Commit Graph

964 Commits

Author SHA1 Message Date
Clément Renault
27b4cab857
Extract and write the documents and words fst in the database 2024-09-04 09:59:19 +02:00
Clément Renault
52d32b4ee9
Move the channel sender in the closure to stop the merger thread 2024-09-03 16:08:33 +02:00
ManyTheFish
da61408e52 Remove unimplemented from document changes 2024-09-03 15:14:16 +02:00
ManyTheFish
fe69385bd7 Fix tokenizer test 2024-09-03 14:24:37 +02:00
Clément Renault
c1557734dc
Use the GlobalFieldsIdsMap everywhere and write it to disk
Co-authored-by: Dureuill <louis@meilisearch.com>
Co-authored-by: ManyTheFish <many@meilisearch.com>
2024-09-03 12:01:01 +02:00
ManyTheFish
c50d3edc4a Integrate first searchable exctrator 2024-09-03 11:02:39 +02:00
Clément Renault
5369bf4a62
Change some lifetimes 2024-09-02 19:51:22 +02:00
Clément Renault
bcb1aa3d22
Find a temporary solution to par into iter on an HashMap
Spoiler: Do not use an HashMap but drain it into a Vec
2024-09-02 19:39:48 +02:00
Clément Renault
9b7858fb90
Expose the new indexer 2024-09-02 15:21:59 +02:00
Clément Renault
ab01679a8f
Remove the useless option from the document changes 2024-09-02 15:21:00 +02:00
Clément Renault
521775f788
I push for Many 2024-09-02 15:10:21 +02:00
Clément Renault
72e7b7846e
Renaming the indexers 2024-09-02 14:42:27 +02:00
Clément Renault
6526ce1208
Fix the merging of documents 2024-09-02 14:41:20 +02:00
Clément Renault
e639ec79d1
Move the indexers into their own modules 2024-09-02 10:42:19 +02:00
Clément Renault
bb885a5810
Fix the merge for roaring bitmap 2024-09-01 23:20:19 +02:00
Clément Renault
b625d31c7d
Introduce the PartialDumpIndexer indexer that generates document ids in parallel 2024-08-30 15:07:21 +02:00
Clément Renault
6487a67f2b
Introduce the ConcurrentAvailableIds struct and rename the other to AvailableIds 2024-08-30 15:06:50 +02:00
Clément Renault
271ce91b3b
Add the rayon Threadpool to the index function parameter 2024-08-30 14:34:24 +02:00
Clément Renault
54f2eb4507
Remove duplication of grenad merger 2024-08-30 14:34:05 +02:00
Clément Renault
794ebcd582
Replace grenad with the new grenad various-improvement branch 2024-08-30 11:53:59 +02:00
Clément Renault
b7c77c7a39
Use the latest version of the obkv crate 2024-08-30 11:53:59 +02:00
Clément Renault
0c57cf7565
Replace obkv with the temporary new version of it 2024-08-30 11:53:58 +02:00
Clément Renault
27df9e6c73
Introduce the indexer::index function that runs the indexation 2024-08-30 11:53:58 +02:00
Clément Renault
45c060831e
Introduce typed channels and the merger loop 2024-08-30 11:53:58 +02:00
Clément Renault
874c1ac538
First channels types 2024-08-30 11:53:58 +02:00
Clément Renault
e6ffa4d454
Implement the document merge function for the replace method 2024-08-30 11:53:58 +02:00
Clément Renault
637a9c8bdd
Implement the document merge function for the update method 2024-08-30 11:53:58 +02:00
Louis Dureuil
c683fa98e6
WIP
Co-authored-by: Kerollmops <clement@meilisearch.com>
Co-authored-by: ManyTheFish <many@meilisearch.com>
2024-08-30 11:53:57 +02:00
meili-bors[bot]
ee62d9ce30
Merge #4845
4845: Fix perf regression facet strings r=ManyTheFish a=dureuill

Benchmarks between v1.9 and v1.10 show a performance regression of about x2 (+3dB regression) for most indexing workloads (+44s for hackernews).

[Benchmark interpretation in the engine weekly meeting](https://www.notion.so/meilisearch/Engine-weekly-4d49560d374c4a87b4e3d126a261d4a0?pvs=4#98a709683276450295fcfe1f8ea5cef3).

- Initial investigation pointed to #4819 as the origin of the regression.
- Further investigation points towards the hypernormalization of each facet value in `extract_facet_string_docids`
- Most of the slowdown is in `normalize_facet_strings`, and precisely in `detection.language()`.

This PR improves the situation (-10s compared with `main` for hackernews, so only +34s regression compared with `v1.9`) by skipping normalization when it can be skipped.

I'm not sure how to fix the root cause though. Should we skip facet locale normalization for now? Cc `@ManyTheFish` 

---

Tentative resolution options:

1. remove locale normalization from facet. I'm not sure why this is required, I believe we weren't doing this before, so maybe we can stop doing that again.
2. don't do language detection when it can be helped: won't help with the regressions in benchmark, but maybe we can skip language detection when the locales contain only one language?
3. use a faster language detection library: `@Kerollmops` told me about https://github.com/quickwit-oss/whichlang which bolsters x10 to x100 throughput compared with whatlang. Should we consider replacing whatlang with whichlang? Now I understand whichlang supports fewer languages than whatlang, so I also suggest:
4. use whichlang when the list of locales is empty (autodetection), or when it only contains locales that whichlang can detect. If the list of locales contains locales that whichlang *cannot* detect, **then** use whatlang instead.

---

> [!CAUTION]
> this PR contains a commit that adds detailed spans, that were used to detect which part of `extract_facet_string_docids` was taking too much time. As this commit adds spans that are called too often and adds 7s overhead, it should be removed before landing.

Co-authored-by: Louis Dureuil <louis@meilisearch.com>
Co-authored-by: ManyTheFish <many@meilisearch.com>
2024-08-19 06:29:48 +00:00
ManyTheFish
0f965d3574 Remove hotloop's spans 2024-08-14 14:33:36 +02:00
ManyTheFish
ade54493ab Only detect language for a facet if several locales have been specified by the user in the settings 2024-08-14 12:03:52 +02:00
Louis Dureuil
c3cdc407ec
Avoid unnecessary clone() 2024-08-08 14:57:02 +02:00
Louis Dureuil
2f10273d14
Group by normalized values, make sure you don't remove a value where there remains at still one value that normalizes towards it 2024-08-08 14:02:53 +02:00
Louis Dureuil
e64d0e0ca8
use insert instead of push for bitmaps 2024-08-01 18:32:45 +02:00
Louis Dureuil
0e68718027
Add detailed spans 2024-07-31 13:05:47 +02:00
Louis Dureuil
7c3fc8c655
Split settings and document facet string extractions 2024-07-31 10:57:46 +02:00
Louis Dureuil
8acd3f50bb
skip normalization when the locales and values are the same 2024-07-31 09:53:00 +02:00
Louis Dureuil
d4ea7cc2a9
fix clippy 👉👈 2024-07-25 12:10:32 +02:00
Louis Dureuil
2413592bbf
Display docid when there are documents without manual embeddings for a manual embedder 2024-07-25 12:10:32 +02:00
Louis Dureuil
553440632e
Introduce Setting::some_or_not_set 2024-07-25 12:01:52 +02:00
Louis Dureuil
7a347966da
Allow explicit dimensions for ollama 2024-07-25 12:01:51 +02:00
Louis Dureuil
4654d51e05
Add custom headers for REST embedder 2024-07-25 12:01:51 +02:00
ManyTheFish
a918561ac1
Fix PR comments 2024-07-25 10:52:56 +02:00
ManyTheFish
04fa44e7eb
Implement localized attributes settings 2024-07-25 10:51:27 +02:00
ManyTheFish
cc02920f2b
Update charabia 2024-07-25 10:51:27 +02:00
Tamo
988552e178
add tests on the rest embedder 2024-07-24 14:34:17 +02:00
Louis Dureuil
0d8199f3b7
Change parameters in milli settings 2024-07-24 14:34:17 +02:00
Louis Dureuil
24240934f9
Improve errors when indexing documents with a user provided embedder 2024-07-16 13:39:01 +02:00
Louis Dureuil
65d0c32aa7
Allow overriding OpenAI's url 2024-07-16 13:39:00 +02:00
Clément Renault
6e80364c50
Apply review comments 2024-07-11 11:00:27 +02:00