Schema metadata

Quick overview


  • immutable (must invoke again to observe changes).
  • getting notifications: CqlSession.builder().withSchemaChangeListener.
  • enabling/disabling: advanced.metadata.schema.enabled in the configuration, or session.setSchemaMetadataEnabled().
  • filtering: advanced.metadata.schema.refreshed-keyspaces in the configuration.
  • schema agreement: wait for the schema to replicate to all nodes (may add latency to DDL statements).

Metadata#getKeyspaces returns a client-side representation of the database schema:

Map<CqlIdentifier, KeyspaceMetadata> keyspaces = session.getMetadata().getKeyspaces();
KeyspaceMetadata system = keyspaces.get(CqlIdentifier.fromCql("system"));
System.out.println("The system keyspace contains the following tables:");
for (TableMetadata table : system.getTables().values()) {
      "  %s (%d columns)%n", table.getName().asCql(true), table.getColumns().size());

Schema metadata is fully immutable (both the map and all the objects it contains). It represents a snapshot of the database at the time of the last metadata refresh, and is consistent with the token map of its parent Metadata object. Keep in mind that Metadata is itself immutable; if you need to get the latest schema, be sure to call session.getMetadata().getKeyspaces() again (and not just getKeyspaces() on a stale Metadata reference).


All schema metadata interfaces accessible through Metadata.getKeyspaces() have a DSE-specific subtype in the package com.datastax.dse.driver.api.core.metadata.schema. The objects returned by the DSE driver implement those types, so you can safely cast:

for (KeyspaceMetadata keyspace : session.getMetadata().getKeyspaces().values()) {
  DseKeyspaceMetadata dseKeyspace = (DseKeyspaceMetadata) keyspace;

If you’re calling a method that returns an optional and want to keep the result wrapped, use this pattern:

Optional<DseFunctionMetadata> f =
        .flatMap(ks -> ks.getFunction("f"))

For future extensibility, there is a DseXxxMetadata subtype for every OSS type. But currently (DSE 6.7), the only types that really add extra information are:

All other types (keyspaces, tables, etc.) are identical to their OSS counterparts.


If you need to follow schema changes, you don’t need to poll the metadata manually; instead, you can register a listener to get notified when changes occur:

SchemaChangeListener listener =
    new SchemaChangeListenerBase() {
      public void onTableCreated(TableMetadata table) {
        System.out.println("New table: " + table.getName().asCql(true));
CqlSession session = CqlSession.builder()

session.execute("CREATE TABLE (k int PRIMARY KEY)");

See SchemaChangeListener for the list of available methods. SchemaChangeListenerBase is a convenience implementation with empty methods, for when you only need to override a few of them.



You can disable schema metadata globally from the configuration:

datastax-java-driver.advanced.metadata.schema.enabled = false

If it is disabled at startup, Metadata#getKeyspaces will stay empty. If you disable it at runtime, it will keep the value of the last refresh.

You can achieve the same thing programmatically with Session#setSchemaMetadataEnabled: if you call it with true or false, it overrides the configuration; if you pass null, it reverts to the value defined in the configuration. One case where that could come in handy is if you are sending a large number of DDL statements from your code:

// Disable temporarily, we'll do a single refresh once we're done 

for (int i = 0; i < 100; i++) {
  session.execute(String.format("CREATE TABLE (k int PRIMARY KEY)", i));


Whenever schema metadata was disabled and becomes enabled again (either through the configuration or the API), a refresh is triggered immediately.


You can also limit the metadata to a subset of keyspaces:

datastax-java-driver.advanced.metadata.schema.refreshed-keyspaces = [ "users", "products" ]

If the property is absent or the list is empty, it is interpreted as “all keyspaces”.

Note that, if you change the list at runtime, onKeyspaceAdded/onKeyspaceDropped will be invoked on your schema listeners for the newly included/excluded keyspaces.

Schema agreement

Due to the distributed nature of Cassandra, schema changes made on one node might not be immediately visible to others. If left unaddressed, this could create race conditions when successive queries get routed to different coordinators:

Text Diagram

To avoid this issue, the driver waits until all nodes agree on a common schema version:

Text Diagram

Schema agreement is checked:

  • before a schema refresh;
  • before completing a successful schema-altering query (like in our example above).

It is done by querying system tables to find out the schema version of all nodes that are currently UP. If all the versions match, the check succeeds, otherwise it is retried periodically, until a given timeout. This process is tunable in the driver’s configuration:

datastax-java-driver.advanced.control-connection.schema-agreement {
  interval = 200 milliseconds
  timeout = 10 seconds
  warn-on-failure = true

After executing a statement, you can check whether schema agreement was successful or timed out with ExecutionInfo#isSchemaInAgreement:

ResultSet rs = session.execute("CREATE TABLE...");
if (rs.getExecutionInfo().isSchemaInAgreement()) {

You can also perform an on-demand check at any time with Session#checkSchemaAgreementAsync (or its synchronous counterpart):

if (session.checkSchemaAgreement()) {

A schema agreement failure is not fatal, but it might produce unexpected results (as explained at the beginning of this section).

Schema agreement in mixed-version clusters

If you’re operating a cluster with different major/minor server releases (for example, Cassandra 2.1 and 2.2), schema agreement will never succeed. This is because the way the schema version is computed changes across releases, so the nodes will report different versions even though they actually agree (see JAVA-750 for the technical details).

This issue would be hard to fix in a reliable way, and shouldn’t be that much of a problem in practice anyway: if you’re in the middle of a rolling upgrade, you’re probably not applying schema changes at the same time.

Relation to token metadata

Some of the data in the token map relies on keyspace metadata (any method that takes a CqlIdentifier argument). If schema metadata is disabled or filtered, token metadata will also be unavailable for the excluded keyspaces.

Performing schema updates from the client

If you issue schema-altering requests from the driver (e.g. session.execute("CREATE TABLE ..")), take a look at the Performance page for a few tips.