Atlas Search Query Performance

Consider Query Operators and Query Complexity

The complexity level of Atlas Search queries and the type of operators used can affect database performance.

Highly complex queries, such as queries with multiple clauses that use the compound operator, or queries which use the regex (regular expression) or the wildcard operator, are resource-intensive.

Compound Queries

If your query includes multiple nested compound statements, ensure that these are not redundant. If the clauses are added programmatically, consider implementing the logic in the application to avoid inclusion of redundant clauses in the queries. Every score calculation per field that mongot performs, such as for the must and should clauses, increases execution time.

Faceted Search

You can use the Atlas Search facet Collector collector to extract metadata and avoid running multiple queries for search results and metadata. For an example, see the Metadata and Search Results example.

Scoring the Results

Atlas Search queries are ranked by score. Queries that return a large number of results are more computationally intensive because they must keep track of all the scores for the result set.

Use `$search` Instead of `$text` or `$regex`

For applications that rely heavily on MongoDB $text and $regex queries, use the following recommendations to determine whether to refactor your applications or migrate your applications to Atlas Search $search. The $search aggregation pipeline stage provides features that are either not available through the MongoDB operators or are available through the MongoDB operators but not as performant as Atlas Search $search.

The following table shows how MongoDB $regex, $text, and Atlas Search $search address your application's requirements.

If your application requires...	Use...	Because...
Datastore to respect write concerns	`$regex`	For transactions with heavy reads after writes, we recommend `$regex`. For `$search`, reads after writes should be rare.
Cluster optimized for write performance	`$search`	Atlas Search indexes don't degrade cluster write performance.
Searching through large data sets	`$search`	Atlas Search uses an inverted index, which enables fast document retrieval at very large scales.
Language awareness	`$search`	Atlas Search supports many language analyzers that can tokenize (create searchable terms) languages, remove stopwords, and interpret diacritics for improved search relevance.
Case-insensitive text search	`$search`	`$search` offers more capabilities than `$regex`.
Highlighting result text	`$search`	Atlas Search highlighting allows you to contextualize the documents in the results, which is essential for natural language queries.
Geospatial-aware search queries	`$regex` or `$search`	MongoDB `$regex` and Atlas Search `$search` treat geospatial parameters differently. In MongoDB, lines between coordinates are spherical, which is well-suited for coordinates for long distance such as air flight. Atlas Search uses Lucene, which draws a straight line between coordinates and is well-suited for short distance.
On-premises or local deployment	`$regex` or `$text`	Atlas Search is unavailable for on-premises or local deployment. Atlas Search is only available for data on the Atlas cluster.
Autocompletion of search queries	`$search`	For autocomplete of characters (nGrams), Atlas Search includes `edgeGrams` for left-to-right autocomplete, `nGrams` for autocomplete of languages that don't have whitespace, and `rightEdgeGram` for autocomplete of languages that you write and read right-to-left. For autocomplete of words (wordGrams), Atlas Search includes shingle token filter, which supports word-based autocomplete by concatenating adjacent words to create a single token.
Fuzzy matching on text input	`$search`	Atlas Search text and autocomplete operators support `fuzzy` matching to filter on input text and address misspelled words (typos).
Filtering based on multiple strings	`$search`	Atlas Search compound supports filtering based on multiple strings.
Relevance score sorted search	`$search`	Atlas Search uses the BM25 algorithm for determining the search relevance score of documents. It supports advanced configuration through `boost` expressions like multiply and gaussian decay, as well as analyzers, search operators, and synonyms. To learn more, see Modify the Score of the Documents in the Results.
Partial indexes	`$regex`	Atlas Search doesn't support partial indexing.
Patial match	`$search`	Atlas Search wildcard and autocomplete operators support partial match queries.
Single compound index on arrays	`$search`	Atlas Search term indexes are intersected in a single Atlas Search index and eliminate the need for compound indexes for filtering on arrays.
Synonyms search	`$search`	Atlas Search supports synonyms defined in a separate collection, which you can reference in your search index for use. To learn more, see the How to Use Synonyms with Atlas Search tutorial.
Faceting for counts	`$search`	Atlas Search provides fast counts of documents based on text criteria, and also supports faceted search for numbers and dates. To learn more, see How to Use Facets with Atlas Search.
Extract metadata	`$search`	Atlas Search `facet` Collector collector returns metadata and doesn't require you to run multiple queries for retrieving metadata. To learn more, see the How to Use Facets with Atlas Search tutorial.
Custom analyzers	`$search`	Atlas Search supports custom analyzers to suit your specific indexing requirements. For example, you can index and search email addresses and HTTP or HTTPS URLs using custom analyzers.
Searching phrases or multiple words	`$search`	Atlas Search phrase operator supports searching for a sequence of terms.
Searching with regular expression	`$search`	Atlas Search provides improved performance when you use the Atlas Search autocomplete operator instead.

Tip

Update $text Queries with Atlas Search for Improved Search Performance - describes how you can replace $text aggregation pipeline stage in your query with $search to improve both the flexibility and performance of these queries.
Use Atlas Search for Full-Text Regex Queries - describes how you can replace inefficient regex matching with $search to improve the performance of text queries.

Use `$limit` Before `$facet`

Using a $limit aggregation pipeline stage after a $facet aggregation pipeline stage might negatively impact query performance. To avoid performance bottlenecks, use $limit before $facet.

Example

{
  {
    "$search": {...}
  },
  { "$limit": 20 },
  {
    "$facet": {
      "results": [],
      "totalCount": $$SEARCH_META
    }
  }
}

For a demonstration, see the following examples:

Minimize Additional MQL Aggregation Stages

Try to encapsulate the entire search logic within the $search stage itself and minimize using additional blocking stages, such as $group, $count, $match, or $sort. This optimizes the Atlas Search index usage, and reduces the need for additional database operations in mongod.

Use `compound.filter` Instead of `$match`

For queries that require multiple filtering operations, use the compound Operator operator with filter clauses. If you must use the $match stage in your aggregation pipeline, consider using the storedSource option to store only the fields that your $match condition needs. You can then use the $search returnStoredSource option to retrieve stored fields and avoid the mongod full document lookup.

Tip

Use `facet` Instead of `$group`

If you use $group to get basic counts for field aggregations, you can use facet Collector inside the $search stage. If you need only metadata results, you can use facet Collector inside inside the $searchMeta stage instead.

Use `count` Instead of `$count`

If you use $count to get a count of the number of documents, we recommend that you use count inside the $search or $searchMeta stage instead.

Use `sort`, `near`, or `returnStoredSource` Instead of `$sort`

For sorting numeric, date, string, boolean, UUID, and objectID fields, use the sort option with the $search stage. To learn more, see Sort Atlas Search Results.
For sorting geo fields, use the near operator.
To sort other fields, use $sort and returnStoredSource fields.

Limit Use of `$skip` and `$limit` After `$search`

Using $skip and $limit to retrieve results non-sequentially might be slow if the results for your query are large. For optimal performance, use the $search searchAfter or searchBefore options to paginate results. To learn more, see Paginate the Results.

To return non-sequential results, such as jumping from page 2 to page 5, you can use the following pipeline stages:

$search searchAfter the last result on Page 2
$skip documents on Pages 3 and 4
$limit results for Page 5

Here, your query is optimized to skip only 2 pages of results, instead of skipping 4 pages if you didn't use searchAfter. For a demonstration of this, see Paginate the Results.

Monitor Performance

You can monitor your Atlas cluster and view charts with performance statistics on the Atlas Metrics tab. These metrics can help you see how Atlas Search queries and index building affect your cluster's performance. To learn more, see Review Atlas Search Metrics.

Atlas might trigger some Atlas alerts when:

Atlas Search queries your clusters, which can impact Atlas performance metrics, such as the query targeting metrics.
The change streams cursors that the Atlas Search process (mongot) uses to keep Atlas Search indexes updated can contribute to the query targeting ratio and trigger query targeting alerts if the ratio is high.
Atlas Search replicates data from MongoDB, which contributes to the metrics measured in Atlas, such as the number of getmore operations.

Note

If your cluster's resources are stretched or near the limits of acceptable performance, consider upgrading to a larger cluster tier before implementing Atlas Search functionality.

Tip

Continue Learning

Follow along with this video to learn how to understand, iterate, and improve your Atlas Search results using explain and $search score details.

Duration: 5 Minutes

Back

Indexes

Deployment Options

Consider Query Operators and Query Complexity

Compound Queries

Faceted Search

Scoring the Results

Use $search Instead of $text or $regex

Tip

Use $limit Before $facet