RussellSpitzer commented on code in PR #7499:
URL: https://github.com/apache/iceberg/pull/7499#discussion_r1186689907
##########
docs/spark-writes.md:
##########
@@ -339,74 +331,55 @@ USING iceberg
PARTITIONED BY (days(ts), category)
```
-To write data to the sample table, your data needs to be sorted by `days(ts),
category`.
-
-If you're inserting data with SQL statement, you can use `ORDER BY` to achieve
it, like below:
+To write data to the sample table, your data needs to be sorted by `days(ts),
category` but this is taken care
+of automatically by the default `hash` distribution.
```sql
INSERT INTO prod.db.sample
SELECT id, data, category, ts FROM another_table
-ORDER BY ts, category
-```
-
-If you're inserting data with DataFrame, you can use either `orderBy`/`sort`
to trigger global sort, or `sortWithinPartitions`
-to trigger local sort. Local sort for example:
-
-```scala
-data.sortWithinPartitions("ts", "category")
- .writeTo("prod.db.sample")
- .append()
```
-You can simply add the original column to the sort condition for the most
partition transformations, except `bucket`.
-
-For `bucket` partition transformation, you need to register the Iceberg
transform function in Spark to specify it during sort.
-
-Let's go through another sample table having bucket partition:
-
-```sql
-CREATE TABLE prod.db.sample (
- id bigint,
- data string,
- category string,
- ts timestamp)
-USING iceberg
-PARTITIONED BY (bucket(16, id))
-```
-
-You need to register the function to deal with bucket, like below:
-
-```scala
-import org.apache.iceberg.spark.IcebergSpark
-import org.apache.spark.sql.types.DataTypes
-
-IcebergSpark.registerBucketUDF(spark, "iceberg_bucket16", DataTypes.LongType,
16)
-```
-
-{{< hint info >}}
-Explicit registration of the function is necessary because Spark doesn't allow
Iceberg to provide functions.
-[SPARK-27658](https://issues.apache.org/jira/browse/SPARK-27658) is filed to
enable Iceberg to provide functions
-which can be used in query.
-{{< /hint >}}
-
-Here we just registered the bucket function as `iceberg_bucket16`, which can
be used in sort clause.
-
-If you're inserting data with SQL statement, you can use the function like
below:
-
-```sql
-INSERT INTO prod.db.sample
-SELECT id, data, category, ts FROM another_table
-ORDER BY iceberg_bucket16(id)
-```
-
-If you're inserting data with DataFrame, you can use the function like below:
-
-```scala
-data.sortWithinPartitions(expr("iceberg_bucket16(id)"))
- .writeTo("prod.db.sample")
- .append()
-```
+There are 3 options for `write.distribution-mode`
+
+* `none` - This is the previous default for Iceberg.<p> This mode does not
require any shuffles or sort to be performed
+automatically by Spark. Because no work is done automatically by Spark, the
data must be either locally or globally
+sorted manually by partition value. To reduce the number of files produced
during writing, using a global sort is recommended.<p>
+A local sort can be avoided by using the Spark [write
fanout](#write-properties) property but this will cause all file handles to
+remain open until each write task has completed.
+* `hash` - This mode is the new default and requests that Spark uses a
hash-based exchange to shuffle the incoming
+write data before writing. Practically, this means that each row is hashed
based on the row's partition value and then placed
+in a corresponding Spark task based upon that value. Further division and
coalescing of tasks may take place based on
+the [Spark's Adaptive Query planning](#controlling-file-sizes).
+* `range` - This mode requests that Spark perform a range based exchanged to
shuffle the data before writing. This is
+a two stage procedure which is more expensive than the `hash` mode. The first
stage samples the data to be written based
+on the partition and sort columns, this information is then used in the second
stage to shuffle data into tasks. Each
+task gets an exclusive range of the input data which clusters the data by
partition and also globally sorts it.
+While this is more expensive than the hash distribution, the global ordering
can be beneficial for read performance if
+sorted columns are used during queries. Further division and coalescing of
tasks may take place based on
+ the [Spark's Adaptive Query planning](#controlling-file-sizes).
+
+
+## Controlling File Sizes
+
+When writing data to Iceberg with Spark, it's important to note that Spark
cannot write a file larger than a Spark
+task. This means although Iceberg will always roll over a file when it grows
to
+[`write.target-file-size-bytes`](../configuration/#write-properties), a file
+will not be able to grow to that size if the task is not large enough. The
+on disk file size will also be much smaller than the Spark task size since the
on disk data will be both compressed
+and in columnar format as opposed to Spark's uncompressed row representation.
This means a 100 megabyte task will
+always corrospond to on an on disk file of much less than 100 megabytes even
when writing to a single Iceberg partition.
+
+To control what data ends up in each task the user must either use a [`write
distribution mode`](#writing-distribution-modes)
+or manually repartition the data.
+
Review Comment:
For future readers I was wrong, the correct thing to do is to end the line
with two spaces
--
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]
