org.apache.beam.sdk.io.contextualtextio

Class ContextualTextIO

java.lang.Object

org.apache.beam.sdk.io.contextualtextio.ContextualTextIO

public class ContextualTextIO
extends java.lang.Object

PTransforms that read text files and collect contextual information of the elements in the input.

Prefer TextIO when not reading files with multi-line records or additional record metadata is not required.

Reading from text files

To read a PCollection from one or more text files, use ContextualTextIO.read(). To instantiate a transform use ContextualTextIO.Read.from(String) and specify the path of the file(s) to be read. Alternatively, if the filenames to be read are themselves in a PCollection you can use FileIO to match them and readFiles() to read them.

read() returns a PCollection of Rows with schema RecordWithMetadata.getSchema(), each corresponding to one line of an input UTF-8 text file (split into lines delimited by '\n', '\r', '\r\n', or specified delimiter via ContextualTextIO.Read.withDelimiter(byte[])).

Filepattern expansion and watching

By default, the filepatterns are expanded only once. The combination of FileIO.Match#continuously(Duration, TerminationCondition) and readFiles() allow streaming of new files matching the filepattern(s).

By default, read() prohibits filepatterns that match no files, and readFiles() allows them in case the filepattern contains a glob wildcard character. Use ContextualTextIO.Read.withEmptyMatchTreatment(org.apache.beam.sdk.io.fs.EmptyMatchTreatment) or FileIO.Match#withEmptyMatchTreatment(EmptyMatchTreatment) plus readFiles() to configure this behavior.

Example 1: reading a file or filepattern.

 Pipeline p = ...;

 // A simple Read of a file:
 PCollection<Row> records = p.apply(ContextualTextIO.read().from("/local/path/to/file.txt"));
 

Example 2: reading a PCollection of filenames.

 Pipeline p = ...;

 // E.g. the filenames might be computed from other data in the pipeline, or
 // read from a data source.
 PCollection<String> filenames = ...;

 // Read all files in the collection.
 PCollection<Row> records =
     filenames
         .apply(FileIO.matchAll())
         .apply(FileIO.readMatches())
         .apply(ContextualTextIO.readFiles());
 

Example 3: streaming new files matching a filepattern.

 Pipeline p = ...;

 PCollection<Row> records = p.apply(ContextualTextIO.read()
     .from("/local/path/to/files/*")
     .watchForNewFiles(
       // Check for new files every minute
       Duration.standardMinutes(1),
       // Stop watching the filepattern if no new files appear within an hour
       afterTimeSinceNewOutput(Duration.standardHours(1))));
 

Example 4: reading a file or file pattern of RFC4180-compliant CSV files with fields that may contain line breaks.

Example of such a file could be:

"aaa","b CRLF bb","ccc" CRLF zzz,yyy,xxx

 Pipeline p = ...;

 PCollection<Row> records = p.apply(ContextualTextIO.read()
     .from("/local/path/to/files/*.csv")
      .withHasMultilineCSVRecords(true));
 

Example 5: reading while watching for new files

 Pipeline p = ...;

 PCollection<Row> records = p.apply(FileIO.match()
      .filepattern("filepattern")
      .continuously(
        Duration.millis(100),
        Watch.Growth.afterTimeSinceNewOutput(Duration.standardSeconds(3))))
      .apply(FileIO.readMatches())
      .apply(ContextualTextIO.readFiles());
 

Example 6: reading with recordNum metadata.

 Pipeline p = ...;

 PCollection<Row> records = p.apply(ContextualTextIO.read()
     .from("/local/path/to/files/*.csv")
      .setWithRecordNumMetadata(true));
 

NOTE: When using ContextualTextIO.Read.withHasMultilineCSVRecords(Boolean), a single reader will be used to process the file, rather than multiple readers which can read from different offsets. For a large file this can result in lower performance.

NOTE: Use ContextualTextIO.Read.withRecordNumMetadata() when recordNum metadata is required. Computing absolute record positions currently introduces a grouping step, which increases the resources used by the pipeline. By default withRecordNumMetadata is set to false, in this case record objects will not contain absolute record positions within the entire file, but will still contain relative positions in respective offsets.

Reading a very large number of files

If it is known that the filepattern will match a very large number of files (e.g. tens of thousands or more), use ContextualTextIO.Read.withHintMatchesManyFiles() for better performance and scalability. Note that it may decrease performance if the filepattern matches only a small number of files.

Nested Class Summary

Nested Classes

Modifier and Type	Class and Description
static class	ContextualTextIO.Read Implementation of read().
static class	ContextualTextIO.ReadFiles Implementation of readFiles().

Method Summary

Modifier and Type	Method and Description
static ContextualTextIO.Read	read() A PTransform that reads from one or more text files and returns a bounded PCollection containing one element for each line in the input files.
static ContextualTextIO.ReadFiles	readFiles() Like read(), but reads each file in a PCollection of FileIO.ReadableFile, returned by FileIO.readMatches().

Methods inherited from class java.lang.Object

clone, equals, finalize, getClass, hashCode, notify, notifyAll, toString, wait, wait, wait

Method Detail

read

public static ContextualTextIO.Read read()

A PTransform that reads from one or more text files and returns a bounded PCollection containing one element for each line in the input files.

readFiles

public static ContextualTextIO.ReadFiles readFiles()

Like read(), but reads each file in a PCollection of FileIO.ReadableFile, returned by FileIO.readMatches().