The DigitalOcean team recently performed a network maintenance to replace their core routing infrastructure with new hardware from Juniper. The upgrade involved replacing edge routers and routing cores, and the team worked closely with Juniper to ensure a smooth transition. However, after the initial deployment, the team encountered several issues, including intermittent networking problems and flapping core routers. After escalating the issue to Juniper's technical support channel, the team discovered that faulty Cisco switches were causing an issue with network re-convergence, which was resolved by removing the spanning-tree protocol on the Cisco hardware. The team also observed a failure of Juniper's self-diagnostics process, which led them to suspect faulty hardware. To resolve this, they temporarily disabled jflow and implemented internal monitoring for DDoS detection and prevention. The team is now meeting with Juniper engineers to review the entire network topology and configurations, and plans to conduct further diagnostics on the Juniper gear. Additionally, the team has decided to move away from a traditional core setup and adopt a pod architecture, which will allow them to scale more efficiently and improve communication with customers.