This version of Meilisearch introduces two much-requested features:

• Nested field support
• Customize typo-tolerance

This version also brings improvements regarding the indexation speed issues. Please consider using this version and share your feedback on this GitHub discussion.

For more information on updating to the latest Meilisearch version, read our dedicated guide.

New features

Nested field support

Meilisearch now supports nested fields in documents.

{
    "id": 0,
    "person": {
         "firstname": "Guillaume"
    }
}

is internally represented as:

{
    "id": 0,
    "person.firstname": "Guillaume"
}

This allows users to express a nested field location in any API parameter accepting document attributes using the . (dot-notation). It can be expressed this way in any API parameter accepting document attributes.

e.g. "filterableAttributes": ["person.firstname"]

Done by @irevoire (#2211)

Customize typo-tolerance

You can customize typo-tolerance via the typoTolerance settings using either of the following endpoints:

• settings (/indexes/:index_uid/settings)
• typo-tolerance settings (/indexes/:index_uid/settings/typo-tolerance)

The typoTolerance setting object can contain the following properties:

• enabled: Whether the typo tolerance feature is enabled. (Default: true)
• disableOnAttributes: Disable typo tolerance on specified document attributes. (Default: [])
• disableOnWords: Disable typo tolerance for a set of query terms given during search. (Default: [])
• minWordSizeForTypos:
  • oneTypo: Customize the minimum size for a word to tolerate 1 typo. (Default: 5 )
  • twoTypos: Customize the minimum size for a word to tolerate 2 typos.(Default: 9)

Improve cropping/highlighting

It's now possible to customize the highlight and crop markers/tags when searching documents:

• highlightPreTag: specifiy the tag to put before the highlighted query terms. (Default: <em>)
• highlightPostTag: specifiy the tag to put after the highlighted query terms. (Default: </em>)
• cropMarker: set the crop marker to apply around the cropped part. (Default: ...)

Other improvements:

• cropLength is now a number of words instead of a number of characters

Done by @ManyTheFish (#2214)

Add Japanese support

New language support: Japanese, by setting up Lindera tokenizer.

Thanks to the community for this integration.

Done by @miiton, @djKooks, @mosuka and @ManyTheFish (#2185)

New instance options

New instance options for RAM and CPU usage during indexation:

• --max-indexing-memory (MEILI_MAX_INDEXING_MEMORY): Sets the maximum RAM size during indexation
• --max-indexing-threads (MEILI_MAX_INDEXING_THREADS): Sets the maximum number of threads available for use during indexation

Done by the amazing @2shiori17 in #2267

Breaking changes

• Stabilize and improve cropping behavior: cropLength is now a number of words instead of a number of characters (#2214) @ManyTheFish
• Typo-tolerance behavior changes: a typo on the first letter will count as 2 typos. This reduces the possibilities during the search regarding the typo tolerance feature and thus make Meilisearch faster as search time in this case. Indeed, users rarely get the first letter wrong. (milli#439) @MarinPostma
• Add a hard limit of 1000 retrievable documents during search (GET/POST /indexes/:uid/search) (#2281) @Kerollmops
  • When calling this endpoint, Meilisearch will not return more than 1000 documents. This limit is a security fix to avoid scraping the Meilisearch DB
  • If you still want to scrap your DB to retrieve all your documents, the GET /indexes/:uid/documents route is still usable since this limit is not applied to it
• Docker now starts from a different working directory, meaning you're supposed to mount your volume in /meili_data.

docker run -it --rm \
-p 7700:7700 \
-v $(pwd)/meili_data:/meili_data \
getmeili/meilisearch:latest

This new command makes the dumps and snapshot features work out-of-the-box with Docker by mounting all the Meilisearch data into meili_data instead of mounting separated volumes.
Done by @irevoire (#2178)

• Following the Docker good practice, the meilisearch binary is not in the current working directory anymore.
  When passing arguments to Meilisearch (like --no-analytics), you cannot call the binary using ./meilisearch but using meilisearch, /meilisearch, or /bin/meilisearch. We recommend using meilisearch.

✅ Correct

docker run -it --rm -p 7700:7700 getmeili/meilisearch:latest meilisearch --no-analytics

❌ Wrong

docker run -it --rm -p 7700:7700 getmeili/meilisearch:latest ./meilisearch --no-analytics

Done by @irevoire (#2178)

Other improvements

• Improve indexation time when inserting documents (#2203) @Kerollmops @MarinPostma
• Improve sortableAttributes and filterableAttributes error messages during search (#2140) @brunoocasali
• Upgrade dependencies of the Meilisearch, Milli and Tokenizer repositories (#2237) @Kerollmops

Fixes

• Fix panic when sending a document update without content type header (#2277) @MarinPostma
• Avoid embedding user input into error response (#2207) @CNLHC

Misc

• Simplify Docker usage (#2271) @Thearas
• Refactor authentication key extraction (#2253) @MarinPostma
• Cargo: use resolver 2 (#2238) @happysalada
• Use milli's heed dependency (#2228) @psvnlsaikumar
• Add test to validate cli (#2245) @MarinPostma
• Add tests with default CLI opts (#2254) @Kerollmops
• Minor change on our CIs (#2233) @curquiza
• Upgrade the config of issue management (#2263) @curquiza

❤️ Thanks again to our external contributors:

• Meilisearch: @2shiori17, @CNLHC, @happysalada, @psvnlsaikumar, and @Thearas
• Milli: @psvnlsaikumar
• Tokenizer: @djKooks, @miiton, @mosuka, and @Thearas