Skip to content
Switch branches/tags
Go to file
This branch is 227 commits ahead, 1 commit behind dork:master.

Latest commit

The chat seems to be abandoned for a long time.

Git stats


Failed to load latest commit information.
Latest commit message
Commit time

Java connector for Tarantool 1.7.4+

Maven Central Build Status Coverage Status

To get the Java connector for Tarantool 1.6.9, visit this GitHub page.

Table of contents

Getting started

  1. Add a dependency to your pom.xml file (look for a last version on the releases page):
  1. Configure TarantoolClientConfig:
TarantoolClientConfig config = new TarantoolClientConfig();
config.username = "test";
config.password = "test";
  1. Create a client:
TarantoolClient client = new TarantoolClientImpl("host:3301", config);

using TarantoolClientImpl(String, TarantoolClientConfig) is equivalent to:

SocketChannelProvider socketChannelProvider = new SingleSocketChannelProviderImpl("host:3301")
TarantoolClient client = new TarantoolClientImpl(socketChannelProvider, config);

You could implement your own SocketChannelProvider. It should return a connected SocketChannel. Feel free to implement get(int retryNumber, Throwable lastError) using your appropriate strategy to obtain the channel. The strategy can take into account current attempt number (retryNumber) and the last transient error occurred on the previous attempt.

The TarantoolClient will be closed if your implementation of a socket channel provider raises exceptions. However, throwing a SocketProviderTransientException or returning null value are handled by the client as recoverable errors. In these cases, the client will make next attempt to obtain the socket channel. Otherwise, you will need a new instance of client to recover. Hence, you should only throw an error different to SocketProviderTransientException in case you have met unrecoverable error.

Below is an example of SocketChannelProvider implementation that tries to connect no more than 3 times, two seconds for each attempt at max.

SocketChannelProvider socketChannelProvider = new SocketChannelProvider() {
    public SocketChannel get(int retryNumber, Throwable lastError) {
        if (retryNumber > 3) {
            throw new RuntimeException("Too many attempts");
        SocketChannel channel = null;
        try {
            channel =;
            channel.socket().connect(new InetSocketAddress("localhost", 3301), 2000);
            return channel;
        } catch (IOException e) {
            if (channel != null) {
                 try {
                 } catch (IOException ignored) { }
            throw new SocketProviderTransientException("Couldn't connect to server", e);

Same behaviour can be achieved using built-in SingleSocketChannelProviderImpl:

TarantoolClientConfig config = new TarantoolClientConfig();
config.connectionTimeout = 2_000; // two seconds timeout per attempt
config.retryCount = 3;            // three attempts at max

SocketChannelProvider socketChannelProvider = new SingleSocketChannelProviderImpl("localhost:3301")
TarantoolClient client = new TarantoolClientImpl(socketChannelProvider, config);

SingleSocketChannelProviderImpl implements ConfigurableSocketChannelProvider that makes possible for the client to configure a socket provider.


  • TarantoolClient is thread-safe and asynchronous, so you should use one client inside the whole application.
  • TarantoolClient does not support name resolution for fields, indexes, spaces and so on. We highly recommend to use server-side Lua when working with named items. For example, you could create a data access object (DAO) with simple CRUD functions. If, for some reason, you do need client name resolution, you could create a function that returns necessary name-to-ID mappings.

TarantoolClient provides four interfaces to execute queries:

  • SyncOps - returns the operation result
  • AsyncOps - returns the operation result as a Future
  • ComposableAsyncOps - return the operation result as a CompletionStage
  • FireAndForgetOps - returns the query ID

Feel free to override any method of TarantoolClientImpl. For example, to hook all the results, you could override this:

protected void complete(TarantoolPacket packet, CompletableFuture<?> future);

Client config options

The client configuration options are represented through the TarantoolClientConfig class.

Supported options are follow:

  1. username is used to authenticate and authorize an user in a Taratool server instance. Default value is null that means client will attempt to auth as a guest.
  2. password is used to authenticate an user in a Taratool server instance. Default value is null.
  3. defaultRequestSize used to be an initial binary buffer size in bytes to send requests. Default value is 4096 (4 KB).
  4. predictedFutures is used to initialize an initial capacity of hash map which stores response futures. The client is asynchronous under the hood even though it provides a synchronous operations using java.concurrent.CompletableFuture. Default value is (1024 * 1024) / 0.75) + 1.
  5. writerThreadPriority describes a priority of writer thread. Default value is Thread.NORM_PRIORITY (5).
  6. readerThreadPriority describes a priority of reader thread. Default value is Thread.NORM_PRIORITY (5).
  7. sharedBufferSize sets a shared buffer size in bytes (place where client collects requests when socket is busy on write). Default value is 8 * 1024 * 1024 (8 MB).
  8. directWriteFactor is used as a factor to calculate a threshold whether request will be accommodated in the shared buffer. If the request size exceeds directWriteFactor * sharedBufferSize request is sent directly. Default value is 0.5.
  9. writeTimeoutMillis sets the max time in ms to perform writing and send the bytes. Default value is 60 * 1000 (1 minute).
  10. useNewCall configures whether client has to use new CALL request signature or old one used to be active in Tarantool 1.6. Default value is true.
  11. initTimeoutMillis sets a max time in ms to establish connection to the server Default values is 60 * 1000L (1 minute).
  12. connectionTimeout is a hint and can be passed to the socket providers which implement ConfigurableSocketChannelProvider interface. This hint should be interpreter as a connection timeout in ms per attempt where 0 means no limit. This options restricts a time budget to perform one connection attempt, while initTimeoutMillis limits an overall time to obtain a connection. Default value is 2 * 1000 (2 seconds).
  13. retryCount is a hint and can be passed to the socket providers which implement ConfigurableSocketChannelProvider interface. This hint should be interpreter as a maximal number of attempts to connect to Tarantool instance. Default value is 3.
  14. operationExpiryTimeMillis is a default request timeout in ms. Default value is 1000 (1 second).

String space/index resolution

Each operation that requires space or index to be executed, can work with number ID as well as string name of a space or an index. Assume, we have my_space space with space ID 512 and its primary index primary with index ID 0. Then, for instance, select operations can be performed using their names:

client.syncOps().select(512, 0, Collections.singletonList(1), 0, 1, Iterator.EQ);
// or using more convenient way
client.syncOps().select("my_space", "primary", Collections.singletonList(1), 0, 1, Iterator.EQ);

Because iproto has not yet supported string spaces and indexes, a client caches current server schema in memory. The client relies on protocol SCHEMA_ID and sends each request with respect to cached schema version. The schema is used primarily to resolve string names of spaces or indexes against its integer IDs.

Schema update

  1. Just after a (re-)connection to the Tarantool instance. The client cannot guarantee that new instance is the same and has same schema, thus, the client drops the cached schema and fetches new one.
  2. Receiving a schema version error as a response to our request. It's possible some request can be rejected by server because of schema mismatching between client and server. In this case the schema will be reloaded and the refused request will be resent using the updated schema version.
  3. Sending a DDL request and receiving a new version in a response.
  4. Sending a request against a non-existent space/index name. The client cannot exactly know whether name was not found because of it does not exist or it has not the latest schema version. A ping request is sent in the case to check a schema version and then a client will reload it if needed. The original request will be retried if a space / an index name will be found in a new schema.

Schema support caveats

  1. Each schema reloading requires at least two extra requests to fetch spaces and indexes metadata respectively. There is also a ping request followed by reloading of the schema to check whether the client has outdated version (see point 4 in Schema update).
  2. In some circumstance, requests can be rejected several times until both client's and server's versions matches. It may take significant amount of time or even be a cause of request timeout.
  3. The client guarantees an order of synchronous requests per thread. Other cases such as asynchronous or multi-threaded requests may be out of order before the execution.

Spring NamedParameterJdbcTemplate usage example

The JDBC driver uses TarantoolClient implementation to provide a communication with server. To configure socket channel provider you should implements SocketChannelProvider and add to connect url.

For example:


Here is an example how you can use the driver covered by Spring DriverManagerDataSource:

NamedParameterJdbcTemplate template = new NamedParameterJdbcTemplate(new DriverManagerDataSource("jdbc:tarantool://localhost:3301?user=test&password=test"));
RowMapper<Object> rowMapper = new RowMapper<Object>() {
    public Object mapRow(ResultSet resultSet, int i) throws SQLException {
        return Arrays.asList(resultSet.getInt(1), resultSet.getString(2));

try {
    System.out.println(template.update("drop table hello_world", Collections.<String, Object>emptyMap()));
} catch (Exception ignored) {

System.out.println(template.update("create table hello_world(hello int not null PRIMARY KEY, world varchar(255) not null)", Collections.<String, Object>emptyMap()));
Map<String, Object> params = new LinkedHashMap<String, Object>();
params.put("text", "hello world");
params.put("id", 1);

System.out.println(template.update("insert into hello_world(hello, world) values(:id,:text)", params));
System.out.println(template.query("select * from hello_world", rowMapper));

System.out.println(template.query("select * from hello_world where hello=:id", Collections.singletonMap("id", 1), rowMapper));

For more implementation details, see API documentation.


Batch updates

Statement and PreparedStatement objects can be used to submit batch updates.

For instance, using Statement object:

Statement statement = connection.createStatement();
statement.addBatch("INSERT INTO student VALUES (30, 'Joe Jones')");
statement.addBatch("INSERT INTO faculty VALUES (2, 'Faculty of Chemistry')");
statement.addBatch("INSERT INTO student_faculty VALUES (30, 2)");

int[] updateCounts = stmt.executeBatch();

or using PreparedStatement:

PreparedStatement stmt = con.prepareStatement("INSERT INTO student VALUES (?, ?)");
stmt.setInt(1, 30);
stmt.setString(2, "Michael Korj");
stmt.setInt(1, 40);
stmt.setString(2, "Linda Simpson");

int[] updateCounts = stmt.executeBatch();

The connector uses a pipelining when it performs a batch request. It means each query is asynchronously sent one-by-one in order they were specified in the batch.

There are a couple of caveats:

  • JDBC spec recommends that auto-commit mode should be turned off to prevent the driver from committing a transaction when a batch request is called. The connector is not support transactions and auto-commit is always enabled, so each statement from the batch is executed in its own transaction.

  • DDL operations aren't transactional in Tarantool. In this way, a batch like this can produce an undefined behaviour (i.e. second statement can fail with an error that student table does not exist).

statement.addBatch("CREATE TABLE student (id INT PRIMARY KEY, name VARCHAR(100))");
statement.addBatch("INSERT INTO student VALUES (1, 'Alex Smith')");
  • If vinyl storage engine is used an execution order of batch statements is not specified. NOTE: This behaviour is incompatible with JDBC spec in the sentence "Batch commands are executed serially (at least logically) in the order in which they were added to the batch"

  • The driver continues processing the remaining commands in a batch once execution of a command fails.

Cluster support

To be more fault-tolerant the connector provides cluster extensions. In particular TarantoolClusterClient and built-in RoundRobinSocketProviderImpl used as a default SocketProvider implementation. When currently connected instance is down then the client will try to reconnect to the first available instance using strategy defined in a socket provider. You need to supply a list of nodes which will be used by the cluster client to provide such ability. Also you can prefer to use a discovery mechanism in order to dynamically fetch and apply the node list.

The RoundRobinSocketProviderImpl class

This cluster-aware provider uses addresses pool to connect to DB server. The provider picks up next address in order the addresses were passed.

Similar to SingleSocketChannelProviderImpl this RR provider also relies on two options from the config: TarantoolClientConfig.connectionTimeout and TarantoolClientConfig.retryCount but in a bit different way. The latter option says how many times the provider should try to establish a connection to one instance before failing an attempt. The provider requires positive retry count to work properly. The socket timeout is used to limit an interval between connections attempts per instance. In other words, the provider follows a pattern connection should succeed after N attempts with M interval between them at max.

Basic cluster client usage

  1. Configure TarantoolClusterClientConfig:
TarantoolClusterClientConfig config = new TarantoolClusterClientConfig();
// fill other settings
config.operationExpiryTimeMillis = 2000;
config.executor = Executors.newSingleThreadExecutor();
  1. Create an instance of TarantoolClusterClientImpl. You need to provide an initial list of nodes:
String[] nodes = new String[] { "myHost1:3301", "myHost2:3302", "myHost3:3301" };
TarantoolClusterClient client = new TarantoolClusterClient(config, nodes);
  1. Work with the client using same API as defined in TarantoolClient:
client.syncOps().insert(23, Arrays.asList(1, 1));


Auto-discovery feature allows a cluster client to fetch addresses of cluster nodes to reflect changes related to the cluster topology. To achieve this you have to create a Lua function on the server side which returns a single array result. Client periodically polls the server to obtain a fresh list and apply it if its content changes.

  1. On the server side create a function which returns nodes:
tarantool> function get_cluster_nodes() return { 'host1:3301', 'host2:3302', 'host3:3301' } end

You need to pay attention to a function contract we are currently supporting:

  • The client never passes any arguments to a discovery function.
  • A discovery function must return an array of strings (i.e return {'host1:3301', 'host2:3301'}).
  • Each string should satisfy the following pattern host[:port] (or more strictly /^[^:]+(:\d+)?$/ - a mandatory host containing any string and an optional colon followed by digits of the port). Also, the port must be in a range between 1 and 65535 if one is presented.
  • A discovery function may return multi-results but the client takes into account only first of them (i.e. return {'host:3301'}, discovery_delay, where the second result is unused). Even more, any extra results are reserved by the client in order to extend its contract with a backward compatibility.
  • A discovery function should NOT return no results, empty result, wrong type result, and Lua errors. The client discards such kinds of results but it does not affect the discovery process for next scheduled tasks.
  1. On the client side configure discovery settings in TarantoolClusterClientConfig:
TarantoolClusterClientConfig config = new TarantoolClusterClientConfig();
// fill other settings
config.clusterDiscoveryEntryFunction = "get_cluster_nodes"; // discovery function used to fetch nodes 
config.clusterDiscoveryDelayMillis = 60_000;                // how often client polls the discovery server  
  1. Create a client using the config made above.
TarantoolClusterClient client = new TarantoolClusterClient(config);
client.syncOps().insert(45, Arrays.asList(1, 1));

Auto-discovery caveats

  • You need to set not empty value to clusterDiscoveryEntryFunction to enable auto-discovery feature.
  • There are only two cases when a discovery task runs: just after initialization of the cluster client and a periodical scheduler timeout defined in TarantoolClusterClientConfig.clusterDiscoveryDelayMillis.
  • A discovery task always uses an active client connection to get the nodes list. It's in your responsibility to provide a function availability as well as a consistent nodes list on all instances you initially set or obtain from the task.
  • Every address which is unmatched with host[:port] pattern will be filtered out from the target addresses list.
  • If some error occurs while a discovery task is running then this task will be aborted without any after-effects for next task executions. These cases, for instance, are a wrong function result (see discovery function contract) or a broken connection. There is an exception if the client is closed then discovery process will stop permanently.
  • It's possible to obtain a list which does NOT contain the node we are currently connected to. It leads the client to try to reconnect to another node from the new list. It may take some time to graceful disconnect from the current node. The client does its best to catch the moment when there are no pending responses and perform a reconnection.

Cluster client config options

In addition to the options for the standard client, cluster config provides some extra options:

  1. executor defines an executor that will be used as a thread of execution to retry writes. Default values is null which means the cluster client will use a single thread executor.
  2. clusterDiscoveryEntryFunction is a name of the stored function to be used to fetch list of instances. Default value is null (not set).
  3. clusterDiscoveryDelayMillis describes how often in ms to poll the server for a new list of cluster nodes. Default value is 60 * 1000 (1 minute).


The connector uses its own logging facade to abstract from any logging libraries which can be used inside the apps where the connector attached. At the moment, the facade supports JUL as a default logging system, SLF4J facade, and Logback directly via SLF4J interface.

Logging integration

The logging facade offers several ways in integrate its internal logging with foreign one in order:

  • Using system property org.tarantool.logging.provider. Supported values are jdk and slf4j for the java util logging and SLF4J/Logback respectively. For instance, use java -Dorg.tarantool.logging.provider=slf4j <...>.

  • Using Java SPI mechanism. Implement your own provider org.tarantool.logging.LoggerProvider To register your provider save file with a single line text contains a fully-qualified class name of the provider implementation.

cat META-INF/services/org.tarantool.logging.LoggerProvider
  • CLASSPATH exploring. Now, the connector will use SLF4J if Logback is also in use.

  • If nothing is successful JUL will be used by default.

Supported loggers

Logger name Level Description
o.t.c.TarantoolClusterStoredFunctionDiscoverer WARN prints out invalid discovery addresses
o.t.TarantoolClusterClient TRACE prints out request retries after transient errors
o.t.TarantoolClientImpl WARN prints out reconnect issues


To run unit tests use:

./mvnw clean test

To run integration tests use:

./mvnw clean verify

Where to get help

Got problems or questions? Post them on Stack Overflow with the tarantool and java tags, or use these tags to search the existing knowledge base for possible answers and solutions.


A Java client for Tarantool




No packages published