Principal Software Engineer, Orchestration

Roblox’s data platform powers analytics, ML, and product decisions for a platform serving 200M+ daily active users. As a Principal Software Engineer in the Data Platform Orchestration team, you will be the technical leader for the system that schedules and runs the company’s data workloads. Our team owns Apache Airflow and the Kubernetes (EKS) infrastructure underneath it, running 8,000+ DAGs every day, most of them Spark jobs. You will set the long-term architecture for how Roblox orchestrates data at scale, and you will be hands-on in making it faster, more reliable, and easier for hundreds of engineers to use.

You Will:

• Define the Orchestration Strategy: Own the multi-year architecture for Airflow and the surrounding platform, including scheduler and executor design, multi-tenancy, upgrades and migrations, and where the platform should evolve next.
• Scale and Harden Airflow: Find and remove scaling limits across the scheduler, metadata database, executors, and DAG parsing. Make sure thousands of daily DAGs run with predictable latency and strong SLAs.
• Own Orchestration on Kubernetes: Lead the design and operation of our EKS clusters for orchestration and Spark workloads, including autoscaling, resource isolation, cost efficiency, and deployment automation.
• Build Platform Tooling for Internal Engineers: Design the APIs, custom operators and providers, libraries, and self-service tooling that let data teams build, deploy, and monitor pipelines safely.
• Drive Observability and Reliability: Establish SLOs, alerting, and diagnostics for pipelines and platform components. Lead incident reviews and fix the root causes in code, not just in runbooks.
• Lead Across Teams: Work with data platform, governance, data science, and product engineering teams to align roadmaps, guide technical decisions, and build consensus on complex initiatives.
• Raise the Engineering Bar: Set standards for code quality, design, and operational readiness. Mentor staff, senior, and mid-level engineers.

You Have:

• B.S. equivalent in CS or sufficient experience.
• [10+] years of experience designing, building, and operating production-grade, large-scale distributed systems.
• Workflow Orchestration Depth: Deep, hands-on experience designing, running, and scaling a production data-related workflow orchestrator or job scheduler (thousands of workflows or DAGs a day). Apache Airflow is ideal, but experience with an equivalent system (Dagster, Prefect, Temporal, or a large in-house scheduler) is also welcome. You understand scheduler and executor design, metadata store performance, task isolation, retries and idempotency, and safe upgrades and migrations.
• Kubernetes Expertise: Proven experience managing and scaling production Kubernetes clusters on AWS (EKS), including Helm, autoscaling (Karpenter or Cluster Autoscaler), and GitOps-style deployment.
• Software Engineering Foundations: Strong proficiency in Python, plus Go or Java, with a record of writing clean, well-tested platform and tooling code. You are a software engineer who builds and improves systems, not someone who mainly handles tickets and operations.
• Data Workload Familiarity: Working knowledge of how Spark jobs run on Kubernetes (Spark on K8s, spark-operator, or EMR on EKS) and how orchestration interacts with them.
• Platform Mindset: Experience building platforms used by other engineers, with strong instincts for APIs, observability, reliability, and developer experience.
• Cross-Organizational Technical Leadership: A track record of influencing direction across a large engineering organization and driving multi-quarter projects from concept to production.
• Extreme Ownership: You define the problem when requirements are unclear, unblock dependencies, and see complex work through to stable production.
• Bonus Points:
  • Contributions to Apache Airflow or related open-source projects (Airflow providers, Helm chart, spark-operator, Karpenter, Argo CD).
  • Experience with other orchestration systems (Dagster, Temporal, Argo Workflows, Flyte) and the judgment to know when each fits.
  • Experience with Terraform or other infrastructure-as-code, Prometheus/OpenTelemetry, and cost optimization or chargeback for shared platforms.
  • Experience operating infrastructure at consumer-internet scale.