[
https://issues.apache.org/jira/browse/BEAM-10100?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17539782#comment-17539782
]
Stephen Patel commented on BEAM-10100:
--------------------------------------
The contents of the files definitely aren't correct. (I mean, they're correct
in the sense that they are records that were output by the pipeline, but
they're only a subset). That is, it's not the case that the 391 files were
compressed down into 120 files (but with the same number of records).
In my case, I loaded the files into AWS Athena (hive basically) to query, and
they only contained a few hundred of the 7000ish records that I expected.
If the files were just being combined, I'd still expect the same number of
records as if I had used a standard write with a single destination (vs a
dynamic destination write).
> FileIO writeDynamic with AvroIO.sink not writing all data
> ---------------------------------------------------------
>
> Key: BEAM-10100
> URL: https://issues.apache.org/jira/browse/BEAM-10100
> Project: Beam
> Issue Type: Bug
> Components: io-java-avro, io-java-files, runner-flink, runner-spark
> Affects Versions: 2.16.0, 2.17.0, 2.18.0, 2.19.0, 2.20.0, 2.21.0, 2.22.0
> Environment: Mac OSX Catalina, tested with SparkRunner - Spark 2.4.5.
> Reporter: Dave Martin
> Priority: P1
> Labels: p1
>
> FileIO writeDynamic with AvroIO.sink is not writing all data in the following
> pipeline. The amount of data written varies between runs but it is
> consistently dropping records. This is with a very small test dataset - 6
> records, which should produce 3 directories.
> {code:java}
> Pipeline p = Pipeline.create(options);
> PCollection<KV<String, AvroRecord>> records =
> p.apply(TextIO.read().from("/tmp/input.csv"))
> .apply(ParDo.of(new StringToDatasetIDAvroRecordFcn()));
> //write out into AVRO in each separate directory
> records.apply("Write avro file per dataset", FileIO.<String, KV<String,
> AvroRecord>>writeDynamic()
> .by(KV::getKey)
> .via(Contextful.fn(KV::getValue), Contextful.fn(x ->
> AvroIO.sink(AvroRecord.class).withCodec(BASE_CODEC)))
> .to(options.getTargetPath())
> .withDestinationCoder(StringUtf8Coder.of())
> .withNaming(key -> defaultNaming(key + "/export",
> PipelinesVariables.Pipeline.AVRO_EXTENSION)));
> p.run().waitUntilFinish();
> {code}
> If i replace AvroIO.sink() with TextIO.sink() (and replace the initial
> mapping function) then the correct number of records are written to the
> separate directories. This is working consistently.
> e.g.
> {code:java}
> // Initialise pipeline
> Pipeline p = Pipeline.create(options);
> PCollection<KV<String, String>> records =
> p.apply(TextIO.read().from("/tmp/input.csv")).apply(ParDo.of(new
> StringToDatasetIDKVFcn()));
> //write out into AVRO in each separate directory
> records.apply("Write CSV file per dataset", FileIO.<String, KV<String,
> String>>writeDynamic()
> .by(KV::getKey)
> .via(Contextful.fn(KV::getValue), TextIO.sink())
> .to(options.getTargetPath())
> .withDestinationCoder(StringUtf8Coder.of())
> .withNaming(datasetID -> defaultNaming(key + "/export", ".csv"));
> p.run().waitUntilFinish();
> {code}
> cc [~timrobertson100]
--
This message was sent by Atlassian Jira
(v8.20.7#820007)
