How does the MongoDB aggregation pipeline work?

Each stage transforms the stream and passes it on. The optimizer can reorder some stages, but you should still filter early and project only the fields you need. Grouping and sorting are the expensive stages because they may need to hold data in memory.

db.orders.aggregate([
  { $match: { status: "paid", createdAt: { $gte: new Date("2026-09-01") } } },
  { $unwind: "$items" },
  { $group: {
      _id: "$items.sku",
      revenue: { $sum: { $multiply: ["$items.price", "$items.qty"] } },
      orders: { $sum: 1 }
  } },
  { $sort: { revenue: -1 } },
  { $limit: 10 }
]);

Stages worth knowing

• $match filters documents and should come as early as possible
• $project shapes the output and can reduce memory use
• $group aggregates by a key and accumulates values
• $unwind expands an array into one document per element
• $lookup performs a left outer join and is the most expensive common stage
• $facet runs several pipelines over the same input for dashboards

A $lookup across a large collection without an index on the foreign field can be very slow. If the joined data is needed on every read, consider denormalizing it instead. Use allowDiskUse for large aggregations, but treat it as a signal to optimize rather than a permanent fix.