Generic Load/Save Functions

• Manually Specifying Options
• Run SQL on files directly
• Save Modes
• Saving to Persistent Tables
• Bucketing, Sorting and Partitioning

In the simplest form, the default data source (parquet unless otherwise configured by spark.sql.sources.default) will be used for all operations.

val usersDF = spark.read.load("examples/src/main/resources/users.parquet")
usersDF.select("name", "favorite_color").write.save("namesAndFavColors.parquet")

Dataset<Row> usersDF = spark.read().load("examples/src/main/resources/users.parquet");
usersDF.select("name", "favorite_color").write().save("namesAndFavColors.parquet");

df = spark.read.load("examples/src/main/resources/users.parquet")
df.select("name", "favorite_color").write.save("namesAndFavColors.parquet")

df <- read.df("examples/src/main/resources/users.parquet")
write.df(select(df, "name", "favorite_color"), "namesAndFavColors.parquet")

Manually Specifying Options

You can also manually specify the data source that will be used along with any extra options that you would like to pass to the data source. Data sources are specified by their fully qualified name (i.e., org.apache.spark.sql.parquet), but for built-in sources you can also use their short names (json, parquet, jdbc, orc, libsvm, csv, text). DataFrames loaded from any data source type can be converted into other types using this syntax.

Please refer the API documentation for available options of built-in sources, for example, org.apache.spark.sql.DataFrameReader and org.apache.spark.sql.DataFrameWriter. The options documented there should be applicable through non-Scala Spark APIs (e.g. PySpark) as well. For other formats, refer to the API documentation of the particular format.

To load a JSON file you can use:

val peopleDF = spark.read.format("json").load("examples/src/main/resources/people.json")
peopleDF.select("name", "age").write.format("parquet").save("namesAndAges.parquet")

Dataset<Row> peopleDF =
  spark.read().format("json").load("examples/src/main/resources/people.json");
peopleDF.select("name", "age").write().format("parquet").save("namesAndAges.parquet");

df = spark.read.load("examples/src/main/resources/people.json", format="json")
df.select("name", "age").write.save("namesAndAges.parquet", format="parquet")

df <- read.df("examples/src/main/resources/people.json", "json")
namesAndAges <- select(df, "name", "age")
write.df(namesAndAges, "namesAndAges.parquet", "parquet")

To load a CSV file you can use:

val peopleDFCsv = spark.read.format("csv")
  .option("sep", ";")
  .option("inferSchema", "true")
  .option("header", "true")
  .load("examples/src/main/resources/people.csv")

Dataset<Row> peopleDFCsv = spark.read().format("csv")
  .option("sep", ";")
  .option("inferSchema", "true")
  .option("header", "true")
  .load("examples/src/main/resources/people.csv");

df = spark.read.load("examples/src/main/resources/people.csv",
                     format="csv", sep=":", inferSchema="true", header="true")

df <- read.df("examples/src/main/resources/people.csv", "csv", sep = ";", inferSchema = TRUE, header = TRUE)
namesAndAges <- select(df, "name", "age")

The extra options are also used during write operation. For example, you can control bloom filters and dictionary encodings for ORC data sources. The following ORC example will create bloom filter and use dictionary encoding only for favorite_color. For Parquet, there exists parquet.enable.dictionary, too. To find more detailed information about the extra ORC/Parquet options, visit the official Apache ORC/Parquet websites.

usersDF.write.format("orc")
  .option("orc.bloom.filter.columns", "favorite_color")
  .option("orc.dictionary.key.threshold", "1.0")
  .option("orc.column.encoding.direct", "name")
  .save("users_with_options.orc")

usersDF.write().format("orc")
  .option("orc.bloom.filter.columns", "favorite_color")
  .option("orc.dictionary.key.threshold", "1.0")
  .option("orc.column.encoding.direct", "name")
  .save("users_with_options.orc");

df = spark.read.orc("examples/src/main/resources/users.orc")
(df.write.format("orc")
    .option("orc.bloom.filter.columns", "favorite_color")
    .option("orc.dictionary.key.threshold", "1.0")
    .option("orc.column.encoding.direct", "name")
    .save("users_with_options.orc"))

df <- read.df("examples/src/main/resources/users.orc", "orc")
write.orc(df, "users_with_options.orc", orc.bloom.filter.columns = "favorite_color", orc.dictionary.key.threshold = 1.0, orc.column.encoding.direct = "name")

CREATE TABLE users_with_options (
  name STRING,
  favorite_color STRING,
  favorite_numbers array<integer>
) USING ORC
OPTIONS (
  orc.bloom.filter.columns 'favorite_color',
  orc.dictionary.key.threshold '1.0',
  orc.column.encoding.direct 'name'
)

Run SQL on files directly

Instead of using read API to load a file into DataFrame and query it, you can also query that file directly with SQL.

val sqlDF = spark.sql("SELECT * FROM parquet.`examples/src/main/resources/users.parquet`")

Dataset<Row> sqlDF =
  spark.sql("SELECT * FROM parquet.`examples/src/main/resources/users.parquet`");

df = spark.sql("SELECT * FROM parquet.`examples/src/main/resources/users.parquet`")

df <- sql("SELECT * FROM parquet.`examples/src/main/resources/users.parquet`")

Save Modes

Save operations can optionally take a SaveMode, that specifies how to handle existing data if present. It is important to realize that these save modes do not utilize any locking and are not atomic. Additionally, when performing an Overwrite, the data will be deleted before writing out the new data.

Saving to Persistent Tables

DataFrames can also be saved as persistent tables into Hive metastore using the saveAsTable command. Notice that an existing Hive deployment is not necessary to use this feature. Spark will create a default local Hive metastore (using Derby) for you. Unlike the createOrReplaceTempView command, saveAsTable will materialize the contents of the DataFrame and create a pointer to the data in the Hive metastore. Persistent tables will still exist even after your Spark program has restarted, as long as you maintain your connection to the same metastore. A DataFrame for a persistent table can be created by calling the table method on a SparkSession with the name of the table.

For file-based data source, e.g. text, parquet, json, etc. you can specify a custom table path via the path option, e.g. df.write.option("path", "/some/path").saveAsTable("t"). When the table is dropped, the custom table path will not be removed and the table data is still there. If no custom table path is specified, Spark will write data to a default table path under the warehouse directory. When the table is dropped, the default table path will be removed too.

Starting from Spark 2.1, persistent datasource tables have per-partition metadata stored in the Hive metastore. This brings several benefits:

• Since the metastore can return only necessary partitions for a query, discovering all the partitions on the first query to the table is no longer needed.
• Hive DDLs such as ALTER TABLE PARTITION ... SET LOCATION are now available for tables created with the Datasource API.

Note that partition information is not gathered by default when creating external datasource tables (those with a path option). To sync the partition information in the metastore, you can invoke MSCK REPAIR TABLE.

Bucketing, Sorting and Partitioning

For file-based data source, it is also possible to bucket and sort or partition the output. Bucketing and sorting are applicable only to persistent tables:

peopleDF.write.bucketBy(42, "name").sortBy("age").saveAsTable("people_bucketed")

peopleDF.write().bucketBy(42, "name").sortBy("age").saveAsTable("people_bucketed");

df.write.bucketBy(42, "name").sortBy("age").saveAsTable("people_bucketed")

CREATE TABLE users_bucketed_by_name(
  name STRING,
  favorite_color STRING,
  favorite_numbers array<integer>
) USING parquet
CLUSTERED BY(name) INTO 42 BUCKETS;

while partitioning can be used with both save and saveAsTable when using the Dataset APIs.

usersDF.write.partitionBy("favorite_color").format("parquet").save("namesPartByColor.parquet")

usersDF
  .write()
  .partitionBy("favorite_color")
  .format("parquet")
  .save("namesPartByColor.parquet");

df.write.partitionBy("favorite_color").format("parquet").save("namesPartByColor.parquet")

CREATE TABLE users_by_favorite_color(
  name STRING,
  favorite_color STRING,
  favorite_numbers array<integer>
) USING csv PARTITIONED BY(favorite_color);

It is possible to use both partitioning and bucketing for a single table:

usersDF
  .write
  .partitionBy("favorite_color")
  .bucketBy(42, "name")
  .saveAsTable("users_partitioned_bucketed")

peopleDF
  .write()
  .partitionBy("favorite_color")
  .bucketBy(42, "name")
  .saveAsTable("people_partitioned_bucketed");

df = spark.read.parquet("examples/src/main/resources/users.parquet")
(df
    .write
    .partitionBy("favorite_color")
    .bucketBy(42, "name")
    .saveAsTable("people_partitioned_bucketed"))

CREATE TABLE users_bucketed_and_partitioned(
  name STRING,
  favorite_color STRING,
  favorite_numbers array<integer>
) USING parquet
PARTITIONED BY (favorite_color)
CLUSTERED BY(name) SORTED BY (favorite_numbers) INTO 42 BUCKETS;

partitionBy creates a directory structure as described in the Partition Discovery section. Thus, it has limited applicability to columns with high cardinality. In contrast bucketBy distributes data across a fixed number of buckets and can be used when the number of unique values is unbounded.