aiborodin commented on code in PR #13340:
URL: https://github.com/apache/iceberg/pull/13340#discussion_r2156145524
##########
flink/v2.0/flink/src/main/java/org/apache/iceberg/flink/sink/dynamic/DynamicRecordProcessor.java:
##########
@@ -142,10 +151,18 @@ private void emit(
Schema schema,
CompareSchemasVisitor.Result result,
PartitionSpec spec) {
- RowData rowData =
- result == CompareSchemasVisitor.Result.SAME
- ? data.rowData()
- : RowDataEvolver.convert(data.rowData(), data.schema(), schema);
+ RowData rowData;
+ if (result == CompareSchemasVisitor.Result.SAME) {
+ rowData = data.rowData();
+ } else {
+ RowDataConverter rowDataConverter =
+ converterCache.get(
+ data.schema(),
+ dataSchema ->
+ new RowDataConverter(
+ FlinkSchemaUtil.convert(dataSchema),
FlinkSchemaUtil.convert(schema)));
Review Comment:
Yes, we did, please see the profile attached. According to the profile,
iceberg `Schema` -> `RowType` conversion takes approximately 51% of our
converter's CPU time, while the static conversion in `RowDataEvolver`, which
recomputes field indices for every record, accounts for about 45%. It is clear
from the profile that caching schemas alone wouldn't be sufficient, and we also
need quasi-code generation.
<img width="1572" alt="Screenshot 2025-06-19 at 3 45 17 pm"
src="https://github.com/user-attachments/assets/baf27122-9520-44dc-a0be-92659bc8dbff";
/>
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: issues-unsubscr...@iceberg.apache.org
For queries about this service, please contact Infrastructure at:
us...@infra.apache.org
---------------------------------------------------------------------
To unsubscribe, e-mail: issues-unsubscr...@iceberg.apache.org
For additional commands, e-mail: issues-h...@iceberg.apache.org
