Delta Sharing: An Open Protocol for Secure Data Sharing

Delta Sharing is an open protocol for secure real-time exchange of large datasets, which enables organizations to share data in real time regardless of which computing platforms they use. It is a simple REST protocol that securely shares access to part of a cloud dataset and leverages modern cloud storage systems, such as S3, ADLS, or GCS, to reliably transfer data.

With Delta Sharing, a user accessing shared data can directly connect to it through pandas, Tableau, Apache Spark, Rust, or other systems that support the open protocol, without having to deploy a specific compute platform first. Data providers can share a dataset once to reach a broad range of consumers, while consumers can begin using the data in minutes.

This repo includes the following components:

• Delta Sharing protocol specification.
• Python Connector: A Python library that implements the Delta Sharing Protocol to read shared tables as pandas DataFrame or Apache Spark DataFrames.
• Apache Spark Connector: An Apache Spark connector that implements the Delta Sharing Protocol to read shared tables from a Delta Sharing Server. The tables can then be accessed in SQL, Python, Java, Scala, or R.
• Delta Sharing Server: A reference implementation server for the Delta Sharing Protocol for development purposes. Users can deploy this server to share existing tables in Delta Lake and Apache Parquet format on modern cloud storage systems.

Python Connector

The Delta Sharing Python Connector is a Python library that implements the Delta Sharing Protocol to read tables from a Delta Sharing Server. You can load shared tables as a pandas DataFrame, or as an Apache Spark DataFrame if running in PySpark with the Apache Spark Connector installed.

System Requirements

Python 3.8+ for delta-sharing version 1.1+, Python 3.6+ for older versions If running Linux, glibc version >= 2.31 (for automatic delta-kernel-rust-sharing-wrapper package installation, please see next section for more details)

Installation

pip3 install delta-sharing

If you are using Databricks Runtime, you can follow Databricks Libraries doc to install the library on your clusters.

If this doesn’t work because of an issue downloading delta-kernel-rust-sharing-wrapper try the following:

• Check python3 version >= 3.8
• Upgrade your pip3 to the latest version
• Check the linux glibc version >= 2.31

If you cannot upgrade glibc or your OS is not supported by the delta-kernel-rust-sharing-wrapper pypi install you will need to install the package manually. See https://pypi.org/project/delta-kernel-rust-sharing-wrapper/0.1.0/#files for supported OS.

To install the delta-kernel-rust-sharing-wrapper package manually:

# you need to use the older version of the delta-sharing package which did not bake delta-kernel-rust-sharing-wrapper into the installation
pip3 install delta-sharing==1.1.0 

cd [delta-sharing-root]/python/delta-kernel-rust-sharing-wrapper
python3 -m venv .venv
source .venv/bin/activate
pip3 install maturin
maturin develop

Accessing Shared Data

The connector accesses shared tables based on profile files, which are JSON files containing a user's credentials to access a Delta Sharing Server. We have several ways to get started:

• Download the profile file to access an open, example Delta Sharing Server that we're hosting here. You can try the connectors with this sample data.
• Start your own Delta Sharing Server and create your own profile file following profile file format to connect to this server.
• Download a profile file from your data provider.

Quick Start

After you save the profile file, you can use it in the connector to access shared tables.

import delta_sharing

# Point to the profile file. It can be a file on the local file system or a file on a remote storage.
profile_file = "<profile-file-path>"

# Create a SharingClient.
client = delta_sharing.SharingClient(profile_file)

# List all shared tables.
client.list_all_tables()

# Create a url to access a shared table.
# A table path is the profile file path following with `#` and the fully qualified name of a table 
# (`<share-name>.<schema-name>.<table-name>`).
table_url = profile_file + "#<share-name>.<schema-name>.<table-name>"

# Fetch 10 rows from a table and convert it to a Pandas DataFrame. This can be used to read sample data 
# from a table that cannot fit in the memory.
delta_sharing.load_as_pandas(table_url, limit=10)

# Load a table as a Pandas DataFrame. This can be used to process tables that can fit in the memory.
delta_sharing.load_as_pandas(table_url)

# Load a table as a Pandas DataFrame explicitly using Delta Format
delta_sharing.load_as_pandas(table_url, use_delta_format = True)

# If the code is running with PySpark, you can use `load_as_spark` to load the table as a Spark DataFrame.
delta_sharing.load_as_spark(table_url)

If the table supports history sharing(tableConfig.cdfEnabled=true in the OSS Delta Sharing Server), the connector can query table changes.

# Load table changes from version 0 to version 5, as a Pandas DataFrame.
delta_sharing.load_table_changes_as_pandas(table_url, starting_version=0, ending_version=5)

# If the code is running with PySpark, you can load table changes as Spark DataFrame.
delta_sharing.load_table_changes_as_spark(table_url, starting_version=0, ending_version=5)

You can try this by running our examples with the open, example Delta Sharing Server.

Details on Profile Paths

• The profile file path for SharingClient and load_as_pandas can be any URL supported by FSSPEC (such as s3a://my_bucket/my/profile/file). If you are using Databricks File System, you can also preface the path with /dbfs/ to access the profile file as if it were a local file.
• The profile file path for load_as_spark can be any URL supported by Hadoop FileSystem (such as s3a://my_bucket/my/profile/file).
• A table path is the profile file path following with # and the fully qualified name of a table (<share-name>.<schema-name>.<table-name>).

Apache Spark Connector

The Apache Spark Connector implements the Delta Sharing Protocol to read shared tables from a Delta Sharing Server. It can be used in SQL, Python, Java, Scala and R.

System Requirements

• Java 8+
• Scala 2.12.x
• Apache Spark 3+ or Databricks Runtime 9+

Accessing Shared Data

The connector loads user credentials from profile files. Please see Accessing Shared Data to download a profile file for our example server or for your own data sharing server.

Configuring Apache Spark

You can set up Apache Spark to load the Delta Sharing connector in the following two ways:

• Run interactively: Start the Spark shell (Scala or Python) with the Delta Sharing connector and run the code snippets interactively in the shell.
• Run as a project: Set up a Maven or SBT project (Scala or Java) with the Delta Sharing connector, copy the code snippets into a source file, and run the project.

If you are using Databricks Runtime, you can skip this section and follow Databricks Libraries doc to install the connector on your clusters.

Set up an interactive shell

To use Delta Sharing connector interactively within the Spark’s Scala/Python shell, you can launch the shells as follows.

PySpark shell

pyspark --packages io.delta:delta-sharing-spark_2.12:3.1.0

Scala Shell

bin/spark-shell --packages io.delta:delta-sharing-spark_2.12:3.1.0

Set up a standalone project

If you want to build a Java/Scala project using Delta Sharing connector from Maven Central Repository, you can use the following Maven coordinates.

Maven

You include Delta Sharing connector in your Maven project by adding it as a dependency in your POM file. Delta Sharing connector is compiled with Scala 2.12.

<dependency>
  <groupId>io.delta</groupId>
  <artifactId>delta-sharing-spark_2.12</artifactId>
  <version>3.1.0</version>
</dependency>

SBT

You include Delta Sharing connector in your SBT project by adding the following line to your build.sbt file:

libraryDependencies += "io.delta" %% "delta-sharing-spark" % "3.1.0"

Quick Start

After you save the profile file and launch Spark with the connector library, you can access shared tables using any language.

SQL

-- A table path is the profile file path following with `#` and the fully qualified name 
-- of a table (`<share-name>.<schema-name>.<table-name>`).
CREATE TABLE mytable USING deltaSharing LOCATION '<profile-file-path>#<share-name>.<schema-name>.<table-name>';
SELECT * FROM mytable;

Python

# A table path is the profile file path following with `#` and the fully qualified name 
# of a table (`<share-name>.<schema-name>.<table-name>`).
table_path = "<profile-file-path>#<share-name>.<schema-name>.<table-name>"
df = spark.read.format("deltaSharing").load(table_path)

Scala

// A table path is the profile file path following with `#` and the fully qualified name 
// of a table (`<share-name>.<schema-name>.<table-name>`).
val tablePath = "<profile-file-path>#<share-name>.<schema-name>.<table-name>"
val df = spark.read.format("deltaSharing").load(tablePath)

Java

// A table path is the profile file path following with `#` and the fully qualified name 
// of a table (`<share-name>.<schema-name>.<table-name>`).
String tablePath = "<profile-file-path>#<share-name>.<schema-name>.<table-name>";
Dataset<Row> df = spark.read.format("deltaSharing").load(tablePath);

R

# A table path is the profile file path following with `#` and the fully qualified name 
# of a table (`<share-name>.<schema-name>.<table-name>`).
table_path <- "<profile-file-path>#<share-name>.<schema-name>.<table-name>"
df <- read.df(table_path, "deltaSharing")

You can try this by running our examples with the open, example Delta Sharing Server.

CDF

Starting from release 0.5.0, querying Change Data Feed is supported with Delta Sharing. Once the provider turns on CDF on the original delta table and shares it through Delta Sharing, the recipient can query CDF of a Delta Sharing table similar to CDF of a delta table.

val tablePath = "<profile-file-path>#<share-name>.<schema-name>.<table-name>"
val df = spark.read.format("deltaSharing")
  .option("readChangeFeed", "true")
  .option("startingVersion", "3")
  .load(tablePath)

Streaming

Starting from release 0.6.0, Delta Sharing table can be used as a data source for Spark Structured Streaming. Once the provider shares a table with history, the recipient can perform a streaming query on the table.

Note: Trigger.AvailableNow is not supported in delta sharing streaming because it's supported since spark 3.3.0, while delta sharing is still using spark 3.1.1.

val tablePath = "<profile-file-path>#<share-name>.<schema-name>.<table-name>"
val df = spark.readStream.format("deltaSharing")
  .option("startingVersion", "1")
  .option("skipChangeCommits", "true")
  .load(tablePath)

Table paths

• A profile file path can be any URL supported by Hadoop FileSystem (such as s3a://my_bucket/my/profile/file).
• A table path is the profile file path following with # and the fully qualified name of a table (<share-name>.<schema-name>.<table-name>).

The Community

Delta Sharing Reference Server

The Delta Sharing Reference Server is a reference implementation server for the Delta Sharing Protocol. This can be used to set up a small service to test your own connector that implements the Delta Sharing Protocol. Please note that this is not a completed implementation of secure web server. We highly recommend you to put this behind a secure proxy if you would like to expose it to public.

Some vendors offer managed services for Delta Sharing too (for example, Databricks). Please refer to your vendor's website for how to set up sharing there. Vendors that are interested in being listed as a service provider should open an issue on GitHub to be added to this README and our project's website.

Here are the steps to setup the reference server to share your own data.

Get the pre-built package

Download the pre-built package delta-sharing-server-x.y.z.zip from GitHub Releases.

Server configuration and adding Shared Data

• Unpack the pre-built package and copy the server config template file conf/delta-sharing-server.yaml.template to create your own server yaml file, such as conf/delta-sharing-server.yaml.
• Make changes to your yaml file. You may also need to update some server configs for special requirements.
• To add Shared Data, add reference to Delta Lake tables you would like to share from this server in this config file.

Config the server to access tables on cloud storage

We support sharing Delta Lake tables on S3, Azure Blob Storage and Azure Data Lake Storage Gen2.

S3

The server is using hadoop-aws to access S3. Table paths in the server config file should use s3a:// paths rather than s3:// paths. There are multiple ways to config S3 authentication.

EC2 IAM Metadata Authentication (Recommended)

Applications running in EC2 may associate an IAM role with the VM and query the EC2 Instance Metadata Service for credentials to access S3.

Authenticating via the AWS Environment Variables

We support configuration via the standard AWS environment variables. The core environment variables are for the access key and associated secret:

export AWS_ACCESS_KEY_ID=my.aws.key
export AWS_SECRET_ACCESS_KEY=my.secret.key

Other S3 authentication methods

You can find other approaches in hadoop-aws doc.

Azure Blob Storage

The server is using hadoop-azure to read Azure Blob Storage. Using Azure Blob Storage requires configuration of credentials. You can create a Hadoop configuration file named core-site.xml and add it to the server's conf directory. Then add the following content to the xml file:

<?xml version="1.0"?>
<?xml-stylesheet type="text/xsl" href="configuration.xsl"?>
<configuration>
  <property>
    <name>fs.azure.account.key.YOUR-ACCOUNT-NAME.blob.core.windows.net</name>
    <value>YOUR-ACCOUNT-KEY</value>
  </property>
</configuration>

YOUR-ACCOUNT-NAME is your Azure storage account and YOUR-ACCOUNT-KEY is your account key.

Azure Data Lake Storage Gen2

The server is using hadoop-azure to read Azure Data Lake Storage Gen2. We support the Shared Key authentication. You can create a Hadoop configuration file named core-site.xml and add it to the server's conf directory. Then add the following content to the xml file:

<?xml version="1.0"?>
<?xml-stylesheet type="text/xsl" href="configuration.xsl"?>
<configuration>
  <property>
    <name>fs.azure.account.auth.type.YOUR-ACCOUNT-NAME.dfs.core.windows.net</name>
    <value>SharedKey</value>
    <description>
    </description>
  </property>
  <property>
    <name>fs.azure.account.key.YOUR-ACCOUNT-NAME.dfs.core.windows.net</name>
    <value>YOUR-ACCOUNT-KEY</value>
    <description>
    The secret password. Never share these.
    </description>
  </property>
</configuration>

YOUR-ACCOUNT-NAME is your Azure storage account and YOUR-ACCOUNT-KEY is your account key.

Google Cloud Storage

We support using Service Account to read Google Cloud Storage. You can find more details in GCP Authentication Doc.

To set up the Service Account credentials, you can specify the environment GOOGLE_APPLICATION_CREDENTIALS before starting the Delta Sharing Server.

export GOOGLE_APPLICATION_CREDENTIALS="KEY_PATH"

Replace KEY_PATH with path of the JSON file that contains your service account key.

Cloudflare R2

We use an R2 implementation of the S3 API and hadoop-aws to read Cloudflare R2. Table paths in the server config file should use the s3a:// scheme. You must generate an API token for usage with existing S3-compatible SDKs. These credentials can be specified in substitute of the S3 credentials in a Hadoop configuration file named core-site.xml within the server's conf directory. For R2 to work, you also need to directly specify the S3 endpoint and reduce fs.s3a.paging.maximum from Hadoop's default of 5000 to 1000 since R2 only supports MaxKeys <= 1000.

<?xml version="1.0"?>
<?xml-stylesheet type="text/xsl" href="configuration.xsl"?>
<configuration>
  <property>
    <name>fs.s3a.access.key</name>
    <value>YOUR-ACCESS-KEY</value>
  </property>
  <property>
    <name>fs.s3a.secret.key</name>
    <value>YOUR-SECRET-KEY</value>
  </property>
  <property>
    <name>fs.s3a.endpoint</name>
    <value>https://YOUR-ACCOUNT-ID.r2.cloudflarestorage.com</value>
  </property>
  <property>
    <name>fs.s3a.paging.maximum</name>
    <value>1000</value>
  </property>
</configuration>

Replace YOUR-ACCESS-KEY with your generated API token's R2 access key ID, YOUR-SECRET-KEY with your generated API token's secret access key, and YOUR-ACCOUNT-ID with your Cloudflare account ID.

Note: S3 and R2 credentials cannot be configured simultaneously.

Authorization

The server supports a basic authorization with pre-configed bearer token. You can add the following config to your server yaml file:

authorization:
  bearerToken: <token>

Then any request should send with the above token, otherwise, the server will refuse the request.

If you don't config the bearer token in the server yaml file, all requests will be accepted without authorization.

To be more secure, you recommend you to put the server behind a secure proxy such as NGINX to set up JWT Authentication.

Start the server

Run the following shell command:

bin/delta-sharing-server -- --config <the-server-config-yaml-file> 

<the-server-config-yaml-file> should be the path of the yaml file you created in the previous step. You can find options to config JVM in sbt-native-packager.

Use the pre-built Docker image

You can use the pre-built docker image from https://hub.docker.com/r/deltaio/delta-sharing-server by running the following command

docker run -p <host-port>:<container-port> \
  --mount type=bind,source=<the-server-config-yaml-file>,target=/config/delta-sharing-server-config.yaml \
  deltaio/delta-sharing-server:0.7.8 -- --config /config/delta-sharing-server-config.yaml

Note that <container-port> should be the same as the port defined inside the config file.

API Compatibility

The REST APIs provided by Delta Sharing Server are stable public APIs. They are defined by Delta Sharing Protocol and we will follow the entire protocol strictly.

The interfaces inside Delta Sharing Server are not public APIs. They are considered internal, and they are subject to change across minor/patch releases.

Delta Sharing Protocol

The Delta Sharing Protocol specification details the protocol.

Building this Project

Python Connector

To execute tests, run

python/dev/pytest

To install in develop mode, run

cd python/
pip install -e .

To install locally, run

cd python/
pip install .

To generate a wheel file, run

cd python/
python setup.py sdist bdist_wheel

It will generate python/dist/delta_sharing-x.y.z-py3-none-any.whl.

Apache Spark Connector and Delta Sharing Server

Apache Spark Connector and Delta Sharing Server are compiled using SBT.

To compile, run

build/sbt compile

To execute tests, run

build/sbt test

To generate the Apache Spark Connector, run

build/sbt spark/package

It will generate spark/target/scala-2.12/delta-sharing-spark_2.12-x.y.z.jar.

To generate the pre-built Delta Sharing Server package, run

build/sbt server/universal:packageBin

It will generate server/target/universal/delta-sharing-server-x.y.z.zip.

To build the Docker image for Delta Sharing Server, run

build/sbt server/docker:publishLocal

This will build a Docker image tagged delta-sharing-server:x.y.z, which you can run with:

docker run -p <host-port>:<container-port> \
  --mount type=bind,source=<the-server-config-yaml-file>,target=/config/delta-sharing-server-config.yaml \
  delta-sharing-server:x.y.z -- --config /config/delta-sharing-server-config.yaml

Note that <container-port> should be the same as the port defined inside the config file.

Refer to SBT docs for more commands.

Reporting Issues

We use GitHub Issues to track community reported issues. You can also contact the community for getting answers.

Contributing

We welcome contributions to Delta Sharing. See our CONTRIBUTING.md for more details.

We also adhere to the Delta Lake Code of Conduct.

License

Apache License 2.0.

Community

We use the same community resources as the Delta Lake project:

• Public Slack Channel

  • Register here
  • Login here

• Public Mailing list