Skip to main content

What are Anyscale jobs?

What are Anyscale jobs?

Anyscale jobs run offline workloads in production with automatic retries, resource management, and comprehensive monitoring. Use jobs for batch processing tasks such as model training, batch inference, and data processing pipelines.

For continuously running applications such as model serving endpoints, see What are Anyscale services?.

Common use cases

The following are examples of workloads suited for Anyscale jobs:

Use caseDescription
Batch inferenceProcess large datasets through ML models for predictions at scale.
Model trainingDistribute training workloads across multiple GPUs or nodes.
Model fine-tuningFine-tune large language models on custom datasets.
Data processingTransform, clean, and prepare large datasets using Ray Data.
Hyperparameter tuningRun parallel experiments to find optimal model configurations.
ETL pipelinesExtract, transform, and load data between systems.
Recurring workloadsSchedule periodic data updates, model retraining, or report generation.

Key features

Anyscale jobs provide the following features for production batch workloads:

FeatureDescription
Automatic retriesConfigure retry policies with max_retries to automatically restart failed jobs. Jobs restart from the beginning with the same configuration.
Job queuesRun multiple jobs on the same cluster to reduce startup times. Supports FIFO, LIFO, and priority-based scheduling. See Use job queues to share clusters.
Job schedulesSchedule recurring workloads using cron expressions with timezone support. Automatically run jobs at specified intervals. See Job schedules.
Cluster managementAnyscale provisions clusters automatically when jobs start and terminates them when jobs complete. Configure custom compute resources or use existing clusters.
Comprehensive monitoringAccess job metrics, logs, Ray Dashboard, and custom Grafana dashboards. Set up alerts for job failures or performance issues. See Monitor a job.
Multi-cloud supportRun jobs with consistent APIs and configurations on AWS, Azure, Google Cloud, neoclouds, or Anyscale-hosted infrastructure.

Getting started

To run your first job, see Get started with jobs.

For detailed information on creating and managing jobs, see Create and manage jobs.

Best practices

Pricing

Jobs use standard Anyscale pricing based on the type of machines used. See the Anyscale pricing page.