NVIDIA Dynamo 影子引擎恢复：秒级恢复 LLM 推理容量
原标题：Restore LLM Inference Capacity in Seconds with Shadow Engine Recovery in NVIDIA Dynamo
AI 摘要
NVIDIA Dynamo 推出影子引擎恢复预览功能，通过 GPU 内存服务（GMS）在活动引擎旁保持一个完全初始化的空闲引擎，并共享权重，避免冷重启。在 GLM-5.2 双工作节点部署中，故障恢复时间从 283 秒降至 7.3 秒，提速约 39 倍。该功能解决了权重与进程绑定及初始化状态不可转移的问题，支持 vLLM、SGLang 和 TensorRT-LLM 集成。
正文节选
When an LLM engine process fails, the standard recovery path involves a cold restart. This requires loading weights into HBM from storage, compiling kernels, and capturing NVIDIA CUDA graphs. For large models, initialization can take several minutes, during which surviving workers must absorb the displaced traffic. Shadow engine recovery, available as a preview feature in NVIDIA Dynamo, moves most of this recovery work off the serving path. It keeps a fully initialized shadow engine idle on the