Kubernetes部署Deepseek(内网环境)
n\u003c/think\u003e\n\n**解答：**\n\n我们来比较一下 \\(3{.}11\\) 和 \\(3{.}3\\) 的大小。\n\n1. **比较整数部分：**\n\\[\n3 \\quad (3{.}11) \\quad \\text{和} \\quad 3 \\quad (3{.}3)\n\\]\n整数部分相同，都为3。如果这里创建的PV不对，ollama-models-
部署Kubernetes集群
略
启动Ollama Operator
官网：https://ollama-operator.ayaka.io/pages/en/guide/getting-started/
#需要科学上网访问github下载：
https://raw.githubusercontent.com/nekomeowww/ollama-operator/v0.10.1/dist/install.yaml
#需要科学上网下载Ollama Operator依赖的两个Image：
ghcr.io/nekomeowww/ollama-operator:0.10.1
gcr.io/kubebuilder/kube-rbac-proxy:v0.15.0
Install operator:
kubectl apply --server-side=true -f install.yamlWait for the operator to be ready:
kubectl wait \
-n ollama-operator-system \
--for=jsonpath='{.status.readyReplicas}'=1 deployment/ollama-operator-controller-manager如下回显说明准备就绪：
deployment.apps/ollama-operator-controller-manager condition met
kubectl get all -n ollama-operator-system
NAME READY STATUS RESTARTS AGEpod/ollama-operator-controller-manager-c896f8956-97tpv 2/2 Running 2 3d1h
NAME TYPE CLUSTER-IP EXTERNAL-IP PORT(S) AGEservice/ollama-operator-controller-manager-metrics-service ClusterIP 10.254.135.104 <none> 8443/TCP 3d1h
NAME READY UP-TO-DATE AVAILABLE AGEdeployment.apps/ollama-operator-controller-manager 1/1 1 1 3d1h
NAME DESIRED CURRENT READY AGEreplicaset.apps/ollama-operator-controller-manager-c896f8956 1 1 1 3d1h
K8s部署大模型
在K8s上部署大模型，可以直接用Ollama Operator的CRD Model资源部署。
Kubernetes上创建所需存储
创建StorageClass：首次部署大模型，会先创建一个store 服务，用于存储ollama的模型文件，store服务会用到PV，需要事先准备好。
kubectl apply -f storage-class.yamlkubectl get sc
NAME PROVISIONER RECLAIMPOLICY VOLUMEBINDINGMODE ALLOWVOLUMEEXPANSION AGE
local-path kubernetes.io/no-provisioner Delete WaitForFirstConsumer true 3d
storage-class.yaml：
apiVersion: storage.k8s.io/v1
kind: StorageClass
metadata:
name: local-path
provisioner: kubernetes.io/no-provisioner # 使用静态 Provisioner
volumeBindingMode: WaitForFirstConsumer # 等待 Pod 调度后再绑定卷
reclaimPolicy: Delete # 卷回收策略（Delete、Retain）
allowVolumeExpansion: true # 允许卷动态扩展创建PV：spec.capacity.storage 大小必须是100G，因为ollama-models-store默认引用的PV就是100G，并且未找到可以修改的地方。如果这里创建的PV不对，ollama-models-store创建的PVC将找不到可以绑定的PV，而一直pending。
| 访问模式 | 定义 | 特点 |
|---|---|---|
| ReadWriteOnce | 卷可以被单个节点以读写方式挂载。 | - 适用于需要独占访问的场景- 常用于单副本应用- 大多数存储系统都支持 |
| ReadOnlyMany | 卷可以被多个节点以只读方式挂载。 | - 适用于内容分发场景- 适合共享只读数据- 支持的存储系统较多 |
| ReadWriteMany | 卷可以被多个节点以读写方式挂载。 | - 适用于需要多副本读写访问的场景- 对存储系统要求较高- 支持的存储系统有限 |
kubectl apply -f pv.yamlkubectl get pv
NAME CAPACITY ACCESS MODES RECLAIM POLICY STATUS CLAIM STORAGECLASS VOLUMEATTRIBUTESCLASS REASON AGE
local-pv-node189 100Gi RWO Delete Bound default/ollama-models-store-pvc local-path <unset> 2d15h
local-pv-node191 100Gi RWO Delete Available local-path <unset> 2d15h
kubectl get pvc
NAME STATUS VOLUME CAPACITY ACCESS MODES STORAGECLASS VOLUMEATTRIBUTESCLASS AGE
ollama-models-store-pvc Bound local-pv-node189 100Gi RWO local-path <unset> 2d15h
pv.yaml：
---
apiVersion: v1
kind: PersistentVolume
metadata:
name: local-pv-node189
spec:
capacity:
storage: 100Gi
volumeMode: Filesystem
accessModes:
- ReadWriteOnce
persistentVolumeReclaimPolicy: Delete
storageClassName: local-path
hostPath:
path: /data/ # 本地存储路径
type: Directory
nodeAffinity:
required:
nodeSelectorTerms:
- matchExpressions:
- key: kubernetes.io/hostname
operator: In
values:
- k8s130-node189 # 替换为你的节点名称启动Deepseek-r1 1.5b
kubectl apply -f deepseek-r1-1.5b.yaml 因为deepseek-r1-1.5b.yaml 中定义的image地址可能不生效。此时，需要进行如下处理
1、编辑statefulset/ollama-models-store
kubectl edit statefulset/ollama-models-store
或者
kubectl describe po/ollama-models-store-0 将
image: ollama/ollama
imagePullPolicy: Always
替换成：
image: 192.168.xxx.101/library/ollama:latest
imagePullPolicy: IfNotPresent
2、编辑deploy/ollama-model-deepseek-r1同上处理
kubectl edit deploy ollama-model-deepseek-r1删除旧的replicaset
kubectl get replicaset
NAME DESIRED CURRENT READY AGE
replicaset.apps/ollama-model-deepseek-r1-644575b7bf 1 1 0 2m55s
replicaset.apps/ollama-model-deepseek-r1-74866c698c 1 1 0 6m15s
kubectl delete replicaset.apps/ollama-model-deepseek-r1-74866c698c
执行kubectl apply -f deepseek-r1-1.5b.yaml命令的创建流程大致如下：
1、首先会创建statefulset/ollama-models-store，调度启动pod/ollama-models-store-0；同时创建：service/ollama-models-store；
pod/ollama-models-store-0启动/bin/ollama serve服务
2、当pod/ollama-models-store准备好，则会创建deployment.apps/ollama-model-deepseek-r1，调度启动pod，首先initContainers，拉取大模型：执行apt update && apt install curl -y && ollama pull deepseek-r1 && curl http://ollama-models-store:11434/api/generate -d '{"model": "%!s(MISSING)"}'deepseek-r1:1.5b
3、initContainers成功后启动，则会创建pod/ollama-model-deepseek-r1-644575b7bf-sstjd ；同时创建service/ollama-model-deepseek-r1
kubectl get all
NAME READY STATUS RESTARTS AGEpod/ollama-model-deepseek-r1-644575b7bf-sstjd 1/1 Running 0 2d15hpod/ollama-models-store-0 1/1 Running 0 2d15h
NAME TYPE CLUSTER-IP EXTERNAL-IP PORT(S) AGEservice/kubernetes ClusterIP 10.254.0.1 <none> 443/TCP 191dservice/ollama-model-deepseek-r1 ClusterIP 10.254.223.236 <none> 11434/TCP 2d15hservice/ollama-models-store ClusterIP 10.254.182.20 <none> 11434/TCP 2d15h
NAME READY UP-TO-DATE AVAILABLE AGEdeployment.apps/ollama-model-deepseek-r1 1/1 1 1 2d15h
NAME DESIRED CURRENT READY AGEreplicaset.apps/ollama-model-deepseek-r1-644575b7bf 1 1 1 2d15h
NAME READY AGEstatefulset.apps/ollama-models-store 1/1 2d15h
## 清理大模型：仅仅delete -f deepseek-r1-1.5b.yaml 并不行。
kubectl delete -f deepseek-r1-1.5b.yaml
kubectl delete StatefulSet/ollama-models-store
kubectl delete pvc/ollama-models-store-pvcdeepseek-r1-1.5b.yaml ：
apiVersion: ollama.ayaka.io/v1
kind: Model
metadata:
name: deepseek-r1
spec:
replicas: 1
image: deepseek-r1:1.5b
imagePullPolicy: IfNotPresent
storageClassName: local-path
persistentVolume:
accessMode: ReadWriteOnce
podTemplate:
spec:
initContainers:
- args:
name: ollama-image-pull
image: 192.168.xxx.101/library/ollama:latest
imagePullPolicy: IfNotPresent
containers:
- args:
name: server
image: 192.168.xxx.101/library/ollama:latest
imagePullPolicy: IfNotPresent调用Deepseek-r1 1.5b
curl http://10.254.56.37:11434/v1/chat/completions -H "Content-Type: application/json" -d '{
"model": "deepseek-r1:latest",
"messages": [
{
"role": "user",
"content": "Hello! 比较3.11和3.3的大小？"
}]}'{"id":"chatcmpl-430","object":"chat.completion","created":1745992155,"model":"deepseek-r1:latest","system_fingerprint":"fp_ollama","choices":[{"index":0,"message":{"role":"assistant","content":"\u003cthink\u003e\n首先，比较两个小数的整数部分。两个数都为3，说明整数部分相同。\n\n接着，比较十分位。第一个数是0.11，第二个数是0.3。明显0.1小于0.3。\n\n因此，可以得出3.11小于3.3。\n\u003c/think\u003e\n\n**解答：**\n\n我们来比较一下 \\(3{.}11\\) 和 \\(3{.}3\\) 的大小。\n\n1. **比较整数部分：**\n \\[\n 3 \\quad (3{.}11) \\quad \\text{和} \\quad 3 \\quad (3{.}3)\n \\]\n 整数部分相同，都为3。\n\n2. **比较十分位：**\n - \\(3{.}11\\) 的十分位是\\(1\\)\n - \\(3{.}3\\) 的十分位是\\(3\\)\n\n 这里，\\(1 \u003c 3\\)，因此：\n \\[\n 3{.}11 \u003c 3{.}3\n \\]\n\n**最终答案：**\n\\[\n\\boxed{3{.}11 \u003c 3{.}3}\n\\]"},"finish_reason":"stop"}],"usage":{"prompt_tokens":20,"completion_tokens":262,"total_tokens":282}}
集成Open WebUI on Kubernetes
创建openwebui所需的PV
kubectl apply -f pv4openwebui.yaml
kubectl apply -f pvc4openwebui.yamlpv4openwebui.yaml
---
apiVersion: v1
kind: PersistentVolume
metadata:
name: open-webui-node191
spec:
capacity:
storage: 10Gi
volumeMode: Filesystem
accessModes:
- ReadWriteOnce
persistentVolumeReclaimPolicy: Delete
storageClassName: local-path
hostPath:
path: /data/open_webui # 本地存储路径
type: Directory
nodeAffinity:
required:
nodeSelectorTerms:
- matchExpressions:
- key: kubernetes.io/hostname
operator: In
values:
- k8s130-node191 # 替换为你的节点名称pvc4openwebui.yaml
---
apiVersion: v1
kind: PersistentVolumeClaim
metadata:
name: open-webui-pvc
spec:
storageClassName: local-path
accessModes:
- ReadWriteOnce
resources:
requests:
storage: 10Gi启动openwebui
kubectl apply -f open-webui-deploy.yaml open-webui-deploy.yaml
apiVersion: apps/v1
kind: Deployment
metadata:
name: open-webui
spec:
replicas: 1
selector:
matchLabels:
app: open-webui
template:
metadata:
labels:
app: open-webui
spec:
restartPolicy: Always
containers:
- name: myweb
image: 192.168.XXX.101/library/open-webui:main
imagePullPolicy: IfNotPresent
ports:
- containerPort: 8080
env:
- name: OLLAMA_BASE_URL
value: 'http://ollama-model-deepseek-r1:11434'
volumeMounts:
- name: open-webui-data
mountPath: /app/backend/data
volumes:
- name: open-webui-data
persistentVolumeClaim:
claimName: open-webui-pvc
---
apiVersion: v1
kind: Service
metadata:
name: open-webui-service
spec:
type: NodePort
ports:
- protocol: TCP
port: 8080
nodePort: 31000
selector:
app: open-webui使用Deepseek
浏览器访问http://node-ip:31000；按照如下提示先注册账号
账号用于登录WEBUI，然后，我们就可以使用本地的Deepseek了：
火山引擎开发者社区是火山引擎打造的AI技术生态平台，聚焦Agent与大模型开发，提供豆包系列模型（图像/视频/视觉）、智能分析与会话工具，并配套评测集、动手实验室及行业案例库。社区通过技术沙龙、挑战赛等活动促进开发者成长，新用户可领50万Tokens权益，助力构建智能应用。
更多推荐
20
0- 0
已为社区贡献7条内容
所有评论(0)