CarbonData as Spark's Datasource

The CarbonData fileformat is now integrated as Spark datasource for read and write operation without using CarbonSession. This is useful for users who wants to use carbondata as spark's data source.

Note: You can only apply the functions/features supported by spark datasource APIs, functionalities supported would be similar to Parquet. The carbon session features are not supported. The result is displayed as byte array format when select query on binary column in spark-sql.

Create Table with DDL

Now you can create Carbon table using Spark's datasource DDL syntax.

 CREATE [TEMPORARY] TABLE [IF NOT EXISTS] [db_name.]table_name
     [(col_name1 col_type1 [COMMENT col_comment1], ...)]
     [OPTIONS (key1=val1, key2=val2, ...)]
     [PARTITIONED BY (col_name1, col_name2, ...)]
     [CLUSTERED BY (col_name3, col_name4, ...) INTO num_buckets BUCKETS]
     [LOCATION path]
     [COMMENT table_comment]
     [TBLPROPERTIES (key1=val1, key2=val2, ...)]
     [AS select_statement]

Supported OPTIONS

Property Default Value Description
table_blocksize 1024 Size of blocks to write onto hdfs. For more details, see Table Block Size Configuration.
table_blocklet_size 64 Size of blocklet to write.
table_page_size_inmb 0 Size of each page in carbon table, if page size crosses this value before 32000 rows, page will be cut to that many rows. Helps in keep page size to fit cache size
local_dictionary_threshold 10000 Cardinality upto which the local dictionary can be generated. For more details, see Local Dictionary Configuration.
local_dictionary_enable false Enable local dictionary generation. For more details, see Local Dictionary Configuration.
sort_columns all dimensions are sorted Columns to include in sort and its order of sort. For more details, see Sort Columns Configuration.
sort_scope local_sort Sort scope of the load.Options include no sort, local sort, batch sort, and global sort. For more details, see Sort Scope Configuration.
long_string_columns null Comma separated string/char/varchar columns which are more than 32k length. For more details, see String longer than 32000 characters.

NOTE: please set long_string_columns for varchar column.



Using DataFrame

Carbon format can be used in dataframe also. Following are the ways to use carbon format in dataframe.

Write carbon using dataframe


Read carbon using dataframe

val df ="carbon").load(path)


import org.apache.spark.sql.SparkSession

val spark = SparkSession
  .appName("Spark SQL basic example")
  .config("spark.some.config.option", "some-value")

// For implicit conversions like converting RDDs to DataFrames
import spark.implicits._
val df = spark.sparkContext.parallelize(1 to 10 * 10 * 1000)
     .map(x => (r.nextInt(100000), "name" + x % 8, "city" + x % 50, BigDecimal.apply(x % 60)))
      .toDF("ID", "name", "city", "age")
// Write to carbon format      

// Read carbon using dataframe
val dfread ="carbon").load("/user/person_table")

Reference : list of carbon properties