Buildbarn Meetup at Datadog
Join us at Datadog's Amsterdam office on October 13th, 10:00–16:00, for a day dedicated to Buildbarn, the open-source remote execution and caching platform for Bazel and other REAPI-compatible build systems. Held alongside BazelCon, this is a chance to connect with fellow build infrastructure engineers, exchange real-world incident lessons and dig into what's next for the project.
Talks, discussion, and lunch included. All are welcome, whether you're running Buildbarn in production or just curious about remote execution at scale. Come share what you're building and meet the humans behind the commits you've been reading.
Schedule
10:00–10:15 — Welcome
10:15–11:00 — Buildbarn at Datadog
James Sharpe and Guilhem Charles — Datadog
It’s 2026 and the Great CI Bottleneck is upon us. Datadog's primary monorepo LOC grew 160% in four months driven by agent-authored code. That growth showed up as billions of Bazel actions executed every week, a scale a handful of the largest Bazel deployments in the world already operate at, but it was uncharted territory for Datadog. Our existing Bazel CI setup wasn't built to absorb that, so we onboarded builds onto Buildbarn remote execution. The migration didn't fail cleanly, it failed repeatedly.
This talk walks through Buildbarn deployment architecture at Datadog, the incident timeline, the root causes underneath each one, and the concrete changes that got us to remote execution at scale. We will go over node and memory resizing, gRPC flow-control capping, size-class learning inversion and more!
11:00–11:45 — The Trio: CI, Bazel, and Buildbarn
Eaugene Thomas — Uber
We’ll cover the challenges of hosting Bazel caches around the world—including five zones across three continents—and scaling them effectively.
11:45–12:30 — bb-storage: Making CDC a First-Class Citizen
Benjamin Ingberg — Meroton
We’ve modified Buildbarn’s bb-storage so that all data in transit and at rest is dynamically processed using Content-Defined Chunking (CDC). We’ll explain how this works under the hood, why it matters, and the impact it can have on a Buildbarn cluster.
12:30–13:30 — Lunch
13:30–14:00 — bb-portal in Action: An Interactive Build Analysis
Isak Stenström and Oscar Falk — Meroton
Bring your laptop and join us for a hands-on walkthrough of the build-analysis tooling offered by bb-portal. Using a prepared Git repository, we’ll work together to analyze common issues that developers may encounter.
14:00–14:30 — Strategies for GPU Workers in Kubernetes
Chris McDonald — DoorDash Labs
14:30–15:00 — Buildbarn Charts
Jake Newfield — Hermetiq
An overview of Hermetiq’s open-source Helm charts for deploying Buildbarn on Kubernetes.
15:00–15:15 — Break
15:15–15:45 — Using buildbarn/bb-action-router to Support User-Specified Runner Docker Images at the Action Level
Marcin Kosiba — Apple
15:45–16:15 — Caching Analysis Data in Bonanza
Ed Schouten — Apple
Bonanza is an experimental build system intended to be compatible with Bazel. It differs from Bazel in that everything happens remotely—not only the execution of build actions and tests, but also parsing BUILD files, downloading external dependencies, and constructing the build graph.
Unlike Bazel, Bonanza does not perform incremental builds. Because a single Bonanza cluster can serve multiple users who may initiate builds in an arbitrary and potentially interleaving order, every build starts from a clean slate. It therefore relies heavily on efficient cache lookups to reuse previously computed parts of the build graph.
This talk will present Bonanza’s storage architecture and evaluation framework, followed by a discussion of ongoing experiments to improve build performance.
16:15–16:30 — Break
16:30–17:00 — Panel Discussion and Q&A