[Bug]: BigQueryIO - unknown repeated fields are merged incorrectly to payload

[stankiewicz]

What happened?

found as part of analysis: #34102 (comment)
problematic line here: StorageApiWriteUnshardedRecords

If we have a row with known fields (name:string, surname:string) and one unknown (foo:string) , this line will concat and not cause issues.

If unknown field is part of repeated structure,
e.g. (name:string, phone:(repeated struct type:string, num:string) ) and we have unknown in repeated struct (phone: [ null, {favourite:yes} ]) concat work well.

to reproduce:

final List<String> elements = Arrays.asList("1234, Foo", "1111, Bar", "3456, Baz");

    List<TableFieldSchema> fields = new ArrayList<>();
    fields.add(new TableFieldSchema().setName("id").setType("INTEGER"));
    fields.add(new TableFieldSchema().setName("name").setType("STRING"));
    fields.add(
        new TableFieldSchema()
            .setName("phone_numbers")
            .setMode("REPEATED")
            .setType("RECORD")
            .setFields(
                ImmutableList.of(
                    new TableFieldSchema()
                        .setName("type")
                        .setType("STRING")
                        .setMode(
                            "REQUIRED"), // remove required mode - BQ IO will insert empty records
                    new TableFieldSchema().setName("number").setType("STRING"))));
    TableSchema schema = new TableSchema().setFields(fields);

    Pipeline p = Pipeline.create(options);

    @UnknownKeyFor
    @NonNull
    @Initialized
    WriteResult apply =
        p
            // Add a source that uses the hard-coded records.
            .apply(Create.of(elements))
            // Add a transform that converts the records (comma-separated strings) to
            // TableRow.
            .apply(
                "to TableRow",
                ParDo.of(
                    new DoFn<String, TableRow>() {
                      @ProcessElement
                      public void processElement(ProcessContext c) {
                        List<String> columns = Splitter.on(", ").splitToList(c.element());

                        TableRow row = new TableRow();
                        row.set("id", Integer.parseInt(columns.get(0)));
                        row.set("name", columns.get(1));
                        if (columns.get(0).equals("1111")) {
                          row.set(
                              "phone_numbers",
                              ImmutableList.of(
                                  new TableCell().set("type", "home").set("number", "88888"),
                                  new TableCell()
                                      .set("type", "work")
                                      .set("number", "99999")
                                      .set("foo", "bar"));  // add unknown field
                          );
                        } else {
                          row.set(
                              "phone_numbers",
                              ImmutableList.of(
                                  new TableCell().set("type", "home").set("number", "12345"),
                                  new TableCell().set("type", "work").set("number", "67890")));
                        }
                        c.output(row);
                      }
                    }))
            .apply(
                "bq",
                BigQueryIO.writeTableRows()
                    .withSchema(schema)
                    .to("<project>:<dataset>.<table>")
                    .withMethod(BigQueryIO.Write.Method.STORAGE_API_AT_LEAST_ONCE)
                    .withCreateDisposition(BigQueryIO.Write.CreateDisposition.CREATE_IF_NEEDED)
                    .withWriteDisposition(BigQueryIO.Write.WriteDisposition.WRITE_APPEND)
                    .withAutoSchemaUpdate(true)
                    .ignoreUnknownValues());

Issue Priority

Priority: 2 (default / most bugs should be filed as P2)

Issue Components

• Component: Python SDK
• Component: Java SDK
• Component: Go SDK
• Component: Typescript SDK
• Component: IO connector
• Component: Beam YAML
• Component: Beam examples
• Component: Beam playground
• Component: Beam katas
• Component: Website
• Component: Infrastructure
• Component: Spark Runner
• Component: Flink Runner
• Component: Samza Runner
• Component: Twister2 Runner
• Component: Hazelcast Jet Runner
• Component: Google Cloud Dataflow Runner

[stankiewicz]

example test with data:

beam/sdks/java/io/google-cloud-platform/src/test/java/org/apache/beam/sdk/io/gcp/bigquery/TableRowToStorageApiProtoTest.java

Line 1738 in b4c03d8

.set("unknown", "valueE")));

test is passing and is producing msg and unknown pair that further down is concatenated.

[stankiewicz]

proto concat is similar in behaviour to mergeFrom - It may replace singular fields (here we are not replacing as unknown placeholder has only new fields), but for repeated fields it will concatenates.
in problematic test unknown placeholder is created with empty singular fields but array has 2 elements - first element is empty row, while second contains new unknown field.

When merged, repeated1 will contain 4 elements which is expected.