Details
-
Sub-task
-
Status: Closed
-
Major
-
Resolution: Fixed
-
None
-
None
Description
In SparkUtil.java, we set the SPARK_REDUCERS by system configuration
public static int getParallelism(List<RDD<Tuple>> predecessors, PhysicalOperator physicalOperator) { String numReducers = System.getenv("SPARK_REDUCERS"); if (numReducers != null) { return Integer.parseInt(numReducers); } int parallelism = physicalOperator.getRequestedParallelism(); if (parallelism <= 0) { // Parallelism wasn't set in Pig, so set it to whatever Spark thinks // is reasonable. parallelism = predecessors.get(0).context().defaultParallelism(); } return parallelism; }
It is better to set it by pig.properties