Mike Pedersen created BEAM-7793:
-----------------------------------

             Summary: BagState drops Rows when triggered by timer
                 Key: BEAM-7793
                 URL: https://issues.apache.org/jira/browse/BEAM-7793
             Project: Beam
          Issue Type: Bug
          Components: runner-direct, sdk-java-core
    Affects Versions: 2.13.0
            Reporter: Mike Pedersen


If a BagState contains rows, it appears to be empty when read in a timer. This 
is for example used by GroupIntoBatches, causing the following to fail:

{code:java}
    @Test
    public void canBatchRows() {
        Schema schema = Schema.builder().addInt32Field("a").build();

        PCollection<KV<Integer, Iterable<Row>>> pcoll = pipeline
                .apply(Create
                        .of(Row.withSchema(schema).addValue(1).build())
                        .withType(TypeDescriptors.rows()))
                .setRowSchema(schema)
                .apply(WithKeys.<Integer, 
Row>of(0).withKeyType(TypeDescriptors.integers()))
                .apply(GroupIntoBatches.ofSize(10));

        PAssert.that(pcoll).containsInAnyOrder(
                KV.of(0, Collections.singletonList(
                        Row.withSchema(schema).addValue(1).build()
                )));

        pipeline.run().waitUntilFinish();
    }
{code}

The above uses GroupIntoBatches as an example, but the problem seem to be with 
all DoFn having BagState.



--
This message was sent by Atlassian JIRA
(v7.6.14#76016)

Reply via email to