[jira] [Commented] (BEAM-10030) Add CSVIO for Java SDK

["Saurabh Joshi (Jira)" <ji...@apache.org>]

    [ https://issues.apache.org/jira/browse/BEAM-10030?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17111013#comment-17111013 ] 

Saurabh Joshi commented on BEAM-10030:
--------------------------------------

Here is a draft implementation of CSVSource: [^CSVSource.java]. There are missing pieces here and there that needs to be worked on, but basically it uses Apache Commons CSV library for parsing the CSV file, with the record number as the offset. When it starts reading, it uses CSVParser to skip records until it reaches the start offset record.

> Add CSVIO for Java SDK
> ----------------------
>
>                 Key: BEAM-10030
>                 URL: https://issues.apache.org/jira/browse/BEAM-10030
>             Project: Beam
>          Issue Type: New Feature
>          Components: io-ideas
>            Reporter: Saurabh Joshi
>            Priority: P2
>         Attachments: CSVSource.java
>
>
> Apache Beam has TextIO class which can read text based files line by line, delimited by either a carriage return, newline, or a carriage return and a newline. This approach does not support CSV files which have records that span multiple lines. This is because there could be fields where there is a newline inside the double quotes.
> This Stackoverflow question is relevant for a feature that should be added to Apache Beam: [https://stackoverflow.com/questions/51439189/how-to-read-large-csv-with-beam]
> I can think of two libraries we could use for handling CSV files. The first one is using Apache Commons CSV library. Here is some example code which can use CSVRecord class for reading and writing CSV records:
> {{{color:#172b4d}{{PipelineOptions options = PipelineOptionsFactory.create();}}
>  {{Pipeline pipeline = Pipeline.create(options);}}
>  {{PCollection<CSVRecord> records = pipeline.apply("ReadCSV", CSVIO.read().from("input.csv"));}}
>  records.apply("WriteCSV", CSVIO.write().to("output.csv"));{color}}}
> Another library we could use is Jackson CSV, which allows users to specify schemas for the columns: [https://github.com/FasterXML/jackson-dataformats-text/tree/master/csv]
> The crux of the problem is this: can we read and write large CSV files in parallel, by splitting the records and distribute it to many workers? If so, would it be good to have a feature where Apache Beam supports reading/writing CSV files?



--
This message was sent by Atlassian Jira
(v8.3.4#803005)