ODIn/spark-instrumented-optimizer
1
0
Fork 0
Code Issues Pull requests Projects Releases Wiki Activity

[SPARK-36089][SQL][DOCS] Update the SQL migration guide about encoding auto-detection of CSV files

Browse source 
### What changes were proposed in this pull request?
In the PR, I propose to update the SQL migration guide, in particular the section about the migration from Spark 2.4 to 3.0. New item informs users about the following issue:

**What**: Spark doesn't detect encoding (charset) in CSV files with BOM correctly. Such files can be read only in the multiLine mode when the CSV option encoding matches to the actual encoding of CSV files. For example, Spark cannot read UTF-16BE CSV files when encoding is set to UTF-8 which is the default mode. This is the case of the current ES ticket.

**Why**: In previous Spark versions, encoding wasn't propagated to the underlying library that means the lib tried to detect file encoding automatically. It could success for some encodings that require BOM presents at the beginning of files. Starting from the versions 3.0, users can specify file encoding via the CSV option encoding which has UTF-8 as the default value. Spark propagates such default to the underlying library (uniVocity), and as a consequence this turned off encoding auto-detection.

**When**: Since Spark 3.0. In particular, the commit 2df34db586 causes the issue.

**Workaround**: Enabling the encoding auto-detection mechanism in uniVocity by passing null as the value of CSV option encoding. A more recommended approach is to set the encoding option explicitly.

### Why are the changes needed?
To improve user experience with Spark SQL. This should help to users in their migration from Spark 2.4.

### Does this PR introduce _any_ user-facing change?
No

### How was this patch tested?
Should be checked by building docs in GA/jenkins.

Closes #33300 from MaxGekk/csv-encoding-migration-guide.

Authored-by: Max Gekk <max.gekk@gmail.com>
Signed-off-by: Hyukjin Kwon <gurwls223@apache.org>
(cherry picked from commit e788a3fa88)
Signed-off-by: Hyukjin Kwon <gurwls223@apache.org>
This commit is contained in:
Max Gekk 2021-07-12 18:54:39 +09:00 committed by Hyukjin Kwon
parent 969c691601
commit 4c7ac5fc90
1 changed files with 2 additions and 0 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

2
docs/sql-migration-guide.md
View file

@ -326,6 +326,8 @@ license: |
- In Spark 3.0, when Avro files are written with user provided non-nullable schema, even the catalyst schema is nullable, Spark is still able to write the files. However, Spark throws runtime NullPointerException if any of the records contains null.
- In Spark version 2.4 and below, CSV datasource can detect encoding of input files automatically when the files have BOM at the beginning. For instance, CSV datasource can recognize UTF-8, UTF-16BE, UTF-16LE, UTF-32BE and UTF-32LE in the multi-line mode (the CSV option `multiLine` is set to `true`). In Spark 3.0, CSV datasource reads input files in encoding specified via the CSV option `encoding` which has the default value of UTF-8. In this way, if file encoding doesn't match to the encoding specified via the CSV option, Spark loads the file incorrectly. To solve the issue, users should either set correct encoding via the CSV option `encoding` or set the option to `null` which fallbacks to encoding auto-detection as in Spark versions before 3.0.
### Others
- In Spark version 2.4, when a Spark session is created via `cloneSession()`, the newly created Spark session inherits its configuration from its parent `SparkContext` even though the same configuration may exist with a different value in its parent Spark session. In Spark 3.0, the configurations of a parent `SparkSession` have a higher precedence over the parent `SparkContext`. You can restore the old behavior by setting `spark.sql.legacy.sessionInitWithConfigDefaults` to `true`.