Skip to content

Fast and the Furious: CICD Pipeline for eBPF Programs at Meta Scale#

KubeCon + CloudNativeCon North America 2025 digest Program Wildcard pick 10–13 November 2025

Recording

Theophilus Benson (CMU) & Prankur Gupta, Meta — Wildcard

Play the recording: Fast and the Furious: CICD Pipeline for eBPF Programs at Meta Scale

  • Why: Not a networking-design talk, but a rare look at the release engineering of eBPF at hyperscale — 150+ programs updated weekly across the fleet. The interesting question is how you test and roll out kernel-attached code without causing outages, which is adjacent to (and quietly underpins) everything Cilium/datapath related.
  • What: How Meta runs CI/CD for its large eBPF program estate spanning observability, load balancing, network-stack specialization, and security. (from abstract only)
  • How: Abstract is truncated in the program; it describes a custom, flexible in-house pipeline and the outage risk introduced by weekly eBPF updates, but specifics aren't fully stated. (from abstract only — partial)
  • Where applicable: Hyperscale eBPF fleets; Meta-specific tooling, but the testing/rollout discipline generalizes to anyone shipping eBPF programs.