Frequently Asked Questions

How do I implement paging?

When using native protocol version 2 or higher, the driver automatically pages large result sets under the hood. You can also save the paging state to resume iteration later. See this page for more information.

Native protocol v1 does not support paging, but you can emulate it in CQL with LIMIT and the token() function. See this conversation on the mailing list.

Can I check if a conditional statement (lightweight transaction) was successful?

When executing a conditional statement, the ResultSet will contain a single Row with a column named “applied” of type boolean. This tells whether the conditional statement was successful or not.

The driver provides a convenience method wasApplied to check this on the result set directly:

ResultSet rset = session.execute(conditionalStatement);

You may also inspect the value yourself:

ResultSet rset = session.execute(conditionalStatement);
Row row =;
row.getBool(0);       // this is equivalent row.getBool("applied")

Note that, unlike manual inspection, wasApplied does not consume the first row.

What is a parameterized statement and how can I use it?

Starting with Cassandra 2.0, normal statements (that is non-prepared statements) do not need to concatenate parameter values inside a query string. Instead you can use ? markers and provide the values separately:

session.execute( "INSERT INTO contacts (email, firstname, lastname)
      VALUES (?, ?, ?)", "", "Barney", "Barton");

See Simple statements for more information.

Does a parameterized statement escape parameters?

A parameterized statement sends the values of parameters separate from the query (similar to the way a prepared statement does) as bytes so there is no need to escape parameters.

What’s the difference between a parameterized statement and a Prepared statement?

The only similarity between a parameterized statement and a prepared statement is in the way that the parameters are sent. The difference is that a prepared statement:

  • is already known on the cluster side (it has been compiled and there is an execution plan available for it) which leads to better performance
  • sends only the statement id and its parameters (thus reducing the amount of data sent to the cluster)

See Prepared statements for more information.

Can I combine PreparedStatements and normal statements in a batch?

Yes. A batch can include both bound statements and simple statements:

PreparedStatement ps = session.prepare( "INSERT INTO contacts (email, firstname, lastname)
      VALUES (?, ?, ?)");
BatchStatement batch = new BatchStatement();
// here's a simple statement
batch.add(new SimpleStatement( "INSERT INTO contacts (email, firstname, lastname) VALUES (?, ?, ?)", ...));

Why do my ‘SELECT *’ PreparedStatement-based queries stop working after a schema change?

Both the driver and Cassandra maintain a mapping of PreparedStatement queries to their metadata. When a change is made to a table, such as a column being added or dropped, there is currently no mechanism for Cassandra to invalidate the existing metadata. Because of this, the driver is not able to properly react to these changes and will improperly read rows after a schema change is made.

See Prepared statements for more information.

Can I get the raw bytes of a text column?

If you need to access the raw bytes of a text column, call the Row.getBytesUnsafe("columnName") method.

Trying to use Row.getBytes("columnName") for the same purpose results in an exception, as the getBytes method can only be used if the column has the CQL type BLOB.

How do I increment counters with QueryBuilder?

Considering the following query:

UPDATE clickstream SET clicks = clicks + 1 WHERE userid = id;

To do this using QueryBuilder:

Statement query = QueryBuilder.update("clickstream")
                              .with(incr("clicks", 1)) // Use incr for counters
                              .where(eq("userid", id));

Is there a way to control the batch size of the results returned from a query?

Use the setFetchSize() method on your Statement object. The fetch size controls how many resulting rows are retrieved simultaneously (the goal being to avoid loading too many results in memory for queries yielding large result sets).

Keep in mind that if your code iterates the ResultSet entirely, the driver may run additional background queries to fetch the rest of the data. The fetch size only affects what is retrieved at a time, not the overall number of rows.

See Paging for more information.

What’s the difference between using setFetchSize() and LIMIT?

Basically, LIMIT controls the maximum number of results returned by the query, while the setFetchSize() method controls the amount of data transferred at a time.

For example, if you limit is 30 and your fetch size is 10, the ResultSet will contain 30 rows, but under the hood the driver will perform 3 requests that will transfer 10 rows each.

See Paging for more information.

I’m reading a BLOB column and the driver returns incorrect data.

Check your code to ensure that you read the returned ByteBuffer correctly. ByteBuffer is a very error-prone API, and we’ve had many reports where the problem turned out to be in user code.

See in the driver-examples module for some examples and explanations.

How do I use the driver in an OSGi application?

Read our OSGi-specific FAQ section to find out.

Why am I seeing messages about tombstone_warn_threshold or tombstone_fail_threshold being exceeded in my Cassandra logs?

Applications which use the object mapper or set null values in their statements may observe that many tombstones are being stored in their tables which subsequently may lead to poor query performance, failed queries, or columns being mysteriously deleted.

This is caused by INSERT/UPDATE statements containing null values for columns that a user does not intend to change. Common circumstances around this come from using the object mapper or writing your own persistence layer and attempting to reuse the same PreparedStatement for inserting data, even with partial updates.

Prior to cassandra 2.2, there was no means of reusing the same PreparedStatement for making partial updates to different columns.

For example, given the following code:

PreparedStatement prepared = session.prepare("INSERT INTO contacts (email, firstname, lastname) VALUES (?, ?, ?)");
BoundStatement bound = prepared.bind();
bound.set("email", "");
bound.set("firstname", "Barney");
// creates a tombstone!!
bound.set("lastname", null);

If one wanted to use this query to update only firstname this would not be achievable without binding the lastname parameter to null. This would have an undesired side effect of creating a tombstone for lastname and thus to the user giving the impression that lastname was deleted.

In cassandra 2.2 and later with protocol v4, bind parameters (?) can optionally be left unset. (CASSANDRA-7304):

 PreparedStatement prepared = session.prepare("INSERT INTO contacts (email, firstname, lastname) VALUES (?, ?, ?)");
 BoundStatement bound = prepared.bind();
 bound.set("email", "");
 bound.set("firstname", "Barney");
 // lastname is left unset.

See Parameters and Binding for more details about unset parameters.

Another possible root cause for this is using the object mapper and leaving fields set to null. This also causes tombstones to be inserted unless setting saveNullFields option to false. See Mapper options for more details.

I am encountering BusyPoolException, what does this mean and how do I avoid it?

Often while writing a bulk loading program or an application that does many concurrent operations a user may encounter exceptions such as the following:

com.datastax.driver.core.exceptions.BusyPoolException: [/X.X.X.X] Pool is busy (no available connection and the queue has reached its max size 256)

This typically means that your application is submitting too many concurrent requests and not enforcing any limitation on the number of requests that are being processed at once. A common mistake users might make is firing off a bunch of queries using executeAsync and not waiting for them to complete before submitting more.

One simple approach to remedying this is to use something like a Semaphore to limit the number of concurrent executeAsync requests at a time. Alternatively, one could submit requests X at a time and collect the returned ResultSetFutures from executeAsync and use Futures.allAsList and wait on completion of the resulting future before submitting the next batch.

See the Acquisition queue section of the Pooling section in the manual for explanation of how the driver enqueues requests when connections are over-utilized.

What is Netty’s native epoll transport and how do I enable or disable it?

Netty provides native transport libraries which generally generate less garbage and improve performance when compared to the default NIO-based transport. By default if the driver detects the netty-transport-native-epoll library in its classpath it will attempt to use EpollEventLoopGroup for its underlying event loop.

In the usual case this works fine in linux environments. On the other hand, many users have run into compatibility issues when the version of netty-transport-native-epoll is not compatible with a version of Netty in an application’s classpath. One such case is where an application depends on a version of netty-all that is different than the version of netty-handler that the driver depends on. In such a case, a user may encounter an exception such as the one described in JAVA-1535.

While the epoll transport may in general improve performance, we expect the improvement to be marginal for a lot of use cases. Therefore, if you don’t want netty-transport-native-epoll to be used by the driver even if the library is present in an application’s classpath, the most direct way to disable this is to provide the system property value -Dcom.datastax.driver.FORCE_NIO=true to your application to force the use of the default Netty NIO-based event loop. If properly used, the following log message will be logged at INFO on startup:

Found Netty’s native epoll transport in the classpath, but NIO was forced through the FORCE_NIO system property.