365 lines
13 KiB
Bash
365 lines
13 KiB
Bash
#!/bin/bash
|
||
|
||
# Color definitions
|
||
RED='\033[0;31m'
|
||
GREEN='\033[0;32m'
|
||
CYAN='\033[36m'
|
||
RESET='\033[0m'
|
||
|
||
:<<'COMMENT'
|
||
检测介质下载路径中是否包含以下介质：模型权重文件和conda环境tar包，如果不存在则退出。
|
||
检测服务器节点目标conda环境是否存在，如果不存在则创建。
|
||
检测模型服务是否已启动，如果模型服务已启动，则终止模型服务。
|
||
启动模型服务
|
||
检查模型服务是否启动成功
|
||
执行curl命令验证模型服务
|
||
将黄块注册脚本和黄块curl命令脚本scp到黄块服务ip
|
||
ssh登录黄块服务ip进行黄块注册，执行黄块curl命令
|
||
COMMENT
|
||
|
||
# 定义帮助信息
|
||
function print_help() {
|
||
echo "Usage: $0"
|
||
echo
|
||
echo "This script reads configuration from config.yaml and performs the necessary setup."
|
||
echo
|
||
echo "Example:"
|
||
echo " bash $0"
|
||
}
|
||
|
||
# 加载 YAML 配置
|
||
function load_config() {
|
||
local config_file="$1"
|
||
MODEL=$(yq eval '.model' "$config_file")
|
||
MODEL_VERSION=$(yq eval '.model_version' "$config_file")
|
||
CONDA_ENV=$(yq eval '.conda_env' "$config_file")
|
||
TENSOR_PARALLEL_SIZE=$(yq eval '.tensor_parallel_size' "$config_file")
|
||
VISIBLE_GPU_INDEX=$(yq eval '.visible_gpu_index' "$config_file")
|
||
MODEL_PORT=$(yq eval '.model_port' "$config_file")
|
||
GPU_MEMORY_UTILIZATION=$(yq eval '.gpu_memory_utilization' "$config_file")
|
||
DTYPE=$(yq eval '.dtype' "$config_file")
|
||
MODEL_SERVER_IP=$(yq eval '.model_server_ip' "$config_file")
|
||
YELLOW_BLOCK_SERVER_IP=$(yq eval '.yellow_block_server_ip' "$config_file")
|
||
YELLOW_BLOCK_SERVER_PORT=$(yq eval '.yellow_block_server_port' "$config_file")
|
||
YELLOW_BLOCK_CONDA_ENV=$(yq eval '.yellow_block_conda_env' "$config_file")
|
||
|
||
MEDIA_DOWNLOAD_PATH=$(yq eval '.media_download_path' "$config_file")
|
||
MODEL_PATH=$(yq eval '.model_path' "$config_file")
|
||
MODEL_NAME=$(yq eval '.model_name' "$config_file" | sed "s|{{model}}|$MODEL|g" | sed "s|{{model_version}}|$MODEL_VERSION|g")
|
||
MODEL_NAME_PATH=$(yq eval '.model_name_path' "$config_file" | sed "s|{{model_path}}|$MODEL_PATH|g" | sed "s|{{model_name}}|$MODEL_NAME|g")
|
||
MODEL_TAR=$(yq eval '.model_tar' "$config_file" | sed "s|{{model_name}}|$MODEL_NAME|g")
|
||
MINICONDA_PATH=$(yq eval '.miniconda_path' "$config_file")
|
||
YELLOW_BLOCK_PATH=$(yq eval '.yellow_block_path' "$config_file")
|
||
AUTOMATED_DEPLOYMENT_PATH=$(yq eval '.automated_deployment_path' "$config_file")
|
||
}
|
||
|
||
# 检查配置文件是否存在
|
||
cd /data/jenkins_script/automated_deployment
|
||
if [ ! -f "config.yaml" ]; then
|
||
echo "Error: config.yaml not found in the current directory."
|
||
exit 1
|
||
fi
|
||
|
||
# 加载配置
|
||
load_config "config.yaml"
|
||
|
||
# 打印参数值（调试用）
|
||
echo "MODEL: $MODEL"
|
||
echo "MODEL_VERSION: $MODEL_VERSION"
|
||
echo "MODEL_NAME: $MODEL_NAME"
|
||
echo "MODEL_NAME_PATH: $MODEL_NAME_PATH"
|
||
echo "MODEL_TAR: $MODEL_TAR"
|
||
echo "CONDA_ENV: $CONDA_ENV"
|
||
echo "TENSOR_PARALLEL_SIZE: $TENSOR_PARALLEL_SIZE"
|
||
echo "VISIBLE_GPU_INDEX: $VISIBLE_GPU_INDEX"
|
||
echo "MODEL_PORT: $MODEL_PORT"
|
||
echo "GPU_MEMORY_UTILIZATION: $GPU_MEMORY_UTILIZATION"
|
||
echo "DTYPE: $DTYPE"
|
||
echo "MODEL_SERVER_IP: $MODEL_SERVER_IP"
|
||
echo "YELLOW_BLOCK_SERVER_IP: $YELLOW_BLOCK_SERVER_IP"
|
||
echo "YELLOW_BLOCK_SERVER_PORT: $YELLOW_BLOCK_SERVER_PORT"
|
||
echo "YELLOW_BLOCK_CONDA_ENV: $YELLOW_BLOCK_CONDA_ENV"
|
||
echo "MEDIA_DOWNLOAD_PATH: $MEDIA_DOWNLOAD_PATH"
|
||
echo "MODEL_PATH: $MODEL_PATH"
|
||
echo "MINICONDA_PATH: $MINICONDA_PATH"
|
||
echo "YELLOW_BLOCK_PATH: $YELLOW_BLOCK_PATH"
|
||
echo "AUTOMATED_DEPLOYMENT_PATH: $AUTOMATED_DEPLOYMENT_PATH"
|
||
|
||
check_and_enter_script() {
|
||
local script_name="$1"
|
||
local path1="$MINICONDA_PATH/$CONDA_ENV/script/$MODEL"
|
||
local path2="$AUTOMATED_DEPLOYMENT_PATH"
|
||
|
||
# 检查第一个路径是否存在脚本
|
||
if [ -f "$path1/$script_name" ]; then
|
||
cd "$path1" || { echo "无法进入目录 $path1"; return 1; }
|
||
return 0
|
||
fi
|
||
|
||
# 检查第二个路径是否存在脚本
|
||
if [ -f "$path2/$script_name" ]; then
|
||
cd "$path2" || { echo "无法进入目录 $path2"; return 1; }
|
||
return 0
|
||
fi
|
||
|
||
# 如果两个路径都不存在脚本
|
||
return 1
|
||
}
|
||
|
||
# 检查介质下载路径中是否包含以下介质：模型权重文件和conda环境tar包。
|
||
check_media_list() {
|
||
echo -e "${CYAN}Checking media files...${RESET}"
|
||
cd "$MEDIA_DOWNLOAD_PATH"
|
||
if [ -f "$MODEL_TAR" ]; then
|
||
echo -e "${GREEN}The media file $MEDIA_DOWNLOAD_PATH/$MODEL_TAR exists.${RESET}"
|
||
else
|
||
echo -e "${RED}The media file $MEDIA_DOWNLOAD_PATH/$MODEL_TAR does not exist.${RESET}"
|
||
fi
|
||
if [ -f "$CONDA_ENV.tar.gz" ]; then
|
||
echo -e "${GREEN}The media file $MEDIA_DOWNLOAD_PATH/$CONDA_ENV.tar.gz exists.${RESET}"
|
||
else
|
||
echo -e "${RED}The media file $MEDIA_DOWNLOAD_PATH/$CONDA_ENV.tar.gz does not exist.${RESET}"
|
||
fi
|
||
}
|
||
|
||
# 检查服务器节点中模型权重文件是否存在
|
||
check_model_path() {
|
||
cd "$MODEL_PATH"
|
||
if [ ! -d "$MODEL_NAME" ]; then
|
||
echo -e "${RED}The model $MODEL_NAME does not exist.${RESET}"
|
||
return 1
|
||
else
|
||
echo -e "${GREEN}The model $MODEL_NAME exists.${RESET}"
|
||
return 0
|
||
fi
|
||
}
|
||
|
||
# 部署模型
|
||
deploy_model() {
|
||
echo -e "${CYAN}Start deploying model.${RESET}"
|
||
cd "$MODEL_PATH"
|
||
cp "$MEDIA_DOWNLOAD_PATH/$MODEL_TAR" ./
|
||
tar -zxf "$MODEL_TAR"
|
||
if [ $? -eq 0 ]; then
|
||
echo -e "${GREEN}Model deployed successfully.${RESET}"
|
||
else
|
||
echo -e "${RED}Model deployment failed.${RESET}"
|
||
exit 1
|
||
fi
|
||
ll "$MODEL_NAME"
|
||
chown -R inspur:inspur "$MODEL_NAME"
|
||
rm -rf "$MODEL_TAR"
|
||
}
|
||
|
||
# 检查服务器节点中conda环境是否存在
|
||
check_conda_env() {
|
||
cd "$MINICONDA_PATH"
|
||
if [ ! -d "$CONDA_ENV" ]; then
|
||
echo -e "${RED}The conda environment $CONDA_ENV does not exist.${RESET}"
|
||
return 1
|
||
else
|
||
echo -e "${GREEN}The conda environment $CONDA_ENV exists.${RESET}"
|
||
return 0
|
||
fi
|
||
}
|
||
|
||
# 将介质下载路径中的conda环境压缩包解压至/data/miniconda3/envs/$CONDA_ENV目录下，修改用户和用户组。
|
||
deploy_conda_env() {
|
||
cd "$MINICONDA_PATH"
|
||
mkdir "$CONDA_ENV"
|
||
cd "$MEDIA_DOWNLOAD_PATH"
|
||
tar -zxf "$CONDA_ENV.tar.gz" -C "$MINICONDA_PATH/$CONDA_ENV"
|
||
if [ $? -eq 0 ]; then
|
||
echo -e "${GREEN}The environment $CONDA_ENV has been created.${RESET}"
|
||
else
|
||
echo -e "${RED}Failed to create environment $CONDA_ENV.${RESET}"
|
||
exit 1
|
||
fi
|
||
chown -R inspur:inspur "$MINICONDA_PATH/$CONDA_ENV"
|
||
}
|
||
|
||
# 调用conda环境中的check.sh脚本，检查模型服务进程
|
||
check_model_service_process() {
|
||
check_and_enter_script "check.sh"
|
||
check_result=$(bash check.sh ${MODEL})
|
||
# 检查输出内容
|
||
if [[ "$check_result" == *"SUCC"* ]]; then
|
||
return 0
|
||
elif [[ "$check_result" == *"FAIL"* ]]; then
|
||
return 1
|
||
else
|
||
exit 1
|
||
fi
|
||
}
|
||
|
||
# 调用业务代码中的stop.sh脚本，停止模型服务
|
||
stop_model_service() {
|
||
check_and_enter_script "stop.sh"
|
||
bash stop.sh ${MODEL} > /dev/null 2>&1
|
||
if [ $? -eq 0 ]; then
|
||
echo -e "${GREEN}$MODEL_NAME service stopped successfully.${RESET}"
|
||
else
|
||
echo -e "${RED}Failed to stop $MODEL_NAME service.${RESET}"
|
||
fi
|
||
}
|
||
|
||
# 激活conda环境，调用业务代码中的start.sh脚本，启动模型服务
|
||
start_model_service(){
|
||
check_and_enter_script "start.sh"
|
||
bash "start.sh" ${VISIBLE_GPU_INDEX} ${CONDA_ENV} ${MODEL} ${MODEL_NAME_PATH} ${TENSOR_PARALLEL_SIZE} ${MODEL_SERVER_IP} ${MODEL_PORT} ${GPU_MEMORY_UTILIZATION} ${DTYPE}
|
||
}
|
||
|
||
# 轮询检测模型服务是否已启动成功
|
||
check_model_started() {
|
||
local start_time=$(date +%s)
|
||
local timeout=300
|
||
local interval=5
|
||
while true; do
|
||
sleep $interval
|
||
if check_model_service_process; then
|
||
echo -e "${GREEN}$MODEL_NAME service started successfully.${RESET}"
|
||
break
|
||
fi
|
||
local current_time=$(date +%s)
|
||
local elapsed_time=$(($current_time - $start_time))
|
||
if [ $elapsed_time -ge $timeout ]; then
|
||
echo -e "${RED}Failed to start $MODEL_NAME service.${RESET}"
|
||
echo -e "${RED}Please check the log under $MINICONDA_PATH/$CONDA_ENV/script/$MODEL${RESET}"
|
||
exit 1
|
||
fi
|
||
done
|
||
}
|
||
|
||
update_model_curl_sh() {
|
||
local script_file="model_curl.sh"
|
||
local model_server_ip=$1
|
||
local model_port=$2
|
||
local model=$3
|
||
|
||
check_and_enter_script $script_file
|
||
# 使用 sed 命令替换变量值
|
||
sed -i "s|^export MODEL_SERVER_IP=.*|export MODEL_SERVER_IP=$model_server_ip|" "$script_file"
|
||
sed -i "s|^export MODEL_PORT=.*|export MODEL_PORT=$model_port|" "$script_file"
|
||
sed -i "s|^export MODEL=.*|export MODEL=$model|" "$script_file"
|
||
echo -e "${GREEN}Updated $script_file with the provided values.${RESET}"
|
||
}
|
||
|
||
# 验证模型服务
|
||
model_curl_verification(){
|
||
echo -e "${GREEN}Start to execute the curl command to verify the model${RESET}"
|
||
update_model_curl_sh $MODEL_SERVER_IP $MODEL_PORT $MODEL
|
||
# check_and_enter_script "model_curl.sh"
|
||
cat model_curl.sh
|
||
echo -e "${GREEN}The curl command returns the following results:${RESET}"
|
||
bash model_curl.sh
|
||
}
|
||
|
||
yellow_block_registration_curl(){
|
||
echo -e "${CYAN}Start to execute yellow block registration and curl command${RESET}"
|
||
if [ ! -d "$YELLOW_BLOCK_PATH" ]; then
|
||
echo -e "${RED}Directory $YELLOW_BLOCK_PATH does not exist. Please deploy the start yellow block first.${RESET}"
|
||
else
|
||
echo -e "${GREEN}Directory $YELLOW_BLOCK_PATH exists.${RESET}"
|
||
cd "$YELLOW_BLOCK_PATH"
|
||
echo -e "${GREEN}Start to execute the yellow block registration${RESET}"
|
||
bash yellow_block_register.sh
|
||
echo -e "${GREEN}Start to execute the curl command to verify the yellow block${RESET}"
|
||
cat yellow_block_curl.sh
|
||
echo -e "${GREEN}The curl command returns the following results:${RESET}"
|
||
bash yellow_block_curl.sh
|
||
fi
|
||
}
|
||
|
||
update_yellow_block_register_sh() {
|
||
local script_file="yellow_block_register.sh"
|
||
local yellow_block_conda_env=$1
|
||
local model_server_ip=$2
|
||
local model_port=$3
|
||
local model=$4
|
||
|
||
check_and_enter_script $script_file
|
||
# 使用 sed 命令替换变量值
|
||
sed -i "s|^export YELLOW_BLOCK_CONDA_ENV=.*|export YELLOW_BLOCK_CONDA_ENV=$yellow_block_conda_env|" "$script_file"
|
||
sed -i "s|^export MODEL_SERVER_IP=.*|export MODEL_SERVER_IP=$model_server_ip|" "$script_file"
|
||
sed -i "s|^export MODEL_PORT=.*|export MODEL_PORT=$model_port|" "$script_file"
|
||
sed -i "s|^export MODEL=.*|export MODEL=$model|" "$script_file"
|
||
# echo -e "${GREEN}Updated $script_file with the provided values.${RESET}"
|
||
}
|
||
|
||
update_yellow_block_curl_sh() {
|
||
local script_file="yellow_block_curl.sh"
|
||
local model=$1
|
||
|
||
check_and_enter_script $script_file
|
||
# 使用 sed 命令替换变量值
|
||
sed -i "s|^export MODEL=.*|export MODEL=$model|" "$script_file"
|
||
# echo -e "${GREEN}Updated $script_file with the provided values.${RESET}"
|
||
}
|
||
|
||
cp_registration_curl(){
|
||
update_yellow_block_register_sh $YELLOW_BLOCK_CONDA_ENV $MODEL_SERVER_IP $MODEL_PORT $MODEL
|
||
update_yellow_block_curl_sh $MODEL
|
||
# Check if the Model IP and Yellow Block IP are the same
|
||
if [ "$MODEL_SERVER_IP" == "$YELLOW_BLOCK_SERVER_IP" ]; then
|
||
echo -e "\n${GREEN}Model IP and Yellow Block IP are the same. Using cp to copy files.${RESET}"
|
||
cp -f yellow_block_* "$YELLOW_BLOCK_PATH"
|
||
yellow_block_registration_curl
|
||
else
|
||
echo -e "\n${GREEN}Model IP and Yellow Block IP are different. Using scp to copy files.${RESET}"
|
||
scp -P "$YELLOW_BLOCK_SERVER_PORT" yellow_block_* "root@$YELLOW_BLOCK_SERVER_IP":"$YELLOW_BLOCK_PATH"
|
||
ssh -p "$YELLOW_BLOCK_SERVER_PORT" "root@$YELLOW_BLOCK_SERVER_IP" << 'EOF'
|
||
# Color definitions
|
||
RED='\033[0;31m'
|
||
GREEN='\033[0;32m'
|
||
CYAN='\033[36m'
|
||
RESET='\033[0m'
|
||
|
||
YELLOW_BLOCK_PATH="/data/app/dev/ihp-model-ops/test/model_service_v2"
|
||
|
||
echo -e "${CYAN}Start to execute yellow block registration and curl command${RESET}"
|
||
if [ ! -d "$YELLOW_BLOCK_PATH" ]; then
|
||
echo -e "${RED}Directory $YELLOW_BLOCK_PATH does not exist. Please deploy the start yellow block first.${RESET}"
|
||
else
|
||
echo -e "${GREEN}Directory $YELLOW_BLOCK_PATH exists.${RESET}"
|
||
cd "$YELLOW_BLOCK_PATH"
|
||
echo -e "${GREEN}Start to execute the yellow block registration${RESET}"
|
||
bash yellow_block_register.sh
|
||
echo -e "${GREEN}Start to execute the curl command to verify the yellow block${RESET}"
|
||
cat yellow_block_curl.sh
|
||
echo -e "${GREEN}The curl command returns the following results:${RESET}"
|
||
bash yellow_block_curl.sh
|
||
fi
|
||
EOF
|
||
fi
|
||
}
|
||
|
||
# Main function
|
||
main(){
|
||
echo -e "${CYAN}======================$MODEL_NAME======================${RESET}"
|
||
check_media_list
|
||
if ! check_model_path; then
|
||
deploy_model
|
||
fi
|
||
if ! check_conda_env; then
|
||
deploy_conda_env
|
||
fi
|
||
if check_model_service_process; then
|
||
echo -e "${GREEN}The model service process already exists, starting to stop the process.${RESET}"
|
||
stop_model_service
|
||
else
|
||
echo -e "${GREEN}The model service process does not exist. Start to start the model service.${RESET}"
|
||
fi
|
||
sleep 5
|
||
start_model_service
|
||
if [ $? -eq 0 ]; then
|
||
echo -e "${CYAN}The model service is starting...${RESET}"
|
||
fi
|
||
check_model_started
|
||
model_curl_verification
|
||
cp_registration_curl
|
||
echo -e "\n${CYAN}======================$MODEL_NAME======================${RESET}"
|
||
}
|
||
|
||
# Execute the main function
|
||
main
|