Native Erlang client for CQL3 over Cassandra's latest binary protocol v4.

Usage · Connecting · Clusters · Performing queries · Query options · Batched queries · Reusable queries · Data types

Installation · Compatibility · Tests · License

CQErl offers a simple Erlang interface to Cassandra using the latest CQL version. The main features include:

• Automatic connection pooling, with hash-based allocation (a la dispcount)
• Single or multi-cluster support
• Batched queries
• Variable bindings in CQL queries (named or positional)
• Automatic query reuse when including variable bindings
• Collection types support
• User-defined type support
• Tunable consistency level
• Automatic pagination support
• Synchronous or asynchronous queries
• Automatic compression (using lz4 or snappy if available)
• SSL support
• Pluggable authentication (as long as it's SASL-based)

CQErl was designed to be as simple as possible on your side. You just provide the configuration you want as environment variables, and ask to get a client everytime you need to perform a transient piece of work (e.g. handle a web request). You do not need to (and in fact should not) keep a client in state for a long time. Under the hood, CQErl maintains a pool of persistent connections with Cassandra and this pattern is the best way to ensure proper load balancing of requests across the pool.

If you installed cassandra and didn't change any configuration related to authentication or SSL, you should be able to connect like this

{ok, Client} = cqerl:get_client({}).

You do not need to close the connection after you've finished using it.

1. The first argument to cqerl:get_client/2,1 is the node to which you wish to connect as {Ip, Port}. If empty, it defaults to {"127.0.0.1", 9042}, and Ip can be given as a string, or as a tuple of components, either IPv4 or IPv6.

   • If the default port is used, you can provide just the IP address as the first argument, either as a tuple, list or binary.
   • If both the default port and localhost are used, you can just provide an empty tuple as the first argument.

2. The second possible argument (when using cqerl:get_client/2) is a list of options, that include:

   • keyspace which determines in which keyspace all subsequent requests operate, on that connection.
   • auth (mentionned below)
   • ssl (which is false by default, but can be set to a list of SSL options) and keyspace (string or binary).
   • protocol_version to connect to older Cassandra instances.

   If you've set simple username/password authentication scheme on Cassandra, you can provide those to CQErl

   {ok, Client} = cqerl:get_client({}, [{auth, {cqerl_auth_plain_handler, [{"test", "aaa"}]}}]).

   Since Cassandra implements pluggable authentication mechanisms, CQErl also allows you to provide custom authentication modules (here cqerl_auth_plain_handler). The options you pass along with it are given to the module's auth_init/3 as its first argument.

3. You can leverage one or more clusters of cassandra nodes by setting up clusters. When set up, you can use

   1. cqerl:get_client() if you have just a single main cluster
   2. cqerl:get_client(ClusterKey) if you want to get a client from a specific, identified cluster

All the options given above can be provided as environment variables, in which case they are used as default (and overridable) values to any cqerl:get_client calls. You can also provide a cassandra_nodes variable containing a list of the tuples used as the first argument to cqerl:get_client (see clusters for more explanations). So for example, in your app.config or sys.config file, you could have the following content:

[
  {cqerl, [
            {cassandra_nodes, [ { "127.0.0.1", 9042 } ]},
            {ssl, [ {cacertfile, "cassandra.pem"} ]},
            {auth, {cqerl_auth_plain_handler, [ {"test", "aaa"} ]}}
          ]},
]

Doing so will fire up connection pools as soon as the CQErl application is started. So when later on you call cqerl:get_client, chances are you will hit a preallocated connection (unless they're so busy that CQErl needs to fire up new ones). In fact, if you provide the cassandra_nodes environment variable, you can call cqerl:get_client/0, which chooses an available client at random.

With CQErl clusters, you can configure either a single set of cassandra nodes from which you can draw a client at any time, or multiple sets that serve different purposes.

You can prepare a single cluster setup using this structure in your sys.config file:

[
  {cqerl, [ {cassandra_nodes, [ 
                % You can use any of the forms below to specify a cassandra node
                { "127.0.0.1", 9042 },
                { {127, 0, 0, 2}, 9042 },
                "127.0.0.3"
            ]},
            { keyspace, dev_keyspace }
          ]},
]

or, equivalently, there's an API you can use to add nodes to the single main cluster:

cqerl_cluster:add_nodes([
    { "127.0.0.1", 9042},
    { {127, 0, 0, 2}, 9042 },
    "127.0.0.3"
]).

or, with connection options:

cqerl_cluster:add_nodes([
    { "127.0.0.1", 9042},
    { {127, 0, 0, 2}, 9042 },
    "127.0.0.3"
], [
    { keyspace, dev_keyspace }
]).

When your main cluster is configured, you can just use cqerl:get_client/0 to get a client random from the cluster.

You can prepare multiple clusters using this structure in your sys.config file:

[
  {cqerl, [ {cassandra_clusters, [
                { config, {
                    [ "127.0.0.1", "127.0.0.3" ], 
                    [ { keyspace, config } ] 
                }},
                { operations, {
                    [ "127.0.0.1:9042", {"127.0.0.1", 9042} ], 
                    [ { keyspace, operations } ] 
                }}
            ]},
          ]},
]

or, equivalently, there's an API you can use to add nodes to particular clusters:

cqerl_cluster:add_nodes(config, [
    { "127.0.0.1", 9042},
    "127.0.0.3"
], [
    { keyspace, config }
]).
cqerl_cluster:add_nodes(operations, [
    { "127.0.0.1", 9042},
    "127.0.0.3"
], [
    { keyspace, operations }
]).

There are two application environment variables that may be set to change query behaviour:

• {maps, true} will cause query result rows to be returned as maps instead of proplists
• {text_uuids, true} will cause timeuuid and uuid fields to be returned as binary strings in canonical form (eg <<"5620c844-e98d-11e5-b97b-08002719e96e">>) rather than pure binary.

Performing a query can be as simple as this:

{ok, Result} = cqerl:run_query(Client, "SELECT * FROM users;").

% Equivalent to
{ok, Result} = cqerl:run_query(Client, <<"SELECT * FROM users;">>).

% Also equivalent to
{ok, Result} = cqerl:run_query(Client, #cql_query{statement = <<"SELECT * FROM users;">>}).

It can also be performed asynchronously using

Tag = cqerl:send_query(Client, "SELECT * FROM users;"),
receive
    {result, Tag, Result} ->
        ok
end.

In situations where you do not need to wait for the response at all, it's perfectly fine to produce this sort of pattern:

{ok, Client} = cqerl:get_client(),
cqerl:send_query(Client, #cql_query{statement="UPDATE secrets SET value = null WHERE id = ?;",
                                    values=[{id, <<"42">>}]}).

That is, you can grab a client only the send a query, then you can get rid of it. CQErl will still perform it, the difference being that no response will be sent back to you.

Here's a rundown of the possible return values

• SELECT queries will yield result of type #cql_result{} (more details below).
• Queries that change the database schema will yield result of type #cql_schema_changed{type, keyspace, table}
• Other queries will yield void if everything worked correctly.
• In any case, errors returned by cassandra in response to a query will be the return value ({error, Reason} in the synchronous case, and {error, Tag, Reason} in the asynchronous case).

The return value of SELECT queries will be a #cql_result{} record, which can be used to obtain rows as proplists and fetch more result if available

{ok, _SchemaChange} = cqerl:run_query(Client, "CREATE TABLE users(id uuid, name varchar, password varchar);"),
{ok, void} = cqerl:run_query(Client, #cql_query{
    statement = "INSERT INTO users(id, name, password) VALUES(?, ?, ?);",
    values = [
        {id, new},
        {name, "matt"},
        {password, "qwerty"}
    ]
}),
{ok, Result} = cqerl:run_query(Client, "SELECT * FROM users;").

Row = cqerl:head(Result),
Tail = cqerl:tail(Result),
{Row, Tail} = cqerl:next(Result),
1 = cqerl:size(Result),
0 = cqerl:size(Tail),
empty_dataset = cqerl:next(Tail),
[Row] = cqerl:all_rows(Result),

<<"matt">> = proplists:get_value(name, Row),
<<"qwerty">> = proplists:get_value(password, Row).

#cql_result{} can also be used to fetch the next page of result, if applicable, synchronously or asynchronously. This uses the automatic paging mechanism described here.

case cqerl:has_more_pages(Result) of
    true -> {ok, Result2} = cqerl:fetch_more(Result);
    false -> ok
end,

Tag2 = cqerl:fetch_more_async(Result),
receive
    {result, Tag2, Result2} -> ok
end.

#cql_schema_changed{} is returned from queries that change the database schema somehow (e.g. ALTER, DROP, CREATE, and so on). It includes:

1. The type of change, either created, updated or dropped
2. The name of the keyspace where the change happened, as a binary
3. If applicable, the name of table on which the change was applied, as a binary

When performing queries, you can provide more information than just the query statement using the #cql_query{} record, which includes the following fields:

1. The query statement, as a string or binary

2. values for binding variables from the query statement (see next section).

3. You can tell CQErl to consider a query reusable or not (see below for what that means). By default, it will detect binding variables and consider it reusable if it contains (named or not) any. Queries containing named binding variables will be considered reusable no matter what you set reusable to. If you explicitely set reusable to false on a query having positional variable bindings (?), you would provide values with in {Type, Value} pairs instead of {Key, Value}.

4. You can specify how many rows you want in every result page using the page_size (integer) field. The devs at Cassandra recommend a value of 100 (which is the default).

5. You can also specify what consistency you want the query to be executed under. Possible values include:

   • any
   • one
   • two
   • three
   • quorum
   • all
   • local_quorum
   • each_quorum
   • local_one

6. In case you want to perform a lightweight transaction using INSERT or UPDATE, you can also specify the serial_consistency that will be use when performing it. Possible values are:

   • serial
   • local_serial

In the #cql_query{} record, you can provide values as a proplists or map, where the keys are all atoms and match the column names or binding variable names in the statement, in lowercase.

Example:

% Deriving the value key from the column name
#cql_query{statement="SELECT * FROM table1 WHERE id = ?", values=[{id, SomeId}]},

% Explicitly providing a binding variable name
#cql_query{statement="SELECT * FROM table1 WHERE id = :id_value", values=[{id_value, SomeId}]},

Special cases include:

• providing TTL and TIMESTAMP option in statements, in which case the proplist key would be [ttl] and [timestamp] respectively. Note that, while values for a column of type timestamp are provided in milliseconds, a value for the TIMESTAMP option is expected in microseconds.

• UPDATE keyspace SET set = set + ? WHERE id = 1;. The name for this variable binding is set, the name of the column, and it's expected to be an erlang list of values.

• UPDATE keyspace SET list = list + ? WHERE id = 1;. The name for this variable binding is list, the name of the column, and it's expected to be an erlang list of values.

• UPDATE keyspace SET map[?] = 1 WHERE id = 1;. The name for this variable binding is key(map), where map is the name of the column.

• UPDATE keyspace SET map['key'] = ? WHERE id = 1;. The name for this variable binding is value(map), where map is the name of the column.

• UPDATE keyspace SET list[?] = 1 WHERE id = 1;. The name for this variable binding is idx(list), where list is the name of the column.

• SELECT * FROM keyspace LIMIT ?. The name for the LIMIT variable is [limit].

  Also, when providing the value for a uuid-type column, you can give the value new, strong or weak, in which case CQErl will generate a random UUID (v4), with either a strong or weak number random generator.

  Finally, when providing the value for a timeuuid or timestamp column, you can give the value now, in which case CQErl will generate a normal timestamp, or a UUID (v1) matching the current date and time.

To perform batched queries (which can include any non-SELECT DML statements), simply put one or more #cql_query{} records in a #cql_query_batch{} record, and run it in place of a normal #cql_query{}. #cql_query_batch{} include the following fields:

1. The consistency level to apply when executing the batch of queries.
2. The mode of the batch, which can be logged, unlogged or counter. Running a batch in unlogged mode removes the performance penalty of enforcing atomicity. The counter mode should be used to perform batched mutation of counter values.
3. Finally, you must specify the list of queries.

InsertQ = #cql_query{statement = "INSERT INTO users(id, name, password) VALUES(?, ?, ?);"},
{ok, void} = cqerl:run_query(Client, #cql_query_batch{
  mode=unlogged,
  queries=[
    InsertQ#cql_query{values = [{id, new},{name, "sean"},{password, "12312"}]},
    InsertQ#cql_query{values = [{id, new},{name, "jenna"},{password, "11111"}]},
    InsertQ#cql_query{values = [{id, new},{name, "kate"},{password, "foobar"}]}
  ]
}).

If any of the following is true:

• you set #cql_query{}'s reusable field to true
• the query contains positional variable bindings (?) and you did not explicitely reusable to false
• the query contains named variable bindings (:name) (ignores the value of reusable)

the query is considered reusable. This means that the first time this query will be performed, CQErl will ask the connected Cassandra node to prepare the query, after which, internally, a query ID will be used instead of the query statement when executing it. That particular cassandra node will hold on to the prepared query on its side and subsequent queries that use exactly the same statement will be performed faster and with less network traffic.

CQErl can tell which query has been previously prepared on which node by keeping a local cache, so all of this happens correctly and transparently.

Here is a correspondance of cassandra column types with their equivalent Erlang types (bold denotes what will used in result sets, the rest is what is accepted).

By default, this client library assumes we're talking to a 2.2+ or 3+ instance of Cassandra. 2.1.x the latest native protocol (v4) which is required to use some of the newest datatypes and optimizations. To tell CQErl to use the older protocol version (v3), which is required to connect to a 2.1.x instance of Cassandra, you can set the protocol_version option to the integer 3, in your configuration file, i.e.

[
  {cqerl, [
            {cassandra_nodes, [ { "127.0.0.1", 9042 } ]},
            {protocol_version, 3}
          ]},
]

or in a cqerl:get_client/2 or cqerl:get_client/2 call

{ok, Client} = cqerl:get_client("127.0.0.1:9042", [{protocol_version, 3}, {keyspace, oper}]).

Just include this repository in your project's rebar.config file and run ./rebar get-deps. See rebar for more details on how to use rebar for Erlang project management.

As said earlier, this library uses Cassandra's newest native protocol versions (v4, or v3 optionally), which is said to perform better than the older Thrift-based interface. It also speaks CQL version 3, and uses new features available in Cassandra 2.X, such as paging, parametrization, query preparation and so on.

All this means is that this library works with Cassandra 2.1.x (2.2+ or 3+ recommended), configured to enable the native protocol. This documentation page gives details about the how to configure this protocol. In the cassandra.yaml configuration file of your Cassandra installation, the start_native_transport need to be set to true and you need to take note of the value for native_transport_port, which is the port used by this library.

CQErl includes a test suite that you can run yourself, especially if you plan to contribute to this project.

1. Clone this repo on your machine
2. Edit test/test.config and put your own cassandra's configurations
3. At the project's top directory, run make test