Skip to content

Use Heata with Anyscale

Anyscale is the managed platform for Ray, built by the creators of Ray. It provides a production-grade environment for scaling AI and ML workloads — including distributed training, batch inference, model serving, and data processing — without managing infrastructure.

Heata can be added as a compute provider within your Anyscale account, giving you access to our low-carbon, distributed infrastructure directly from the Anyscale platform. Your workloads run on Heata hardware while you keep the full Anyscale experience — workspace management, job scheduling, service deployments, and observability.

How it works

  1. You keep using Anyscale — same dashboard, same APIs, same workflows
  2. We provide the compute — Heata appears as an available cloud in your Anyscale account
  3. Your workloads run on low-carbon infrastructure — the heat from your compute provides free hot water to UK households

What you can do with Anyscale on Heata

Workspaces

Interactive development environments with a scalable Ray cluster behind them. Write and test code in a managed VS Code environment, then promote to production jobs or services with one command.

  • Autoscales from a single node to multi-node clusters as your workload grows
  • Persistent storage for notebooks, data, and code
  • Git integration and dependency management built in

Jobs

Production batch workloads — training runs, data pipelines, batch inference.

  • Job queues: share a persistent cluster across multiple jobs with FIFO, LIFO, or PRIORITY execution
  • Cron scheduling: run jobs on a schedule with cron_expression
  • Timeouts: set timeout_s per job to prevent runaway workloads
  • Retries: configure max_retries for automatic recovery from transient failures
  • Concurrency: control with max_concurrency (up to 100 concurrent jobs per queue)

Services (Model Serving)

Deploy models as production API endpoints via Ray Serve.

  • Autoscaling replicas: scale based on request load with configurable min_replicas, max_replicas, and target_ongoing_requests
  • Zero-downtime upgrades: incremental rollouts with automatic rollback
  • Canary deployments: run up to 10 versions behind a single endpoint for A/B testing
  • Multi-model hosting: serve multiple models on shared infrastructure

Compute configuration

When Heata is set up as a cloud in your Anyscale account, you configure compute using Anyscale's standard compute config. This controls cluster size, autoscaling, and instance selection:

cloud: heata
head_node:
  instance_type: m5.8xlarge
worker_nodes:
  - instance_type: m5.4xlarge
    min_nodes: 1
    max_nodes: 10
  - instance_type: g5.4xlarge
    min_nodes: 0
    max_nodes: 5
flags:
  idle_termination_seconds: 60

Key options:

Setting Description
instance_type Compute instance type per node group
min_nodes / max_nodes Autoscaling bounds per worker group
min_resources / max_resources Cluster-wide resource limits (e.g. max GPUs)
idle_termination_seconds How long idle workers stay up before scaling down
timeout_s Per-job timeout to prevent runaway workloads
max_retries Automatic job retries on failure

Full details: Anyscale Compute Configuration docs

Enterprise security

Anyscale is SOC 2 Type 2 certified and designed for enterprise workloads:

  • Isolation: each Anyscale Cloud is a fully isolated environment — separate dev, staging, and production with distinct resource boundaries
  • Private networking: VPC support for network-level isolation
  • SSO and access control: integrate with your identity provider
  • Audit logs: track who did what, when
  • Shared responsibility model: Anyscale manages the control plane with 24/7 monitoring, automated alerting, and disaster recovery

For full details, see the Anyscale Trust Center and Security & Compliance docs.

Anyscale documentation

Getting started

Adding Heata as a compute provider in your Anyscale account requires a short setup on our side. Contact us and we'll get you connected:

Your compute on Heata provides free hot water for families around the UK.