Github user Ben-Zvi commented on a diff in the pull request:
https://github.com/apache/drill/pull/938#discussion_r138236250
--- Diff:
exec/java-exec/src/main/java/org/apache/drill/exec/ExecConstants.java ---
@@ -92,18 +92,20 @@
// Hash Aggregate Options
- String HASHAGG_NUM_PARTITIONS = "drill.exec.hashagg.num_partitions";
String HASHAGG_NUM_PARTITIONS_KEY = "exec.hashagg.num_partitions";
LongValidator HASHAGG_NUM_PARTITIONS_VALIDATOR = new
RangeLongValidator(HASHAGG_NUM_PARTITIONS_KEY, 1, 128); // 1 means - no spilling
- String HASHAGG_MAX_MEMORY = "drill.exec.hashagg.mem_limit";
String HASHAGG_MAX_MEMORY_KEY = "exec.hashagg.mem_limit";
LongValidator HASHAGG_MAX_MEMORY_VALIDATOR = new
RangeLongValidator(HASHAGG_MAX_MEMORY_KEY, 0, Integer.MAX_VALUE);
// min batches is used for tuning (each partition needs so many batches
when planning the number of partitions,
// or reserve this number when calculating whether the remaining
available memory is too small and requires a spill.)
// Low value may OOM (e.g., when incoming rows become wider), higher
values use fewer partitions but are safer
- String HASHAGG_MIN_BATCHES_PER_PARTITION =
"drill.exec.hashagg.min_batches_per_partition";
- String HASHAGG_MIN_BATCHES_PER_PARTITION_KEY =
"drill.exec.hashagg.min_batches_per_partition";
- LongValidator HASHAGG_MIN_BATCHES_PER_PARTITION_VALIDATOR = new
RangeLongValidator(HASHAGG_MIN_BATCHES_PER_PARTITION_KEY, 2, 5);
+ String HASHAGG_MIN_BATCHES_PER_PARTITION_KEY =
"exec.hashagg.min_batches_per_partition";
+ LongValidator HASHAGG_MIN_BATCHES_PER_PARTITION_VALIDATOR = new
RangeLongValidator(HASHAGG_MIN_BATCHES_PER_PARTITION_KEY, 1, 5);
+ // Can be turns off mainly for testing. Memory prediction is used to
decide on when to spill to disk; with this option off,
--- End diff --
Done
---
