从数据格式转换的角度 flink cdc 如何写入paimon？

从数据格式转换的角度 flink cdc 如何同步数据，写入paimon？

从一个测试用例着手

org/apache/flink/cdc/connectors/paimon/sink/v2/PaimonSinkITCase.java

java 复制代码

public void testSinkWithDataChange(String metastore, boolean enableDeleteVector)
            throws IOException, InterruptedException, Catalog.DatabaseNotEmptyException,
                    Catalog.DatabaseNotExistException, SchemaEvolveException {
         // 初始化paimon的元数据
        initialize(metastore);
                      // 创建paimonSink
        PaimonSink<Event> paimonSink =
                new PaimonSink<>(
                        catalogOptions, new PaimonRecordEventSerializer(ZoneId.systemDefault()));
        PaimonWriter<Event> writer = paimonSink.createWriter(new MockInitContext());
        Committer<MultiTableCommittable> committer = paimonSink.createCommitter();

        // insert
        writeAndCommit(
                writer, committer, createTestEvents(enableDeleteVector).toArray(new Event[0]));
        Assertions.assertThat(fetchResults(table1))
                .containsExactlyInAnyOrder(
                        Row.ofKind(RowKind.INSERT, "1", "1"), Row.ofKind(RowKind.INSERT, "2", "2"));

        // delete
        writeAndCommit(
                writer,
                committer,
                generateDelete(
                        table1, Arrays.asList(Tuple2.of(STRING(), "1"), Tuple2.of(STRING(), "1"))));

        Assertions.assertThat(fetchResults(table1))
                .containsExactlyInAnyOrder(Row.ofKind(RowKind.INSERT, "2", "2"));

        // update
        writeAndCommit(
                writer,
                committer,
                generateUpdate(
                        table1,
                        Arrays.asList(Tuple2.of(STRING(), "2"), Tuple2.of(STRING(), "2")),
                        Arrays.asList(Tuple2.of(STRING(), "2"), Tuple2.of(STRING(), "x"))));
        Assertions.assertThat(fetchResults(table1))
                .containsExactlyInAnyOrder(Row.ofKind(RowKind.INSERT, "2", "x"));

        if (enableDeleteVector) {
            Assertions.assertThat(fetchMaxSequenceNumber(table1.getTableName()))
                    .containsExactlyInAnyOrder(
                            Row.ofKind(RowKind.INSERT, 1L), Row.ofKind(RowKind.INSERT, 3L));
        } else {
            Assertions.assertThat(fetchMaxSequenceNumber(table1.getTableName()))
                    .containsExactlyInAnyOrder(
                            Row.ofKind(RowKind.INSERT, 1L),
                            Row.ofKind(RowKind.INSERT, 2L),
                            Row.ofKind(RowKind.INSERT, 3L));
        }
    }

整体流程：

创建变更事件。
创建PaimonWriter
对于每一个事件event，都调用paimonWriter中的write。具体又包括以下几个步骤
- 序列化event为paimon 中的数据格式genericRow
- 创建paimon的write算子（StoreSinkWrite），将genericRow写入paimon表. 进入paimon内部的写入逻辑。
  - 从每一条row中提取rowkind（INSERT,UPDATE_BEFORE,UPDATE_AFTER,DELETE）
  - 提取data，MergeTreeWriter中执行两个操作，内存够，就把数据put到内存，内存不够就flush到磁盘