Features
The following are new features added in MongoDB 3.6 and supported in the Node.js driver.
Retryable Writes
Support has been added for retryable writes through the connection string. MongoDB 3.6 will utilize server sessions to allow some write commands to specify a transaction ID to enforce at-most-once semantics for the write operation(s) and allow for retrying the operation if the driver fails to obtain a write result (e.g. network error or "not master" error after a replica set failover)Full details can be found in the Retryable Writes Specification.
DNS Seedlist Support
Support has been added for DNS Seedlists. Users may now configure a single domain to return a list of host names. Full details can be found in the Seedlist Discovery Specification.
Change Streams
Support has been added for creating a stream to track changes to a particular collection. This is a new feature in MongoDB 3.6. Full details can be found in the Change Stream Specification as well as examples in the test directory.
Sessions
Version 3.6 of the server introduces the concept of logical sessions for clients. In this driver,
MongoClient now tracks all sessions created on the client, and explicitly cleans them up upon
client close. More information can be found in the Driver Sessions Specification.
API Changes
We removed the following API methods.
Db.prototype.authenticateDb.prototype.logoutDb.prototype.openDb.prototype.dbDb.prototype.closeAdmin.prototype.authenticateAdmin.prototype.logoutAdmin.prototype.profilingLevelAdmin.prototype.setProfilingLevelAdmin.prototype.profilingInfoCursor.prototype.nextObject
We've added the following API methods.
MongoClient.prototype.logoutMongoClient.prototype.isConnectedMongoClient.prototype.dbMongoClient.prototype.closeMongoClient.prototype.connectDb.prototype.profilingLevelDb.prototype.setProfilingLevelDb.prototype.profilingInfo
In core we have removed the possibility of authenticating multiple credentials against the same connection pool. This is to avoid problems with MongoDB 3.6 or higher where all users will reside in the admin database and thus database level authentication is no longer supported.
The legacy construct
var db = var Db('test', new Server('localhost', 27017));
db.open((err, db) => {
// Authenticate
db.admin().authenticate('root', 'root', (err, success) => {
....
});
});
is replaced with
new MongoClient(new Server('localhost', 27017), {
user: 'root'
, password: 'root'
, authSource: 'adming'}).connect((err, client) => {
....
})
MongoClient.connect works as expected but it returns the MongoClient instance instead of a
database object.
The legacy operation
MongoClient.connect('mongodb://localhost:27017/test', (err, db) => {
// Database returned
});
is replaced with
MongoClient.connect('mongodb://localhost:27017/test', (err, client) => {
// Client returned
var db = client.db('test');
});
Other Changes
Below are more updates to the driver in the 3.0.0 release.
Connection String
Following changes to the MongoDB connection string specification, authentication and hostname details in connection strings must now be URL-encoded. These changes reduce ambiguity in connection strings.
For example, whereas before mongodb://u$ername:pa$$w{}rd@/tmp/mongodb-27017.sock/test would have
been a valid connection string (with username u$ername, password pa$$w{}rd, host /tmp/mongodb-27017.sock
and auth database test), the connection string for those details would now have to be provided to
MongoClient as mongodb://u%24ername:pa%24%24w%7B%7Drd@%2Ftmp%2Fmongodb-27017.sock/test.
Unsupported URL options in a connection string now log a warning instead of throwing an error.
For more information about connection strings, read the connection string specification.
BulkWriteResult & BulkWriteError
When errors occured with bulk write operations in the past, the driver would callback or reject with
the first write error, as well as passing the resulting BulkWriteResult. For example:
MongoClient.connect('mongodb://localhost', function(err, client) {
const collection = client.db('foo').collection('test-collection')
collection
.insert({ id: 1 })
.then(() => collection.insertMany([ { id: 1 }, { id: 1 } ]))
.then(result => /* deal with errors in `result */)
.catch(err => /* no error is thrown for bulk errors */);
});
becomes:
MongoClient.connect('mongodb://localhost', function(err, client) {
const collection = client.db('foo').collection('test-collection')
collection
.insert({ id: 1 })
.then(() => collection.insertMany([ { id: 1 }, { id: 1 } ]))
.then(() => /* this will not be called in the event of a bulk write error */)
.catch(err => /* deal with errors in `err` */);
});
Where the result of the failed operation is a BulkWriteError which has a child value result
which is the original BulkWriteResult. Similarly, the callback form no longer calls back with an
(Error, BulkWriteResult), but instead just a (BulkWriteError).
mapReduce inlined results
When Collection.prototype.mapReduce is invoked with a callback that includes out: 'inline',
it would diverge from the Promise-based variant by returning additional data as positional
arguments to the callback ((err, result, stats, ...)). This is no longer the case, both variants
of the method will now return a single object for all results - a single value for the default case,
and an object similar to the existing Promise form for cases where there is more data to pass to
the user.
Find
find and findOne no longer support the fields parameter. You can achieve the same results as
the fields parameter by using Cursor.prototype.project or by passing the projection property
in on the options object . Additionally, find does not support individual options like skip and
limit as positional parameters. You must either pass in these parameters in the options object,
or add them via Cursor methods like Cursor.prototype.skip.
Collection.prototype.aggregate
Collection.prototype.aggregate no longer accepts variadic arguments. While this
was originally added to improve compatibility with the mongo shell, it has never
been a documented feature, and has led to more bugs and maintenance burden.
Pipeline stages are now only accepted as an Array of stages as the first argument.
2.x syntax:
collection.prototype.aggregate(
{$match: {a: 1}},
{$project: {b: 1, _id: 0}},
function (err, result) {
...
}
);
3.x syntax
collection.prototype.aggregate(
[
{$match: {a: 1}},
{$project: {b: 1, _id: 0}}
],
function (err, cursor) {
...
}
);
Collection.prototype.aggregate now returns a cursor if a callback is provided. It used to return
the resulting documents which is the same as calling cursor.toArray() on the cursor we now pass to
the callback.
2.x syntax
collection.prototype.aggregate(
[
{$match: {a: 1}},
{$project: {b: 1, _id: 0}}
],
function (err, result) {
console.log(result);
}
);
3.x syntax
collection.prototype.aggregate(
[
{$match: {a: 1}},
{$project: {b: 1, _id: 0}}
],
function (err, cursor) {
cursor.toArray(function(err, result) {
console.log(result);
});
}
);
Support added for comment in the aggregation command. Support also added for a hint field in the
aggregation options.
If you use aggregation and try to use the explain flag while you have a readConcern or
writeConcern, your query will now fail.
updateOne & updateMany
The driver now ensures that updated documents contain atomic operators. For instance, if a user
tries to update an existing document but passes in no operations (such as $set, $unset, or
$rename), the driver will now error:
let testCollection = db.collection('test');
testCollection.updateOne({_id: 'test'}, {});
// An error is returned: The update operation document must contain at least one atomic operator.
keepAlive
Wherever it occurs, the option keepAlive has been changed. keepAlive is now a boolean that enables/disables keepAlive, while keepAliveInitialDelay specifies how long to wait before initiating keepAlive. This brings the API in line with NodeJS's socket api
insertMany
Now insertMany returns insertedIds in a map of the index of the inserted document to the id of the inserted document:
{
"0": 2,
"1": 3
}
Previously an array of ids was returned: [ 2, 3 ]. This change occurs with both ordered and unordered insertMany calls, see the CRUD specifications for more details.
geoNear command helper
The functionality of the geoNear command is duplicated elsewhere in the language, in the $near/$nearSphere query operators on unsharded collections, and in the $geoNear aggregation stage on all collections. Maintaining this command increases our test surface, and creates additional work when adding features that must be supported on all read commands. As a result, the command will be fully
removed in the MongoDB 4.0 release, and we are choosing to remove it in this
major release of the node driver.
Tests
We have updated all of the tests to use Mocha and a new test runner, mongodb-test-runner, which
sets up topologies for the test scenarios.