[security] make csv/parquet engine work with dal

[sundy-li]

Summary

Now csv/parquet engine can only work with a local file. However, it allows users to read any file with permission in the query node.

This is dangerous, such as: /etc/passwd, /etc/sudoers.

We should make these engines work with dal.

[BohuTANG]

1. For local file
   we should introduce a tmp_dir, all the files in this directory can be accessed.
   During reading:
   read('file_label_name') -- read from '/tmp_dir/file_label_name'

2. For the cloud
   user puts their local file to a s3 storage bucket(perhaps give a http api to handle it), and using the COPY ... FROM S3(...) command to read.

Please see the references in this issue: #2627

[sundy-li]

How about reuse the Config.Storage in query node, there is a config named data_path that can be used.

If we delopy as cloud-native serverless , the local file did not make sense.

[BohuTANG]

How about reuse the Config.Storage in query node, there is a config named data_path that can be used.

If we delopy as cloud-native serverless , the local file did not make sense.

Looks great 💯

[Xuanwo]

Any updates on this?

I'm working on the implementation of #3677, is there any thing I need to know?

[BohuTANG]

Any updates on this?

We don't have plan for it in the future time, I think we should close it first.