We are building connectors to bring Delta Lake to popular big-data engines outside Apache Spark (e.g., Apache Hive, Presto).
This is the repository for Delta Lake Connectors. It includes a library for querying Delta Lake metadata and connectors to popular big-data engines (e.g., Apache Hive, Presto). Please refer to the main Delta Lake repository if you want to learn more about the Delta Lake project.
The project is compiled using SBT. It has the following subprojects.
Delta Standalone Reader is a JVM library to read Delta Lake tables. Unlike https://github.com/delta-io/delta, this project doesn't use Spark to read tables and it has only a few transitive dependencies. It can be used to any application that cannot use a Spark cluster.
- To compile the project, run
build/sbt standalone/compile - To test the project, run
build/sbt standalone/test - To generate the JAR, run
build/sbt standalone/package
You can add the Delta Standalone Reader library as a dependency using your favorite build tool.
Scala 2.12:
<dependency>
<groupId>io.delta</groupId>
<artifactId>delta-standalone_2.12</artifactId>
<version>0.2.0</version>
</dependency>Scala 2.11:
<dependency>
<groupId>io.delta</groupId>
<artifactId>delta-standalone_2.11</artifactId>
<version>0.2.0</version>
</dependency>libraryDependencies += "io.delta" %% "delta-standalone" % "0.2.0"
This project is a library to make Hive read Delta Lake tables. The project provides a uber JAR delta-hive-assembly_<scala_version>-0.2.0.jar to use in Hive. You can use either Scala 2.11 or 2.12. The released JARs are available in the releases page. Please download the uber JAR for the corresponding Scala version you would like to use.
You can also use the following instructions to build it as well.
Please skip this section if you have downloaded the connector JARs.
- To compile the project, run
build/sbt hive/compile - To test the project, run
build/sbt hive/test - To generate the uber JAR that contains all libraries needed for Hive, run
build/sbt hive/assembly
The above commands will generate the following JAR:
hive/target/scala-2.12/delta-hive-assembly_2.12-0.2.0.jar
This uber JAR includes the Hive connector and all its dependencies. They need to be put in Hive’s classpath.
Note: if you would like to build using Scala 2.11, you can run the SBT command build/sbt "++ 2.11.12 hive/assembly" to generate the following JAR:
hive/target/scala-2.11/delta-hive-assembly_2.11-0.2.0.jar
This section describes how to set up Hive to load the Delta Hive connector.
Before starting your Hive CLI or running your Hive script, add the following special Hive config to the hive-site.xml file (Its location is /etc/hive/conf/hive-site.xml in a EMR cluster).
<property>
<name>hive.input.format</name>
<value>io.delta.hive.HiveInputFormat</value>
</property>
<property>
<name>hive.tez.input.format</name>
<value>io.delta.hive.HiveInputFormat</value>
</property>Alternatively, you can also run the following SQL commands in Hive CLI before reading Delta tables to set io.delta.hive.HiveInputFormat:
SET hive.input.format=io.delta.hive.HiveInputFormat;
SET hive.tez.input.format=io.delta.hive.HiveInputFormat;
The second step is to upload the above uber JAR to the machine that runs Hive. Finally, add the path of the uber JAR to Hive’s environment variable, HIVE_AUX_JARS_PATH. You can find this environment variable in the hive-env.sh file, whose location is /etc/hive/conf/hive-env.sh on an EMR cluster. This setting will tell Hive where to find the connector JAR.
After finishing setup, you should be able to create a Delta table in Hive.
Right now the connector supports only EXTERNAL Hive tables. The Delta table must be created using Spark before an external Hive table can reference it.
Here is an example of a CREATE TABLE command that defines an external Hive table pointing to a Delta table on s3://foo-bucket/bar-dir.
CREATE EXTERNAL TABLE deltaTable(col1 INT, col2 STRING)
STORED BY 'io.delta.hive.DeltaStorageHandler'
LOCATION '/delta/table/path'io.delta.hive.DeltaStorageHandler is the class that implements Hive data source APIs. It will know how to load a Delta table and extract its metadata. The table schema in the CREATE TABLE statement must be consistent with the underlying Delta metadata. Otherwise, the connector will throw an error to tell you about the inconsistency.
/delta/table/path in LOCATION is a normal path. If there is no scheme in the path, it will use the default file system specified in your Hadoop configuration.
You can add an explicit scheme to specify which file system you would like to use, such as file:///delta/table/path, s3://your-s3-bucket/delta/table/path.
Hive 2.x.
No. The connector must be used with Apache Hive. It doesn't work in other systems, such as Apache Spark or Presto.
- This connector does not provide the support for defining Hive Metastore tables in Apache Spark. It will be added in Delta Lake core repository. It is tracked by the issue delta-io/delta#85.
- This Hive connector does not native connectivity for Presto. But you can generate a manifest file to load a Delta table in Presto. See https://docs.delta.io/latest/presto-integration.html.
- Other system support can be found in https://docs.delta.io/latest/integrations.html.
No. The table created by this connector in Hive cannot be read in any other systems right now. We recommend to create different tables in different systems but point to the same path. Although you need to use different table names to query the same Delta table, the underlying data will be shared by all of systems.
No. The connector doesn't support writing to a Delta table.
No. The partition columns are read from the underlying Delta metadata. The connector will know the partition columns and use this information to do the partition pruning automatically.
Why do I need to specify the table schema? Shouldn’t it exist in the underlying Delta table metadata?
Unfortunately, the table schema is a core concept of Hive and Hive needs it before calling the connector.
If the schema in the underlying Delta metadata is not consistent with the schema specified by CREATE TABLE statement, the connector will report an error when loading the table and ask you to fix the schema. You must drop the table and recreate it using the new schema. Hive 3.x exposes a new API to allow a data source to hook ALTER TABLE. You will be able to use ALTER TABLE to update a table schema when the connector supports Hive 3.x.
The connector supports MapReduce and Tez. It doesn't support Spark execution engine in Hive.
We use GitHub Issues to track community reported issues. You can also contact the community for getting answers.
We welcome contributions to Delta Lake Connectors repository. We use GitHub Pull Requests for accepting changes.
There are two mediums of communication within the Delta Lake community.
-
Public Slack Channel
-
Public Mailing list