[{"id":"IL-0481","title":{"en":"RoCEv2 GIDs changed after RNIC port flaps, collapsing the communication collective and crashing LLM training jobs","ko":"RoCEv2 학습 클러스터에서 RNIC 포트가 플랩한 뒤 GID 가 바뀌어 집합 통신이 깨지고 LLM 학습 작업에 크래시가 났다"},"org":"Infrawaves","org_slug":"infrawaves","date":"2026-06-01","severity":"unknown","root_cause":"software-bug","url":"https://benchlet.com/incidentlib/app/cases/IL-0481","shared_components":["nccl"]}]