Dart Driver for Apache Cassandra

Build Status

Dart driver for Apache Cassandra that supports Cassandra Query Language version 3.0+ (CQL3).

The driver has a small dependency tree and implements Cassandra binary protocol (versions 2 and 3) for communicating with Cassandra servers. The protocol and CQL versions to be used are both configurable by the user.


  • Asynchronous API based on Future and Streams
  • Connection management via connection pools
  • Connection load-balancing and failover
  • Server event handling (node/topology/schema change events)
  • Query multiplexing on each connection
  • Batch and prepared queries with either positional or named placeholders
  • Query result streaming
  • Support for all Cassandra types including user defined types (UDT), tuples and custom types (via user-defined Codecs)


First, add the library to the list of dependencies in your pubspec.yaml file:

  dart_cassandra_cql: ">=0.1.0 <0.2.0"

Then, run pub install to install the package.

If your package is an application package you should use any as the version constraint.

Quick start

For a more in-depth explanation please check out the Api section.

import "dart:async";
import 'package:dart_cassandra_cql/dart_cassandra_cql.dart' as cql;

void main() {
  // Create a client for connecting to our cluster using native
  // protocol V2 and sensible defaults. The client will setup
  // a connection pool for you and connect automatically when
  // you execute a query.
  cql.Client client = new cql.Client.fromHostList([
      , ""

  // Perform a select with positional bindings
      new cql.Query("SELECT * from test.type_test WHERE id=?", bindings : [123])
  ).then((Iterable<Map<String, Object>> rows) {
    // ...

  // Perform an prepared insert with named bindings, a time-based uuid and tuneable consistency
      new cql.Query("INSERT INTO test.type_test (id, uuid_value) VALUES (:id, :uuid)", bindings : {
          "id" : 1
          , "uuid" : new cql.Uuid.timeBased()
      }, consistency : cql.Consistency.LOCAL_QUORUM
       , prepared : true)
  ).then((cql.ResultMessage res) {
    // ...

  // Perform a batch insert query
      new cql.BatchQuery()
          new cql.Query("INSERT INTO test.type_test (id, uuid_value) VALUES (:id, :uuid)", bindings : {
              "id" : 1
              , "uuid" : new cql.Uuid.timeBased()
          new cql.Query("INSERT INTO test.type_test (id, uuid_value) VALUES (:id, :uuid)", bindings : {
              "id" : 2
              , "uuid" : new cql.Uuid.timeBased()
        ..consistency = cql.Consistency.TWO
  ).then((cql.ResultMessage res) {
    // ...
  }).catchError((e) {
    // Handle errors

  // Stream (paginated) query
  StreamSubscription sub;
  sub = client.stream(
      new cql.Query("SELECT * from test.type_test")
      , pageSize : 200
  ).listen((Map<String, Object> row) {
    // Handle incoming row
    print("Next row: ${row}");
    // ... or manipulate stream



Cassandra to dart type mapping

The following table summarizes type mappings between Cassandra and dart\_cassandra\_cql. Due to the fact we cannot establish a one-to-one mapping between the Dart and Cassandra types, the following conversions are only applied when:

  • the driver parses query responses from the server
  • the driver encodes data prior to execution of a prepared query

In all other cases, the driver performs automatic inline expansion of each bound query parameter.

|Cassandra type | Dart type |:------------- |:--------- |ascii | String |bigint | int |blob | Uint8List |boolea | bool |counter | int |decimal | int or double |double | double |float | double |inet | InternetAddress |int | int |list<X> | List<X> |map<X, Y> | LinkedHashMap<X, Y> |set<X> | Set<X> |text | String |timestamp | DateTime |uuid | Uuid |timeuuid | Uuid |varchar | String |varint | int |UDT | LinkedHashMap. See the section on UDTs |tuple | Tuple |custom | Uint8List or type instance implementing CustomType


The Uuid class provides a wrapper around UUIDs and provides factory constructors for generating simple and time-based UUIDs.

Uuid simpleUuid = new Uuid.simple();
Uuid timeUuid = new Uuid.timeBased();

If you have some externally generated UUIDs that you wish to pass to a Cassandra query you can either pass them as a String or wrap them with a Uuid object:

Uuid externalUuid = new Uuid("550e8400-e29b-41d4-a716-446655440000");

User defined types

The driver supports UDTs with arbitrary nesting. The driver will parse UDTs as LinkedHashMap<String, Object> objects.


Whenever you need to use a tuple type in your queries or read it from a query result you need to use the driver-provided Tuple class.

This class is essentially a decorated List<Object>. You can instanciate a Tuple object from any Iterable using the fromIterable named constructor. Here is an example:

Tuple tuple = new Tuple.fromIterable([1, "test", 3.14]);

Custom types

Custom types are user-defined server-side Java classes that extend the types supported by Cassandra. These classes provide mechanisms for handing custom type serialization/de-serialization.

By default, custom types are parsed and returned as a Uint8List. The driver however allows you to register a user-defined Codec for handling custom type serialization/de-serialization to a Dart class instance.

To use this feature, the Dart class that represents the custom type needs to implement thee CustomType interface:

abstract class CustomType {
  String get customTypeClass;

This interface defines a getter for querying the fully qualified Java class name that implements this custom type at the server and is used by the driver to select the appropriate Codec when it encounters an instance of this class.

You will also need to register a Codec<Uint8List, CustomType> for handing the actual serialization/de-serialization. To register the codec you will need to invoke the globally available registerCodec method as follows:

registerCodec('fully.qualified.java.class.name', new MyCustomTypeCodec() );

After this step, the driver will automatically invoke the codec whenever it encounters a custom type with this class name while parsing query results or whenever an CustomType object instance of this type is bound to a query.

The connection pool

A connection pool is used to keep track of active connections to Cassandra nodes and to provide load-balancing, fault-tolerance and server event subscription to Cassandra clients.

To instanciate a connection pool you need to provide a PoolConfiguration object to the pool constructor. The following named parameters may be passed to the PoolConfiguration object constructor to override the default values:

|Option name | Default value | Description | |:------------------------|:------------------------|:--------------- | protocolVersion | ProtocolVersion.V2 | The binary protocol version to use. Its value should be one of the ProtocolVersion enums | cqlVersion | "3.0.0" | The CQL version to use. To use the new CQL features you should request at least version "3.1.0" | connectionsPerHost | 1 | The number of concurrent connections to each host | streamsPerConnection | 128 | The max number of requests that can be multiplexed over each connection. According to the binary protocol specification, each connection can multiplex up to 128 streams in V2 mode and 32768 streams in V3 mode | maxConnectionAttempts | 10 | The max number of reconnection attempts before declaring a connection as unusable | reconnectWaitTime | 1500ms | A Duration object specifying the time to wait between reconnection attempts | streamReservationTimeout| 0ms | A Duration object specifying the time to wait for a connection stream to become available when all connection streams are in use. In case of a timeout, the driver will try the next available connection | autoDiscoverNodes | true | If this flag is set to true, the connection pool will listen for server topology change events and automatically update the pool when new nodes come online/go offline. If set to false, the pool will only process UP/DOWN events for nodes already in the pool | authenticator | null | An authentication provider instance implementing the Authenticator interface or null if no authentication is required. For more information see the section on authentication | compression | null | The compression algorithm to be used with Cassandra nodes. Its value should be one of the Compression enums or null if no compression should be used. For more information see the section on compression

Simple connection pool

The driver provides the SimpleConnectionPool as its default ConnectionPool implementation. This pool should be adequate for most projects although you can always roll up your own pool if you want.

The pool manages a list of connectionsPerHost open connections and tries to load-balance between them. In addition to that, the pool subscribes to server generated events and can add/remove connections from the pool whenever Cassandra nodes come online/go offline or nodes enter/exit the cluster.

If the autoDiscoverNodes pool configuration option is set to true then the pool will automatically open connections to nodes not originally present in the pool when they come online.

Connection selection

The simple connection pool uses the following algorithm for selecting connections:

  • Select the first healthy connection with available connection streams.
  • If no connection satisfies these criteria, then the first healthy connection with no available connection streams is selected.

In the later case, the driver will attempt to reserve a connection stream. The streamReservationTimeout pool configuration option controls the reservation timeout. If the reservation times out then another connection is selected from the pool and the process is repeated.

In any case, whenever a connection is selected from the pool, it will be removed from its current position in the pool and moved to its end. This allows the driver to perform load-balancing.

Custom connection pools

To create a custom connection pool you need to extend the abstract ConnectionPool class and implement its abstract methods to apply your custom connection selection logic. An instance of your custom connection pool can then be passed to the client during construction. Here are a few ideas for a custom connection pool:

  • A DC-aware connection pool that tries local DC connections first and then falls back to a secondary, tertiary e.t.c connection pool.
  • A connection pool that discovers Cassandra hosts automatically via an external service (EC2 API queries, etcd)

Cassandra authentication

The driver provides the built-in PasswordAuthenticator class that allows you to authenticate with Cassandra servers requiring the org.apache.Cassandra.auth.PasswordAuthenticator provider. To use this provider, you need to instanciate it with your user/pass credentials and pass it to the pool configuration.

If you need to implement any other form of authentication you can roll your own by implementing the Authenticator interface.

abstract class Authenticator {
  String get authenticatorClass;

  Uint8List answerChallenge(Uint8List challenge);

The implementation should define: + a getter for retrieving the fully qualified Cassandra authentication class that the authenticator handles. This value is used to match the authenticator class requested by Cassandra during the initial handshake. + an answerChallenge method that will be invoked each time the remote server sends an authentication challenge.

If the Cassandra cluster requires authentication but none is supplied or an incompatible authenticator instance is specified, then connection attempts will fail with an AuthenticationException error.


In order to keep external dependencies to a minimum, the driver does not ship with native implementations for the two compression schemes currently supported by Cassandra (lz4 and snappy).

The driver however allows provides a mechanism for registering a Codec\<Uint8List, Uint8List> that implements one of the above compression schemes. This allows you to use a third-party dart package for handling compression if your particular application requires it.

To use this feature you need to invoke the public method registerCodec(String, Codec<Object, Uint8List>) and specify one of the Compression enum values (e.g. Compression.LZ4.value) as the first argument and a class instance implementing Codec<Uint8, Uint8> as the second argument.

After the codec is registered you may then enable compression when you define your pool configuration. Keep in mind that the codec needs to be registered before the pool configuration is instanciated or an exception will be thrown.

Here is an example for registering a codec and requesting it when creating the client:

import "package:dart_cassandra_cql/dart_cassandra_cql.dart" as cql;
import "mocks/lz4.dart" as compress;

int main() {
  cql.registerCodec(cql.Compression.LZ4.value, new compress.LZ4Codec());

  // This client will now use LZ4 compression when talking to Cassandra
  cql.Client client = new cql.Client.fromHostList(
      , poolConfig : new cql.PoolConfiguration(
          protocolVersion : cql.ProtocolVersion.V2
          , compression : cql.Compression.LZ4

The client

To create a new client instance you may use one of the two available named constructors:

  • Client.fromHostList(List<String> hosts, {String defaultKeyspace, PoolConfiguration poolConfig}) which should be used when a host list is available. You may also specify the default keyspace to be used as well as a specific pool configuration (if no configuration is specified a default one will be used). This method will instanciate a SimpleConnectionPool with the supplied parameters and bind it to the client.
  • Client.withPool(ConnectionPool this.connectionPool, {String defaultKeyspace}) which should be used with an already instanciated connection pool (simple or custom).

Single and batch queries

To define a single query, the driver provides the Query class. Its constructor accepts a query string that may contain either positional (indicated by the ? character) or named (indicated by : followed by the parameter name) arguments. Mixing positional and named parameters is not currently supported. The following named parameters may also be passed to the constructor:

| Parameter Name | Description |:---------------|:---------------- | bindings | The query string bindings. The parameter value value must be a Iterable<String> when using positional parameters or a Map<String, Object> when using named parameters | consistency | One of the Consistency enum values to select the consistency level for this query. Defaults to Consistency.QUORUM | serialConsistency | One of Consistency.SERIAL, Consistency.LOCAL_SERIAL or Consistency.LOCAL_ONE enum values to select the serial consistency value of this call. This value will be ignored when not using ProtocolVersion.V3 | prepared | A boolean flag specifying whether this query should be prepared or not. Defaults to false. For more info see the section on prepared queries

Here are some examples:

new cql.Query(
  "SELECT * FROM test.test_table WHERE id=? AND alt_id=?"
  , bindings : [ 1, 2 ]

new cql.Query(
  "SELECT * FROM test.test_table WHERE id=:id AND alt_id=:id"
  , bindings : { "id" : 1 }
  , consistency : cql.Consistency.ONE

If you need to execute a batch query, the driver provides the BatchQuery class. The class constructor accepts the following optional params:

| Parameter Name | Description |:---------------|:---------------- | consistency | One of the Consistency enum values to select the consistency level for the batch query. Defaults to Consistency.QUORUM. | serialConsistency | One of Consistency.SERIAL, Consistency.LOCAL_SERIAL or Consistency.LOCAL_ONE enum values to select the serial consistency value of this call. This value will be ignored when not using ProtocolVersion.V3. | batchType | One of the BatchType enum values to select the batch type. Defaults to BatchType.LOGGED. This value will be ignored when not using ProtocolVersion.V3.

The BatchQuery class provides the add( Query query) method for appending individual Query instances to the batch. Keep in mind that when useing batch queries, the consistency and serialConsistency settings of the BatchQuery object override any individual consistency settings specified by the appended Query objects. Here is an example:

new cql.BatchQuery(consistency: cql.Consistency.TWO)
  ..add(new cql.Query(
    "INSERT INTO test.test_table (id, alt_id) VALUES (?, ?)"
    , bindings : [ 1, 2 ]
  ..add(new cql.Query(
    "INSERT INTO test.test_table (id, alt_id) VALUES (:id, :id)"
    , bindings : {"id" : 1}

Executing queries

The client provides two methods for executing single queries: query() and execute().

To execute a single select Query, the query method should be used. It returns back a Future that evaluates to an Iterable<Map<String, Object>> with the result rows. Each row is modeled as a Map<String, Object> where the key is the column name and the value is the unserialized column value.

In all other cases (single queries or batch queries) the execute method should be used. This method accepts either a Query or BatchQuery as its argument. The method also accepts the optional pageSize and pagingState named parameters to enable pagination for single select queries. It returns back a Future that evaluates to one of the following concrete implementations of the ResultMessage class depending on the query type:

|Message type | Returnes when | |:--------------------------|:--------------- | VoidResultMessage | If the query returns no value. (e.g. insert queries) | RowsResultMessage | When executing a select query. | SetKeyspaceResultMessage | When a set keyspace query is executed. | SchemaChangeResultMessage | When an alter query is executed.

Paginated queries

In some use cases, you may need to perform client-side pagination. One way to achieve this is to encode the pagination parameters (e.g. fromDate - toDate) inside the where clause of your selection query. Another way to achieve this is to use Cassandra's native pagination support.

To use native pagination you need to invoke the execute method with your selection query and supply the pageSize named parameter. After the returned Future completes, you receive a RowsResultMessage . You can access the returned rows (a List<Map<String, Object>>) via the rows getter. The message also contains a ResultMetadata instance which can be obtained via the metadata getter. The metadata object contains the pagingState attribute whose value is generated by Cassandra and serves as a pointer for obtaining the next page of results.

To retrieve the next set of rows you need to invoke once again the execute method with the same pageSize value as before and the pagingState named parameter set to the value obtained by the previous method invocation. Here is an example:

cql.Query query = new cql.Query("SELECT * from really_big_dataset");
  .execute(query, pageSize : 10)
  .then((cql.RowsResultMessage result) {
      print("Page 1");

      // Fetch next page
      return client.execute(query, pageSize : 10, pagingState : result.metadata.pagingState);
  .then((cql.RowsResultMessage result) {
	  print("Page 2");

Streaming query results

If you need to iterate through all rows of a large dataset without loading the entire dataset to memory you can use the stream method. This method accepts a Query object and the named parameter pageSize (defaults to 100) and returns back a Dart Stream. The stream supports the usual stream-related operations (pause, resume, stop e.t.c) and can be combined with other Dart streams for additional processing.

The underlying StreamController buffers the rows for each page on-demand and emits each row to the stream as a Map<String, Object> where the key is the column name and the value is the unserialized column value. Here is an example that streams all rows from a dataset and prints each one to the console:

       new cql.Query("SELECT * FROM really_big_dataset")
       , pageSize: 20
).listen( (Map<String, Object> row) => print );

Prepared queries

If you are executing the same single Query multiple times you can increase your query throughput by converting it to a prepared query. To do this, set the prepared named constructor parameter to true when you instanciate your query object. When executing a prepared query, the driver is aware of the actual Cassandra type of each bound argument and will properly serialize the bound arguments instead of performing automatic inline expansion as with normal queries.

Prepared queries are associated with a randomly selected node from the connection pool and they can only be executed via connections to that particular node. If during execution time no connection to the associated node is available, the driver will automatically prepare the query on another random node from the pool and execute it there.

Listening for Cassandra events

The connection pool registers itself as a listener for events broadcasted by Cassandra nodes. These events include notifications about: - schema changes (keyspace/table create, update, drop) - node status changes (node came up or went down) - cluster topology changes (node added or removed)

The connection pool uses these events to add new nodes to the pool (if the autoDiscoverNodes pool configuration option is true) to remove dead nodes from the pool or to attempt to reconnect to offline nodes when they go online.

If your application needs to process these kinds of events, you can use the listenForServerEvents method of the connection pool. This method accepts a List of EventRegistrationType values and returns back a Dart Stream that emits EventMessage objects. Here is an example:

]).listen( (cql.EventMessage message ){
  print("Got a ${message.type} event with sub type ${message.subType} for keyspace ${message.keyspace}");

Shutting down the client

The shutdown the client you need to invoke the shutdown() method. This method accepts the named parameters drain (defaults to true) and drainTimeout (defaults to Duration(seconds : 5) that control how the shutdown should be performed and returns a Future.

To perform a graceful shutdown, set the drain named parameter to true. The client will mark all connections as unhealthy so that no more queries can be performed and wait for any pending queries to complete (or the drainTimeout expires) before shutting down each connection. The returned Future will complete once all active connections shut down.

If the drain named parameter is set to false then the client will mark all connections as unhealthy so that no more queries can be performed and abort any pending queries.

Handling errors

Whenever you invoke one of the available client methods, you should always watch out for errors as unhandled errors will probably terminate your application or current zone.

For more granular error handling you can test for the following exception types:

| Exception type | Thrown |:------------------------------|:------------------- | NoHealthyConnectionsException | When the connection pool contains no healthy connections | AuthenticationException | When authentication failed, missing Authentication provider or unsupported Authentication provider | CassandraException | When cassandra reports an exception (invalid query, cannot meet consistency requirement e.t.c) | DriverException | General driver exception

All of the above exceptions include a message getter for retrieving the exception cause and a stackTrace getter for accessing the stack trace (if available).


See the Contributing Guide.



dart\_cassandra\_cql is distributed under the MIT license.