This code achieves an interface, receiving the POST and GET then scheduling several docker containers where adversarial samples are generated.
Now it implemented to Task 16, docker link and security-enhance modes have been implemented.
- setting with ip and port in gunicorn.conf.py, as follows:
workers = 5
worker_class = "gevent"
bind = "127.0.0.1:5901" // ip and portCaution
为方便日志保存，run.sh 中的运行逻辑有所修改
- 运行
./run.sh来启动调度中心。此时调度中心进程将运行于后台，./run.sh将立即返回。原先的控制台输出将保存在log文件夹下的access_log和error_log中。
./run.shlog
├── access_log
└── error_log
- 利用
stop.sh停止调度中心进程。run.sh会将调度中心进程的pid写入文件中，stop.sh读取该pid并停止调度中心进程。
./stop.sh//in the url blank:
-
for testing mode1 获取被测对象的数据源:
http://127.0.0.1:5901/test_model
curl -v -X GET "http://127.0.0.1:5901/test_model" --noproxy 127.0.0.1 -
for testing mode2 获取内置依赖库及其版本的数据源
http://127.0.0.1:5901/depn_lib
curl -v -X GET "http://127.0.0.1:5901/depn_lib" --noproxy 127.0.0.1 -
for testing mode3 获取被测对象的模型权重文件数量
http://127.0.0.1:5901/weight_number?test_model=Vgg16
curl -v -X GET "http://127.0.0.1:5901/weight_number?test_model=Vgg16" --noproxy 127.0.0.1 -
for testing mode4 被测对象的模型权重文件zip包下载
http://127.0.0.1:5901/weight_download?test_model=Vgg16
curl -v -X GET "http://127.0.0.1:5901/weight_download?test_model=Vgg16" -o "Vgg16_weights.zip" --noproxy 127.0.0.1
-
for testing mode5 获取被测对象的模型权重文件列表、对抗方法列表的数据源
http://127.0.0.1:5901/check_model?test_model=Vgg16
curl -v -X GET "http://127.0.0.1:5901/check_model?test_model=Vgg16" --noproxy 127.0.0.1 -
for testing mode6 启动对抗样本生成
- leveraging curl for POST:
curl -X POST http://127.0.0.1:5901/adver_gen \
-H "Content-Type: application/x-www-form-urlencoded" \
-d "mission_id=123&test_model=Vgg16&test_weight=weightA&test_seed=seed1&test_method=FGSM&timeout=3600" \
--noproxy 127.0.0.1 ***if you has set proxy, this option should be added***-
for testing mode7 对抗样本生成过程中数据轮询
-
for testing mode8 停止对抗样本生成
curl -X POST http://127.0.0.1:5901/adver_gen_stop \
-H "Content-Type: application/x-www-form-urlencoded" \
-d "mission_id=123" \
--noproxy 127.0.0.1-
for testing mode9 生成的对抗样本zip包下载
-
for testing mode10 获取不同被测对象下的评估配置指标
http://127.0.0.1:5901/adver_metrics?test_model=Vgg16
- for testing mode11 启动测试任务评估
curl -X POST http://127.0.0.1:5901/adver_eval \
-H "Content-Type: application/x-www-form-urlencoded" \
-d "mission_id=123&eval_metric=abc" \
--noproxy 127.0.0.1-
for testing mode12 评估过程中数据轮询
-
for testing mode13 启动安全加固任务
curl -X POST http://127.0.0.1:5901/sec_enhance \
-H "Content-Type: application/x-www-form-urlencoded" \
-d "enhance_id=777&test_model=Vgg16&mission_id=12" \
--noproxy 127.0.0.1-
for testing mode14 安全加固过程数据轮询
现版本代码需要调试运行的话请将 interface_main.py中model6-model8中的诸如：
model_dict = init_read_yaml_for_model_duplicate()
docker_shell_run = model_dict[test_model].get('docker_container_run_shell')
container_id, script_path = docker_shell_run.split(":", 1)
shell_command = f"{script_path} {test_model} {test_weight} {test_seed} {test_method}"
shell_path = f"{container_id}:{shell_command}"
exec_docker_container_shell(shell_path)
这部分代码注释掉后（后端还未对接docker引擎）即可正常的调试运行。
若当前后端存在docker引擎的运行环境，即可使用对应的curl命令进行调试，注意参数--noproxy 127.0.0.1最好加上，不然可能会报错
调试运行之前注意yaml文件中各个路径等参数的配置，特别是诸如docker_id之类的，随环境变化而变化的部分，如：
weight_download_addr: 5334338a5dfe:/usr/weight.txt注意这里的5334338a5dfe就要改成自己docker环境的docker_id