[SPARK-39375] SPIP: Spark Connect - A client and server interface for Apache Spark

• Type: Epic
• Status: Reopened
• Priority: Critical
• Resolution: Unresolved
• Affects Version/s: 3.4.0
• Component/s: Connect
• Labels: SPIP
• Epic Name: Spark Connect
• Assignee: Martin Grund
• Reporter: Martin Grund
• Shepherd: Hyukjin Kwon
• Created: 03/Jun/22 17:30
• Updated: 24/Oct/23 00:21

Description

Please find the full document for discussion here: Spark Connect SPIP Below, we have just referenced the introduction.

What are you trying to do?

While Spark is used extensively, it was designed nearly a decade ago, which, in the age of serverless computing and ubiquitous programming language use, poses a number of limitations. Most of the limitations stem from the tightly coupled Spark driver architecture and fact that clusters are typically shared across users: (1) Lack of built-in remote connectivity: the Spark driver runs both the client application and scheduler, which results in a heavyweight architecture that requires proximity to the cluster. There is no built-in capability to  remotely connect to a Spark cluster in languages other than SQL and users therefore rely on external solutions such as the inactive project Apache Livy. (2) Lack of rich developer experience: The current architecture and APIs do not cater for interactive data exploration (as done with Notebooks), or allow for building out rich developer experience common in modern code editors. (3) Stability: with the current shared driver architecture, users causing critical exceptions (e.g. OOM) bring the whole cluster down for all users. (4) Upgradability: the current entangling of platform and client APIs (e.g. first and third-party dependencies in the classpath) does not allow for seamless upgrades between Spark versions (and with that, hinders new feature adoption).

 

We propose to overcome these challenges by building on the DataFrame API and the underlying unresolved logical plans. The DataFrame API is widely used and makes it very easy to iteratively express complex logic. We will introduce Spark Connect, a remote option of the DataFrame API that separates the client from the Spark server. With Spark Connect, Spark will become decoupled, allowing for built-in remote connectivity: The decoupled client SDK can be used to run interactive data exploration and connect to the server for DataFrame operations. 

 

Spark Connect will benefit Spark developers in different ways: The decoupled architecture will result in improved stability, as clients are separated from the driver. From the Spark Connect client perspective, Spark will be (almost) versionless, and thus enable seamless upgradability, as server APIs can evolve without affecting the client API. The decoupled client-server architecture can be leveraged to build close integrations with local developer tooling. Finally, separating the client process from the Spark server process will improve Spark’s overall security posture by avoiding the tight coupling of the client inside the Spark runtime environment.

 

Spark Connect will strengthen Spark’s position as the modern unified engine for large-scale data analytics and expand applicability to use cases and developers we could not reach with the current setup: Spark will become ubiquitously usable as the DataFrame API can be used with (almost) any programming language.
 

	SPARK-41282	Feature parity: Column API in Spark Connect		REOPENED	Ruifeng Zheng	Actions
	SPARK-41283	Feature parity: Functions API in Spark Connect		RESOLVED	Ruifeng Zheng	Actions
	SPARK-41279	Feature parity: DataFrame API in Spark Connect		OPEN	Ruifeng Zheng	Actions
	SPARK-41281	Feature parity: SparkSession API in Spark Connect		OPEN	Ruifeng Zheng	Actions
	SPARK-41284	Feature parity: I/O in Spark Connect		REOPENED	Rui Wang	Actions
	SPARK-41289	Feature parity: Catalog API		RESOLVED	Hyukjin Kwon	Actions
	SPARK-41286	Build, package and infrastructure for Spark Connect		RESOLVED	Hyukjin Kwon	Actions
	SPARK-40451	Type annotations for Spark Connect Python client		RESOLVED	Hyukjin Kwon	Actions
	SPARK-40452	Developer documentation		OPEN	Unassigned	Actions
	SPARK-41285	Test basework and improvement of test coverage in Spark Connect		OPEN	Hyukjin Kwon	Actions
	SPARK-41288	Server-specific improvement, error handling and API		OPEN	Martin Grund	Actions
	SPARK-41305	Connect Proto Completeness		REOPENED	Rui Wang	Actions
	SPARK-41531	Debugging and Stability		OPEN	Unassigned	Actions
	SPARK-41625	Feature parity: Streaming support		OPEN	Unassigned	Actions
	SPARK-41627	Spark Connect Server Development		OPEN	Unassigned	Actions
	SPARK-41642	Deduplicate docstrings in Python Spark Connect		RESOLVED	Hyukjin Kwon	Actions
	SPARK-41651	Test parity: pyspark.sql.tests.test_dataframe		RESOLVED	Sandeep Singh	Actions
	SPARK-41652	Test parity: pyspark.sql.tests.test_functions		RESOLVED	Sandeep Singh	Actions
	SPARK-41661	Support for User-defined Functions in Python		RESOLVED	Xinrong Meng	Actions
	SPARK-41653	Test parity: enable doctests in Spark Connect		RESOLVED	Sandeep Singh	Actions
	SPARK-41932	Bootstrapping Spark Connect		OPEN	Hyukjin Kwon	Actions
	SPARK-41997	Test parity: pyspark.sql.tests.test_readwriter		RESOLVED	Unassigned	Actions
	SPARK-42006	Test parity: pyspark.sql.tests.test_group, test_serde, test_datasources and test_column		RESOLVED	Hyukjin Kwon	Actions
	SPARK-42018	Test parity: pyspark.sql.tests.test_types		RESOLVED	Unassigned	Actions
	SPARK-42156	Support client-side retries in Spark Connect Python client		RESOLVED	Martin Grund	Actions
	SPARK-42264	Test Parity: pyspark.sql.tests.test_udf and pyspark.sql.tests.pandas.test_pandas_udf		RESOLVED	Xinrong Meng	Actions
	SPARK-42374	User-facing documentation		OPEN	Haejoon Lee	Actions
	SPARK-42393	Support for Pandas/Arrow Functions API		RESOLVED	Xinrong Meng	Actions
	SPARK-42471	Distributed ML <> spark connect		OPEN	Unassigned	Actions
	SPARK-42497	Support of pandas API on Spark for Spark Connect		IN PROGRESS	Unassigned	Actions
	SPARK-42499	Support for Runtime SQL configuration		RESOLVED	Takuya Ueshin	Actions
	SPARK-43289	PySpark UDF supports python package dependencies		OPEN	Weichen Xu	Actions
	SPARK-43612	Python: Artifact transfer from Scala/JVM client to Server		RESOLVED	Hyukjin Kwon	Actions
	SPARK-43747	Implement the pyfile support in SparkSession.addArtifacts		RESOLVED	Hyukjin Kwon	Actions
	SPARK-43768	Implement the archive support in SparkSession.addArtifacts		RESOLVED	Hyukjin Kwon	Actions
	SPARK-43795	Remove parameters not used for SparkConnectPlanner		RESOLVED	jiaan.geng	Actions
	SPARK-43829	Improve SparkConnectPlanner by reuse Dataset and avoid construct new Dataset		OPEN	Unassigned	Actions
	SPARK-44135	Document Spark Connect only API in PySpark		RESOLVED	Hyukjin Kwon	Actions
	SPARK-44290	Session-based files and archives in Spark Connect		RESOLVED	Hyukjin Kwon	Actions

Issue Links

• incorporates: SPARK-42554 Spark Connect Scala Client

Sub-Tasks

• High-Level design doc for Spark Connect: Resolved
• Initial protobuf definition for Spark Connect API: Resolved
• Initial prototype implementation: Resolved