213: Fix the benchmarks script and names r=Kerollmops a=Kerollmops
The benchmarks compare script was not using the `--output` flag and was therefore failing the download of the JSON reports. We also modified the criterion benchmarks to use shorter names, it helps in looking at the benchmarks in the terminal.
Co-authored-by: Kerollmops <clement@meilisearch.com>
193: Fix primary key behavior r=Kerollmops a=MarinPostma
this pr:
- Adds early returns on empty document additions, avoiding error messages to be returned when adding no documents and no primary key was set.
- Changes the primary key inference logic to match that of legacy meilisearch.
close#194
Co-authored-by: Marin Postma <postma.marin@protonmail.com>
Co-authored-by: marin postma <postma.marin@protonmail.com>
204: Decorrelate Distinct, Asc/Desc, Filterable fields from the faceted fields r=Kerollmops a=Kerollmops
This PR decorrelates the fields that need to be stored in facet databases (big inverted indexes for fast access) from the filterable fields, the previously named faceted fields are now named filterable fields and are the union of the distinct attribute, all the Asc/Desc criteria and, the filterable fields.
I added two tests to make sure that the engine was correctly generating the faceted databases when a distinct attribute or an Asc/Desc criteria were added, and one to make sure that it was impossible to filter on a non-filterable field even if it was a faceted one.
Note that the `AttributesForFacetting` has also been renamed into `FilterableAttributes`. But it will be the Transplant's job to do that on the API, this change is only visible to the milli's library users.
- Related to https://github.com/meilisearch/transplant/issues/187.
- Fixes#161 by returning the documents that don't have the Asc/Desc field at the end of the bucket.
- Fixes#168.
- Fixes#152.
Co-authored-by: Kerollmops <clement@meilisearch.com>
Co-authored-by: Marin Postma <postma.marin@protonmail.com>
Co-authored-by: many <maxime@meilisearch.com>
158: Implements the dumps r=irevoire a=irevoire
closes#20
divergence from legacy meilisearch:
- dump v2 added, support loading of pending updates (only works dumps created from v2)
- added time stamps to the dump info
- Dump info are only persisted in an internal data structure, and they are not fetched from fs on demand anymore. This was a potential security flaw. This means that the dump infos are flushed on every restart.
Co-authored-by: tamo <tamo@meilisearch.com>
Co-authored-by: Marin Postma <postma.marin@protonmail.com>
