Merge pull request 'fix: bugs for interuption of downloading models' (#2) from vllm into main

Reviewed-on: #2

charts.list

@@ -1,4 +1,4 @@
 # List each Helm chart directory to package and push (one per line)
-# vllm-serve
+vllm-serve
 # code-server-chart
-open-webui
+# open-webui

vllm-serve/Chart.yaml

@@ -5,4 +5,4 @@ appVersion: 1.16.0
 description: A Helm chart for deploying vLLM with NFS storage
 name: vllm-serve
 type: application
-version: 0.2.0
+version: 0.2.1

vllm-serve/templates/lws.yaml

@@ -31,19 +31,31 @@ spec:
               DEST_DIR="{{ .Values.model.localMountPath }}/Weight/$MODEL_NAME"
               # DEST_DIR="{{ .Values.model.localMountPath }}/{{ .Values.model.huggingfaceName }}"
               # 检查模型是否存在，不存在则下载
-              echo "DEST_DIR= $DEST_DIR"
+              # echo "DEST_DIR= $DEST_DIR"
-              if [ ! -f "$DEST_DIR/config.json" ]; then
+              # if [ ! -f "$DEST_DIR/config.json" ]; then
-                ls -l {{ .Values.model.localMountPath }}
+              #   ls -l {{ .Values.model.localMountPath }}
-                echo "Downloading model {{ .Values.model.huggingfaceName }} to $DEST_DIR"
+              #   echo "Downloading model {{ .Values.model.huggingfaceName }} to $DEST_DIR"
-                wget https://hf-mirror.com/hfd/hfd.sh 
+              #   wget https://hf-mirror.com/hfd/hfd.sh 
-                chmod a+x hfd.sh
+              #   chmod a+x hfd.sh
-                apt update && apt upgrade
+              #   apt update && apt upgrade
-                apt install aria2 -y
+              #   apt install aria2 -y
-                ./hfd.sh {{ .Values.model.huggingfaceName }} --local-dir "$DEST_DIR"
+              #   ./hfd.sh {{ .Values.model.huggingfaceName }} --local-dir "$DEST_DIR"
-                # huggingface-cli download {{ .Values.model.huggingfaceName }} --local-dir "$DEST_DIR"
+              #   # huggingface-cli download {{ .Values.model.huggingfaceName }} --local-dir "$DEST_DIR"
-              else
+              # else
-                echo "Model already exists at $DEST_DIR"
+              #   echo "Model already exists at $DEST_DIR"
+              # fi
+              SUCCESS_FLAG="${DEST_DIR}/.success_flag"
+              if [ -f "$SUCCESS_FLAG" ]; then
+                echo "✅ Success flag found. Skipping download."
+                exit 0
               fi
+              echo "⬇️ Starting download..."
+              apk add --no-cache bash aria2 wget ca-certificates
+              wget https://hf-mirror.com/hfd/hfd.sh -O hfd.sh && chmod +x hfd.sh
+              ./hfd.sh {{ .Values.model.huggingfaceName }} --tool aria2c -x 8 --local-dir "$DEST_DIR"
+
+              touch "$SUCCESS_FLAG"
+              echo "🎉 Done."
           volumeMounts:
           - name: weight-volume
             mountPath: {{ .Values.model.localMountPath }}
@@ -94,9 +106,9 @@ spec:
               - containerPort: 8000
                 name: http
             readinessProbe:
-              tcpSocket:
+              # tcpSocket:
-                #httpGet:
+              httpGet:
-                #path: /health
+                path: /health
                 port: 8000
               initialDelaySeconds: 120
               periodSeconds: 20
@@ -190,4 +202,4 @@ spec:
         tolerations:
           {{- toYaml . | nindent 10 }}
         {{- end }} 
-{{- end }}
+{{- end }}

vllm-serve/templates/single.yaml

@@ -16,7 +16,7 @@ spec:
       initContainers:
      # 模型下载作为第一个 initContainer
       - name: download-model
-        image: {{ .Values.model.download.image }}
+        image: alpine:latest
         imagePullPolicy: IfNotPresent
         env:
           - name: HF_ENDPOINT
@@ -24,33 +24,46 @@ spec:
           - name: HUGGING_FACE_HUB_TOKEN
             value: {{ .Values.model.huggingfaceToken }}
         command:
-          - sh
+          - /bin/sh
           - -c
           - |
+            set -e
             MODEL_NAME=$(basename "{{ .Values.model.huggingfaceName }}")
             DEST_DIR="{{ .Values.model.localMountPath }}/Weight/$MODEL_NAME"
+            SUCCESS_FLAG="${DEST_DIR}/.success_flag"
             # DEST_DIR="{{ .Values.model.localMountPath }}/{{ .Values.model.huggingfaceName }}"
             # 检查模型是否存在，不存在则下载
-            echo "DEST_DIR= $DEST_DIR"
+            # echo "DEST_DIR= $DEST_DIR"
-            if [ ! -f "$DEST_DIR/config.json" ]; then
+            # if [ ! -f "$DEST_DIR/config.json" ]; then
-              ls -l {{ .Values.model.localMountPath }}
+            #   ls -l {{ .Values.model.localMountPath }}
-              echo "Downloading model {{ .Values.model.huggingfaceName }} to $DEST_DIR"
+            #   echo "Downloading model {{ .Values.model.huggingfaceName }} to $DEST_DIR"
-              wget https://hf-mirror.com/hfd/hfd.sh 
+            #   wget https://hf-mirror.com/hfd/hfd.sh 
-              chmod a+x hfd.sh
+            #   chmod a+x hfd.sh
-              apt update && apt upgrade
+            #   apt update && apt upgrade
-              apt install aria2 -y
+            #   apt install aria2 -y
-              ./hfd.sh {{ .Values.model.huggingfaceName }} --local-dir "$DEST_DIR"
+            #   ./hfd.sh {{ .Values.model.huggingfaceName }} --local-dir "$DEST_DIR"
-              # huggingface-cli download {{ .Values.model.huggingfaceName }} --local-dir "$DEST_DIR"
+            #   # huggingface-cli download {{ .Values.model.huggingfaceName }} --local-dir "$DEST_DIR"
-            else
+            # else
-              echo "Model already exists at $DEST_DIR"
+            #   echo "Model already exists at $DEST_DIR"
+            # fi
+            if [ -f "$SUCCESS_FLAG" ]; then
+              echo "✅ Success flag found. Skipping download."
+              exit 0
             fi
+            echo "⬇️ Starting download..."
+            apk add --no-cache bash aria2 wget ca-certificates
+            wget https://hf-mirror.com/hfd/hfd.sh -O hfd.sh && chmod +x hfd.sh
+            ./hfd.sh {{ .Values.model.huggingfaceName }} --tool aria2c -x 8 --local-dir "$DEST_DIR"
+
+            touch "$SUCCESS_FLAG"
+            echo "🎉 Done."
         volumeMounts:
         - name: weight-volume
           mountPath: {{ .Values.model.localMountPath }}
       containers:
       - name: vllm-pod
         image: {{ .Values.vllm.image }}
-        imagePullPolicy:  IfNotPresent
+        imagePullPolicy:  {{ .Values.imagePullPolicy }}
         env:
         - name: HUGGING_FACE_HUB_TOKEN
           value: {{ .Values.vllm.huggingfaceToken }}
@@ -81,6 +94,7 @@ spec:
             nvidia.com/gpu: "{{ .Values.resources.gpuLimit }}"
             memory: {{ .Values.resources.memoryLimit }}
             ephemeral-storage: 10Gi
+            cpu: {{ .Values.resources.cpuRequest }}
           requests:
             ephemeral-storage: 10Gi
             cpu: {{ .Values.resources.cpuRequest }}
@@ -124,4 +138,4 @@ spec:
       tolerations:
         {{- toYaml . | nindent 8 }}
       {{- end }}
-{{- end }}
+{{- end }}