创建 Atlas Search 索引

Atlas Search is a fully managed, Apache Lucene-based search engine embedded directly into MongoDB Atlas. Unlike MongoDB's native text indexes, Atlas Search provides relevance scoring, fuzzy matching, autocomplete, faceted navigation, and sophisticated analyzers—all accessible through an aggregation pipeline stage called $search. You do not need to run a separate Elasticsearch or Solr cluster; Atlas Search lives alongside your data.

Atlas Search runs a Lucene engine as a separate process alongside each MongoDB node. When documents are inserted, updated, or deleted, MongoDB streams the changes to the Lucene index asynchronously. This means Atlas Search indexes are eventually consistent with the MongoDB data—there is a small lag (typically milliseconds) between a write and its appearance in search results. For most applications this is perfectly acceptable.

In the MongoDB Atlas UI, navigate to your cluster, click the Search tab, and choose Create Search Index. You can use the Visual Editor (point-and-click field mapping) or the JSON Editor to write the index definition directly. The index applies to a specific database and collection. After clicking Create, Atlas builds the index asynchronously—a green status indicator signals when it is ready to query.

The simplest Atlas Search index uses dynamic mapping: MongoDB automatically indexes all string, number, date, and boolean fields without requiring you to enumerate them. The default index definition is just { mappings: { dynamic: true } }. This is the fastest way to get started and is suitable for exploration. For production workloads, switch to static mappings to control which fields are indexed and how they are analyzed.

Static mapping gives you explicit control over which fields are indexed and how they are analyzed. Set dynamic: false and list each field with its type and analyzer. Fields not listed in a static mapping are not indexed. This reduces index size and improves indexing throughput for large collections by avoiding unnecessary indexing of every field.

An analyzer defines how text is broken into tokens and normalized before storage in the Lucene index. lucene.standard lowercases text and splits on whitespace and punctuation. lucene.english also applies stemming (reducing 'running' to 'run') for better recall. lucene.keyword treats the entire string as a single token, useful for exact-match fields like email addresses or SKU codes. The choice of analyzer significantly impacts search quality.

For infrastructure-as-code and CI/CD pipelines, you can create Atlas Search indexes using the Atlas CLI or the Atlas Administration API. This lets you version-control your search index definitions alongside your application code. The Atlas CLI command is atlas clusters search indexes create with a JSON configuration file.

After creating a search index, it goes through the BUILDING state while Lucene processes existing documents. During this time, queries may return incomplete results or fail with an error. Once the status shows READY (green in Atlas UI), the index covers all existing documents. New documents added after index creation are indexed with a small lag. Monitor index status programmatically using the Atlas Administration API or the Atlas CLI.

Atlas Search can index nested document fields and array elements. Use dot notation in the field path to reference nested fields, just like regular MongoDB queries. For arrays of subdocuments, each element is indexed separately. This allows you to search inside embedded objects—for example, finding products where any color option matches a search term.

Once your Atlas Search index is READY, test it with a $search aggregation stage. The simplest query uses the text operator to search for a keyword across indexed string fields. A successful response confirms the index is built and queries are working. Examine the searchScore in the projected output to understand relevance ranking.

You can create multiple Atlas Search indexes on the same collection, each with a different name, field configuration, and analyzers. This allows different use cases to use specialized indexes—a general search index with dynamic mapping, a strict-matching index with keyword analyzers, and an autocomplete index. When running a $search query, specify the index field to select which search index to use.

In this lesson you learned: Atlas Search is a Lucene-based full-text search engine embedded in MongoDB Atlas, dynamic mapping indexes all fields automatically while static mapping requires explicit field definitions for finer control, and analyzers like lucene.standard, lucene.english, and lucene.keyword control how text is tokenized and stored. Next up we explore writing $search aggregation queries including text, phrase, and wildcard operators.