[
https://issues.apache.org/jira/browse/FLINK-9025?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16405162#comment-16405162
]
Fabian Hueske commented on FLINK-9025:
--------------------------------------
Rows have a schema with fixed data types like most types in Flink.
This behavior was deliberately chosen to improve performance. Since the {{Row}}
type is internally used by Flink's SQL feature and Table API, we cannot make
the serialization and copy logic more heavy to avoid performance implications.
If you'd like to use the {{Row}} data type, you should ensure that all input
records follow the same schema.
If that is not possible, you should chose a different data type.
Closing this issue as "Not a problem".
Feel free to reopen if you'd like to discuss the issue further or reach out to
the user mailing list.
> Problem with convert String to Row
> ----------------------------------
>
> Key: FLINK-9025
> URL: https://issues.apache.org/jira/browse/FLINK-9025
> Project: Flink
> Issue Type: Bug
> Components: Java API
> Reporter: ChenDonghe
> Priority: Major
>
> when we convert json string into Row, and set type of filed in the row, code
> below:
> {code:java}
> // code placeholder
> DataStream<String> stream = env.addSource(consumer.getInstance(sourceTopic,
> new SimpleStringSchema()).setStartFromLatest());
> DataStream<Row> dataStreamRow = stream.map(new
> ConvertDataStream()).returns(typeinfo);
> {code}
> this convert process calls the copy function in RowSerializer class, code
> below:
>
> {code:java}
> // public final class RowSerializer extends TypeSerializer<Row>
> @Override
> public Row copy(Row from) {
> int len = fieldSerializers.length;
> if (from.getArity() != len) {
> throw new RuntimeException("Row arity of from does not match
> serializers.");
> }
> Row result = new Row(len);
> for (int i = 0; i < len; i++) {
> Object fromField = from.getField(i);
> if (fromField != null) {
> Object copy = fieldSerializers[i].copy(fromField);
> result.setField(i, copy);
> } else {
> result.setField(i, null);
> }
> }
> return result;
> }
> {code}
> the json string type message from kafka convert to the row type, in this
> process, RowSerrializer copy the from-row to a new result-row, but the type
> of result-row filed is Object type, for example, the first message from kafka
> filed0 is Integer type, the second message from kafka filed0 is Long type, if
> we set the filed0 is Long, we hope Integer can be compatible, but we got an
> exception : can not convert Interger to Long, so we hope RowSerrializer copy
> function can be more flexible, it can be act as a table, we can insert an
> Integer type value into Long type filed.
>
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
