::ErrorKind::InvalidData, msg)
@@ -19,33 +19,6 @@ pub struct ColumnarReader {
column_dictionary: Dictionary<RangeSSTable>,
column_data: FileSlice,
num_rows: RowId,
format_version: Version,
}
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.
@@ -54,7 +27,6 @@ impl fmt::Debug for ColumnarReader {
fn read_all_columns_in_stream(
mut stream: sstable::Streamer<'_, RangeSSTable>,
column_data: &FileSlice,
format_version: Version,
) -> io::Result<Vec<DynamicColumnHandle>> {
let mut results = Vec::new();
while stream.advance() {
@@ -69,7 +41,6 @@ fn read_all_columns_in_stream(
let dynamic_column_handle = DynamicColumnHandle {
file_slice,
column_type,
format_version,
};
results.push(dynamic_column_handle);
}
@@ -91,7 +62,7 @@ impl ColumnarReader {
let num_rows = u32::deserialize(&mut &footer_bytes[8..12])?;
let version_footer_bytes: [u8; format_version::VERSION_FOOTER_NUM_BYTES] =
footer_bytes[12..].try_into().unwrap();
let format_version = format_version::parse_footer(version_footer_bytes)?;
let _version = format_version::parse_footer(version_footer_bytes)?;
let (column_data, sstable) =
file_slice_without_sstable_len.split_from_end(sstable_len as usize);
let column_dictionary = Dictionary::open(sstable)?;
@@ -99,49 +70,36 @@ impl ColumnarReader {
column_dictionary,
column_data,
num_rows,
format_version,
})
}
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,
format_version: self.format_version,
};
Some((column_name, column_handle))
} else {
None
}
}))
}
// TODO Add unit tests
pub fn list_columns(&self) -> io::Result<Vec<(String, DynamicColumnHandle)>> {
Ok(self.iter_columns()?.collect())
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)
}
fn stream_for_column_range(&self, column_name: &str) -> sstable::StreamerBuilder<RangeSSTable> {
@@ -172,7 +130,7 @@ impl ColumnarReader {
.stream_for_column_range(column_name)
.into_stream_async()
.await?;
read_all_columns_in_stream(stream, &self.column_data, self.format_version)
read_all_columns_in_stream(stream, &self.column_data)
}
/// Get all columns for the given column name.
@@ -181,7 +139,7 @@ impl ColumnarReader {
/// different types.
pub fn read_columns(&self, column_name: &str) -> io::Result<Vec<DynamicColumnHandle>> {
let stream = self.stream_for_column_range(column_name).into_stream()?;
read_all_columns_in_stream(stream, &self.column_data, self.format_version)
read_all_columns_in_stream(stream, &self.column_data)
}
/// Return the number of columns in the columnar.
@@ -200,7 +158,7 @@ mod tests {
columnar_writer.record_column_type("col1", ColumnType::Str, false);
columnar_writer.record_column_type("col2", ColumnType::U64, false);
let mut buffer = Vec::new();
columnar_writer.serialize(1, &mut buffer).unwrap();
columnar_writer.serialize(1, None, &mut buffer).unwrap();
let columnar = ColumnarReader::open(buffer).unwrap();
let columns = columnar.list_columns().unwrap();
assert_eq!(columns.len(), 2);
@@ -216,7 +174,7 @@ mod tests {
columnar_writer.record_column_type("count", ColumnType::U64, false);
columnar_writer.record_numerical(1, "count", 1u64);
let mut buffer = Vec::new();
columnar_writer.serialize(2, &mut buffer).unwrap();
columnar_writer.serialize(2, None, &mut buffer).unwrap();
let columnar = ColumnarReader::open(buffer).unwrap();
let columns = columnar.list_columns().unwrap();
assert_eq!(columns.len(), 1);

32
columnar/src/columnar/writer/column_writers.rs
View File

@@ -41,10 +41,31 @@ impl ColumnWriter {
pub(super) fn operation_iterator<'a, V: SymbolValue>(
&self,
arena: &MemoryArena,
old_to_new_ids_opt: Option<&[RowId]>,
buffer: &'a mut Vec<u8>,
) -> impl Iterator<Item = ColumnOperation<V>> + 'a {
buffer.clear();
self.values.read_to_end(arena, buffer);
if let Some(old_to_new_ids) = old_to_new_ids_opt {
// TODO avoid the extra deserialization / serialization.
let mut sorted_ops: Vec<(RowId, ColumnOperation<V>)> = Vec::new();
let mut new_doc = 0u32;
let mut cursor = &buffer[..];
for op in std::iter::from_fn(|| ColumnOperation::<V>::deserialize(&mut cursor)) {
if let ColumnOperation::NewDoc(doc) = &op {
new_doc = old_to_new_ids[*doc as usize];
sorted_ops.push((new_doc, ColumnOperation::NewDoc(new_doc)));
} else {
sorted_ops.push((new_doc, op));
}
}
// stable sort is crucial here.
sorted_ops.sort_by_key(|(new_doc_id, _)| *new_doc_id);
buffer.clear();
for (_, op) in sorted_ops {
buffer.extend_from_slice(op.serialize().as_ref());
}
}
let mut cursor: &[u8] = &buffer[..];
std::iter::from_fn(move || ColumnOperation::deserialize(&mut cursor))
}
@@ -210,9 +231,11 @@ impl NumericalColumnWriter {
pub(super) fn operation_iterator<'a>(
self,
arena: &MemoryArena,
old_to_new_ids: Option<&[RowId]>,
buffer: &'a mut Vec<u8>,
) -> impl Iterator<Item = ColumnOperation<NumericalValue>> + 'a {
self.column_writer.operation_iterator(arena, buffer)
self.column_writer
.operation_iterator(arena, old_to_new_ids, buffer)
}
}
@@ -246,17 +269,18 @@ impl StrOrBytesColumnWriter {
dictionaries: &mut [DictionaryBuilder],
arena: &mut MemoryArena,
) {
let unordered_id =
dictionaries[self.dictionary_id as usize].get_or_allocate_id(bytes, arena);
let unordered_id = dictionaries[self.dictionary_id as usize].get_or_allocate_id(bytes);
self.column_writer.record(doc, unordered_id, arena);
}
pub(super) fn operation_iterator<'a>(
&self,
arena: &MemoryArena,
old_to_new_ids: Option<&[RowId]>,
byte_buffer: &'a mut Vec<u8>,
) -> impl Iterator<Item = ColumnOperation<UnorderedId>> + 'a {
self.column_writer.operation_iterator(arena, byte_buffer)
self.column_writer
.operation_iterator(arena, old_to_new_ids, byte_buffer)
}
}

283
columnar/src/columnar/writer/mod.rs
View File

@@ -8,13 +8,14 @@ use std::net::Ipv6Addr;
use column_operation::ColumnOperation;
pub(crate) use column_writers::CompatibleNumericalTypes;
use common::json_path_writer::JSON_END_OF_PATH;
use common::CountingWriter;
pub(crate) use serializer::ColumnarSerializer;
use stacker::{Addr, ArenaHashMap, MemoryArena};
use crate::column_index::{SerializableColumnIndex, SerializableOptionalIndex};
use crate::column_values::{MonotonicallyMappableToU128, MonotonicallyMappableToU64};
use crate::column_index::SerializableColumnIndex;
use crate::column_values::{
ColumnValues, MonotonicallyMappableToU128, MonotonicallyMappableToU64, VecColumn,
};
use crate::columnar::column_type::ColumnType;
use crate::columnar::writer::column_writers::{
ColumnWriter, NumericalColumnWriter, StrOrBytesColumnWriter,
@@ -44,7 +45,7 @@ struct SpareBuffers {
/// columnar_writer.record_str(1u32 /* doc id */, "product_name", "Apple");
/// columnar_writer.record_numerical(0u32 /* doc id */, "price", 10.5f64); //< uh oh we ended up mixing integer and floats.
/// let mut wrt: Vec<u8> = Vec::new();
/// columnar_writer.serialize(2u32, &mut wrt).unwrap();
/// columnar_writer.serialize(2u32, None, &mut wrt).unwrap();
/// ```
#[derive(Default)]
pub struct ColumnarWriter {
@@ -60,8 +61,25 @@ pub struct ColumnarWriter {
buffers: SpareBuffers,
}
#[inline]
fn mutate_or_create_column<V, TMutator>(
arena_hash_map: &mut ArenaHashMap,
column_name: &str,
updater: TMutator,
) where
V: Copy + 'static,
TMutator: FnMut(Option<V>) -> V,
{
assert!(
!column_name.as_bytes().contains(&0u8),
"key may not contain the 0 byte"
);
arena_hash_map.mutate_or_create(column_name.as_bytes(), updater);
}
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()
@@ -69,11 +87,48 @@ 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`
/// column.
///
/// If the column is multivalued, use the first value for scoring.
/// If no value is associated to a specific row, the document is assigned
/// the lowest possible score.
///
/// 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 mut symbols_buffer = Vec::new();
let mut values = Vec::new();
let mut last_doc_opt: Option<RowId> = None;
for op in numerical_col_writer.operation_iterator(&self.arena, None, &mut symbols_buffer) {
match op {
ColumnOperation::NewDoc(doc) => {
last_doc_opt = Some(doc);
}
ColumnOperation::Value(numerical_value) => {
if let Some(last_doc) = last_doc_opt {
let score: f32 = f64::coerce(numerical_value) as f32;
values.push((score, last_doc));
}
}
}
}
for doc in values.len() as u32..num_docs {
values.push((0.0f32, doc));
}
values.sort_by(|(left_score, _), (right_score, _)| {
if reversed {
right_score.partial_cmp(left_score).unwrap()
} else {
left_score.partial_cmp(right_score).unwrap()
}
});
values.into_iter().map(|(_score, doc)| doc).collect()
}
/// Records a column type. This is useful to bypass the coercion process,
@@ -103,8 +158,9 @@ impl ColumnarWriter {
},
&mut self.dictionaries,
);
hash_map.mutate_or_create(
column_name.as_bytes(),
mutate_or_create_column(
hash_map,
column_name,
|column_opt: Option<StrOrBytesColumnWriter>| {
let mut column_writer = if let Some(column_writer) = column_opt {
column_writer
@@ -119,21 +175,24 @@ impl ColumnarWriter {
);
}
ColumnType::Bool => {
self.bool_field_hash_map.mutate_or_create(
column_name.as_bytes(),
mutate_or_create_column(
&mut self.bool_field_hash_map,
column_name,
|column_opt: Option<ColumnWriter>| column_opt.unwrap_or_default(),
);
}
ColumnType::DateTime => {
self.datetime_field_hash_map.mutate_or_create(
column_name.as_bytes(),
mutate_or_create_column(
&mut self.datetime_field_hash_map,
column_name,
|column_opt: Option<ColumnWriter>| column_opt.unwrap_or_default(),
);
}
ColumnType::I64 | ColumnType::F64 | ColumnType::U64 => {
let numerical_type = column_type.numerical_type().unwrap();
self.numerical_field_hash_map.mutate_or_create(
column_name.as_bytes(),
mutate_or_create_column(
&mut self.numerical_field_hash_map,
column_name,
|column_opt: Option<NumericalColumnWriter>| {
let mut column: NumericalColumnWriter = column_opt.unwrap_or_default();
column.force_numerical_type(numerical_type);
@@ -141,8 +200,9 @@ impl ColumnarWriter {
},
);
}
ColumnType::IpAddr => self.ip_addr_field_hash_map.mutate_or_create(
column_name.as_bytes(),
ColumnType::IpAddr => mutate_or_create_column(
&mut self.ip_addr_field_hash_map,
column_name,
|column_opt: Option<ColumnWriter>| column_opt.unwrap_or_default(),
),
}
@@ -155,8 +215,9 @@ impl ColumnarWriter {
numerical_value: T,
) {
let (hash_map, arena) = (&mut self.numerical_field_hash_map, &mut self.arena);
hash_map.mutate_or_create(
column_name.as_bytes(),
mutate_or_create_column(
hash_map,
column_name,
|column_opt: Option<NumericalColumnWriter>| {
let mut column: NumericalColumnWriter = column_opt.unwrap_or_default();
column.record_numerical_value(doc, numerical_value.into(), arena);
@@ -166,6 +227,10 @@ impl ColumnarWriter {
}
pub fn record_ip_addr(&mut self, doc: RowId, column_name: &str, ip_addr: Ipv6Addr) {
assert!(
!column_name.as_bytes().contains(&0u8),
"key may not contain the 0 byte"
);
let (hash_map, arena) = (&mut self.ip_addr_field_hash_map, &mut self.arena);
hash_map.mutate_or_create(
column_name.as_bytes(),
@@ -179,30 +244,24 @@ impl ColumnarWriter {
pub fn record_bool(&mut self, doc: RowId, column_name: &str, val: bool) {
let (hash_map, arena) = (&mut self.bool_field_hash_map, &mut self.arena);
hash_map.mutate_or_create(
column_name.as_bytes(),
|column_opt: Option<ColumnWriter>| {
let mut column: ColumnWriter = column_opt.unwrap_or_default();
column.record(doc, val, arena);
column
},
);
mutate_or_create_column(hash_map, column_name, |column_opt: Option<ColumnWriter>| {
let mut column: ColumnWriter = column_opt.unwrap_or_default();
column.record(doc, val, arena);
column
});
}
pub fn record_datetime(&mut self, doc: RowId, column_name: &str, datetime: common::DateTime) {
let (hash_map, arena) = (&mut self.datetime_field_hash_map, &mut self.arena);
hash_map.mutate_or_create(
column_name.as_bytes(),
|column_opt: Option<ColumnWriter>| {
let mut column: ColumnWriter = column_opt.unwrap_or_default();
column.record(
doc,
NumericalValue::I64(datetime.into_timestamp_nanos()),
arena,
);
column
},
);
mutate_or_create_column(hash_map, column_name, |column_opt: Option<ColumnWriter>| {
let mut column: ColumnWriter = column_opt.unwrap_or_default();
column.record(
doc,
NumericalValue::I64(datetime.into_timestamp_micros()),
arena,
);
column
});
}
pub fn record_str(&mut self, doc: RowId, column_name: &str, value: &str) {
@@ -227,6 +286,10 @@ impl ColumnarWriter {
}
pub fn record_bytes(&mut self, doc: RowId, column_name: &str, value: &[u8]) {
assert!(
!column_name.as_bytes().contains(&0u8),
"key may not contain the 0 byte"
);
let (hash_map, arena, dictionaries) = (
&mut self.bytes_field_hash_map,
&mut self.arena,
@@ -246,13 +309,17 @@ impl ColumnarWriter {
},
);
}
pub fn serialize(&mut self, num_docs: RowId, wrt: &mut dyn io::Write) -> io::Result<()> {
pub fn serialize(
&mut self,
num_docs: RowId,
old_to_new_row_ids: Option<&[RowId]>,
wrt: &mut dyn io::Write,
) -> io::Result<()> {
let mut serializer = ColumnarSerializer::new(wrt);
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();
@@ -262,67 +329,67 @@ impl ColumnarWriter {
columns.extend(
self.bytes_field_hash_map
.iter()
.map(|(column_name, addr)| (column_name, 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));
let (arena, buffers, dictionaries) = (&self.arena, &mut self.buffers, &self.dictionaries);
let mut symbol_byte_buffer: Vec<u8> = Vec::new();
for (column_name, column_type, addr) in columns {
if column_name.contains(&JSON_END_OF_PATH) {
// Tantivy uses b'0' as a separator for nested fields in JSON.
// Column names with a b'0' are not simply ignored by the columnar (and the inverted
// index).
continue;
}
match column_type {
ColumnType::Bool => {
let column_writer: ColumnWriter = self.bool_field_hash_map.read(addr);
let cardinality = column_writer.get_cardinality(num_docs);
let mut column_serializer =
serializer.start_serialize_column(column_name, column_type);
serializer.serialize_column(column_name, column_type);
serialize_bool_column(
cardinality,
num_docs,
column_writer.operation_iterator(arena, &mut symbol_byte_buffer),
column_writer.operation_iterator(
arena,
old_to_new_row_ids,
&mut symbol_byte_buffer,
),
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.start_serialize_column(column_name, ColumnType::IpAddr);
serializer.serialize_column(column_name, ColumnType::IpAddr);
serialize_ip_addr_column(
cardinality,
num_docs,
column_writer.operation_iterator(arena, &mut symbol_byte_buffer),
column_writer.operation_iterator(
arena,
old_to_new_row_ids,
&mut symbol_byte_buffer,
),
buffers,
&mut column_serializer,
)?;
column_serializer.finalize()?;
}
ColumnType::Bytes | ColumnType::Str => {
let str_or_bytes_column_writer: StrOrBytesColumnWriter =
@@ -337,51 +404,58 @@ impl ColumnarWriter {
.column_writer
.get_cardinality(num_docs);
let mut column_serializer =
serializer.start_serialize_column(column_name, column_type);
serializer.serialize_column(column_name, column_type);
serialize_bytes_or_str_column(
cardinality,
num_docs,
str_or_bytes_column_writer.sort_values_within_row,
dictionary_builder,
str_or_bytes_column_writer
.operation_iterator(arena, &mut symbol_byte_buffer),
str_or_bytes_column_writer.operation_iterator(
arena,
old_to_new_row_ids,
&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.start_serialize_column(column_name, column_type);
serializer.serialize_column(column_name, column_type);
let numerical_type = column_type.numerical_type().unwrap();
serialize_numerical_column(
cardinality,
num_docs,
numerical_type,
numerical_column_writer.operation_iterator(arena, &mut symbol_byte_buffer),
numerical_column_writer.operation_iterator(
arena,
old_to_new_row_ids,
&mut symbol_byte_buffer,
),
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.start_serialize_column(column_name, ColumnType::DateTime);
serializer.serialize_column(column_name, ColumnType::DateTime);
serialize_numerical_column(
cardinality,
num_docs,
NumericalType::I64,
column_writer.operation_iterator(arena, &mut symbol_byte_buffer),
column_writer.operation_iterator(
arena,
old_to_new_row_ids,
&mut symbol_byte_buffer,
),
buffers,
&mut column_serializer,
)?;
column_serializer.finalize()?;
}
};
}
@@ -392,7 +466,6 @@ 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,
@@ -400,7 +473,6 @@ 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 {
@@ -409,8 +481,7 @@ fn serialize_bytes_or_str_column(
..
} = buffers;
let mut counting_writer = CountingWriter::wrap(wrt);
let term_id_mapping: TermIdMapping =
dictionary_builder.serialize(arena, &mut counting_writer)?;
let term_id_mapping: TermIdMapping = dictionary_builder.serialize(&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>| {
@@ -546,7 +617,10 @@ fn send_to_serialize_column_mappable_to_u128<
value_index_builders: &mut PreallocatedIndexBuilders,
values: &mut Vec<T>,
mut wrt: impl io::Write,
) -> io::Result<()> {
) -> io::Result<()>
where
for<'a> VecColumn<'a, T>: ColumnValues<T>,
{
values.clear();
// TODO: split index and values
let serializable_column_index = match cardinality {
@@ -562,16 +636,16 @@ fn send_to_serialize_column_mappable_to_u128<
let optional_index_builder = value_index_builders.borrow_optional_index_builder();
consume_operation_iterator(op_iterator, optional_index_builder, values);
let optional_index = optional_index_builder.finish(num_rows);
SerializableColumnIndex::Optional(SerializableOptionalIndex {
SerializableColumnIndex::Optional {
num_rows,
non_null_row_ids: Box::new(optional_index),
})
}
}
Cardinality::Multivalued => {
let multivalued_index_builder = value_index_builders.borrow_multivalued_index_builder();
consume_operation_iterator(op_iterator, multivalued_index_builder, values);
let serializable_multivalued_index = multivalued_index_builder.finish(num_rows);
SerializableColumnIndex::Multivalued(serializable_multivalued_index)
let multivalued_index = multivalued_index_builder.finish(num_rows);
SerializableColumnIndex::Multivalued(Box::new(multivalued_index))
}
};
crate::column::serialize_column_mappable_to_u128(
@@ -582,6 +656,15 @@ fn send_to_serialize_column_mappable_to_u128<
Ok(())
}
fn sort_values_within_row_in_place(multivalued_index: &[RowId], values: &mut [u64]) {
let mut start_index: usize = 0;
for end_index in multivalued_index.iter().copied() {
let end_index = end_index as usize;
values[start_index..end_index].sort_unstable();
start_index = end_index;
}
}
fn send_to_serialize_column_mappable_to_u64(
op_iterator: impl Iterator<Item = ColumnOperation<u64>>,
cardinality: Cardinality,
@@ -590,7 +673,10 @@ fn send_to_serialize_column_mappable_to_u64(
value_index_builders: &mut PreallocatedIndexBuilders,
values: &mut Vec<u64>,
mut wrt: impl io::Write,
) -> io::Result<()> {
) -> io::Result<()>
where
for<'a> VecColumn<'a, u64>: ColumnValues<u64>,
{
values.clear();
let serializable_column_index = match cardinality {
Cardinality::Full => {
@@ -605,22 +691,19 @@ fn send_to_serialize_column_mappable_to_u64(
let optional_index_builder = value_index_builders.borrow_optional_index_builder();
consume_operation_iterator(op_iterator, optional_index_builder, values);
let optional_index = optional_index_builder.finish(num_rows);
SerializableColumnIndex::Optional(SerializableOptionalIndex {
SerializableColumnIndex::Optional {
non_null_row_ids: Box::new(optional_index),
num_rows,
})
}
}
Cardinality::Multivalued => {
let multivalued_index_builder = value_index_builders.borrow_multivalued_index_builder();
consume_operation_iterator(op_iterator, multivalued_index_builder, values);
let serializable_multivalued_index = multivalued_index_builder.finish(num_rows);
let multivalued_index = multivalued_index_builder.finish(num_rows);
if sort_values_within_row {
sort_values_within_row_in_place(
serializable_multivalued_index.start_offsets.boxed_iter(),
values,
);
sort_values_within_row_in_place(multivalued_index, values);
}
SerializableColumnIndex::Multivalued(serializable_multivalued_index)
SerializableColumnIndex::Multivalued(Box::new(multivalued_index))
}
};
crate::column::serialize_column_mappable_to_u64(
@@ -631,18 +714,6 @@ fn send_to_serialize_column_mappable_to_u64(
Ok(())
}
fn sort_values_within_row_in_place(
multivalued_index: impl Iterator<Item = RowId>,
values: &mut [u64],
) {
let mut start_index: usize = 0;
for end_index in multivalued_index {
let end_index = end_index as usize;
values[start_index..end_index].sort_unstable();
start_index = end_index;
}
}
fn coerce_numerical_symbol<T>(
operation_iterator: impl Iterator<Item = ColumnOperation<NumericalValue>>,
) -> impl Iterator<Item = ColumnOperation<u64>>
@@ -690,7 +761,7 @@ mod tests {
assert_eq!(column_writer.get_cardinality(3), Cardinality::Full);
let mut buffer = Vec::new();
let symbols: Vec<ColumnOperation<NumericalValue>> = column_writer
.operation_iterator(&arena, &mut buffer)
.operation_iterator(&arena, None, &mut buffer)
.collect();
assert_eq!(symbols.len(), 6);
assert!(matches!(symbols[0], ColumnOperation::NewDoc(0u32)));
@@ -719,7 +790,7 @@ mod tests {
assert_eq!(column_writer.get_cardinality(3), Cardinality::Optional);
let mut buffer = Vec::new();
let symbols: Vec<ColumnOperation<NumericalValue>> = column_writer
.operation_iterator(&arena, &mut buffer)
.operation_iterator(&arena, None, &mut buffer)
.collect();
assert_eq!(symbols.len(), 4);
assert!(matches!(symbols[0], ColumnOperation::NewDoc(1u32)));
@@ -742,7 +813,7 @@ mod tests {
assert_eq!(column_writer.get_cardinality(2), Cardinality::Optional);
let mut buffer = Vec::new();
let symbols: Vec<ColumnOperation<NumericalValue>> = column_writer
.operation_iterator(&arena, &mut buffer)
.operation_iterator(&arena, None, &mut buffer)
.collect();
assert_eq!(symbols.len(), 2);
assert!(matches!(symbols[0], ColumnOperation::NewDoc(0u32)));
@@ -761,7 +832,7 @@ mod tests {
assert_eq!(column_writer.get_cardinality(1), Cardinality::Multivalued);
let mut buffer = Vec::new();
let symbols: Vec<ColumnOperation<NumericalValue>> = column_writer
.operation_iterator(&arena, &mut buffer)
.operation_iterator(&arena, None, &mut buffer)
.collect();
assert_eq!(symbols.len(), 3);
assert!(matches!(symbols[0], ColumnOperation::NewDoc(0u32)));

35
columnar/src/columnar/writer/serializer.rs
View File

@@ -1,7 +1,6 @@
use std::io;
use std::io::Write;
use common::json_path_writer::JSON_END_OF_PATH;
use common::{BinarySerializable, CountingWriter};
use sstable::value::RangeValueWriter;
use sstable::RangeSSTable;
@@ -20,7 +19,7 @@ pub struct ColumnarSerializer<W: io::Write> {
fn prepare_key(key: &[u8], column_type: ColumnType, buffer: &mut Vec<u8>) {
buffer.clear();
buffer.extend_from_slice(key);
buffer.push(JSON_END_OF_PATH);
buffer.push(0u8);
buffer.push(column_type.to_code());
}
@@ -35,12 +34,11 @@ impl<W: io::Write> ColumnarSerializer<W> {
}
}
/// Creates a ColumnSerializer.
pub fn start_serialize_column<'a>(
pub fn serialize_column<'a>(
&'a mut self,
column_name: &[u8],
column_type: ColumnType,
) -> ColumnSerializer<'a, W> {
) -> impl io::Write + 'a {
let start_offset = self.wrt.written_bytes();
prepare_key(column_name, column_type, &mut self.prepare_key_buffer);
ColumnSerializer {
@@ -62,21 +60,20 @@ impl<W: io::Write> ColumnarSerializer<W> {
}
}
pub struct ColumnSerializer<'a, W: io::Write> {
struct ColumnSerializer<'a, W: io::Write> {
columnar_serializer: &'a mut ColumnarSerializer<W>,
start_offset: u64,
}
impl<'a, W: io::Write> ColumnSerializer<'a, W> {
pub fn finalize(self) -> io::Result<()> {
impl<'a, W: io::Write> Drop for ColumnSerializer<'a, W> {
fn drop(&mut self) {
let end_offset: u64 = self.columnar_serializer.wrt.written_bytes();
let byte_range = self.start_offset..end_offset;
self.columnar_serializer.sstable_range.insert(
self.columnar_serializer.sstable_range.insert_cannot_fail(
&self.columnar_serializer.prepare_key_buffer[..],
&byte_range,
)?;
);
self.columnar_serializer.prepare_key_buffer.clear();
Ok(())
}
}
@@ -93,3 +90,19 @@ impl<'a, W: io::Write> io::Write for ColumnSerializer<'a, W> {
self.columnar_serializer.wrt.write_all(buf)
}
}
#[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], 0u8);
assert_eq!(buffer[11], ColumnType::Str.to_code());
}
}

81
columnar/src/columnar/writer/value_index.rs
View File

@@ -1,4 +1,3 @@
use crate::column_index::{SerializableMultivalueIndex, SerializableOptionalIndex};
use crate::iterable::Iterable;
use crate::RowId;
@@ -60,47 +59,31 @@ impl IndexBuilder for OptionalIndexBuilder {
#[derive(Default)]
pub struct MultivaluedIndexBuilder {
doc_with_values: Vec<RowId>,
start_offsets: Vec<u32>,
start_offsets: Vec<RowId>,
total_num_vals_seen: u32,
current_row: RowId,
current_row_has_value: bool,
}
impl MultivaluedIndexBuilder {
pub fn finish(&mut self, num_docs: RowId) -> SerializableMultivalueIndex<'_> {
self.start_offsets.push(self.total_num_vals_seen);
let non_null_row_ids: Box<dyn Iterable<RowId>> = Box::new(&self.doc_with_values[..]);
SerializableMultivalueIndex {
doc_ids_with_values: SerializableOptionalIndex {
non_null_row_ids,
num_rows: num_docs,
},
start_offsets: Box::new(&self.start_offsets[..]),
}
pub fn finish(&mut self, num_docs: RowId) -> &[u32] {
self.start_offsets
.resize(num_docs as usize + 1, self.total_num_vals_seen);
&self.start_offsets[..]
}
fn reset(&mut self) {
self.doc_with_values.clear();
self.start_offsets.clear();
self.start_offsets.push(0u32);
self.total_num_vals_seen = 0;
self.current_row = 0;
self.current_row_has_value = false;
}
}
impl IndexBuilder for MultivaluedIndexBuilder {
fn record_row(&mut self, row_id: RowId) {
self.current_row = row_id;
self.current_row_has_value = false;
self.start_offsets
.resize(row_id as usize + 1, self.total_num_vals_seen);
}
fn record_value(&mut self) {
if !self.current_row_has_value {
self.current_row_has_value = true;
self.doc_with_values.push(self.current_row);
self.start_offsets.push(self.total_num_vals_seen);
}
self.total_num_vals_seen += 1;
}
}
@@ -158,32 +141,6 @@ mod tests {
);
}
#[test]
fn test_multivalued_value_index_builder_simple() {
let mut multivalued_value_index_builder = MultivaluedIndexBuilder::default();
{
multivalued_value_index_builder.record_row(0u32);
multivalued_value_index_builder.record_value();
multivalued_value_index_builder.record_value();
let serialized_multivalue_index = multivalued_value_index_builder.finish(1u32);
let start_offsets: Vec<u32> = serialized_multivalue_index
.start_offsets
.boxed_iter()
.collect();
assert_eq!(&start_offsets, &[0, 2]);
}
multivalued_value_index_builder.reset();
multivalued_value_index_builder.record_row(0u32);
multivalued_value_index_builder.record_value();
multivalued_value_index_builder.record_value();
let serialized_multivalue_index = multivalued_value_index_builder.finish(1u32);
let start_offsets: Vec<u32> = serialized_multivalue_index
.start_offsets
.boxed_iter()
.collect();
assert_eq!(&start_offsets, &[0, 2]);
}
#[test]
fn test_multivalued_value_index_builder() {
let mut multivalued_value_index_builder = MultivaluedIndexBuilder::default();
@@ -192,15 +149,17 @@ mod tests {
multivalued_value_index_builder.record_value();
multivalued_value_index_builder.record_row(2u32);
multivalued_value_index_builder.record_value();
let SerializableMultivalueIndex {
doc_ids_with_values,
start_offsets,
} = multivalued_value_index_builder.finish(4u32);
assert_eq!(doc_ids_with_values.num_rows, 4u32);
let doc_ids_with_values: Vec<u32> =
doc_ids_with_values.non_null_row_ids.boxed_iter().collect();
assert_eq!(&doc_ids_with_values, &[1u32, 2u32]);
let start_offsets: Vec<u32> = start_offsets.boxed_iter().collect();
assert_eq!(&start_offsets[..], &[0, 2, 3]);
assert_eq!(
multivalued_value_index_builder.finish(4u32).to_vec(),
vec![0, 0, 2, 3, 3]
);
multivalued_value_index_builder.reset();
multivalued_value_index_builder.record_row(2u32);
multivalued_value_index_builder.record_value();
multivalued_value_index_builder.record_value();
assert_eq!(
multivalued_value_index_builder.finish(4u32).to_vec(),
vec![0, 0, 0, 2, 2]
);
}
}

183
columnar/src/compat_tests.rs
View File

@@ -1,183 +0,0 @@
use std::path::PathBuf;
use itertools::Itertools;
use crate::{
merge_columnar, Cardinality, Column, ColumnarReader, DynamicColumn, StackMergeOrder,
CURRENT_VERSION,
};
const NUM_DOCS: u32 = u16::MAX as u32;
fn generate_columnar(num_docs: u32, value_offset: u64) -> Vec<u8> {
use crate::ColumnarWriter;
let mut columnar_writer = ColumnarWriter::default();
for i in 0..num_docs {
if i % 100 == 0 {
columnar_writer.record_numerical(i, "sparse", value_offset + i as u64);
}
if i % 5 == 0 {
columnar_writer.record_numerical(i, "dense", value_offset + i as u64);
}
columnar_writer.record_numerical(i, "full", value_offset + i as u64);
columnar_writer.record_numerical(i, "multi", value_offset + i as u64);
columnar_writer.record_numerical(i, "multi", value_offset + i as u64);
}
let mut wrt: Vec<u8> = Vec::new();
columnar_writer.serialize(num_docs, &mut wrt).unwrap();
wrt
}
#[test]
/// Writes a columnar for the CURRENT_VERSION to disk.
fn create_format() {
let version = CURRENT_VERSION.to_string();
let file_path = path_for_version(&version);
if PathBuf::from(file_path.clone()).exists() {
return;
}
let columnar = generate_columnar(NUM_DOCS, 0);
std::fs::write(file_path, columnar).unwrap();
}
fn path_for_version(version: &str) -> String {
format!("./compat_tests_data/{}.columnar", version)
}
#[test]
fn test_format_v1() {
let path = path_for_version("v1");
test_format(&path);
}
#[test]
fn test_format_v2() {
let path = path_for_version("v2");
test_format(&path);
}
fn test_format(path: &str) {
let file_content = std::fs::read(path).unwrap();
let reader = ColumnarReader::open(file_content).unwrap();
check_columns(&reader);
// Test merge
let reader2 = ColumnarReader::open(generate_columnar(NUM_DOCS, NUM_DOCS as u64)).unwrap();
let columnar_readers = vec![&reader, &reader2];
let merge_row_order = StackMergeOrder::stack(&columnar_readers[..]);
let mut out = Vec::new();
merge_columnar(&columnar_readers, &[], merge_row_order.into(), &mut out).unwrap();
let reader = ColumnarReader::open(out).unwrap();
check_columns(&reader);
}
fn check_columns(reader: &ColumnarReader) {
let column = open_column(reader, "full");
check_column(&column, |doc_id| vec![(doc_id, doc_id as u64).into()]);
assert_eq!(column.get_cardinality(), Cardinality::Full);
let column = open_column(reader, "multi");
check_column(&column, |doc_id| {
vec![
(doc_id * 2, doc_id as u64).into(),
(doc_id * 2 + 1, doc_id as u64).into(),
]
});
assert_eq!(column.get_cardinality(), Cardinality::Multivalued);
let column = open_column(reader, "sparse");
check_column(&column, |doc_id| {
if doc_id % 100 == 0 {
vec![(doc_id / 100, doc_id as u64).into()]
} else {
vec![]
}
});
assert_eq!(column.get_cardinality(), Cardinality::Optional);
let column = open_column(reader, "dense");
check_column(&column, |doc_id| {
if doc_id % 5 == 0 {
vec![(doc_id / 5, doc_id as u64).into()]
} else {
vec![]
}
});
assert_eq!(column.get_cardinality(), Cardinality::Optional);
}
struct RowIdAndValue {
row_id: u32,
value: u64,
}
impl From<(u32, u64)> for RowIdAndValue {
fn from((row_id, value): (u32, u64)) -> Self {
Self { row_id, value }
}
}
fn check_column<F: Fn(u32) -> Vec<RowIdAndValue>>(column: &Column<u64>, expected: F) {
let num_docs = column.num_docs();
let test_doc = |doc: u32| {
if expected(doc).is_empty() {
assert_eq!(column.first(doc), None);
} else {
assert_eq!(column.first(doc), Some(expected(doc)[0].value));
}
let values = column.values_for_doc(doc).collect_vec();
assert_eq!(values, expected(doc).iter().map(|x| x.value).collect_vec());
let mut row_ids = Vec::new();
column.row_ids_for_docs(&[doc], &mut vec![], &mut row_ids);
assert_eq!(
row_ids,
expected(doc).iter().map(|x| x.row_id).collect_vec()
);
let values = column.values_for_doc(doc).collect_vec();
assert_eq!(values, expected(doc).iter().map(|x| x.value).collect_vec());
// Docid rowid conversion
let mut row_ids = Vec::new();
let safe_next_doc = |doc: u32| (doc + 1).min(num_docs - 1);
column
.index
.docids_to_rowids(&[doc, safe_next_doc(doc)], &mut vec![], &mut row_ids);
let expected_rowids = expected(doc)
.iter()
.map(|x| x.row_id)
.chain(expected(safe_next_doc(doc)).iter().map(|x| x.row_id))
.collect_vec();
assert_eq!(row_ids, expected_rowids);
let rowid_range = column
.index
.docid_range_to_rowids(doc..safe_next_doc(doc) + 1);
if expected_rowids.is_empty() {
assert!(rowid_range.is_empty());
} else {
assert_eq!(
rowid_range,
expected_rowids[0]..expected_rowids.last().unwrap() + 1
);
}
};
test_doc(0);
test_doc(num_docs - 1);
test_doc(num_docs - 2);
test_doc(65000);
}
fn open_column(reader: &ColumnarReader, name: &str) -> Column<u64> {
let column = reader.read_columns(name).unwrap()[0]
.open()
.unwrap()
.coerce_numerical(crate::NumericalType::U64)
.unwrap();
let DynamicColumn::U64(column) = column else {
panic!();
};
column
}

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>,
@@ -31,38 +31,26 @@ pub struct OrderedId(pub u32);
/// mapping.
#[derive(Default)]
pub(crate) struct DictionaryBuilder {
dict: SharedArenaHashMap,
dict: FnvHashMap<Vec<u8>, UnorderedId>,
}
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], 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)
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
}
/// Serialize the dictionary into an fst, and returns the
/// `UnorderedId -> TermOrdinal` map.
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();
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();
terms.sort_unstable_by_key(|(key, _)| *key);
// TODO Remove the allocation.
let mut unordered_to_ord: Vec<OrderedId> = vec![OrderedId(0u32); terms.len()];
@@ -75,10 +63,6 @@ impl DictionaryBuilder {
sstable_builder.finish()?;
Ok(TermIdMapping { unordered_to_ord })
}
pub(crate) fn mem_usage(&self) -> usize {
self.dict.mem_usage()
}
}
#[cfg(test)]
@@ -87,13 +71,12 @@ 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", &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 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 mut buffer = Vec::new();
let id_mapping = dictionary_builder.serialize(&arena, &mut buffer).unwrap();
let id_mapping = dictionary_builder.serialize(&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));

132
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::{ByteCount, DateTime, HasLen, OwnedBytes};
use common::{DateTime, HasLen, OwnedBytes};
use crate::column::{BytesColumn, Column, StrColumn};
use crate::column_values::{monotonic_map_column, StrictlyMonotonicFn};
use crate::columnar::ColumnType;
use crate::{Cardinality, ColumnIndex, ColumnValues, NumericalType, Version};
use crate::{Cardinality, NumericalType};
#[derive(Clone)]
pub enum DynamicColumn {
@@ -22,54 +22,19 @@ pub enum DynamicColumn {
Str(StrColumn),
}
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(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(),
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(),
}
}
pub fn column_type(&self) -> ColumnType {
match self {
DynamicColumn::Bool(_) => ColumnType::Bool,
@@ -108,11 +73,11 @@ impl DynamicColumn {
fn coerce_to_f64(self) -> Option<DynamicColumn> {
match self {
DynamicColumn::I64(column) => Some(DynamicColumn::F64(Column {
index: column.index,
idx: column.idx,
values: Arc::new(monotonic_map_column(column.values, MapI64ToF64)),
})),
DynamicColumn::U64(column) => Some(DynamicColumn::F64(Column {
index: column.index,
idx: column.idx,
values: Arc::new(monotonic_map_column(column.values, MapU64ToF64)),
})),
DynamicColumn::F64(_) => Some(self),
@@ -126,7 +91,7 @@ impl DynamicColumn {
return None;
}
Some(DynamicColumn::I64(Column {
index: column.index,
idx: column.idx,
values: Arc::new(monotonic_map_column(column.values, MapU64ToI64)),
}))
}
@@ -141,7 +106,7 @@ impl DynamicColumn {
return None;
}
Some(DynamicColumn::U64(Column {
index: column.index,
idx: column.idx,
values: Arc::new(monotonic_map_column(column.values, MapI64ToU64)),
}))
}
@@ -228,11 +193,10 @@ static_dynamic_conversions!(StrColumn, Str);
static_dynamic_conversions!(BytesColumn, Bytes);
static_dynamic_conversions!(Column<Ipv6Addr>, IpAddr);
#[derive(Clone, Debug)]
#[derive(Clone)]
pub struct DynamicColumnHandle {
pub(crate) file_slice: FileSlice,
pub(crate) column_type: ColumnType,
pub(crate) format_version: Version,
}
impl DynamicColumnHandle {
@@ -248,12 +212,7 @@ impl DynamicColumnHandle {
}
/// Returns the `u64` fast field reader reader associated with `fields` of types
/// 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`.
/// Str, u64, i64, f64, or datetime.
///
/// If not, the fastfield reader will returns the u64-value associated with the original
/// FastValue.
@@ -261,24 +220,13 @@ impl DynamicColumnHandle {
let column_bytes = self.file_slice.read_bytes()?;
match self.column_type {
ColumnType::Str | ColumnType::Bytes => {
let column: BytesColumn =
crate::column::open_column_bytes(column_bytes, self.format_version)?;
let column: BytesColumn = crate::column::open_column_bytes(column_bytes)?;
Ok(Some(column.term_ord_column))
}
ColumnType::IpAddr => {
let column = crate::column::open_column_u128_as_compact_u64(
column_bytes,
self.format_version,
)?;
Ok(Some(column))
}
ColumnType::Bool
| ColumnType::I64
| ColumnType::U64
| ColumnType::F64
| ColumnType::DateTime => {
let column =
crate::column::open_column_u64::<u64>(column_bytes, self.format_version)?;
ColumnType::Bool => Ok(None),
ColumnType::IpAddr => Ok(None),
ColumnType::I64 | ColumnType::U64 | ColumnType::F64 | ColumnType::DateTime => {
let column = crate::column::open_column_u64::<u64>(column_bytes)?;
Ok(Some(column))
}
}
@@ -286,38 +234,22 @@ impl DynamicColumnHandle {
fn open_internal(&self, column_bytes: OwnedBytes) -> io::Result<DynamicColumn> {
let dynamic_column: DynamicColumn = match self.column_type {
ColumnType::Bytes => {
crate::column::open_column_bytes(column_bytes, self.format_version)?.into()
}
ColumnType::Str => {
crate::column::open_column_str(column_bytes, self.format_version)?.into()
}
ColumnType::I64 => {
crate::column::open_column_u64::<i64>(column_bytes, self.format_version)?.into()
}
ColumnType::U64 => {
crate::column::open_column_u64::<u64>(column_bytes, self.format_version)?.into()
}
ColumnType::F64 => {
crate::column::open_column_u64::<f64>(column_bytes, self.format_version)?.into()
}
ColumnType::Bool => {
crate::column::open_column_u64::<bool>(column_bytes, self.format_version)?.into()
}
ColumnType::IpAddr => {
crate::column::open_column_u128::<Ipv6Addr>(column_bytes, self.format_version)?
.into()
}
ColumnType::Bytes => crate::column::open_column_bytes(column_bytes)?.into(),
ColumnType::Str => crate::column::open_column_str(column_bytes)?.into(),
ColumnType::I64 => crate::column::open_column_u64::<i64>(column_bytes)?.into(),
ColumnType::U64 => crate::column::open_column_u64::<u64>(column_bytes)?.into(),
ColumnType::F64 => crate::column::open_column_u64::<f64>(column_bytes)?.into(),
ColumnType::Bool => crate::column::open_column_u64::<bool>(column_bytes)?.into(),
ColumnType::IpAddr => crate::column::open_column_u128::<Ipv6Addr>(column_bytes)?.into(),
ColumnType::DateTime => {
crate::column::open_column_u64::<DateTime>(column_bytes, self.format_version)?
.into()
crate::column::open_column_u64::<DateTime>(column_bytes)?.into()
}
};
Ok(dynamic_column)
}
pub fn num_bytes(&self) -> ByteCount {
self.file_slice.len().into()
pub fn num_bytes(&self) -> usize {
self.file_slice.len()
}
pub fn column_type(&self) -> ColumnType {

9
columnar/src/iterable.rs
View File

@@ -1,7 +1,4 @@
use std::ops::Range;
use std::sync::Arc;
use crate::{ColumnValues, RowId};
pub trait Iterable<T = u64> {
fn boxed_iter(&self) -> Box<dyn Iterator<Item = T> + '_>;
@@ -20,9 +17,3 @@ where Range<T>: Iterator<Item = T>
Box::new(self.clone())
}
}
impl Iterable for Arc<dyn crate::ColumnValues<RowId>> {
fn boxed_iter(&self) -> Box<dyn Iterator<Item = u64> + '_> {
Box::new(self.iter().map(|row_id| row_id as u64))
}
}

50
columnar/src/lib.rs
View File

@@ -1,22 +1,3 @@
//! # 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)]
@@ -26,12 +7,10 @@ extern crate more_asserts;
#[cfg(all(test, feature = "unstable"))]
extern crate test;
use std::fmt::Display;
use std::io;
mod block_accessor;
mod column;
pub mod column_index;
mod column_index;
pub mod column_values;
mod columnar;
mod dictionary;
@@ -40,15 +19,12 @@ 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, EmptyColumnValues, MonotonicallyMappableToU128, MonotonicallyMappableToU64,
};
pub use column_values::{ColumnValues, MonotonicallyMappableToU128, MonotonicallyMappableToU64};
pub use columnar::{
merge_columnar, ColumnType, ColumnarReader, ColumnarWriter, HasAssociatedColumnType,
MergeRowOrder, ShuffleMergeOrder, StackMergeOrder, Version, CURRENT_VERSION,
MergeRowOrder, ShuffleMergeOrder, StackMergeOrder,
};
use sstable::VoidSSTable;
pub use value::{NumericalType, NumericalValue};
@@ -58,7 +34,7 @@ pub use self::dynamic_column::{DynamicColumn, DynamicColumnHandle};
pub type RowId = u32;
pub type DocId = u32;
#[derive(Clone, Copy, Debug)]
#[derive(Clone, Copy)]
pub struct RowAddr {
pub segment_ord: u32,
pub row_id: RowId,
@@ -95,17 +71,6 @@ 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)
@@ -113,12 +78,10 @@ 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),
@@ -131,6 +94,3 @@ impl Cardinality {
#[cfg(test)]
mod tests;
#[cfg(test)]
mod compat_tests;

716
columnar/src/tests.rs
View File

@@ -1,19 +1,10 @@
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, ColumnTypeCategory};
use crate::columnar::ColumnType;
use crate::dynamic_column::{DynamicColumn, DynamicColumnHandle};
use crate::value::{Coerce, NumericalValue};
use crate::{
BytesColumn, Cardinality, Column, ColumnarReader, ColumnarWriter, RowAddr, RowId,
ShuffleMergeOrder, StackMergeOrder,
};
use crate::value::NumericalValue;
use crate::{Cardinality, ColumnarReader, ColumnarWriter};
#[test]
fn test_dataframe_writer_str() {
@@ -21,12 +12,12 @@ fn test_dataframe_writer_str() {
dataframe_writer.record_str(1u32, "my_string", "hello");
dataframe_writer.record_str(3u32, "my_string", "helloeee");
let mut buffer: Vec<u8> = Vec::new();
dataframe_writer.serialize(5, &mut buffer).unwrap();
dataframe_writer.serialize(5, None, &mut buffer).unwrap();
let columnar = ColumnarReader::open(buffer).unwrap();
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(), 73);
assert_eq!(cols[0].num_bytes(), 158);
}
#[test]
@@ -35,12 +26,12 @@ fn test_dataframe_writer_bytes() {
dataframe_writer.record_bytes(1u32, "my_string", b"hello");
dataframe_writer.record_bytes(3u32, "my_string", b"helloeee");
let mut buffer: Vec<u8> = Vec::new();
dataframe_writer.serialize(5, &mut buffer).unwrap();
dataframe_writer.serialize(5, None, &mut buffer).unwrap();
let columnar = ColumnarReader::open(buffer).unwrap();
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(), 73);
assert_eq!(cols[0].num_bytes(), 158);
}
#[test]
@@ -49,7 +40,7 @@ fn test_dataframe_writer_bool() {
dataframe_writer.record_bool(1u32, "bool.value", false);
dataframe_writer.record_bool(3u32, "bool.value", true);
let mut buffer: Vec<u8> = Vec::new();
dataframe_writer.serialize(5, &mut buffer).unwrap();
dataframe_writer.serialize(5, None, &mut buffer).unwrap();
let columnar = ColumnarReader::open(buffer).unwrap();
assert_eq!(columnar.num_columns(), 1);
let cols: Vec<DynamicColumnHandle> = columnar.read_columns("bool.value").unwrap();
@@ -57,9 +48,7 @@ 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,]);
}
@@ -74,16 +63,14 @@ fn test_dataframe_writer_u64_multivalued() {
dataframe_writer.record_numerical(6u32, "divisor", 2u64);
dataframe_writer.record_numerical(6u32, "divisor", 3u64);
let mut buffer: Vec<u8> = Vec::new();
dataframe_writer.serialize(7, &mut buffer).unwrap();
dataframe_writer.serialize(7, None, &mut buffer).unwrap();
let columnar = ColumnarReader::open(buffer).unwrap();
assert_eq!(columnar.num_columns(), 1);
let cols: Vec<DynamicColumnHandle> = columnar.read_columns("divisor").unwrap();
assert_eq!(cols.len(), 1);
assert_eq!(cols[0].num_bytes(), 50);
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
@@ -97,7 +84,7 @@ fn test_dataframe_writer_ip_addr() {
dataframe_writer.record_ip_addr(1, "ip_addr", Ipv6Addr::from_u128(1001));
dataframe_writer.record_ip_addr(3, "ip_addr", Ipv6Addr::from_u128(1050));
let mut buffer: Vec<u8> = Vec::new();
dataframe_writer.serialize(5, &mut buffer).unwrap();
dataframe_writer.serialize(5, None, &mut buffer).unwrap();
let columnar = ColumnarReader::open(buffer).unwrap();
assert_eq!(columnar.num_columns(), 1);
let cols: Vec<DynamicColumnHandle> = columnar.read_columns("ip_addr").unwrap();
@@ -105,9 +92,7 @@ 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,
@@ -128,7 +113,7 @@ fn test_dataframe_writer_numerical() {
dataframe_writer.record_numerical(2u32, "srical.value", NumericalValue::U64(13u64));
dataframe_writer.record_numerical(4u32, "srical.value", NumericalValue::U64(15u64));
let mut buffer: Vec<u8> = Vec::new();
dataframe_writer.serialize(6, &mut buffer).unwrap();
dataframe_writer.serialize(6, None, &mut buffer).unwrap();
let columnar = ColumnarReader::open(buffer).unwrap();
assert_eq!(columnar.num_columns(), 1);
let cols: Vec<DynamicColumnHandle> = columnar.read_columns("srical.value").unwrap();
@@ -140,10 +125,8 @@ 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.index.get_cardinality(), Cardinality::Optional);
let DynamicColumn::I64(column_i64) = column else { panic!(); };
assert_eq!(column_i64.idx.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));
@@ -161,14 +144,12 @@ fn test_dictionary_encoded_str() {
columnar_writer.record_str(3, "my.column", "c");
columnar_writer.record_str(3, "my.column2", "different_column!");
columnar_writer.record_str(4, "my.column", "b");
columnar_writer.serialize(5, &mut buffer).unwrap();
columnar_writer.serialize(5, None, &mut buffer).unwrap();
let columnar_reader = ColumnarReader::open(buffer).unwrap();
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);
@@ -195,14 +176,12 @@ fn test_dictionary_encoded_bytes() {
columnar_writer.record_bytes(3, "my.column", b"c");
columnar_writer.record_bytes(3, "my.column2", b"different_column!");
columnar_writer.record_bytes(4, "my.column", b"b");
columnar_writer.serialize(5, &mut buffer).unwrap();
columnar_writer.serialize(5, None, &mut buffer).unwrap();
let columnar_reader = ColumnarReader::open(buffer).unwrap();
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();
@@ -231,656 +210,3 @@ 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 => (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 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)>]) -> 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, &mut buffer).unwrap();
ColumnarReader::open(buffer).unwrap()
}
fn build_columnar(docs: &[Vec<(&'static str, ColumnValue)>]) -> ColumnarReader {
build_columnar_with_mapping(docs)
}
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);
}
}
#[track_caller]
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),
}
}
}
}
// 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()) {
test_columnar_merge_and_remap(columnar_docs, shuffle_merge_order);
}
}
fn test_columnar_merge_and_remap(
columnar_docs: Vec<Vec<Vec<(&'static str, ColumnValue)>>>,
shuffle_merge_order: Vec<RowAddr>,
) {
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_ref: 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_ref[..],
&[],
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_and_remap_bug_1() {
let columnar_docs = vec![vec![
vec![
("c1", ColumnValue::Numerical(NumericalValue::U64(0))),
("c1", ColumnValue::Numerical(NumericalValue::U64(0))),
],
vec![],
]];
let shuffle_merge_order: Vec<RowAddr> = vec![
RowAddr {
segment_ord: 0,
row_id: 1,
},
RowAddr {
segment_ord: 0,
row_id: 0,
},
];
test_columnar_merge_and_remap(columnar_docs, shuffle_merge_order);
}
#[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_nanos(timestamp_micros)
DateTime::from_timestamp_micros(timestamp_micros)
}
}

7
common/Cargo.toml
View File

@@ -1,6 +1,6 @@
[package]
name = "tantivy-common"
version = "0.7.0"
version = "0.5.0"
authors = ["Paul Masurel <paul@quickwit.io>", "Pascal Seitz <pascal@quickwit.io>"]
license = "MIT"
edition = "2021"
@@ -9,17 +9,16 @@ documentation = "https://docs.rs/tantivy_common/"
homepage = "https://github.com/quickwit-oss/tantivy"
repository = "https://github.com/quickwit-oss/tantivy"
# See more keys and their definitions at https://doc.rust-lang.org/cargo/reference/manifest.html
[dependencies]
byteorder = "1.4.3"
ownedbytes = { version= "0.7", path="../ownedbytes" }
ownedbytes = { version= "0.5", path="../ownedbytes" }
async-trait = "0.1"
time = { version = "0.3.10", features = ["serde-well-known"] }
serde = { version = "1.0.136", features = ["derive"] }
[dev-dependencies]
binggan = "0.8.1"
proptest = "1.0.0"
rand = "0.8.4"

64
common/benches/bench.rs
View File

@@ -1,64 +0,0 @@
use binggan::{black_box, BenchRunner};
use rand::seq::IteratorRandom;
use rand::thread_rng;
use tantivy_common::{serialize_vint_u32, BitSet, TinySet};
fn bench_vint() {
let mut runner = BenchRunner::new();
let vals: Vec<u32> = (0..20_000).collect();
runner.bench_function("bench_vint", move |_| {
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]);
}
black_box(out);
});
let vals: Vec<u32> = (0..20_000).choose_multiple(&mut thread_rng(), 100_000);
runner.bench_function("bench_vint_rand", move |_| {
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]);
}
black_box(out);
});
}
fn bench_bitset() {
let mut runner = BenchRunner::new();
runner.bench_function("bench_tinyset_pop", move |_| {
let mut tinyset = TinySet::singleton(black_box(31u32));
tinyset.pop_lowest();
tinyset.pop_lowest();
tinyset.pop_lowest();
tinyset.pop_lowest();
tinyset.pop_lowest();
tinyset.pop_lowest();
black_box(tinyset);
});
let tiny_set = TinySet::empty().insert(10u32).insert(14u32).insert(21u32);
runner.bench_function("bench_tinyset_sum", move |_| {
assert_eq!(black_box(tiny_set).into_iter().sum::<u32>(), 45u32);
});
let v = [10u32, 14u32, 21u32];
runner.bench_function("bench_tinyarr_sum", move |_| {
black_box(v.iter().cloned().sum::<u32>());
});
runner.bench_function("bench_bitset_initialize", move |_| {
black_box(BitSet::with_max_value(1_000_000));
});
}
fn main() {
bench_vint();
bench_bitset();
}

49
common/src/bitset.rs
View File

@@ -1,10 +1,9 @@
use std::convert::TryInto;
use std::io::Write;
use std::{fmt, io};
use std::{fmt, io, u64};
use ownedbytes::OwnedBytes;
use crate::ByteCount;
#[derive(Clone, Copy, Eq, PartialEq)]
pub struct TinySet(u64);
@@ -387,8 +386,8 @@ impl ReadOnlyBitSet {
}
/// Number of bytes used in the bitset representation.
pub fn num_bytes(&self) -> ByteCount {
self.data.len().into()
pub fn num_bytes(&self) -> usize {
self.data.len()
}
}
@@ -696,3 +695,43 @@ mod tests {
}
}
}
#[cfg(all(test, feature = "unstable"))]
mod bench {
use test;
use super::{BitSet, TinySet};
#[bench]
fn bench_tinyset_pop(b: &mut test::Bencher) {
b.iter(|| {
let mut tinyset = TinySet::singleton(test::black_box(31u32));
tinyset.pop_lowest();
tinyset.pop_lowest();
tinyset.pop_lowest();
tinyset.pop_lowest();
tinyset.pop_lowest();
tinyset.pop_lowest();
});
}
#[bench]
fn bench_tinyset_sum(b: &mut test::Bencher) {
let tiny_set = TinySet::empty().insert(10u32).insert(14u32).insert(21u32);
b.iter(|| {
assert_eq!(test::black_box(tiny_set).into_iter().sum::<u32>(), 45u32);
});
}
#[bench]
fn bench_tinyarr_sum(b: &mut test::Bencher) {
let v = [10u32, 14u32, 21u32];
b.iter(|| test::black_box(v).iter().cloned().sum::<u32>());
}
#[bench]
fn bench_bitset_initialize(b: &mut test::Bencher) {
b.iter(|| BitSet::with_max_value(1_000_000));
}
}

114
common/src/byte_count.rs
View File

@@ -1,114 +0,0 @@
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

@@ -1,31 +1,25 @@
use std::fmt;
use std::io::{Read, Write};
use serde::{Deserialize, Serialize};
use time::format_description::well_known::Rfc3339;
use time::{OffsetDateTime, PrimitiveDateTime, UtcOffset};
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.
/// DateTime Precision
#[derive(
Clone, Copy, Debug, Hash, PartialEq, Eq, PartialOrd, Ord, Serialize, Deserialize, Default,
)]
#[serde(rename_all = "lowercase")]
pub enum DateTimePrecision {
/// Second precision.
pub enum DatePrecision {
/// Seconds precision
#[default]
Seconds,
/// Millisecond precision.
/// Milli-seconds precision.
Milliseconds,
/// Microsecond precision.
/// Micro-seconds precision.
Microseconds,
/// Nanosecond precision.
Nanoseconds,
}
/// A date/time value with nanoseconds precision.
/// A date/time value with microsecond precision.
///
/// This timestamp does not carry any explicit time zone information.
/// Users are responsible for applying the provided conversion
@@ -35,48 +29,31 @@ pub enum DateTimePrecision {
/// 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, Serialize, Deserialize)]
#[derive(Clone, Default, Copy, PartialEq, Eq, PartialOrd, Ord, Hash)]
pub struct DateTime {
// Timestamp in nanoseconds.
pub(crate) timestamp_nanos: i64,
// Timestamp in microseconds.
pub(crate) timestamp_micros: 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_nanos: seconds * 1_000_000_000,
timestamp_micros: seconds * 1_000_000,
}
}
/// Create new from UNIX timestamp in milliseconds
pub const fn from_timestamp_millis(milliseconds: i64) -> Self {
Self {
timestamp_nanos: milliseconds * 1_000_000,
timestamp_micros: milliseconds * 1_000,
}
}
/// Create new from UNIX timestamp in microseconds.
pub const fn from_timestamp_micros(microseconds: i64) -> Self {
Self {
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,
timestamp_micros: microseconds,
}
}
@@ -84,9 +61,9 @@ impl DateTime {
///
/// The given date/time is converted to UTC and the actual
/// time zone is discarded.
pub fn from_utc(dt: OffsetDateTime) -> Self {
let timestamp_nanos = dt.unix_timestamp_nanos() as i64;
Self { timestamp_nanos }
pub const fn from_utc(dt: OffsetDateTime) -> Self {
let timestamp_micros = dt.unix_timestamp() * 1_000_000 + dt.microsecond() as i64;
Self { timestamp_micros }
}
/// Create new from `PrimitiveDateTime`
@@ -100,27 +77,23 @@ impl DateTime {
/// Convert to UNIX timestamp in seconds.
pub const fn into_timestamp_secs(self) -> i64 {
self.timestamp_nanos / 1_000_000_000
self.timestamp_micros / 1_000_000
}
/// Convert to UNIX timestamp in milliseconds.
pub const fn into_timestamp_millis(self) -> i64 {
self.timestamp_nanos / 1_000_000
self.timestamp_micros / 1_000
}
/// Convert to UNIX timestamp in microseconds.
pub const fn into_timestamp_micros(self) -> i64 {
self.timestamp_nanos / 1_000
}
/// Convert to UNIX timestamp in nanoseconds.
pub const fn into_timestamp_nanos(self) -> i64 {
self.timestamp_nanos
self.timestamp_micros
}
/// Convert to UTC `OffsetDateTime`
pub fn into_utc(self) -> OffsetDateTime {
let utc_datetime = OffsetDateTime::from_unix_timestamp_nanos(self.timestamp_nanos as i128)
let timestamp_nanos = self.timestamp_micros as i128 * 1000;
let utc_datetime = OffsetDateTime::from_unix_timestamp_nanos(timestamp_nanos)
.expect("valid UNIX timestamp");
debug_assert_eq!(UtcOffset::UTC, utc_datetime.offset());
utc_datetime
@@ -143,34 +116,21 @@ impl DateTime {
}
/// Truncates the microseconds value to the corresponding precision.
pub fn truncate(self, precision: DateTimePrecision) -> Self {
pub fn truncate(self, precision: DatePrecision) -> Self {
let truncated_timestamp_micros = match precision {
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,
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,
};
Self {
timestamp_nanos: truncated_timestamp_micros,
timestamp_micros: 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

@@ -1,4 +1,3 @@
use std::fs::File;
use std::ops::{Deref, Range, RangeBounds};
use std::sync::Arc;
use std::{fmt, io};
@@ -6,7 +5,7 @@ use std::{fmt, io};
use async_trait::async_trait;
use ownedbytes::{OwnedBytes, StableDeref};
use crate::{ByteCount, HasLen};
use crate::HasLen;
/// Objects that represents files sections in tantivy.
///
@@ -33,62 +32,6 @@ 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> {
@@ -124,30 +67,6 @@ 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`.
@@ -297,11 +216,6 @@ 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 + Clone,
Self::Item: Clone,
K: PartialEq + Copy,
Self::Item: Copy,
{
GroupByIterator::new(self, key)
}
}
impl<I: Iterator> GroupByIteratorExtended for I {}
pub struct GroupByIterator<I, F, K: Clone>
pub struct GroupByIterator<I, F, K: Copy>
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: Clone>
struct GroupByShared<I, F, K: Copy>
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: Clone,
K: Copy,
{
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: Clone,
I::Item: Copy,
F: FnMut(&I::Item) -> K,
K: Clone,
K: Copy,
{
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()?.clone();
let value = *inner.iter.peek()?;
let key = (inner.group_by_fn)(&value);
let inner = self.inner.clone();
let group_iter = GroupIterator {
inner,
group_key: key.clone(),
group_key: key,
};
Some((key, group_iter))
}
}
pub struct GroupIterator<I, F, K: Clone>
pub struct GroupIterator<I, F, K: Copy>
where
I: Iterator,
F: FnMut(&I::Item) -> K,
@@ -110,10 +110,10 @@ where
group_key: K,
}
impl<I, F, K: PartialEq + Clone> Iterator for GroupIterator<I, F, K>
impl<I, F, K: PartialEq + Copy> Iterator for GroupIterator<I, F, K>
where
I: Iterator,
I::Item: Clone,
I::Item: Copy,
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()?.clone();
let peek_val = *inner.iter.peek()?;
if (inner.group_by_fn)(&peek_val) == self.group_key {
inner.iter.next()
} else {

144
common/src/json_path_writer.rs
View File

@@ -1,144 +0,0 @@
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]) };
/// Separates the json path and the value in
/// a JSON term binary representation.
pub const JSON_END_OF_PATH: u8 = 0u8;
pub const JSON_END_OF_PATH_STR: &str =
unsafe { std::str::from_utf8_unchecked(&[JSON_END_OF_PATH]) };
/// 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 with_expand_dots(expand_dots: bool) -> Self {
JsonPathWriter {
path: String::new(),
indices: Vec::new(),
expand_dots,
}
}
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.indices.len() > 1 {
self.path.push(JSON_PATH_SEGMENT_SEP as char);
}
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())
};
}
}
/// Set the end of JSON path marker.
#[inline]
pub fn set_end(&mut self) {
self.path.push_str(JSON_END_OF_PATH_STR);
}
/// 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.set_expand_dots(false);
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");
}
#[test]
fn test_json_path_expand_dots_enabled_pop_segment() {
let mut json_writer = JsonPathWriter::with_expand_dots(true);
json_writer.push("hello");
assert_eq!(json_writer.as_str(), "hello");
json_writer.push("color.hue");
assert_eq!(json_writer.as_str(), "hello\x01color\x01hue");
json_writer.pop();
assert_eq!(json_writer.as_str(), "hello");
}
}

10
common/src/lib.rs
View File

@@ -5,23 +5,20 @@ use std::ops::Deref;
pub use byteorder::LittleEndian as Endianness;
mod bitset;
mod byte_count;
mod datetime;
pub mod file_slice;
mod group_by;
pub mod json_path_writer;
mod serialize;
mod vint;
mod writer;
pub use bitset::*;
pub use byte_count::ByteCount;
pub use datetime::{DateTime, DateTimePrecision};
pub use datetime::{DatePrecision, DateTime};
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::{
read_u32_vint, read_u32_vint_no_advance, serialize_vint_u32, write_u32_vint, VInt, VIntU128,
deserialize_vint_u128, read_u32_vint, read_u32_vint_no_advance, serialize_vint_u128,
serialize_vint_u32, write_u32_vint, VInt, VIntU128,
};
pub use writer::{AntiCallToken, CountingWriter, TerminatingWrite};
@@ -116,7 +113,6 @@ 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

@@ -1,4 +1,3 @@
use std::borrow::Cow;
use std::io::{Read, Write};
use std::{fmt, io};
@@ -250,47 +249,11 @@ 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::*;
use super::{VInt, *};
use crate::serialize::BinarySerializable;
pub fn fixed_size_test<O: BinarySerializable + FixedSize + Default>() {
let mut buffer = Vec::new();
O::default().serialize(&mut buffer).unwrap();

79
common/src/vint.rs
View File

@@ -1,6 +1,8 @@
use std::io;
use std::io::{Read, Write};
use byteorder::{ByteOrder, LittleEndian};
use super::BinarySerializable;
/// Variable int serializes a u128 number
@@ -17,6 +19,26 @@ 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);
@@ -58,13 +80,17 @@ 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;
@@ -73,29 +99,25 @@ 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) = if val < START_2 {
(val | STOP_BIT, 1)
} else if val < START_3 {
(
let (res, num_bytes) = match val {
0..=STOP_1 => (val | STOP_BIT, 1),
START_2..=STOP_2 => (
(val & MASK_1) | ((val & MASK_2) << 1) | (STOP_BIT << (8)),
2,
)
} else if val < START_4 {
(
),
START_3..=STOP_3 => (
(val & MASK_1) | ((val & MASK_2) << 1) | ((val & MASK_3) << 2) | (STOP_BIT << (8 * 2)),
3,
)
} else if val < START_5 {
(
),
START_4..=STOP_4 => (
(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)
@@ -103,9 +125,9 @@ pub fn serialize_vint_u32(val: u32, buf: &mut [u8; 8]) -> &[u8] {
| ((val & MASK_5) << 4)
| (STOP_BIT << (8 * 4)),
5,
)
),
};
*buf = res.to_le_bytes();
LittleEndian::write_u64(&mut buf[..], res);
&buf[0..num_bytes]
}
@@ -151,7 +173,7 @@ pub fn read_u32_vint_no_advance(data: &[u8]) -> (u32, usize) {
(result, vlen)
}
/// Write a `u32` as a vint payload.
pub fn write_u32_vint<W: io::Write + ?Sized>(val: u32, writer: &mut W) -> io::Result<()> {
pub fn write_u32_vint<W: io::Write>(val: u32, writer: &mut W) -> io::Result<()> {
let mut buf = [0u8; 8];
let data = serialize_vint_u32(val, &mut buf);
writer.write_all(data)
@@ -223,6 +245,7 @@ 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];
@@ -261,7 +284,27 @@ 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}");
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);
}
#[test]

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

Binary file not shown.
Side by Side

Before

Width:  |  Height:  |  Size: 30 KiB

5
doc/src/index_sorting.md
View File

@@ -7,11 +7,6 @@
- [Other](#other)
- [Usage](#usage)
# Index Sorting has been removed!
More infos here:
https://github.com/quickwit-oss/tantivy/issues/2352
# Index Sorting
Tantivy allows you to sort the index according to a property.

71
examples/aggregation.rs
View File

@@ -7,12 +7,17 @@
// ---
use serde_json::{Deserializer, Value};
use tantivy::aggregation::agg_req::Aggregations;
use tantivy::aggregation::agg_req::{
Aggregation, Aggregations, BucketAggregation, BucketAggregationType, MetricAggregation,
RangeAggregation,
};
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, IndexWriter, TantivyDocument};
use tantivy::Index;
fn main() -> tantivy::Result<()> {
// # Create Schema
@@ -37,7 +42,7 @@ fn main() -> tantivy::Result<()> {
.set_index_option(IndexRecordOption::WithFreqs)
.set_tokenizer("raw"),
)
.set_fast(None)
.set_fast()
.set_stored();
schema_builder.add_text_field("category", text_fieldtype);
schema_builder.add_f64_field("stock", FAST);
@@ -132,10 +137,10 @@ fn main() -> tantivy::Result<()> {
let stream = Deserializer::from_str(data).into_iter::<Value>();
let mut index_writer: IndexWriter = index.writer(50_000_000)?;
let mut index_writer = index.writer(50_000_000)?;
let mut num_indexed = 0;
for value in stream {
let doc = TantivyDocument::parse_json(&schema, &serde_json::to_string(&value.unwrap())?)?;
let doc = schema.parse_document(&serde_json::to_string(&value.unwrap())?)?;
index_writer.add_document(doc)?;
num_indexed += 1;
if num_indexed > 4 {
@@ -187,11 +192,58 @@ fn main() -> tantivy::Result<()> {
//
let agg_req: Aggregations = serde_json::from_str(agg_req_str)?;
let collector = AggregationCollector::from_aggs(agg_req, Default::default());
let collector = AggregationCollector::from_aggs(agg_req, None);
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 res: Value = serde_json::to_value(agg_res)?;
let res1: Value = serde_json::to_value(agg_res)?;
// ### Aggregation Result
//
@@ -209,7 +261,8 @@ fn main() -> tantivy::Result<()> {
}
"#;
let expected_json: Value = serde_json::from_str(expected_res)?;
assert_eq!(expected_json, res);
assert_eq!(expected_json, res1);
assert_eq!(expected_json, res2);
// ### Request 2
//
@@ -234,7 +287,7 @@ fn main() -> tantivy::Result<()> {
let agg_req: Aggregations = serde_json::from_str(agg_req_str)?;
let collector = AggregationCollector::from_aggs(agg_req, Default::default());
let collector = AggregationCollector::from_aggs(agg_req, None);
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, IndexWriter, ReloadPolicy};
use tantivy::{doc, Index, 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: IndexWriter = index.writer(50_000_000)?;
let mut index_writer = 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 = TantivyDocument::default();
let mut old_man_doc = Document::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::OnCommitWithDelay)
.reload_policy(ReloadPolicy::OnCommit)
.try_into()?;
// We now need to acquire a searcher.
@@ -217,23 +217,9 @@ fn main() -> tantivy::Result<()> {
// the document returned will only contain
// a title.
for (_score, doc_address) in top_docs {
let retrieved_doc: TantivyDocument = searcher.doc(doc_address)?;
println!("{}", retrieved_doc.to_json(&schema));
let retrieved_doc = searcher.doc(doc_address)?;
println!("{}", schema.to_json(&retrieved_doc));
}
// 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(())
}

5
examples/custom_collector.rs
View File

@@ -11,10 +11,9 @@ use columnar::Column;
// ---
// Importing tantivy...
use tantivy::collector::{Collector, SegmentCollector};
use tantivy::index::SegmentReader;
use tantivy::query::QueryParser;
use tantivy::schema::{Schema, FAST, INDEXED, TEXT};
use tantivy::{doc, Index, IndexWriter, Score};
use tantivy::{doc, Index, Score, SegmentReader};
#[derive(Default)]
struct Stats {
@@ -143,7 +142,7 @@ fn main() -> tantivy::Result<()> {
// this example.
let index = Index::create_in_ram(schema);
let mut index_writer: IndexWriter = index.writer(50_000_000)?;
let mut index_writer = 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 \

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