release 1.8.5

README.cn.md

@@ -9,16 +9,12 @@
 
 ---
 
-**Vega ver1.8.4 发布**
+**Vega ver1.8.5 发布**
 
 - 错误修正
 
-  - 修正ASHA算法更新数据时失败的问题。
-  - 修正HCCL+Apex下，loss不更新的问题。
-  - 增加字典类指标。
-  - 更新安全配置文档。
-  - 移除安全模式下对Horovod和TensorFlow的支持。
-  - 增加安全模型下对Python3.9及以上版本的要求。
+  - 修正SPNAS算法集群训练失败时的问题。
+  - 修正了安全模式下模型拷贝失败等问题。
 
 ---
 

README.md

@@ -8,16 +8,13 @@
 
 ---
 
-**Vega ver1.8.4 released**
+**Vega ver1.8.5 released**
 
 - Bug Fixed:
 
-  - Fixed bug that ASHA failed to update data.
-  - Fixed bug that loss is not updated on HCCL+Apex.
-  - Add dictionary metrics.
-  - Update the security configuration document.
-  - Not Allowed Horovod and TensorFlow in safe mode.
-  - Python 3.9 or later is required in the security model.  
+  - Fixed a bug when the SPNAS algorithm cluster training fails.
+  - Fixed bugs such as model copy failure in safe mode.
+
 
 ---
 

RELEASE.md

@@ -1,4 +1,4 @@
-**Vega ver1.8.4 released:**
+**Vega ver1.8.5 released:**
 
 **Introduction**
 

docs/cn/user/ascend_910.md

@@ -137,10 +137,3 @@ pip3 install --user --no-deps noah-vega
 ```bash
 pip3 show noah-vega
 ```
-
-另外要注意的是，dask和distributed这两个包，需要安装如下版本：
-
-```bash
-pip3 install --user distributed==2021.7.0
-pip3 install --user dask==2021.7.0
-```

docs/cn/user/security_configure.md

@@ -17,6 +17,7 @@ Vega的安全配置，包括如下步骤：
 1. **Python3.9及以上**
 2. **dask和distributed版本为2022.2.0**
 
+
 ## 1.安装OpenSSL
 
 首先要安装OpenSSL 1.1.1，从源码编译安装，或者直接安装编译后的发行包。
@@ -87,7 +88,7 @@ openssl x509 -req -in server.csr -CA ca.crt -CAkey ca.key -CAcreateserial -out s
 rm server.csr
 ```
 
-执行如下脚本生成评估服务客户端所使用的证书的加密私钥，执行该命令时，会提示输入加密密码，密码的强度要求如服务器端私钥，且和服务器段私钥密码不同，请记录好改密码，后继还需使用：
+执行如下脚本生成评估服务客户端所使用的证书的加密私钥，执行该命令时，会提示输入加密密码，密码的强度要求如服务器端私钥，且和服务器端私钥密码不同，请记录好该密码，后继还需使用：
 
 ```shell
 openssl genrsa -aes-256-ofb -out client.key 4096
@@ -172,6 +173,17 @@ chmod 600 ~/.vega/*
 1. 如上的秘钥、证书、加密材料也可以放到其他目录位置，注意访问权限要设置为`600`，并在后继的配置文件中同步修改该文件的位置，需要使用绝对路径。
 2. 在训练集群上，需要保留`ca.crt`、`client.key`、`client.crt`、`ksmaster_client.dat`、`ksstandby_client.dat`、`server_dask.key`、`server_dask.crt`、`client_dask.key`、`client_dask.crt`，并删除其他文件。
 3. 评估服务上，需要保留`ca.crt`、`server.key`、`server.crt`、`ksmaster_server.dat`、`ksstandby_server.dat`，并删除其他文件。
+4. 以下为默认配置的加密套件:
+
+   ```txt
+   ECDHE-ECDSA-AES128-CCM:ECDHE-ECDSA-AES256-CCM:ECDHE-ECDSA-AES128-GCM-SHA256:ECDHE-ECDSA-AES256-GCM-SHA384:ECDHE-RSA-AES128-GCM-SHA256:ECDHE-RSA-AES256-GCM-SHA384:DHE-RSA-AES128-GCM-SHA256:DHE-RSA-AES256-GCM-SHA384:DHE-DSS-AES128-GCM-SHA256:DHE-DSS-AES256-GCM-SHA384:DHE-RSA-AES128-CCM:DHE-RSA-AES256-CCM
+   ```
+
+   如需缩小范围，可在`client.ini`与`vega.ini`中加入配置：
+
+   ```ini
+   ciphers=ECDHE-ECDSA-AES128-CCM:ECDHE-ECDSA-AES256-CCM
+   ```
 
 在`~/.vega`目录下创建`server.ini`和`client.ini`。
 
@@ -290,3 +302,18 @@ find ~/.local/ -name *.pem
 ### 9.5 Horovod 和 TensorFlow
 
 在安全模式下，Vega不支持Horovod数据并行，也不支持TensorFlow框架，Vega在运行前检查若是Horovod数据并行程序，或者TensorFlow框架，会自动退出。
+
+### 9.6 限定Distributed仅使用tls1.3协议进行通信
+
+若需要限定开源软件Distributed的组件间的通信仅使用tls1.3协议，需要配置`~/.config/dask/distributed.yaml`
+
+distributed.yaml：
+
+```yaml
+distributed:
+    comm:
+        tls:
+            min-version: 1.3
+```
+
+请参考Dask的[配置指导](https://docs.dask.org/en/stable/configuration.html)。

docs/en/user/ascend_910.md

@@ -144,10 +144,3 @@ Run the following command to view the Vega dependency package:
 ```bash
 pip3 show noah-vega
 ```
-
-Note that the following versions must be installed for the dask and distributed packages:
-
-```bash
-pip3 install --user distributed==2021.7.0
-pip3 install --user dask==2021.7.0
-```

docs/en/user/security_configure.md

@@ -17,6 +17,7 @@ requirements:
 1. **Python 3.9 or later.**
 2. **Dask and Distributed version is 2022.2.0.**
 
+
 ## 1. Install OpenSSL
 
 You need to install OpenSSL 1.1.1, compile and install from the source code, or directly install the compiled release package.
@@ -173,6 +174,17 @@ Description:
 1. The preceding keys, certificates, and encryption materials can also be stored in other directories. The access permission must be set to 600, and the file location must be changed to an absolute path in subsequent configuration files.
 2. In the train cluster, reserve `ca.crt`, `client.key`, `client.crt`, `ksmaster_client.dat`, `ksstandby_client.dat`, and `server_dask.key`, `server_dask.crt`, `client_dask.key`, `client_dask.crt`, and delete other files.
 3. In the evaluate service, reserve `ca.crt`, `server.key`, `server.crt`, `ksmaster_server.dat`, and `ksstandby_server.dat` files, and delete other files.
+4. The default cipher suites are as follows::
+
+   ```txt
+   ECDHE-ECDSA-AES128-CCM:ECDHE-ECDSA-AES256-CCM:ECDHE-ECDSA-AES128-GCM-SHA256:ECDHE-ECDSA-AES256-GCM-SHA384:ECDHE-RSA-AES128-GCM-SHA256:ECDHE-RSA-AES256-GCM-SHA384:DHE-RSA-AES128-GCM-SHA256:DHE-RSA-AES256-GCM-SHA384:DHE-DSS-AES128-GCM-SHA256:DHE-DSS-AES256-GCM-SHA384:DHE-RSA-AES128-CCM:DHE-RSA-AES256-CCM
+   ```
+
+   To narrow down the scope, add configurations to the `client.ini` and `vega.ini` files:
+
+   ```ini
+   ciphers=ECDHE-ECDSA-AES128-CCM:ECDHE-ECDSA-AES256-CCM
+   ```
 
 Create `server.ini` and `client.ini` in the `~/.vega` directory.
 
@@ -292,3 +304,18 @@ Find the private key file of the open-source software on which Vega depends amon
 ### 9.5 Horovod and TensorFlow
 
 In security mode, Vega does not support Horovod or the TensorFlow framework. Vega automatically exits if Vega run on Horovod or the TensorFlow framework.
+
+### 9.6 Only TLS 1.3 can be used for Distributed
+
+If only the tls1.3 protocol needs to be used for communication between distributed components，configure `~/.config/dask/distributed.yaml`
+
+distributed.yaml：
+
+```yaml
+distributed:
+    comm:
+        tls:
+            min-version: 1.3
+```
+
+For details, see the [Configuration Guide](https://docs.dask.org/en/stable/configuration.html)。

evaluate_service/RELEASE.md

@@ -1,4 +1,4 @@
-**Evaluate Service ver1.8.4 released:**
+**Evaluate Service ver1.8.5 released:**
 
 **Introduction**
 

evaluate_service/evaluate_service/__init__.py

@@ -16,4 +16,4 @@
 
 """Evaluate service."""
 
-__version__ = "1.8.4"
+__version__ = "1.8.5"

evaluate_service/evaluate_service/main.py

@@ -176,8 +176,10 @@ def upload_files(self):
         logging.warning("The timestamp is {}.".format(self.now_time))
         self.upload_file_path = os.path.join(self.current_path, "out", self.now_time)
         self.share_dir = os.path.join(self.current_path, "out", self.job_id)
-        os.makedirs(self.upload_file_path)
-        os.makedirs(self.share_dir)
+        if not os.path.exists(self.upload_file_path):
+            os.makedirs(self.upload_file_path)
+        if not os.path.exists(self.share_dir):
+            os.makedirs(self.share_dir)
         patterns = [".pkl", ".pth", ".pt", ".pb", ".ckpt", ".air", '.om',
                     ".onnx", ".caffemodel", ".pbtxt", ".prototxt"]
         model_file = request.files.get("model_file")

evaluate_service/evaluate_service/run_flask.py

@@ -18,6 +18,7 @@
 
 import configparser
 import logging
+import ssl
 import os
 from multiprocessing import Process
 import gevent
@@ -76,13 +77,28 @@ def run_flask(app, host, port, security_mode):
         encrypted_password = config.get('security').get('encrypted_password')
         key_component_1 = config.get('security').get('key_component_1')
         key_component_2 = config.get('security').get('key_component_2')
+        ciphers = config.get('security').get('ciphers')
+        cipher_suites = "ECDHE-ECDSA-AES128-CCM:ECDHE-ECDSA-AES256-CCM:ECDHE-ECDSA-AES128-GCM-SHA256" \
+                        ":ECDHE-ECDSA-AES256-GCM-SHA384:ECDHE-RSA-AES128-GCM-SHA256:ECDHE-RSA-AES256-GCM-SHA384" \
+                        ":DHE-RSA-AES128-GCM-SHA256:DHE-RSA-AES256-GCM-SHA384:DHE-DSS-AES128-GCM-SHA256" \
+                        ":DHE-DSS-AES256-GCM-SHA384:DHE-RSA-AES128-CCM:DHE-RSA-AES256-CCM"
+
+        if ciphers:
+            ciphersList = [cipher for cipher in ciphers.split(':') if cipher in cipher_suites.split(':')]
+            if ciphersList == []:
+                raise ssl.SSLError("The ciphers are invalid, please check.")
+            else:
+                ciphers = ':'.join(ciphersList)
+        else:
+            ciphers = cipher_suites
+
         if not check_risky_files((ca_cert, server_cert, server_secret_key, key_component_1, key_component_2)):
             return
         try:
             if encrypted_password == "":
-                ssl_context = create_context(ca_cert, server_cert, server_secret_key)
+                ssl_context = create_context(ca_cert, server_cert, server_secret_key, ciphers)
             else:
-                ssl_context = create_context(ca_cert, server_cert, server_secret_key,
+                ssl_context = create_context(ca_cert, server_cert, server_secret_key, ciphers,
                                              encrypted_password, key_component_1, key_component_2)
         except Exception:
             logging.error("Fail to create context.")

evaluate_service/setup.py

@@ -60,7 +60,7 @@ def run(self):
 
 setuptools.setup(
     name="evaluate-service",
-    version="1.8.4",
+    version="1.8.5",
     packages=["evaluate_service"],
     include_package_data=True,
     python_requires=">=3.6",
@@ -80,6 +80,7 @@ def run(self):
         "Flask-RESTful",
         "Flask-Limiter",
         "gevent",
+        "PyYAML",
     ],
     cmdclass={
         "build_py": custom_build_py,

examples/data_augmentation/cyclesr/cyclesr.yml

@@ -1,5 +1,6 @@
 general:
     backend: pytorch
+    requires: ["tensorboardX"]
 
 
 pipeline: [fully_train]

examples/nas/modnas/darts.yml

@@ -1,5 +1,6 @@
 general:
     backend: pytorch
+    requires: ["tensorboardX"]
 
 
 pipeline: [nas, fully_train]

examples/nas/modnas/mbv2.yml

@@ -1,5 +1,6 @@
 general:
     backend: pytorch
+    requires: ["tensorboardX"]
 
 
 pipeline: [fully_train]

examples/nas/modnas/ps.yml

@@ -1,5 +1,6 @@
 general:
     backend: pytorch
+    requires: ["tensorboardX"]
 
 
 pipeline: [nas, fully_train]

examples/nas/modnas/pxl.yml

@@ -1,5 +1,6 @@
 general:
     backend: pytorch
+    requires: ["tensorboardX"]
 
 
 pipeline: [nas, fully_train]

setup.py

@@ -29,7 +29,7 @@
 
 setuptools.setup(
     name="noah-vega",
-    version="1.8.4",
+    version="1.8.5",
     packages=["vega"],
     include_package_data=True,
     python_requires=">=3.6",
@@ -59,7 +59,6 @@
         "dill",
         "scikit-learn",
         "opencv-python",
-        "tensorboardX",
     ],
     entry_points="""
         [console_scripts]

vega/__init__.py

@@ -34,7 +34,7 @@
     "get_quota",
 ]
 
-__version__ = "1.8.4"
+__version__ = "1.8.5"
 
 
 import sys

vega/algorithms/nas/sp_nas/spnas_trainer_callback.py

@@ -41,6 +41,7 @@
 def valid():
     """Construct the trainer of SpNas."""
     config_val = DatasetConfig().to_dict()
+    dataset_type = config_val.type
     config_val = config_val['_class_data'].val
     prefix = "FasterRcnn_eval.mindrecord"
     mindrecord_dir = config_val.mindrecord_dir
@@ -49,7 +50,7 @@ def valid():
     if not os.path.exists(mindrecord_file):
         if not os.path.isdir(mindrecord_dir):
             os.makedirs(mindrecord_dir)
-        if config_val.dataset == "coco":
+        if dataset_type == "CocoDataset":
             if os.path.isdir(config_val.coco_root):
                 data_to_mindrecord_byte_image(config_val, "coco", False, prefix, file_num=1)
             else:
@@ -67,6 +68,7 @@ def valid():
 def train():
     """Train fasterrcnn dataset."""
     config_train = DatasetConfig().to_dict()
+    dataset_type = config_train.type
     config_train = config_train['_class_data'].train
     prefix = "FasterRcnn.mindrecord"
     mindrecord_dir = config_train.mindrecord_dir
@@ -78,7 +80,7 @@ def train():
     if rank == 0 and not os.path.exists(mindrecord_file):
         if not os.path.isdir(mindrecord_dir):
             os.makedirs(mindrecord_dir)
-        if config.dataset == "coco":
+        if dataset_type == "CocoDataset":
             if os.path.isdir(config_train.coco_root):
                 if not os.path.exists(config_train.coco_root):
                     logging.info("Please make sure config:coco_root is valid.")

vega/algorithms/nas/sp_nas/src/dataset.py

@@ -361,10 +361,10 @@ def create_coco_label(is_training, config):
     from pycocotools.coco import COCO
 
     coco_root = config.coco_root
-    data_type = config.val_data_type
     if is_training:
         data_type = config.train_data_type
-
+    else:
+        data_type = config.val_data_type
     # Classes need to train or test.
     train_cls = config.coco_classes
     train_cls_dict = {}

vega/algorithms/nas/sp_nas/src/model_utils/config.py

@@ -66,7 +66,7 @@ def parse_cli_to_yaml(parser, cfg, helper=None, choices=None, cfg_path="default_
             else:
                 parser.add_argument("--" + item, type=type(cfg[item]), default=cfg[item], choices=choice,
                                     help=help_description)
-    args = parser.parse_args()
+    args, _ = parser.parse_known_args()
     return args
 
 

vega/common/backend_register.py

@@ -17,7 +17,8 @@
 """Backend Register."""
 
 import os
-import sys
+import logging
+import traceback
 
 __all__ = [
     "set_backend",
@@ -135,7 +136,8 @@ def get_devices():
 
 def import_extension_module():
     """Import extension module."""
-    try:
-        import ascend_automl
-    except ImportError:
-        pass
+    if is_npu_device():
+        try:
+            import ascend_automl
+        except ImportError:
+            logging.debug(traceback.format_exc())