Glob support on file access

[Etienne Chauchot]

Hi all,

In case it is useful to some of you:

I have a big batch that needs to use globs (*.parquet for example) to read input files. It seems that globs do not work out of the box (see https://issues.apache.org/jira/browse/FLINK-6417)

But there is a workaround:

final FileInputFormat inputFormat = new FileInputFormat(new Path(extractDir(filePath))); /* or any subclass of FileInputFormat*/ /*extact parent dir*/
inputFormat.setFilesFilter(new GlobFilePathFilter(Collections.singletonList(filePath), Collections.emptyList())); /*filePath contains glob, the whole path needs to be provided to GlobFilePathFilter*/
inputFormat.setNestedFileEnumeration(true);

Hope, it helps some people

Etienne Chauchot

[Etienne Chauchot]

But still this workaround would only work when you have access to the underlying FileInputFormat. For SQL and Table APIs, you don't so you'll be unable to apply this workaround. So what we could do is make a PR to support glob at the FileInputFormat level to profit for all APIs.

I'm gonna do it if everyone agrees.

Best

Etienne Chauchot

On 25/03/2021 13:12, Etienne Chauchot wrote:

Hi all,

In case it is useful to some of you:

I have a big batch that needs to use globs (*.parquet for example) to read input files. It seems that globs do not work out of the box (see https://issues.apache.org/jira/browse/FLINK-6417)

But there is a workaround:

final FileInputFormat inputFormat = new FileInputFormat(new Path(extractDir(filePath))); /* or any subclass of FileInputFormat*/ /*extact parent dir*/
inputFormat.setFilesFilter(new GlobFilePathFilter(Collections.singletonList(filePath), Collections.emptyList())); /*filePath contains glob, the whole path needs to be provided to GlobFilePathFilter*/
inputFormat.setNestedFileEnumeration(true);

Hope, it helps some people

Etienne Chauchot

[Arvid Heise-4]

Hi Etienne,

In general, any small PR on this subject is very welcome. I don't think that the community as a whole will invest much into FileInputFormat as the whole DataSet API is phasing out.

Afaik SQL and Table API are only using InputFormat for the legacy compatibility layer (e.g. when it comes to translating into DataSet). All the new batchy stuff is based on BulkFormat and unified source/sink interface. I'm CC'ing Timo who can correct me if I'm wrong.

So if you just want to add glob support on FileInputFormat /only/ for SQL and Table API, I don't think it's worth the effort. It would be more interesting to see if the new FileSource does support it properly and rather add it there.

On Mon, Mar 29, 2021 at 4:57 PM Etienne Chauchot <[hidden email]> wrote:

But still this workaround would only work when you have access to the
underlying /FileInputFormat/. For//SQL and Table APIs, you don't so
you'll be unable to apply this workaround. So what we could do is make a
PR to support glob at the FileInputFormat level to profit for all APIs.

I'm gonna do it if everyone agrees.

Best

Etienne Chauchot

On 25/03/2021 13:12, Etienne Chauchot wrote:
>
> Hi all,
>
> In case it is useful to some of you:
>
> I have a big batch that needs to use globs (*.parquet for example) to
> read input files. It seems that globs do not work out of the box (see
> https://issues.apache.org/jira/browse/FLINK-6417)
>
> But there is a workaround:
>
>
> final  FileInputFormat inputFormat =new  FileInputFormat(new  Path(extractDir(filePath)));/* or any subclass of FileInputFormat*/  /*extact parent dir*/
> inputFormat.setFilesFilter(new  GlobFilePathFilter(Collections.singletonList(filePath), Collections.emptyList()));/*filePath contains glob, the whole path needs to be provided to
> GlobFilePathFilter*/
> inputFormat.setNestedFileEnumeration(true);
>
> Hope, it helps some people
>
> Etienne Chauchot
>
>