How do I fix MongoDB "Sort exceeded memory limit"?

db.orders.createIndex({ status: 1, createdAt: -1 })

Now MongoDB seeks to status = "shipped", walks the index already in createdAt order, and stops after 50 entries. There's no SORT stage, no memory limit to hit, and the query is fast whatever the collection size.

Why the order matters: with { createdAt: -1, status: 1 } MongoDB can sort from the index but has to walk every order to filter by status. With a range field before the sort field, { total: 1, createdAt: -1 }, the results come out ordered by total first, so an in-memory sort is needed again.

Direction also matters for multi-field sorts. An index { a: 1, b: -1 } supports sort({ a: 1, b: -1 }) and its exact inverse sort({ a: -1, b: 1 }), but not sort({ a: 1, b: 1 }).

Aggregation pipelines

In an aggregation, $sort can only use an index when it comes at the start of the pipeline (after a $match the index also covers). A $sort after $group, $unwind or $lookup always sorts in memory, because those stages produce new documents.

db.orders.aggregate([
  { $match: { status: "shipped" } },     // uses the index
  { $sort:  { createdAt: -1 } },         // still uses the index
  { $limit: 1000 },                      // cut early
  { $lookup: { from: "customers", localField: "customerId", foreignField: "_id", as: "c" } },
])

$sort directly followed by $limit is optimized into a top-k sort that keeps only k documents in memory, so always limit as early as you can. $group has its own 100 MB limit and spills the same way.

When allowDiskUse is the right answer

For one-off analytics, exports and exploratory pipelines where building an index isn't worth it, opting in is fine:

db.orders.find({}).sort({ total: -1 }).allowDiskUse()
db.orders.aggregate(pipeline, { allowDiskUse: true })

Just don't use it to paper over a hot API query. It turns a clear error into a slow query that uses a lot of I/O.

Reduce what you sort

• Project only the fields you need before sorting. The memory limit counts document size, so smaller documents mean more fit in 100 MB.
• Paginate with a range on the sort key (createdAt < lastSeen) instead of skip(), which has to sort everything it skips.

Checklist

1. explain("executionStats"): a SORT stage means an in-memory sort.
2. Build a compound index in ESR order: equality, sort, range.
3. In pipelines, $match and $sort first and $limit early.
4. Use allowDiskUse for analytics, not as a fix for hot queries.

MongoDB, Inc. Goes Public

MongoDB has started trading publicly on NASDAQ.

#news