Temporal Serverless Workers on Cloud Run: Scale workers ahead of demand

Temporal’s open-source platform is widely used for orchestrating distributed systems and AI workloads, but deploying workers on Kubernetes or VMs requires continuous management and incurs costs even when idle. Traditional serverless platforms are request-driven and ill-suited for Temporal’s pull-based task polling model, leading to scaling delays and operational overhead. Cloud Run worker pools address this by running instances that connect to task queues, process items, and scale to zero when idle, eliminating the need to maintain persistent clusters or autoscaling configurations.

Temporal Serverless Workers on Cloud Run, now in preview, automate the scaling of worker pools directly through Temporal Cloud. The system tracks task queue backlogs and adjusts worker pool sizes via the Cloud Run Admin API, using customer-provided IAM permissions. Workers run in the customer’s project under their service account, with Temporal managing capacity without opening connections to instances or requiring firewall adjustments.

The integration enables several use cases, including autoscaled AI agent workers that handle bursts of LLM steps or API calls and scale to zero when idle. Long-running workflows, such as those tracking real-world resources, benefit from flexible compute that scales down during idle periods. Bursty batch jobs, like billing runs or media transcoding, see costs reduced as worker pools scale to zero between tasks, with maximum instance counts capping expenses.

Running Temporal workers on Cloud Run worker pools removes the need for public endpoints, load balancers, or health-check ports, while offering steady CPU at a lower price. Worker pools list at nearly 40% less per vCPU-hour and GB-hour than instance-billed Cloud Run services. Temporal’s scaling avoids revision churn, as instance count changes do not trigger new deployments, and scaling decisions are based on task queue backlog-to-worker latency.