Incident Library?

Incident Library?

Component ?

InfiniBand fabric InfiniBand 패브릭 · `infiniband`

2 cases · 2024 to 2026

Root cause ?Dependency failure 2

Contributing factor ?Monitoring blind spot 2 Unverified assumption 1 Unreviewed defaults 1 Partial rollout 1

Symptom ?Link flapping 2 NCCL timeout 2 GPU unavailable 1 Throughput degradation 1 Training job failure 1

Cases, by severity

IL-0023 A training run failed on an NCCL timeout from one brief InfiniBand port drop — Databricks, 2026, sev4
Root cause Dependency failure
IL-0025 Link flapping, hangs and slowdowns while bringing up a 4,088-GPU cluster — Imbue, 2024
Root cause Dependency failure

View on the graph