stevenzwu commented on a change in pull request #2064:
URL: https://github.com/apache/iceberg/pull/2064#discussion_r557548243
##########
File path: core/src/main/java/org/apache/iceberg/TableProperties.java
##########
@@ -138,6 +138,9 @@ private TableProperties() {
public static final String ENGINE_HIVE_ENABLED = "engine.hive.enabled";
public static final boolean ENGINE_HIVE_ENABLED_DEFAULT = false;
+ public static final String WRITE_SHUFFLE_BY_PARTITION =
"write.shuffle-by.partition";
Review comment:
@openinx those limitations are the result of hash shuffling. If we use
traffic distribution stats per bucket/group, we can shuffle the traffic
relatively evenly to all downstream writer tasks. Table partition (like number
of buckets) doesn't have to tie with writer parallelism, which is inflexible.
Weight based shuffle can make use of all writer tasks. It won't be perfect, but
should be reasonably good.
Regarding the example of bucket transformation for table partition, it works
for table that are partitioned by some well distributed "id" column. We
actually have a super large table using this pattern. But that is not a general
scenario though, as not every table can to want to have a bucket partition.
E.g., we may have table partitioned only by "ts" and "country" columns.
----------------------------------------------------------------
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.
For queries about this service, please contact Infrastructure at:
[email protected]
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
