Senior Software Engineer (Data Infrastructure Compute)
Posted Updated
Summary
Senior Software Engineer building and operating Roblox's data compute platform on Kubernetes (EKS) for Spark workloads, including job submission, interactive notebooks, cluster automation, and scheduling at petabyte scale.
- Roblox’s data platform processes petabytes of data daily, powering analytics, ML, and product decisions. The Data Platform Compute team owns the compute layer underneath it: the Kubernetes (EKS) clusters where all of Roblox’s Spark workloads run, and the software that lets engineers, analysts, and ML practitioners submit and run that work reliably and cost-effectively
- As a Senior Software Engineer on this team, you will build and operate the systems that sit between our users and the cluster. These include the Spark job submission service, our internal Spark libraries, interactive notebook compute, and cluster lifecycle automation. You’ll have real ownership, you’ll work on problems where the design is not yet settled, and you’ll help decide which technologies Roblox adopts next
- Build and own core compute platform services: Design, build, and operate the software behind Spark job submission, including APIs, scheduling integration, and the libraries users depend on
- Launch interactive Spark at scale: Deliver Spark Connect to support notebook workflows, including provisioning and managing the EKS compute that backs them, with sensible isolation, resource limits, and cost controls
- Run a safe, automated cluster fleet: Build fleet management automation for EKS, such as blue-green rollouts that validate new AMIs and Kubernetes versions before they reach production workloads, with minimal disruption to running jobs
- Make jobs run better without user effort: Develop our auto-tuner, which analyzes Spark history and event logs to recommend or apply job configuration changes that improve performance and reduce cost
- Own batch scheduling on Kubernetes: Operate and upgrade our YuniKorn-based scheduling, including queue design, quotas, fairness across tenants, and version upgrades
- Shape what we adopt next: Evaluate emerging technology such as native Spark acceleration (e.g., Apache Comet). Run rigorous benchmarks on real Roblox workloads and make a clear recommendation backed by data
- Be an owner of reliability and cost: Define SLOs and metrics, build observability, and drive down toil, incidents, and compute spend across a multi-tenant platform
- Partner across the company: Work with data engineering, ML, analytics, and infrastructure teams to understand their needs, communicate changes, and migrate users to better paths
- Kubernetes depth: Strong experience operating production Kubernetes clusters at scale, such as upgrades, node lifecycle, autoscaling, and operators or controllers
- Spark on Kubernetes: Hands-on experience running Apache Spark on Kubernetes (EKS, GKE, or self-managed) as a platform, including executor and driver behavior, resource tuning, and troubleshooting failed or slow jobs. We are looking for platform builders, not just Spark users
- Engineering foundations: Strong proficiency in at least one of Go, Java, Scala, or Clojure, with a track record of writing clean, well-tested, maintainable backend code. Python is a plus, since much of our tuning and analysis tooling uses it
- Technical leadership: Ability to influence design decisions across teams, mentor engineers, and drive alignment on cross-cutting changes
- B.S. in Computer Science or equivalent practical experience
- 5+ years of experience building and operating production-grade distributed systems or infrastructure platforms
- Cloud fluency: Experience with AWS (EKS, EC2, IAM, S3) or GCP equivalents, and infrastructure-as-code
- Operational ownership: Experience taking a system from design through production operation, including on-call, incident response, and observability
- Experience with YuniKorn, Volcano, Kueue, or other batch schedulers on Kubernetes
- Experience with Spark Connect, Livy, Kyuubi, or other interactive and gateway layers for Spark
- Experience with the Spark Operator, Karpenter, Helm, Terraform, or Argo
- Experience maintaining internal Spark builds or driving Spark version upgrades
- Experience building tooling that uses Spark event logs or metrics for tuning, cost attribution, or anomaly detection
- Familiarity with native execution engines (Comet, Gluten, Velox) or JVM performance tuning
- Experience migrating workloads from YARN/Hadoop to Kubernetes
- Contributions to open-source projects in the data or Kubernetes ecosystem
- Experience operating at consumer-internet scale (100M+ users)