HashWriteDistributionSortedDataframeSpec: + See https://iceberg.apache.org/docs/1.6.0/spark-writes/#writing-distribution-modes Using write.distribution-mode - should create the appropriate number of Iceberg files + Given a table that has a distribution mode of hash and is created with: CREATE TABLE polaris.my_namespace.HashWriteDistributionSortedDataframeSpec ( id int, label String, partitionKey long, date Date, timestamp Timestamp ) USING iceberg TBLPROPERTIES ( 'format-version' = '2', 'write.spark.fanout.enabled' = 'true', 'write.distribution-mode' = 'hash' ) PARTITIONED BY (partitionKey); + And the data is sorted on the partitionKey column + And a query plan that looks like: == Optimized Logical Plan == Sort [partitionKey#3208L ASC NULLS FIRST], true, Statistics(sizeInBytes=6.3 MiB) +- Repartition 6, true, Statistics(sizeInBytes=6.3 MiB) +- Project [id#3206, partitionKey#3208L, date#3209, timestamp#3210, concat(cast(otherId#3224L as string), xxx) AS label#3243], Statistics(sizeInBytes=6.3 MiB) +- Join Inner, (partitionKey#3208L = otherId#3224L), Statistics(sizeInBytes=4.9 MiB) :- LocalRelation [id#3206, partitionKey#3208L, date#3209, timestamp#3210], Statistics(sizeInBytes=640.0 B, rowCount=20) +- Project [value#3221L AS otherId#3224L], Statistics(sizeInBytes=7.8 KiB) +- SerializeFromObject [input[0, bigint, false] AS value#3221L], Statistics(sizeInBytes=7.8 KiB) +- MapElements uk.co.odinconsultants.iceberg.distributions.AbstractWriteDistributionSpec$$Lambda$5586/0x0000000802566990@15f87946, class java.lang.Long, [StructField(value,LongType,true)], obj#3220: bigint, Statistics(sizeInBytes=7.8 KiB) +- DeserializeToObject staticinvoke(class java.lang.Long, ObjectType(class java.lang.Long), valueOf, id#3216L, true, false, true), obj#3219: java.lang.Long, Statistics(sizeInBytes=7.8 KiB) +- Range (0, 1000, step=1, splits=Some(4)), Statistics(sizeInBytes=7.8 KiB, rowCount=1.00E+3) == Physical Plan == AdaptiveSparkPlan isFinalPlan=false +- Sort [partitionKey#3208L ASC NULLS FIRST], true, 0 +- Exchange rangepartitioning(partitionKey#3208L ASC NULLS FIRST, 200), ENSURE_REQUIREMENTS, [plan_id=5239] +- Exchange RoundRobinPartitioning(6), REPARTITION_BY_NUM, [plan_id=5237] +- Project [id#3206, partitionKey#3208L, date#3209, timestamp#3210, concat(cast(otherId#3224L as string), xxx) AS label#3243] +- BroadcastHashJoin [partitionKey#3208L], [otherId#3224L], Inner, BuildLeft, false :- BroadcastExchange HashedRelationBroadcastMode(List(input[1, bigint, false]),false), [plan_id=5234] : +- LocalTableScan [id#3206, partitionKey#3208L, date#3209, timestamp#3210] +- Project [value#3221L AS otherId#3224L] +- SerializeFromObject [input[0, bigint, false] AS value#3221L] +- MapElements uk.co.odinconsultants.iceberg.distributions.AbstractWriteDistributionSpec$$Lambda$5586/0x0000000802566990@15f87946, obj#3220: bigint +- DeserializeToObject staticinvoke(class java.lang.Long, ObjectType(class java.lang.Long), valueOf, id#3216L, true, false, true), obj#3219: java.lang.Long +- Range (0, 1000, step=1, splits=4) + And it has 20 rows over 5 data file(s) when writing with 4 executor threads + When we add another 20 rows of the same data that is logically distributed over 5 partition(s) + And the data is sorted on the partitionKey column + Then there are now 5 more data files + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + +