Meilisearch v1.56.2 removes a leftover debug statement that could impede performance and was a regression from v1.56.1. Upgrading from previous versions of Meilisearch is strongly recommended.

It also updates several dependencies.

🦋 Fixes

• remove unwanted log & warn about future unwanted dbg by @dureuill in #6725

🔩 Miscellaneous

• Bump docker/build-push-action from 6.19.2 to 7.4.0 by @dependabot[bot] in #6712
• Bump docker/login-action from 4.2.0 to 4.7.0 by @dependabot[bot] in #6714
• Fix outdated CA certificates in the Docker images by @curquiza in #6705
• Bump actions/download-artifact from 8.0.1 to 8.0.2 by @dependabot[bot] in #6713
• Bump dart-lang/setup-dart from 1.7.2 to 1.8.1 by @dependabot[bot] in #6711
• Bump mongodb to v9.0 in rails SDKs tests by @Strift in #6716

Full Changelog: v1.56.1...v1.56.2

Meilisearch v1.56.1 introduces important security fixes. Updating from v1.56.0 is recommended for all users.

🦋 Fixes

• Fix issue with ranking scores greater than 1 by @dureuill in #6703

• Make sure that all nested fields are inserted in field id map by @dureuill in #6700

  • Fixes a panic when modifying the list of searchable attributes after an existing document has been updated to insert new nested fields.

  • Existing databases encountering this issue can be fixed by adding a new document containing the new nested fields.

  • Fixes #6636 and fixes #6657

🔒 Security

• Fix the memory consumption and runtime of long filters by @dureuill in #6695
• Bump rustls from 0.23.45 to 0.23.46 by @dependabot[bot] in #6706

🔩 Miscellaneous

• Clamp map_size to page size when removing tasks during snapshot creation by @mavrodikomsd-eng in #6699
• Reduce the number of cargo flaky iterations to 60 by @dureuill in #6689
• Fix Debian CI by @curquiza in #6698
• Bump mislav/bump-homebrew-formula-action from 3.6 to 4.2 by @dependabot[bot] in #6567
• Bump actions/setup-node from 6.5.0 to 7.1.0 by @dependabot[bot] in #6569
• Bump actions/setup-python from 6.3.0 to 7.1.0 by @dependabot[bot] in #6570

• Bump actions/checkout from 5.1.0 to 7.0.1 by @dependabot[bot] in #6568

New Contributors

• @mavrodikomsd-eng made their first contribution in #6699

Full Changelog: v1.56.0...v1.56.1

Meilisearch v1.55.1 introduces important security fixes. Upgrading from v1.55.0 and earlier is recommended for all users.

See the release notes for Meilisearch v1.56.1 for details.

Full Changelog: v1.55.0...v1.55.1

✨ Enhancement

• Introduce sampledCount to compute the facet distribution by @Kerollmops in #6649
  It is a new sorting option when retrieving the facet distribution. It is less precise than by count but provides a rough estimate of the facet distribution. Note that the count of documents for each facet is exhaustive but the list of facets is not and some facets that would have been listed by using the by count method can be ignored/skipped.
• Give up indexing when the embedder can't handle the load by @ManyTheFish in #6681
  • An adaptive strategy for rate limiting: For small usage with few documents, we accept a slow embedder service, and for big usage with updates containing thousands of documents, Meilisearch will give up early
  • A better outage handling, where Meilisearch will give up early if the embedder server is unavailable for too long
  • Support the Retry-After header when the value is in seconds, avoiding useless retries in a short timespan
• Optimization of general search performance by @itsamine27 in #6571

🪲 Bug fixes

• Fix wrong binary quantization state after deleting embedders by @dureuill in #6679

🔩 Miscellaneous

• Improve code correctness by using the latest version of heed by @Kerollmops in #6687
• Update README by @mayasssh in #6692
• Disallow AI Contributions by @Kerollmops in #6684
• Fix empty request body section in the settings API reference by @curquiza in #6678
• Bump rustls from 0.23.43 to 0.23.45 by @dependabot[bot] in #6672

🌈 Improvements

Preview the effect of a dynamic search rule

Apply an inline-specified dynamic search rule to a search query.

Search query objects in the following endpoints now have an additional, optional inlineRule parameter.

When present and not null, then Meilisearch will check if query satisfies the conditions of the dynamic search rule specified in inlineRule, and if so will apply the actions of the rule as if it had been sent to the PATCH /dynamicSearchRule/{:uid} endpoint.

Additionally, if the Meili-Include-Metadata header is present in the request with the value true, then the metadata in the response will contain information about whether the conditions of the inline rule were satisfied and if not why.

Meilisearch Cloud will use this backend feature to implement a preview UI for dynamic search rules.

Using inlineRule requires Meilisearch Enterprise Edition and the dynamicSearchRules experimental feature.

By @dureuill in #6669

Details

1. There is a new InlineRule object with the following field:
   1. uid, string, mandatory: unique ID of the rule. If there is already a rule with this id, then the inline rule will replace the existing rule for the duration of this search query.
   2. active, boolean, optional, defaults to true: whether the rule is active. The actions of an inactive inline rule never apply to the search, but its conditions are still checked
   3. conditions: RuleConditions object, identical to the conditions field in the body of PATCH /dynamic-search-rules/{:uid}
   4. actions: RuleActions object, identical to the actions field in the body of PATCH /dynamic-search-rules/{:uid}
   5. precedence: integer, optional, identical to the precedence field in the body of PATCH /dynamic-search-rules/{:uid}
2. The POST /indexes/{:indexUid}/search body has an additional, optional (defaults to null) inlineRule field which is an InlineRule object.
3. In the POST /multi-search body, the objects inside the queries array have an additional, optional (defaults to null) inlineRule field which is an InlineRule object.

Examples

// POST /indexes/movies/search
{
  "q": "super hero",
  "inlineRule": {
    "uid": "user-likes-batman",
    "actions": {
      "scale": [
        {
          // filter recent Batman movies
          "filter": "hero = batman AND release_time = fresh",
          "weight": 1.2 // moderate relevancy boost
        }
      ]
    }
  }
}

// POST /multi-search -H 'Meili-Include-Metadata: true`

{
  "queries": [
    // base query without inline rule
    {
      "q": "fashion",
      "filter": "color = blue AND category = shirt"
    },
    // same query with inline rule
    {
      "q": "fashion",
      "filter": "color = blue AND category = shirt",
      "inlineRule": {
        "uid": "special-sales",
        "conditions": {
          "query": {
            "words": "pants",
          },
            "filter": {
              "values": {
                "color":  "yellow",
                "stock_intent": "destock"
            }
          }
        }
      }
    }
  ]
}

{
  // [ ... ] some other fields
  "results": [
    // results without the rule
    {
      "hits": [ ... ] // documents returned without the rule
    },
    // results with the rule
    {
      // [...] some other fields
      "hits": [...], // documents returned with the rule. Here the same, because the conditions of the rules are not satisfied
      "metadata": {
        // [...] some other fields
        "inlineRuleConditionOutcomes":  {
          "satisfiesActiveCondition": true,
          "satisfiesQueryWordsCondition": {
            "missingWord": "pants"
          },
        "satifiesFilterCondition": {
          "missingConstraintOnField": {
            "field":  "stock_intent"
          },
          "unmetConstraintOnField": 
            "field": "color"
          }
        }
      }
    }
  ]
}

Improve showPerformanceDetails at search time

When setting showPerformanceDetails to true in a search request, the response's performanceDetails will now display an entry for the facet distribution computation.

curl -X POST 'http://localhost:7700/indexes/movies/search' \
  -H 'Content-Type: application/json' \
  -d '{
    "q": "batman",
    "facets": ["genres"],
    "showPerformanceDetails": true,
    "limit": 3
  }'

{
    "hits": [
		# ...
    ],
    "query": "batman",
    "processingTimeMs": 4,
    "limit": 3,
    "offset": 0,
    "estimatedTotalHits": 68,
    "facetDistribution": {
        "genres": {
			#...
        }
    },
    "facetStats": {},
    "requestUid": "01a12f80-0e1b-7632-b4f3-c8d4914048c9",
    "performanceDetails": {
        "wait in queue": "68.58µs",
        "preprocess filters > prepare": "100.25µs",
        "preprocess filters": "100.42µs",
        "process > prepare": "115.54µs",
        "process > send to remote": "102.71µs",
        "process > execute local > load field ids map": "30.88µs",
        "process > execute local > query[0] > tokenize query": "265.38µs",
        "process > execute local > query[0] > evaluate query": "890.96µs",
        "process > execute local > query[0] > keyword ranking": "425.04µs",
        "process > execute local > query[0]": "2.95ms",
        "process > execute local > pin hits": "9.92µs",
        "process > execute local > format": "94.62µs",
        "process > execute local > facet distribution > compute facet distribution": "315.75µs",
        "process > execute local > facet distribution > compute facet stats": "27.71µs",
        "process > execute local > facet distribution": "432.62µs",
        "process > execute local": "3.81ms",
        "process > wait for remote": "50.96µs",
        "process > merge": "905.62µs",
        "process": "5.07ms"
    }
}

        "process > execute local > facet distribution > compute facet distribution": "315.75µs",
        "process > execute local > facet distribution > compute facet stats": "27.71µs",
        "process > execute local > facet distribution": "432.62µs",

By @ManyTheFish in #6644

🦋 Fixes

Avoid panic at search time during formatting

Documents containing raw unicode codepoints (e.g. \u03b1) would cause a panic at search time when formatting results for highlighting when using some segmenters.

A failure at the segmenter's level no longer causes a panic.

By @dureuill in #6673

Fix search performance of ranking score threshold

To avoid a performance sink when using the rankingScoreThreshold parameter in a search, Meilisearch now guarantees that the search will stop after evaluating the first document whose score is below the threshold. Previously Meilisearch would try to evaluate more documents, which could lead to evaluating all documents in the database, causing a very long search time.

By @ManyTheFish in #6661

Prevent unsupported binaryQuantization deactivation in embedder settings

Since the exact value of embeddings is not retained when the embedder is configured for binary quantize embeddings, disabling the binary quantization on an index is not allowed.

Unfortunately, an oversight in the API allowed to accidentally disable binary quantization by passing binaryQuantization: null in a setting update. This now results in an error.

By @dureuill in #6656

🔩 Miscellaneous

• Maintainability: replace some unreachable!() code paths with static typing by @dureuill in #6670
• Add GHA to enforce benchmarks label on each PR by @curquiza in #6663
• Add link to MCP guide in generated API reference by @Strift in #6650
• Add link to agentic search demo & fix links by @Strift in #6655
• Small changes to CONTRIBUTING.md by @dureuill in #6668
• Bump actions/setup-java from 5.7.0 to 6.0.1 by @dependabot[bot] in #6662
• Fix OpenAPI docs by @curquiza in #6667
• CI: Fix flaky tests by @dureuill in #6674

Full Changelog: v1.54.3...v1.55.0

This release contains an important stability fix for Meilisearch. Update from v1.54.2 is recommended.

See #6658 for details

This release contains an important stability fix for Meilisearch. Update from v1.53.2 is recommended.

See #6658 for details

This release contains an important stability fix for Meilisearch. Update from v1.52.3 is recommended.

See #6658 for details

Meilisearch v1.54.2 contains an important fix for users of the dynamicSearchRule experimental feature.

🦋 Fixes

• Trying to process, display, delete or cancel dynamic search rules update tasks that were created in Meilisearch v1.53 or prior no longer causes internal errors. By @dureuill in #6651. Thanks to @connorabbas for their report in #6646 ❤️

🪲 Bug fixes

• Make the embedder setting change succeed when the embedder's API key is expired by @ManyTheFish in #6640
• Ignore geo errors when updating or deleting existing docs by @dureuill in #6638

🔩 Miscellaneous

• Aggregate the analytics from the MCP route by @Kerollmops in #6639