virtual column _row_number, _error, _raw_value for file, s3, hdfs functions.

[UnamedRus]

Use case

Describe the solution you'd like

Give ability to get corresponding row number of file being ingested via s3/hdfs/file table function.

Describe alternatives you've considered

 SELECT row_number() OVER (PARTITION BY _path) FROM s3(...)

But it's too slow.

Additional context

Probably we can have some other columns like _raw in case of parsing error.

[tolmalev]

Hi!
Just wanted to bump this feature request.
For parquet files some virtual columns could be quite useful:

• _row_group_number - number of the row group in current file for the given row
• _row_number_in_row_group - number of row within current row group
• _row_number - row number within the file

Right now it's possible to use row_number() over () but it has some huge disadvantages. It limits thread count to 1 and also requires memory to materialize full query results before any aggregation can be applied on top of such table.

Such virtual columns would unblock some interesting options for parquet files processing