session.getMetadata(): node states, schema and token map.
- immutable, provides a consistent view at a given point in time (e.g. token map always matches schema).
- pitfall: holding onto a stale instance; must call
session.getMetadata()again to observe changes.
The driver exposes metadata about the Cassandra cluster via the Session#getMetadata method. It returns a Metadata object, which contains three types of information:
Metadata is mostly immutable (except for the fields of the Node class, see the "node metadata"
link above for details). Each call to getMetadata() will return a new copy if something has
changed since the last call. Do not cache the result across usages:
Metadata metadata = session.getMetadata();
session.execute("CREATE TABLE test.foo (k int PRIMARY KEY)");
// WRONG: the metadata was retrieved before the CREATE TABLE call, it does not reflect the new table
TableMetadata fooMetadata =
metadata
.getKeyspace(CqlIdentifier.fromCql("test"))
.getTable(CqlIdentifier.fromCql("foo"));
assert fooMetadata == null;On the other hand, the advantage of immutability is that a Metadata instance provides a
consistent view of the cluster at a given point in time. In other words, the token map is
guaranteed to be in sync with the node and schema metadata:
Metadata metadata = session.getMetadata();
// Pick up any node and keyspace:
Node node = metadata.getNodes().values().iterator().next();
KeyspaceMetadata keyspace = metadata.getKeyspaces().values().iterator().next();
TokenMap tokenMap = metadata.getTokenMap().get();
// The token map is guaranteed to have the corresponding data:
Set<TokenRange> tokenRanges = tokenMap.getTokenRanges(keyspace.getName(), node);This is a big improvement over previous versions of the driver, where it was possible to observe a new keyspace in the schema metadata before the token metadata was updated.
Schema and node state events are debounced. This allows you to control how often the metadata gets refreshed. See the Performance page for more details.