Snowflake Task-based orchestration for containerized Jobs

12 March 2026

Today we’re announcing improved Snowflake Task-based orchestration for containeriazed Jobs (for ML Jobs and Snowpark Container Services customers). Tasks let users orchestrate Jobs natively inside Snowflake: scheduling or event-triggering containerized runs, chaining them into end-to-end pipelines, and keeping orchestration close to the data and the same governance and security model teams already rely on.

With task graphs/DAGs customers can express multi-step dependencies (including parallel branches and fan-in/fan-out patterns) as a single coordinated workflow rather than stitching together external schedulers and custom glue code, and because execution is surfaced directly in Snowsight, operators get built-in observability — visual graph views, run history, node-level status, and fast drill-down straight into SPCS telemetry for troubleshooting and reruns.

This feature set takes batch and ML/GPU-based workloads in Snowflake to the next level. Here are the specific improvements that underpin this new functionality:

Job execution in a serverless Task - lets Snowflake orchestrate long-running Jobs using GS and Compute pools only, so a Warehouse never has to be resumed or kept running for the Job’s duration. This feature is key for compute-intensive workloads such as ML, where jobs could run for a long period of time on powerful GPU-based hardware, and price/performance is key for customers.
Task context propagation in Jobs - easy access to return values and graph configuration inside Job code via get_predecessor_return_value(), set_return_value(), and get_task_graph_config() methods (Python)
ML Jobs with Tasks - use an ML Job definition to define what a Task executes by passing in a definition, say a training function, into a DAGTask()
Snowsight observability - easy linking between Task runs / Query details and Job details for improved debugging of failed Tasks and robust operational monitoring

Here is a short video demonstrating the new functionality:

Here are some additional resources to help you get started.

We can’t wait for you to try this out!