Help Center/ MapReduce Service/ Component Operation Guide (Normal)/ Using CarbonData (for MRS 3.x or Later)/ CarbonData FAQ/ Why INSERT INTO/LOAD DATA Task Distribution Is Incorrect and the Opened Tasks Are Less Than the Available Executors when the Number of Initial ExecutorsIs Zero?
Updated on 2022-09-15 GMT+08:00
View PDF

Why INSERT INTO/LOAD DATA Task Distribution Is Incorrect and the Opened Tasks Are Less Than the Available Executors when the Number of Initial Executors Is Zero?

Question

Why INSERT INTO or LOAD DATA task distribution is incorrect, and the openedtasks are less than the available executors when the number of initial executors is zero?

Answer

In case of INSERT INTO or LOAD DATA, CarbonData distributes one task per node. If the executors are not allocated from the distinct nodes then CarbonData will launch fewer tasks.

Solution:

Configure higher value for the executor memory and core so that the yarn can launch only one executor per node.

  1. Configure the number of the Executor cores.
    • Configure the spark.executor.cores in spark-defaults.conf or the SPARK_EXECUTOR_CORES in spark-env.sh appropriately.
    • Add --executor-cores NUM parameter to configure the cores during use the spark-submit command.
  2. Configure the Executor memory.
    • Configure the spark.executor.memory in spark-defaults.conf or the SPARK_EXECUTOR_MEMORY in spark-env.sh appropriately.
    • Add --executor-memory MEM parameter to configure the memory during use the spark-submit command.
Parent topic: CarbonData FAQ