$unwind：拆解数组字段

The $unwind stage deconstructs an array field in a document into multiple output documents—one per array element. Each output document is a copy of the original document with the array field replaced by a single element from the array. This is essential for per-element analytics, like counting how many times each tag appears across all articles in a collection.

The simplest form of $unwind is a string with the dollar-prefixed field path: { $unwind: '$arrayField' }. The extended object syntax allows additional options like preserving empty arrays and including the array index. For most cases, the simple string form is sufficient.

By default, $unwind removes documents where the specified field is missing, null, or an empty array. This behavior is like an inner join: only documents with non-empty arrays pass through. To preserve documents with missing or empty arrays, set preserveNullAndEmptyArrays: true. This is important when the array field is optional and you don't want to lose the parent document.

The includeArrayIndex option adds a new field to each output document containing the 0-based index of the element in the original array. This is useful when the position in the array carries meaning—for example, the order of steps in a process, the rank of a search result, or the sequence of events in a log.

The most common pattern in MongoDB aggregation is $unwind + $group: first flatten the array to get one document per element, then group to compute per-element statistics. For example, unwinding the tags array and then grouping by tag name gives you the count of articles per tag.

After a $lookup, the joined field is always an array. When the join is one-to-one (joining by a unique _id), you typically $unwind immediately to flatten the single-element array into an embedded object. This is such a common pattern that you'll see it in almost every pipeline that uses $lookup.

For documents with nested arrays (arrays inside arrays), you can use multiple consecutive $unwind stages. The first $unwind deconstructs the outer array, and the second deconstructs the inner array. Each stage multiplies the number of output documents, so be cautious with deep nesting to avoid explosive document expansion.

A powerful use case for $unwind is computing statistics on individual array elements. For example, in an e-commerce database with orders containing line items (an array of sub-documents), unwinding line items lets you compute total revenue, average quantity, or top-selling items across all line items in all orders—not just per order.

$unwind can dramatically multiply the number of documents in the pipeline (a document with a 100-element array becomes 100 documents after unwind). This increases memory usage and processing time for all subsequent stages. Always ask: 'Can I achieve this with an array expression operator like $size, $filter, or $map in a $project stage instead of unwinding?' Array expression operators are often faster because they don't expand the document count.

Because $unwind multiplies documents, it is usually the most expensive stage in a pipeline when applied to large arrays. Performance tips: apply $match and $project before $unwind to reduce the input document size and count; add a $match immediately after $unwind if you only care about specific elements; and limit the number of documents entering the unwind with an early $limit when applicable.

After a $unwind + $group pipeline, you can use $push in the $group stage to rebuild an array from the processed elements. This pattern is useful for filtering or transforming array elements: unwind to get individual elements, apply per-element transformations or filters, then push back into a new array. This is more flexible than using $filter or $map for complex per-element logic.

In this lesson you learned: $unwind creates one output document per array element, preserveNullAndEmptyArrays: true keeps documents with missing or empty arrays, and the classic $unwind + $group pattern enables per-element analytics like tag frequency counts. Next up we explore $addFields, $replaceRoot, and $mergeObjects.