Documentation Request Summary:

The page below will need to be updated:
https://docs.mongodb.com/manual/reference/command/count/

The following paragraphs should be changed:
---------------------------------------------------
On a sharded cluster, count can result in an inaccurate count if orphaned documents exist or if a chunk migration is in progress.

To avoid these situations, on a sharded cluster, use the $group stage of the db.collection.aggregate() method to $sum the documents. For example, the following operation counts the documents in a collection:
---------------------------------------------------

The new behavior is that, when running a sharded cluster, a fast count may return inaccurate results. A count() with a predicate will not (as of 4.0). See the "Behavior of "fast count" and non-"fast count" in the description of the ticket. If you have any questions feel free to slack/email/comment on a ticket!

Scope of changes:

• count (various)

Impact to other docs outside of this product:

none
 

MVP:

Resources:

Engineering Ticket Description:

Summary

Count does not filter out unowned (orphaned) documents and can therefore report larger values than one will find via a normal query, or using itcount() in the shell.

Causes

The following conditions can lead to counts being off:

• Active migrations
• Orphaned documents (left from failed migrations)
• Non-Primary read preferences (see SERVER-5931)

Workaround

A workaround to get accurate counts is to ensure all migrations have been cleaned up and no migrations are active. To query non-primaries you must also ensure that there is no replication lag including any migration data, in addition to the above requirements.

Non-Primary Reads

For issues with counts/reads from non-primaries please see SERVER-5931

Behavior of "fast count" and non-"fast count"

A "fast count" is a count run without a predicate. It is "fast" because the implementation only reads the metadata, without fetching any documents.

The problem of count() reporting inaccurate results has been fixed for non-"fast counts," that is, starting in 4.0, counts which are run with a predicate are accurate when run on sharded clusters. "Fast counts" (count() run without a predicate) may still report too many documents (see SERVER-33753).

In general, if one needs an accurate count of how many documents are in a collection, we do not recommend using the count command. Instead, we suggest using the $count aggregation stage, like this:

db.foo.aggregate([{$count: "nDocs"}]);

See the docs.

For users who need the performance of "fast count", and are okay with approximate results, we suggest using $collStats instead of the count command:

db.matrices.aggregate( [ { $collStats: { count: { } } } ] )