Python/Flask在开发中工作，但在Docker中不工作

Question

我不知道发生了什么事。该体系结构是，Spring Boot（Java）后端将请求（映像）发送到另一个Spring Boot服务器，该服务器将请求发送到Flask，以预测它是猫还是狗。当我手动启动每台服务器时，它工作正常。但作为一个docker复合体（docker compose），只有在这种猫或狗的方法中，flask以某种方式失败了。这是同样的方法。我们还得到了一个OCR方法，我们在Spring Boot中做了完全相同的事情，效果很好。所以它必须和烧瓶和这个方法有关。。。但是什么呢

@app.route('/cat-or-dog/predict', methods=['POST'])
def image_classifier_cat_or_dog():
    img_codes = json.loads(request.form['images'])

    # Decoding base64 images
    imgs = []
    for img_code in img_codes:
        imgs.append(image.img_to_array(image.load_img(BytesIO(base64.b64decode(img_code)),
                                                      target_size=(150, 150, 3))))

    # Pre-processing images
    X_imgs = preprocess_images(imgs)
    X_imgs = np.array(X_imgs)
    n_imgs = len(X_imgs)

    # Predicting content of images
    predictions = []
    i = 0
    for batch in datagen.flow(X_imgs, batch_size=1):
        pred = model.predict(batch)
        if pred > 0.5:
            predictions.append("dog")
        else:
            predictions.append("cat")
        i += 1
        if i % n_imgs == 0:
            break

    return jsonify(predictions=predictions)

# Image preprocessing for cat or dog classifier


def preprocess_images(new_imgs):
    X = []

    for img in new_imgs:
        X.append(cv2.resize(img, (150, 150), interpolation=cv2.INTER_CUBIC))

    return X

错误：

aiproject-flask           | ERROR:app:Exception on /cat-or-dog/predict [POST]
aiproject-flask           | Traceback (most recent call last):
aiproject-flask           |   File "/usr/local/lib/python3.6/dist-packages/flask/app.py", line 2447, in wsgi_app
aiproject-flask           |     response = self.full_dispatch_request()
aiproject-flask           |   File "/usr/local/lib/python3.6/dist-packages/flask/app.py", line 1952, in full_dispatch_request
aiproject-flask           |     rv = self.handle_user_exception(e)
aiproject-flask           |   File "/usr/local/lib/python3.6/dist-packages/flask_cors/extension.py", line 161, in wrapped_function
aiproject-flask           |     return cors_after_request(app.make_response(f(*args, **kwargs)))
aiproject-flask           |   File "/usr/local/lib/python3.6/dist-packages/flask/app.py", line 1821, in handle_user_exception
aiproject-flask           |     reraise(exc_type, exc_value, tb)
aiproject-flask           |   File "/usr/local/lib/python3.6/dist-packages/flask/_compat.py", line 39, in reraise
aiproject-flask           |     raise value
aiproject-flask           |   File "/usr/local/lib/python3.6/dist-packages/flask/app.py", line 1950, in full_dispatch_request
aiproject-flask           |     rv = self.dispatch_request()
aiproject-flask           |   File "/usr/local/lib/python3.6/dist-packages/flask/app.py", line 1936, in dispatch_request
aiproject-flask           |     return self.view_functions[rule.endpoint](**req.view_args)
aiproject-flask           |   File "/app/app.py", line 85, in image_classifier_cat_or_dog
aiproject-flask           |     target_size=(150, 150, 3))))
aiproject-flask           |   File "/usr/local/lib/python3.6/dist-packages/keras_preprocessing/image/utils.py", line 113, in load_img
aiproject-flask           |     with open(path, 'rb') as f:
aiproject-flask           | TypeError: expected str, bytes or os.PathLike object, not _io.BytesIO
aiproject-flask           | INFO:werkzeug:172.30.0.2 - - [09/Jun/2020 17:45:12] "POST /cat-or-dog/predict HTTP/1.1" 500 -

Answer 1

根据Keras docs，load_img函数只接受类似路径的对象，这与您正在经历的行为一致

您正试图在容器内提供一些烧瓶应用程序，对吗？您确定您使用的是在其他地方正常运行的同一个代码分支吗

通过查看以下内容的输出，找到正确的容器名称：

docker container ls

使用此选项，登录到容器：

docker container exec -it <container_name> bash

接下来，导航到应用程序目录（假设您在此处使用git…）：

cd /path/to/your/application/repository

并检查输出中的分支名称：

git status

然后，在原始存储库中执行同样的操作，您将“手动启动每台服务器”。比较这两个输出

我几乎可以肯定，您使用了来自过时/未更新分支或类似分支的错误代码库。想办法弄清楚

如果确定要修改所显示的代码，则可以通过替换以下语句来修复该错误：

imgs.append(image.img_to_array(image.load_img(BytesIO(base64.b64decode(img_code)),
                                                      target_size=(150, 150, 3))))

为此：

# move this import somewere else...
from PIL import Image as PIL_Image


bytes_stream = BytesIO(base64.b64decode(img_code))
img = PIL_Image.open(bytes_stream)
img = img.convert('RGB')
img = img.resize((150, 150, 3), PIL_Image.NEAREST)

imgs.append(image.img_to_array(img))

上述变更是在this issue中提出的，作为解决与您非常相似的问题的一种方法