Introducing Multi-Cloud and Multi-Region Resilience in Orkes Conductor

Protecting critical agents and workflows from regional or cloud provider outages shouldn’t require your team to build its own redundancy system. We’re excited to introduce multi-region and multi-cloud resilience in Orkes Conductor so you don’t have to operate this yourself. Orkes continuously replicates execution state, monitors cluster health and handles failover automatically, all backed by a 99.99% availability SLA.

Orkes continuously replicates agent workflow execution state and configuration to a second Conductor cluster and monitors their health. This can be in another region, across a continent, or another cloud. When the primary becomes unavailable and a failover is needed, Orkes handles it automatically by redirecting the same multi-region endpoint to the second cluster. Your client applications, agents, workers and tools keep their existing configuration, with no customer action needed during failover.

Mission-critical business continuity

Your critical agents and applications need both an available orchestration service and a way to continue the work already in progress. Multi-region and multi-cloud resilience gives the orchestration layer a recovery and continuity path outside the primary environment in which it runs. Let’s explore a real-world use case and how your workflows are protected.

For example, in a payment processing use case, an agent might gather transaction records, assess a discrepancy and propose a correction. A person reviews the proposal before the workflow calls a payment service to apply it. The records, task results and approval decision are all part of the execution state.

In the example above, progress in each task of this workflow is replicated to the secondary cluster from the primary. Before the “Apply correction” task begins execution, the primary cluster goes offline. Thanks to automated multi-region failover, Conductor resumes the same investigation on the second cluster, exactly where it was left off, and proceeds with the authorized correction.

Agents and conventional workflows use the same recovery model. Some in-flight tasks may be retried during recovery, so tasks that perform business actions, such as issuing a payment, should handle retries without duplicating the action.

Managed recovery across regions and clouds

Orkes Conductor already preserves execution state through worker failures and restarts, while multi-AZ deployments for single-region clusters protect against availability-zone outages. A multi-region deployment extends that protection with two running Conductor clusters. The primary serves traffic and runs agents and workflows. The replica continuously receives execution state and configuration, ready to take over when necessary.

A replica in a second region lets the platform recover from a regional outage. Placing it in another cloud extends recovery across providers. Replication runs within Conductor, so both types of deployments use the same replication and recovery models. Workflow definitions, prompts and access settings are replicated with execution state, which prepares the second cluster to continue the process in the event of an outage.

Replication is encrypted end-to-end in transit between the clusters, and data is encrypted at rest in both locations. Replicated data stays in the regions, clouds and accounts selected for your deployment.

Enterprise grade recovery targets

Our recovery time objective (RTO) is 10 minutes, but in practice recovery is typically faster. For asynchronous workflow executions, we target a maximum 90-second data-loss window recovery point objective (RPO) for recent execution progress.

For workflows started in synchronous execution mode, our RPO target is near zero when failovers happen. That matters for critical applications where restoring service also means preserving the progress of requests already being processed.

Available for new and existing clusters

Multi-region and multi-cloud resilience is available across AWS, Microsoft Azure and Google Cloud, in both Orkes-hosted and customer-hosted deployments. Orkes manages replication and recovery whether the clusters run in your cloud accounts or ours.

You can add a replica to an existing cluster in another region, with the same cloud provider or a different cloud provider. Once it has caught up, the pair is ready for multi-region protection. The capability is available as an add-on for enterprise deployments.

Get a demo to see how Orkes provides mission-critical availability for your agentic workflows.