如何使用Paramiko getfo将文件从SFTP服务器下载到内存以进行处理
我正在尝试使用Paramiko从SFTP下载CSV文件(内存中),并将其导入到pandas数据框中.
I am trying to download a CSV file (in-memory) from SFTP using Paramiko and import it into a pandas dataframe.
transport = paramiko.Transport((server, 22))
transport.connect(username=username, password=password)
sftp = paramiko.SFTPClient.from_transport(transport)
with open(file_name, 'wb') as fl:
sftp.getfo(file_name, fl, callback=printTotals)
df = pd.read_csv(fl, sep=' ')
下面的代码失败,告诉我:
The code below fails, telling me:
OSError:文件未打开以供读取
OSError: File is not open for reading
我假设我需要某种缓冲区或文件,例如fl
的对象,因为open需要一个文件.我对这一切还比较陌生,所以如果有人可以帮忙,我会很高兴.
I assume that I need some kind of buffer or file like object for fl
instead, since open needs a file. I am relatively new to all of this, so I would be happy it if someone could help.
仍然允许您使用进度回调的简单解决方案是:
A simple solution that still allows you to use progress callback is:
- 使用
BytesIO
类似于文件的对象将下载的文件存储到内存中; - 下载文件后,您必须先查找文件指针,使其返回文件开始位置,然后再开始阅读文件.
- Use
BytesIO
file-like object to store a downloaded file to memory; - You have to seek file pointer back to file start after downloading it, before you start reading it.
with io.BytesIO() as fl:
sftp.getfo(file_name, fl, callback=printTotals)
fl.seek(0)
df = pd.read_csv(fl, sep=' ')
尽管使用此解决方案,您最终将文件两次加载到内存中.
Though with this solution, you will end up having the file loaded to memory twice.
更好的解决方案是实现类似文件的对象.它甚至可以让您同时下载和解析文件.
Better solution is to implement a file-like object. It will even allow you to download and parse the file at the same time.
class FileWithProgress:
def __init__(self, fl):
self.fl = fl
self.size = fl.stat().st_size
self.p = 0
def read(self, blocksize):
r = self.fl.read(blocksize)
self.p += len(r)
print(str(self.p) + " of " + str(self.size))
return r
并像这样使用它:
with sftp.open(file_name, "rb") as fl:
fl.prefetch()
df = pd.read_csv(FileWithProgress(fl), sep=' ')
有关SFTPFile.prefetch
调用,请参阅:
读取使用Python Paramiko SFTPClient.open方法打开的文件很慢.
For the SFTPFile.prefetch
call, refer to:
Reading file opened with Python Paramiko SFTPClient.open method is slow.