Skip to content

[QUESTION] Is it possible to use metaflow.S3 outside of standard metaflow use? #183

@manugarri

Description

@manugarri

First of all, my apologies if this is not the appropriate channel for questions.

After researching metaflow, I believe metaflow's implementation of parallel s3 reading is significantly faster than other alternatives (looking for a replacement of dask internal s3fs reading logic).

However, I cant seem to be able to read valid data from metaflow.S3. Here is a snippet that shows my issue:

S3_PATH = "s3://s3-bucket/path/"
s3 = S3(s3root=S3_PATH) 
s3 = s3.__enter__() # issue is the same with context manager
data = s3.get_all() # read all files in the root

first_file = data[0]
first_file.text   

The text representation of the files seems to be encoded in a way I cant figure out how to deserialize? How can we properly deserialize these string representations?

I also tried deserializing the locally downloaded data using metaflow datastore

with gzip.GzipFile(data[0].path, mode="rb") as f: 
    r = f.read()

Yields Not a gzipped file (b'15')

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions