Compatibility Changes in MongoDB 3.4

Sharded Cluster Changes

Initial Sync and renameCollection

If a collection is renamed on the sync source while an initial sync is running, the initial sync process fails and restarts to avoid possible data corruption. See SERVER-26117.

Operations that rename collections include:

• renameCollection command and db.collection.renameCollection() method.

• Aggregation (db.collection.aggregate() method or aggregate command) with the $out stage.

• Map-reduce (db.collection.mapReduce() method or mapReduce command) with the out option.

• convertToCapped command.

As such, when upgrading from 3.2.11 or earlier versions to 3.4, initial syncs may start failing if they encounter renameCollection operations.

In MongoDB versions 3.2.11 or earlier versions, initial sync process would proceed when encountering renameCollection operations, which could potentially corrupt data. See SERVER-4941.

Deprecated Operations

Stricter Validation of Collection and Index Specifications

db.createCollection( "myCappedCollection", { cappedtypo: true, size: 5242880 } )

General Compatibility Changes

• Updates to namespace restrictions: in MongoDB 3.4, the $ character is no longer supported in database names.

  Important

  You must drop any databases that contain a $ in its name before upgrading to MongoDB 3.4.

• Removal of deprecated textSearchEnabled parameter. Starting from version 2.6, MongoDB enables the text search feature by default.

• Removal of mongosniff. In MongoDB 3.4, mongosniff is replaced by mongoreplay, which offers a more flexible superset of mongosniff's functionality.

• Updates to $project specification behavior: empty documents in $project specifications produce an error.

• If you include a hint() that specifies a sparse index when you perform a count() of all documents in a collection (i.e. with an empty query predicate), the sparse index is used even if the sparse index results in an incorrect count.

  db.collection.insertOne( { _id: 1, y: 1 } );
  db.collection.createIndex( { x: 1 }, { sparse: true } );
  db.collection.find().hint( { x: 1 } ).count();

  To obtain the correct count, do not hint() with a sparse index when performing a count of all documents in a collection.

  db.collection.find().count();
  db.collection.createIndex( { y: 1 } );
  db.collection.find().hint( { y: 1 } ).count();

  Previous versions ignored the hint if the use of the sparse index would result in an incomplete count.

User Roles Changes

The privileges of the following built-in roles no longer apply to the local and config databases:

Correspondingly, the following built-in roles include additional read and write privileges on local and config databases:

• clusterManager

• clusterMonitor

• backup

• restore.

Backwards Incompatible Features

The following 3.4 features persist data that earlier MongoDB versions cannot correctly handle and require that featureCompatibilityVersion be set to "3.4":

• Views

• Collation and Case-Insensitive Indexes

• Decimal Type

• Index version v: 2. v:2 indexes add support for collation and decimal data type. Earlier index versions support neither collation nor the decimal data type.

  If featureCompatibilityVersion: "3.4", indexes created in MongoDB 3.4 default to v: 2 . Otherwise, new indexes default to v: 1.

To set the featureCompatibilityVersion, see setFeatureCompatibilityVersion command.

3.4 deployments have the following default featureCompatibilityVersion values:

Earlier versions of MongoDB will not start if the database contains views, collation specifications, or v:2 indexes. If the data contains the decimal data type, operations against these documents may fail. See Downgrade MongoDB 3.4 to 3.2 for details. If you need to downgrade, you must remove data related to these incompatible features from your database before downgrading the binaries.

Driver Compatibility Changes

To use the various new features such as the new Decimal Type and Collation with a MongoDB driver, an upgrade to a driver version that supports these features is necessary.

Single Element $in With upsert

When an upsert operation finds no matching documents, it creates a document to insert based on the equality statements in the query and then applies the update modifiers to this seeded document. For example:

db.c.drop()
db.c.update( { a : 3, b: "foo" }, { $set : { c : 15 } }, { upsert : true } )
db.c.find()
{ "_id" : ObjectId("59c03009529946822d0afb8c"), "a" : 3, "b" : "foo", "c" : 15 }

Prior to 3.4, a single-element $in query did not seed the upsert document. In the example below, the $addToSet update expression is successful because of this behavior:

db.c.drop()
db.c.update( { a : { $in : [1] } }, { $addToSet : { a : 2 } }, { upsert : true } )
db.c.find()
{ "_id" : ObjectId("58bdb00eb39e8f87607e9222"), "a" : [ 2 ] }

In 3.4 and newer versions, however, a single-element $in behaves like an equality statement for upserts. If the query includes this condition on a field, the field value is set to the element.

As a result of this behavior, certain upsert operations may fail in 3.4. In example above, the $addToSet upsert would fail because the a field would be seeded with a single value, and $addToSet cannot be applied to a scalar field. To avoid this error, you must wrap the $in expression in an $elemMatch expression:

db.c.drop()
db.c.update(
   { a : { $elemMatch : { $in : [ 2 ] } } },
   { $addToSet : { a: 3 } },
   { upsert: true } )
db.c.find()
{ "_id" : ObjectId("..."), "a" : [ 3 ] }