stevenzwu commented on a change in pull request #2898:
URL: https://github.com/apache/iceberg/pull/2898#discussion_r799192961
##########
File path:
flink/v1.14/flink/src/main/java/org/apache/iceberg/flink/sink/FlinkSink.java
##########
@@ -412,18 +417,38 @@ private String operatorName(String suffix) {
switch (writeMode) {
case NONE:
+ if (!equalityFieldIds.isEmpty()) {
+ LOG.info("Distribute rows by equality fields in '{}' distribution
mode", DistributionMode.NONE.modeName());
+ return input.keyBy(new EqualityFieldKeySelector(equalityFieldIds,
iSchema, flinkRowType));
+ }
+
return input;
case HASH:
if (partitionSpec.isUnpartitioned()) {
+ if (!equalityFieldIds.isEmpty()) {
+ LOG.info("Distribute rows by equality fields in '{}'
distribution mode, because table is unpartitioned",
+ DistributionMode.HASH.modeName());
+ return input.keyBy(new
EqualityFieldKeySelector(equalityFieldIds, iSchema, flinkRowType));
+ }
+
+ LOG.warn("Fallback to use '{}' distribution mode, because table is
unpartitioned",
+ DistributionMode.NONE.modeName());
return input;
} else {
+ LOG.info("Distribute rows by partition fields in '{}' distribution
mode", DistributionMode.HASH.modeName());
Review comment:
hash distribution was trying to bring all rows for the same partition
value to a single writer. if we include equality field, that would defeat the
purpose. On the other hand, if we don't include equality field, we would still
suffer the issue that this PR is trying to address. Maybe with equality field,
we shouldn't use `HASH` distribution?
--
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: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
