CSV file, filter with list of columns

[ypynda]

Hi team, and thanks for a great product!

I’m using FilePulseSourceConnector with CSVFilter type. My source files don’t have any headers.

When I have the following configuration, everything is working fine

  "filters.ParseCSVLine.type"       : "io.streamthoughts.kafka.connect.filepulse.filter.CSVFilter",
  "filters.ParseCSVLine.seperator"  : ",",
  "filters.ParseCSVLine.ignore.leading.whitespace": "true",
  "filters.ParseCSVLine.auto.generate.column.names": "true",

When I try to specify column name:type, I get unexpected results. It looks like the column list is sorted alphabetically and the column names don't match

  "filters.ParseCSVLine.type"       : "io.streamthoughts.kafka.connect.filepulse.filter.CSVFilter",
  "filters.ParseCSVLine.seperator"  : ",",
  "filters.ParseCSVLine.ignore.leading.whitespace": "true",
  "filters.ParseCSVLine.auto.generate.column.names": "false",
  "filters.ParseCSVLine.columns"    : "DataID1:string;TextField1:string; DataID2:string;TextField2:string;Date1:string;Date2:string;diffDate:string",

Can you please check it out?
Thank you

[mjseid]

We are experiencing this as well on the latest 2.12.0 release

[fhussonnois]

Hi @ypynda, @mjseid thank you for reporting this issue. I will check the filter and release new version if necessary.

[abgoswami]

Hi @fhussonnois ,
Can you please assign this bug to me? Or add me as collaborator? I have forked the repository and want to submit a Pull Request for the fix, but want to make sure that someone else is not working simultaneously on this issue.
Thanks

[abgoswami]

@fhussonnois , Thanks.
I'll submit a PR and drop a note here.

[thutch]

I'm seeing the same issue. Has this been resolved?
The latest version that does not have this issue is 2.8.0

[kirito73]

The code indeed take the order in which it is provided in the config and doesn't reorder based on actual position in the csv.
So if you provide the same ordering of columns in your config and csv both it will work as expected.
But I think this must be handled in the code itself.

I can raise a PR for the fix if that is fine with @fhussonnois and @abgoswami

[pmizenin]

I'm facing a similar issue, with "filters.ParseCSVLine.extract.column.name": "headers" setting the columns of the CSV come out of order.

Where this bites me is when I try to publish in protobuf and attempt to use "value.converter": "io.confluent.connect.protobuf.ProtobufConverter",. In this case the schema I provide doesn't work with the structure that gets produced by the filter.

Is somebody looking into this problem? If not, should I attempt to?

[github-actions]

This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you for your contributions.