Author: ACatSmiling
Since: 2024-11-23
内网机加载 Docker 镜像，并使用 Flask 封装接口。
-
此步骤针对内网机，首先，上传打包好的基础镜像文件到服务器，然后加载。
[root@zeloud ~]# docker load -i similar.tar 9853575bc4f9: Loading layer [==================================================>] 77.83MB/77.83MB 1505ca8b8119: Loading layer [==================================================>] 9.539MB/9.539MB 3548b2e74780: Loading layer [==================================================>] 32.89MB/32.89MB 2b728766068e: Loading layer [==================================================>] 4.608kB/4.608kB 63d299102447: Loading layer [==================================================>] 12.08MB/12.08MB 9967d022da90: Loading layer [==================================================>] 2.048kB/2.048kB e8f0676a25c7: Loading layer [==================================================>] 2.56kB/2.56kB d495473888f3: Loading layer [==================================================>] 6.656kB/6.656kB 9ebdfa55c24e: Loading layer [==================================================>] 241.8MB/241.8MB 24ac2bd8b6a9: Loading layer [==================================================>] 15.67MB/15.67MB 77faf2b53f78: Loading layer [==================================================>] 1.025GB/1.025GB b8e9b30bceef: Loading layer [==================================================>] 820.6MB/820.6MB 5a00afc95457: Loading layer [==================================================>] 829.3MB/829.3MB Loaded image: 10.3.170.88:5000/user/admin/similar:latest -
根据实际情况，确定是否需要重新打标签。
[root@zeloud ~]# docker images | grep /user/admin 10.3.164.57:5000/user/admin/similar flask 3d4e06e095fa 3 days ago 3.03GB 10.3.170.88:5000/user/admin/onnxruntime wenet3 03298f22daff 5 days ago 4.83GB registry.zeloud.com:5000/user/admin/onnxruntime wenet3 03298f22daff 5 days ago 4.83GB 10.3.170.88:5000/user/admin/esr_train_img v1.1.0 18d583b4bbaf 6 weeks ago 24.2GB 10.3.170.88:5000/user/admin/segan_2.2_docker v1.1 b2118a02a24d 6 weeks ago 31.7GB 10.3.170.88:5000/user/admin/segan_2.2_docker v1.0 040f7f705966 6 weeks ago 31.7GB 10.3.170.88:5000/user/admin/esr_train_img v1.0.0 0212bfcc9ba6 6 months ago 24.2GB 10.3.170.88:5000/user/admin/pytorch-py38-cuda11 base-centos7 c806bde76f8c 7 months ago 5.95GB 10.3.170.88:5000/user/admin/fairseq-tools-202103 v1.0 ed750c76036b 13 months ago 21.3GB 10.3.170.88:5000/user/admin/obj_detect_v1 latest d64244ed90a3 14 months ago 5.64GB 10.3.170.88:5000/user/admin/segan_2.1_docker latest d9565fdef5ad 17 months ago 31.7GB [root@zeloud ~]# docker tag 10.3.164.57:5000/user/admin/similar:flask 10.3.170.88:5000/user/admin/image-similar:dinov2 [root@zeloud ~]# docker images | grep /user/admin 10.3.164.57:5000/user/admin/image-similar dinov2 3d4e06e095fa 3 days ago 3.03GB 10.3.164.57:5000/user/admin/similar flask 3d4e06e095fa 3 days ago 3.03GB 10.3.170.88:5000/user/admin/onnxruntime wenet3 03298f22daff 5 days ago 4.83GB registry.zeloud.com:5000/user/admin/onnxruntime wenet3 03298f22daff 5 days ago 4.83GB 10.3.170.88:5000/user/admin/esr_train_img v1.1.0 18d583b4bbaf 6 weeks ago 24.2GB 10.3.170.88:5000/user/admin/segan_2.2_docker v1.1 b2118a02a24d 6 weeks ago 31.7GB 10.3.170.88:5000/user/admin/segan_2.2_docker v1.0 040f7f705966 6 weeks ago 31.7GB 10.3.170.88:5000/user/admin/esr_train_img v1.0.0 0212bfcc9ba6 6 months ago 24.2GB 10.3.170.88:5000/user/admin/pytorch-py38-cuda11 base-centos7 c806bde76f8c 7 months ago 5.95GB 10.3.170.88:5000/user/admin/fairseq-tools-202103 v1.0 ed750c76036b 13 months ago 21.3GB 10.3.170.88:5000/user/admin/obj_detect_v1 latest d64244ed90a3 14 months ago 5.64GB 10.3.170.88:5000/user/admin/segan_2.1_docker latest d9565fdef5ad 17 months ago 31.7GB [root@zeloud ~]# docker push 10.3.170.88:5000/user/admin/image-similar:dinov2 The push refers to repository [10.3.170.88:5000/user/admin/image-similar] Get "https://10.3.170.88:5000/v2/": http: server gave HTTP response to HTTPS client -
查看 Docker 基础镜像，是否已安装 Flask，如果没有，则 pip 安装。
[root@zeloud ~]# docker images | grep similar registry.zeloud.com:5000/user/admin/similar latest e7edddd7af1c 2 days ago 3.02GB [root@zeloud ~]# docker run -it --rm --entrypoint /bin/bash e7edddd7af1c root@bb3bfdda299e:/app# pip --version pip 24.2 from /usr/local/lib/python3.10/site-packages/pip (python 3.10) root@bb3bfdda299e:/app# python --version Python 3.10.14 root@bb3bfdda299e:/app# pip list | grep Flask root@bb3bfdda299e:/app# pip install flask -i https://depend.zeloud.com/artifactory/api/pypi/pypi-repo/simple Looking in indexes: https://depend.zeloud.com/artifactory/api/pypi/pypi-repo/simple Collecting flask Downloading https://depend.zeloud.com/artifactory/api/pypi/pypi-repo/packages/packages/af/47/93213ee66ef8fae3b93b3e29206f6b251e65c97bd91d8e1c5596ef15af0a/flask-3.1.0-py3-none-any.whl (102 kB) Collecting Werkzeug>=3.1 (from flask) Downloading https://depend.zeloud.com/artifactory/api/pypi/pypi-repo/packages/packages/52/24/ab44c871b0f07f491e5d2ad12c9bd7358e527510618cb1b803a88e986db1/werkzeug-3.1.3-py3-none-any.whl (224 kB) Requirement already satisfied: Jinja2>=3.1.2 in /usr/local/lib/python3.10/site-packages (from flask) (3.1.3) Collecting itsdangerous>=2.2 (from flask) Downloading https://depend.zeloud.com/artifactory/api/pypi/pypi-repo/packages/packages/04/96/92447566d16df59b2a776c0fb82dbc4d9e07cd95062562af01e408583fc4/itsdangerous-2.2.0-py3-none-any.whl (16 kB) Requirement already satisfied: click>=8.1.3 in /usr/local/lib/python3.10/site-packages (from flask) (8.1.7) Collecting blinker>=1.9 (from flask) Downloading https://depend.zeloud.com/artifactory/api/pypi/pypi-repo/packages/packages/10/cb/f2ad4230dc2eb1a74edf38f1a38b9b52277f75bef262d8908e60d957e13c/blinker-1.9.0-py3-none-any.whl (8.5 kB) Requirement already satisfied: MarkupSafe>=2.0 in /usr/local/lib/python3.10/site-packages (from Jinja2>=3.1.2->flask) (2.1.5) Installing collected packages: Werkzeug, itsdangerous, blinker, flask Successfully installed Werkzeug-3.1.3 blinker-1.9.0 flask-3.1.0 itsdangerous-2.2.0 WARNING: Running pip as the 'root' user can result in broken permissions and conflicting behaviour with the system package manager, possibly rendering your system unusable.It is recommended to use a virtual environment instead: https://pip.pypa.io/warnings/venv. Use the --root-user-action option if you know what you are doing and want to suppress this warning. [notice] A new release of pip is available: 24.2 -> 24.3.1 [notice] To update, run: pip install --upgrade pip root@bb3bfdda299e:/app# pip list | grep Flask Flask 3.1.0安装 Flask 如果碰到依赖冲突，可以强制安装：
root@bb3bfdda299e:/# pip install blinker==1.9 --ignore-installed -i https://depend.zeloud.com/artifactory/api/pypi/pypi-repo/simple -
新打开一个终端，查看上一步生成的容器，commit 为新镜像，并 push 到仓库，然后，退出上一步生成的容器。
[root@zeloud ~]# docker ps | grep e7edddd7af1c bb3bfdda299e e7edddd7af1c "bash" 8 minutes ago Up 8 minutes dazzling_clarke [root@zeloud ~]# docker commit bb3bfdda299e 10.3.170.88:5000/user/admin/similar:flask sha256:3d4e06e095fa5e222ac2e3798ac0642221356fc1191a759f34bc0d451ad9985c [root@zeloud ~]# docker push 10.3.170.88:5000/user/admin/similar:flask The push refers to repository [10.3.170.88:5000/user/admin/similar] 96894b0a9bbe: Pushed 5a00afc95457: Pushed b8e9b30bceef: Pushed 77faf2b53f78: Pushed 24ac2bd8b6a9: Pushed 9ebdfa55c24e: Pushed d495473888f3: Pushed e8f0676a25c7: Pushed 9967d022da90: Pushed 63d299102447: Pushed 2b728766068e: Pushed 3548b2e74780: Pushed 1505ca8b8119: Pushed 9853575bc4f9: Pushed flask: digest: sha256:8822d2a1476f0b58852052074e9165de27db8a90e93d3dd0b2ab5cdde5a8e718 size: 3266 [root@zeloud onnx-wenet-predict]# docker images | grep similar 10.3.170.88:5000/user/admin/similar flask 3d4e06e095fa 6 minutes ago 3.03GB 10.3.170.88:5000/user/admin/similar latest e7edddd7af1c 2 days ago 3.02GB -
如果需要迁移到其他内网机，可以将新镜像保存到本地，拷贝之后，使用 docker load 命令加载。
# 打包保存 [root@zeloud ~]# docker save -o similar_flask.tar 10.3.170.88:5000/user/admin/similar:flask # 或者打包之后进行压缩，docker load 之前，需要先 gunzip 解压 [root@zeloud ~]# docker save 10.3.170.88:5000/user/admin/similar:flask | gzip > ./similar_flask.tar.gz -
使用新镜像启动容器，挂载 Flask 封装接口的路径。如果镜像没有默认启动 bash，则在 -c 参数前面加上 /bin/bash。如果希望后台启动，则添加 -d 参数。
[root@xd-dev2 admin]# docker run -it --rm -p 31700:7000 -v /zeloud/volume/turing/gv1/admin/similar-predict/:/zeloud/similar-predict 3d4e06e095fa -c "cd /zeloud/similar-predict/ && python flask-app.py" * Serving Flask app 'flask-app' * Debug mode: off WARNING: This is a development server. Do not use it in a production deployment. Use a production WSGI server instead. * Running on all addresses (0.0.0.0) * Running on http://127.0.0.1:7000 * Running on http://172.17.0.7:7000 Press CTRL+C to quit查看镜像的信息：
$ docker inspect <imageId> -
postman 请求调用示例：
-
Flask 封装接口示例：
from flask import Flask, request, jsonify import subprocess import os import io import torch from transformers import AutoImageProcessor, AutoModel from PIL import Image import torch.nn as nn app = Flask(__name__) def convert(image_data): image_stream = io.BytesIO(image_data) return Image.open(image_stream) @app.route("/predict", methods=["POST"]) def predict(): try: # 获取请求中的图片数据 if 'image' not in request.files: return jsonify({"error": "No image file in the request"}), 400 if 'image2' not in request.files: return jsonify({"error": "No pk image file in the request"}), 400 image_data = request.files['image'].read() image2_data = request.files['image2'].read() # 将图片数据转换为 PIL 的 Image 格式 image1 = convert(image_data) image2 = convert(image2_data) # 设置设备为GPU，如果不可用则使用CPU device = torch.device('cuda' if torch.cuda.is_available() else "cpu") # 加载预训练的图像处理器 processor = AutoImageProcessor.from_pretrained('./models/dinov2-base', local_files_only=True) # 加载预训练的模型，并将其移至相应的设备（GPU或CPU） model = AutoModel.from_pretrained('./models/dinov2-base', local_files_only=True).to(device) # 打开第一张图片 # image1 = Image.open('5.jpg') # 不计算梯度，用于推断 with torch.no_grad(): # 使用处理器处理图像，转换为PyTorch张量，并移至相应设备 inputs1 = processor(images=image1, return_tensors="pt").to(device) # 通过模型获取输出 outputs1 = model(**inputs1) # 获取最后一层的隐藏状态 image_features1 = outputs1.last_hidden_state # 对特征取平均，以得到单个向量表示 image_features1 = image_features1.mean(dim=1) print(f"outputs1 is {image_features1.shape}") # 重复上述过程，处理第二张图片 # image2 = Image.open('3.jpg') with torch.no_grad(): inputs2 = processor(images=image2, return_tensors="pt").to(device) outputs2 = model(**inputs2) image_features2 = outputs2.last_hidden_state image_features2 = image_features2.mean(dim=1) print(f"outputs2 is {image_features2.shape}") # 使用余弦相似度计算两个向量的相似度 cos = nn.CosineSimilarity(dim=0) sim = cos(image_features1[0],image_features2[0]).item() # 将相似度值调整到[0,1]范围内 sim = (sim+1)/2 print('similarity:', sim) return jsonify({'status': 'success', 'similarity': sim}) except Exception as e: return jsonify({"error": str(e)}) if __name__ == "__main__": app.run(host="0.0.0.0", port=7000)