Filtrado de eventos con un pipeline de agregación

Without filtering, a change stream delivers every change event on a collection. In a busy production collection, this can mean thousands of events per second, most of which your application does not care about. Filtering at the server using an aggregation pipeline reduces network traffic, lowers CPU usage in your application, and ensures your event handler only processes relevant events. Filtering happens before events leave MongoDB—only matching events are transmitted to your client.

The first argument to watch() is an aggregation pipeline array. MongoDB applies this pipeline to each change event document before deciding whether to deliver it to your application. Not all aggregation stages are permitted in change stream pipelines—only a specific subset is allowed, primarily $match, $project, $addFields, $replaceRoot, and $redact. The $group and $lookup stages are not allowed.

Filtering by operationType is the most common pipeline filter. You can use $match with a single operation type string or with $in to match multiple types. This is useful when your application cares about inserts and updates but not deletes, or when different microservices subscribe to different operation types on the same collection.

You can filter update events based on which fields were modified by querying the updateDescription.updatedFields object in the $match stage. This lets you subscribe only to specific field changes—for example, only when a document's status field transitions to a particular value. This is more efficient than receiving all updates and filtering in application code.

For insert events, you can filter based on fields in the fullDocument sub-document. For example, receive only inserts where fullDocument.priority is 'high' or fullDocument.region equals 'US-WEST'. This server-side filtering is especially powerful in multi-tenant architectures where different application instances need events for different subsets of data.

The $project stage in a change stream pipeline reshapes the event document before it is delivered to your application. You can include only the fields your handler needs, rename fields, or compute derived fields. This reduces the payload size transmitted over the network and simplifies your event handler code by presenting only the data it needs.

The $addFields stage lets you add computed fields to the change event document. You can add a timestamp when the event was processed, derive a category from the operation type, or compute a routing key. These enriched fields are included in the event your application receives, allowing downstream code to use pre-computed values without recalculating them.

You can chain multiple pipeline stages in a change stream pipeline for powerful composition. A common pattern is: $match to filter events → $addFields to enrich → $project to trim. Each stage processes the output of the previous one. Remember that the order of stages matters—apply the most selective $match first to minimize the documents processed by later stages.

Server-side pipeline filtering in change streams is significantly more efficient than receiving all events and filtering in application code. Without server-side filtering, every event must be serialized and transmitted over the network. With a $match stage, MongoDB evaluates the filter internally and transmits only matching events. For high-traffic collections, this can reduce network usage and application CPU by orders of magnitude.

MongoDB restricts which aggregation stages can be used in change stream pipelines. Permitted: $match, $project, $addFields, $replaceRoot, $replaceWith, $redact. Forbidden: $group, $lookup, $unwind, $geoNear, $out, $merge, and several others. Attempting to use a forbidden stage causes an error when opening the stream. If you need complex transformations, do them in application code after receiving the (pre-filtered) events.

In multi-tenant applications, multiple tenants share one collection with a tenantId field. Rather than running one change stream per tenant (expensive), run one stream per service instance with a $match filter on fullDocument.tenantId scoped to the tenants that instance serves. This scales to hundreds of tenants with far fewer open cursors on the MongoDB server.

Test your understanding of MongoDB & NoSQL Databases concepts from this lesson.

In this lesson you learned: pass an aggregation pipeline as the first argument to watch() to filter events server-side, permitted stages include $match, $project, $addFields, $replaceRoot, and $redact — but not $group or $lookup, and server-side filtering dramatically reduces network traffic and application CPU compared to application-side filtering. Next up we explore resuming change streams after an interruption using resume tokens.