BoF: Training the Next Generation of Open Source Developers

### Overview In this Birds of a Feather (BoF) session, Jeremy, Director of Technical Community and head of the Open Source Programs Office at Datadog, discusses the creation, structure, and outcomes of Datadog's Open Source Co-op Program. Recognizing that many university graduates enter the workforce lacking practical open source experience and observability background, Datadog established a dedicated 15-to-20-week paid internship focused on contributing upstream to open source projects. The initiative aims to train emerging engineers, give back to the open source community, dogfood internal tooling, and establish a viable recruiting pipeline.

### Key Topics - **Open Source at Scale:** Datadog's dual role as a proprietary SaaS monitoring and observability platform processing trillions of data points daily while heavily utilizing, contributing upstream to, and open-sourcing internal tools (e.g., Kafka Kit, libraries, and OpenTelemetry contributions). - **The Co-op Program Model:** A 15-to-20-week paid internship providing full benefits and employee status, where co-ops identify, report, and fix upstream bugs and performance bottlenecks in projects relevant both to the student's technical interests and the company's tech stack. - **Engineering Communication:** Requiring co-ops to document their technical journeys and performance findings on the Datadog engineering blog, reinforcing written communication skills. - **Mutual Benefits:** Co-ops gain a publicly referenceable portfolio of merged pull requests, review discussions, and community engagement. Meanwhile, Datadog identifies onboarding friction in its own products, builds an engineering recruitment pipeline, and bolsters its open source profile.

### Technical Details - **Case Study: Andrew McBurney:** A Waterloo student who joined in 2018 as the inaugural co-op to validate the program. He investigated performance issues in Homebrew's `brew linkage` command. - **Homebrew Performance Optimization:** Before the optimization, `brew linkage` took approximately 8 seconds without caching. Andrew's initial proposed solution utilized SQLite, but was rejected upstream due to introducing an unwanted external dependency. Mentorship helped him adopt a DBM-based solution, avoiding extra dependencies. The merged change introduced a negligible ~0.1-second overhead on the first run to build the cache, while speeding up subsequent runs to approximately 200 milliseconds. Flame graphs and metrics were used to validate the improvement. - **Internal Tooling and Other Work:** Beyond Homebrew, Andrew developed "Jello," an internal Trello and GitHub integration addressing internal workflow inefficiencies, which was open-sourced and later expanded/ported toward Jira by subsequent co-ops (such as Ursula). Co-ops also touched projects such as Dgraph and Fastlane.

### Challenges and Takeaways - **Community Alignment & Pre-engagement:** Maintainers must be engaged early so they understand the intern's goals and timeline constraints; open source maintainer responsiveness varies greatly across projects and over time. - **Realistic Scoping & Time-Boxing:** Avoid wildly optimistic goals (such as tackling all of Kubernetes or Rails); projects must be scoped to match academic rubric requirements and 16-to-20-week timeframes. Program lengths under 8–16 weeks were found insufficient. - **Crash Courses & Mentorship:** Interns need structured guidance not just on technical architecture, but on upstream etiquette (e.g., participating in read-only mode first to understand project norms) and understanding why their open source contributions deliver direct value back to the sponsoring enterprise.