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>
4786: Update dependencies r=Kerollmops a=irevoire
# Pull Request
## Related issue
Fixes#4753
## What does this PR do?
- Update all dependencies except rustls
- [x] Release charabia
- [x] Update charabia
- [x] Double check that the docker build works after updating charabia
Co-authored-by: Tamo <tamo@meilisearch.com>
Co-authored-by: Clément Renault <clement@meilisearch.com>
4713: Speed up facet distribution r=ManyTheFish a=Kerollmops
This PR is akin to #4682, but this time, the same logic is applied to the facets. Bitmaps are not decoded, and we do an intersection on the bytes with the search candidates instead of materializing the RoaringBitmap to destroy it just after the operation.
A prospect raised some slow requests when performing facet searches, and I found out that the disk optimization intersection wasn't performed on the facets.
Co-authored-by: Clément Renault <clement@meilisearch.com>
4646: Reduce `Transform`'s disk usage r=Kerollmops a=Kerollmops
This PR implements what is described in #4485. It reduces the number of disk writes and disk usage.
Co-authored-by: Clément Renault <clement@meilisearch.com>
4621: Bring back changes from v1.8.0 into main r=curquiza a=curquiza
Co-authored-by: ManyTheFish <many@meilisearch.com>
Co-authored-by: Tamo <tamo@meilisearch.com>
Co-authored-by: meili-bors[bot] <89034592+meili-bors[bot]@users.noreply.github.com>
Co-authored-by: Clément Renault <clement@meilisearch.com>
4597: Fix embeddings settings update r=ManyTheFish a=ManyTheFish
# Pull Request
- add some conditions reducing the work done when changing the settings
- add some benchmarks on embedders
## Related issue
Fixes#4585
Co-authored-by: ManyTheFish <many@meilisearch.com>
4543: Bring back changes from v1.7.4 into main r=Kerollmops a=dureuill
Co-authored-by: Louis Dureuil <louis@meilisearch.com>
Co-authored-by: meili-bors[bot] <89034592+meili-bors[bot]@users.noreply.github.com>
Co-authored-by: dureuill <dureuill@users.noreply.github.com>
4466: Implements the search cutoff r=irevoire a=irevoire
# Pull Request
## Related issue
Fixes https://github.com/meilisearch/meilisearch/issues/4488
## What does this PR do?
- Adds a cutoff to the bucket sort after 150ms has been spent
- Adds a new setting to customize the default value of 150ms
- When the time is exceeded, we exit early with what we had the time to sort
- If the cutoff has been reached, the search details are updated with a new `Skip` ranking details for the ranking rules that were skipped
- Adds analytics to measure the total number of degraded search requests
- Adds the number of degraded search requests to the Prometheus metrics and Grafana dashboard
- The cutoff **must not** skip the filters; otherwise, we would leak documents to people who don’t have the right to see them
Co-authored-by: Tamo <tamo@meilisearch.com>
Co-authored-by: Louis Dureuil <louis@meilisearch.com>
4476: Make the `/facet-search` route use the `sortFacetValuesBy` setting r=irevoire a=Kerollmops
This PR fixes#4423 by ensuring that the `/facet-search` route uses the `sortFacetValuesBy` setting.
Note for the documentation team (to be moved in the tracking issue): Using the new `sortFacetValuesBy` setting can slow down the facet-search requests as Meilisearch iterates over the whole list of facet values and computes the count of documents on every entry. That is hardly or even impossible to optimize correctly.
### TODO
- [x] Create a custom HashMap wrapper for the facet `OrderBy` settings.
This wrapper will return the `OrderBy` setting of the facet, if not defined will use the default `*` one, and if not there either (strange) will fall back on the lexicographic one.
- [x] Create a `ValuesCollection` wrapper that implements the logic for the lexicographic and count order by.
- [x] Use it when there is no search query.
- [x] Use it when there is a search query with and without allowed typos.
- [x] Do not change the original logic, only use a wrapper.
- [x] Add tests
Co-authored-by: Clément Renault <clement@meilisearch.com>
4456: Add Ollama as an embeddings provider r=dureuill a=jakobklemm
# Pull Request
## Related issue
[Related Discord Thread](https://discord.com/channels/1006923006964154428/1211977150316683305)
## What does this PR do?
- Adds Ollama as a provider of Embeddings besides HuggingFace and OpenAI under the name `ollama`
- Adds the environment variable `MEILI_OLLAMA_URL` to set the embeddings URL of an Ollama instance with a default value of `http://localhost:11434/api/embeddings` if no variable is set
- Changes some of the structs and functions in `openai.rs` to be public so that they can be shared.
- Added more error variants for Ollama specific errors
- It uses the model `nomic-embed-text` as default, but any string value is allowed, however it won't automatically check if the model actually exists or is an embedding model
Tested against Ollama version `v0.1.27` and the `nomic-embed-text` model.
## PR checklist
Please check if your PR fulfills the following requirements:
- [x] Does this PR fix an existing issue, or have you listed the changes applied in the PR description (and why they are needed)?
- [x] Have you read the contributing guidelines?
- [x] Have you made sure that the title is accurate and descriptive of the changes?
Co-authored-by: Jakob Klemm <jakob@jeykey.net>
Co-authored-by: Louis Dureuil <louis.dureuil@gmail.com>
Instead of the user manually specifying the model dimensions it will now automatically get determined
Just like with hf.rs the word "test" gets embedded to determine the dimensions of the output
Add a dedicated error type for if the model doesn't exist (don't automatically pull it though) and set the fault of that error to be the user
4462: Divide threshold by ten r=dureuill a=ManyTheFish
Change the facet incremental vs bulk indexing threshold to better fit our user needs, it might be changed in the future if we have more insights
Co-authored-by: ManyTheFish <many@meilisearch.com>
Initial prototype of Ollama embeddings actually working, error handlign / retries still missing.
Allow model to be any String and require dimensions parameter
Fixed rustfmt formatting issues
There were some formatting issues in the initial PR and this should not make the changes comply with the Rust style guidelines
Because I accidentally didn't follow the style guide for commits in my commit messages I squashed them into one to comply
4391: Tracing r=dureuill a=irevoire
# Pull Request
- [ ] Hide the parameters of the process batch
- [x] Make actix-web trace every call on every route
- [x] Remove all `env_logger`/`logs` dependencies
- [x] Be able to enable or disable the memory measurement using the `/logs` route parameters
See the following product discussion: https://github.com/orgs/meilisearch/discussions/721
Supersedes https://github.com/meilisearch/meilisearch/pull/4338
## Related issue
Fixes https://github.com/meilisearch/meilisearch/issues/4317
## What does this PR do?
Update the format of the logs from:
```
[2024-02-06T14:54:11Z INFO actix_server::builder] starting 10 workers
```
to
```
2024-02-06T13:58:14.710803Z INFO actix_server::builder: 200: starting 10 workers
```
First, run meilisearch with the route enabled via the feature flag:
- `cargo run --experimental-enable-logs-route`
- Or at runtime by sending the following payload:
```
curl \
-X PATCH 'http://localhost:7700/experimental-features/' \
-H 'Content-Type: application/json' \
--data-binary '{
"logsRoute": true
}'
```
Then gather data from meilisearch by calling for example:
```
curl \
-X POST http://localhost:7700/logs \
-H 'Content-Type: application/json' \
--data-binary '{
"mode": "fmt",
"target": "milli=trace"
}'
```
Once your operation is over, tell meilisearch to stop the route:
```
curl \
-X DELETE http://localhost:7700/logs
```
----
In the case you’re profiling code, you will be interested by the next command that converts the output of the route to a format that the firefox profiler can understand.
```bash
cargo run --release --bin trace-to-firefox -- 2024-01-17_17:07:55-indexing-trace.json
```
Then go to https://profiler.firefox.com and load it.
Note that we can also share the profiles using the https://share.firefox.dev website.
Co-authored-by: Louis Dureuil <louis@meilisearch.com>
Co-authored-by: Clément Renault <clement@meilisearch.com>
Co-authored-by: Tamo <tamo@meilisearch.com>
4388: Cap the maximum memory of the grenad sorters r=curquiza a=Kerollmops
This PR clamps the memory usage of the grenad sorters to a reasonable maximum. Grenad sorters are opened on multiple threads at a time. This can result in higher memory usage than expected, even though it shouldn't consume more than the memory available.
Fixes#4152.
Co-authored-by: Clément Renault <clement@meilisearch.com>
-> make sure the settings change is rejected or the settings task fails when the specified model doesn't support
overriding `dimensions` and the passed `dimensions` differs from the model's default dimensions.
