Configuring the Default Number of Data Blocks Divided by SparkSQL

Scenarios

By default, SparkSQL divides data into 200 data blocks during shuffle. In data-intensive scenarios, each data block may have excessive size. If a single data block of a task is larger than 2 GB, an error similar to the following will be reported while Spark attempts to fetch the data block:

Adjusted frame length exceeds 2147483647: 2717729270 - discarded

For example, setting the number of default data blocks to 200 causes SparkSQL to encounter an error in running a TPCDS 500-GB test. To avoid this, increase the number of default blocks in data-intensive scenarios.

Configuration parameters

1. Log in to FusionInsight Manager.

   For details, see Accessing FusionInsight Manager.

2. Choose Cluster > Services > Spark2x, and click Configurations and then All Configurations. Enter a parameter name in the search box.

   Table 1 Parameter description

   Parameter	Description	Example Value
   spark.sql.shuffle.partitions	Indicates the default number of blocks divided during shuffle.	200