Netflix Reworks Conductor Workflow Engine for Increased Scale and Efficiency

Conductor Reworked for Scale

Netflix has re-engineered its Conductor workflow orchestration engine to manage growing demands across its Content and Studio Engineering, Ads, and Games organizations. The updated system now supports approximately 200,000 workflow definitions across 150 applications and executes about 420 million workflows per month.

Increased Workflow Capacity and Performance

The redesign significantly increases the supported workflow size, allowing for approximately 30,000 tasks per workflow, up from 2,500. This change also reduced p99 workflow evaluation latency by about 40%, indicating a substantial improvement in processing efficiency.

Architectural Changes in Conductor 4.0

Conductor 4.0 addresses previous bottlenecks by separating workflow metadata from task and user data. The evaluator now uses a lightweight workflow blueprint and loads only the necessary task data for each decision, rather than the entire workflow state. This change mitigates memory pressure that occurred with larger workflows in earlier versions.

Further architectural changes include removing locking from task state coordination by storing pending and terminal task states separately. Workflow evaluation has also been moved out of the synchronous request path, with updates processed asynchronously via exclusive Timestone queues.

Addressing Prior Scaling Challenges

Community discussions in 2022 highlighted scaling issues, such as high JVM heap usage with large numbers of workflows and tasks, and slow definition loading times for complex workflows. These issues stemmed from the engine loading entire running workflows into memory for evaluation. The current redesign directly addresses these previously identified limitations.