[Proposal] Introducing PL/Container to Cloudberry – Feedback Requested

[yjhjstz]

Proposers

@yjhjstz @my-ship-it

Proposal Status

Under Discussion

Abstract

Hi everyone,

We’re proposing to bring PL/Container into the Cloudberry ecosystem and would like your feedback.

Motivation

🧩 What is PL/Container?

PL/Container is a trusted language execution engine that enables sandboxed execution of user-defined functions by launching Docker containers. This ensures strong isolation of the runtime environment from the host OS, supporting safe and flexible language extensions such as Python and R.

“This is an implementation of a trusted language execution engine capable of bringing up Docker containers to isolate executors from the host OS, i.e., implement sandboxing.”
— from the original PL/Container project

For more details, see the PL/Container Architecture.

Implementation

🚀 What’s happening now?

We’ve submitted an initial pull request to restore and import the original code into Cloudberry’s contrib space:

🔗 cloudberry-contrib/plcontainer#1

This PR brings the archived code from the Greenplum repository and sets the stage for:

• Integration with Cloudberry's evolving UDF execution model

Rollout/Adoption Plan

💬 We’d love to hear your thoughts:

• Is this approach aligned with Cloudberry’s goals?
• Are there alternative sandboxing/execution models we should consider?
• Would you be interested in testing or contributing to future development?

Let’s collaborate to enable safe, powerful, and containerized language support in Cloudberry!

Thanks in advance for your input 🙌

Are you willing to submit a PR?

• Yes I am willing to submit a PR!

[jianlirong]

This is a very useful feature and an excellent start. By inheriting from PostgreSQL and Greenplum Database, Cloudberry has had very comprehensive UDF support from day one. Nevertheless, aside from the security issues mentioned above, compared to supporting a new procedural language, PL/Container can more agilely and flexibly support various user-defined computations, allowing the use of any familiar languages and tools.

We can also see this trend from other data warehousing products/services. Although Snowflake provides the ability to write UDFs in various languages through Snowpark, they also offer Snowflake Container Service to make it more convenient for users to use custom computations, especially in the current context where LLMs are increasingly being integrated into data analysis scenarios.