How to do Snapshotted Queries in the Mongo Database

This document refers to query snapshots. For backup snapshots of the database's datafiles, see the fsync lock page.

MongoDB does not support full point-in-time snapshotting. However, some functionality is available which is detailed below.

Cursors

A MongoDB query returns data as well as a cursor ID for additional lookups, should more data exist.  Drivers lazily perform a "getMore" operation as needed on the cursor to get more data.  Cursors may have latent getMore accesses that occurs after an intervening write operation on the database collection (i.e., an insert, update, or delete).

Conceptually, a cursor has a current position. If you delete the item at the current position, the cursor automatically skips its current position forward to the next item.

MongoDB cursors do not provide a snapshot: if other write operations occur during the life of your cursor, it is unspecified if your application will see the results of those operations. In fact, it is even possible (although unlikely) to see the same object returned twice if the object were updated and grew in size (and thus moved in the datafile). To assure no update duplications, use snapshot() mode (see below).

Snapshot Mode

snapshot() mode assures that objects which update during the lifetime of a query are returned once and only once. This is most important when doing a find-and-update loop that changes the size of documents that are returned ($inc does not change size).

> // mongo shell example
> var cursor = db.myCollection.find({country:'uk'}).snapshot();

Even with snapshot mode, items inserted or deleted during the query may or may not be returned; that is, this mode is not a true point-in-time snapshot.

Because snapshot mode traverses the _id index, it may not be used with sorting or explicit hints. It also cannot use any other index for the query.

You can get the same effect as snapshot by using any unique index on a field(s) that will not be modified (probably best to use explicit hint() too). If you want to use a non-unique index (such as creation time), you can make it unique by appending _id to the index at creation time.