BentoML Kubernetes 部署
io.github.bentoml/BentoML/bentoml-k8s-deploy
使用纯 kubectl manifest(不用 Helm、operator 或 BentoCloud/Yatai)将容器化的 BentoML 服务部署到 vanilla Kubernetes 集群。接收已推送的容器镜像(来自 bentoml-containerize 技能),探测 bento 的服务拓扑,为部署编写一份 `config.yml`,并据此为每个 BentoML 服务渲染一个 Deployment + Service(外加可选 HPA/Ingress),按依赖顺序应用,并用真实推理请求验证 rollout。当用户说“deploy my BentoML service to Kubernetes”“deploy this bento image to my cluster”“run my bento on k8s”“create k8s manifests for my bento”“split my bento services into separate pods”“expose my BentoML service in Kubernetes”等时使用。也可诊断失败的部署——“my BentoML pods are crashing”“ImagePullBackOff”“pod stuck Pending”“readiness probe failing”“rollout stuck”“can't reach my service on Kubernetes”“inference 4xx/5xx”。