Dataset, View, and Repository URIs

Datasets, views, and repositories are identified by URI. As a general rule, you can use a URI any time you would specify a dataset or view. If you attempt to perform an action on a view that is not allowed, the action fails.

Dataset URIs

You construct a dataset URI using one of the following patterns, depending on your chosen dataset scheme.

Dataset patterns always begin with the dataset: prefix. Any of these patterns can be modified to create a View URI.

Hive

Hive manages your datatables for you. You only have to provide the dataset name. You also have the option of providing a namespace.

dataset:hive:<namespace>/<dataset>

If you want to use external Hive datatables, you must also provide a path to the dataset. If you don’t explicitly set a namespace, Kite uses the default namespace. For a Hive dataset, a Kite namespace maps one-to-one to a Hive database.

dataset:hive:/<path>/<namespace>/<dataset-name>

In earlier versions of Kite, dataset:hive:a/b meant directory ./a/b Now, it has changed to namespace=a dataset=b.

To create an external table, add location=/path/to/data/dir to the dataset URI.

dataset:hive:namespace/dataset?location=/path/to/data/dir

HDFS

The URI for a dataset in HDFS uses the following pattern. You provide a path to the dataset. dataset:hdfs:/<path>/<namespace>/<dataset-name>

While it is not a typical use case, there might be times where it is useful to specify the HDFS host and port. You can insert the host and port before the path in the URI. You can use these URIs to select between HDFS instances.

The host and port are required if your Hadoop configuration files aren’t on your classpath. This can happen if your application is running outside the cluster and you haven’t taken the step of deploying client configuration files to the server running the application. The host is the hostname for the namenode; the port is the port for the namenode (typically 8020). If you have an HA configuration, you always need client configuration files, and the host should be the NameService ID.

dataset:hdfs://<host>[:port]/<path>/<namespace>/<dataset-name>

Local File System Datasets

The local file system dataset URI follows a pattern similar to the HDFS URI, with the file: scheme.

dataset:file:/<path>/<namespace>/<dataset-name>

HBase

hbase:<zookeeper>/<dataset-name>

The zookeeper argument is a comma separated list of hosts. For example

hbase:host1,host2:9999,host3/myDataset

View URIs

A view URI is constructed by changing the prefix of a dataset URI from dataset: to view:. You then add query arguments as name/value pairs, similar to query arguments in an HTTP URL. Query arguments place constraints on the information returned in the view.

view:<scheme-specific-URI>?<field>=<constraint>

For example, you can restrict values returned from a table of users to users whose favorite color is pink.

view:hdfs:/default/cloudera/users?favoriteColor=pink

You can insert records in a view, and the changes are reflected in the source dataset.

You can also set constraints based on dataset partitions. For example, if a dataset of movie ratings were partitioned by date, you might create the view URI this way, constraining the ratings returned to March, 2014.

view:hdfs:/default/cloudera/ratings?year=2014&month=3

If the URI begins with dataset:, any constraints are ignored.

There are three formats used to set constraint values. The values can be numbers or strings, but the values you specify must match the schema definition for the field.

See Interval Notation for more examples of defining ranges of values.

Repository URIs

Repository URI patterns always begin with the repo: prefix. In the Kite Dataset API, you use a repository URI with the Datasets.list method to retrieve a list of valid datasets.

For example, to list the dataset URIs for the Hive repository, use Datasets.list("repo:hive");.