Github user ppadma commented on a diff in the pull request:
https://github.com/apache/drill/pull/1101#discussion_r166141274
--- Diff:
exec/java-exec/src/main/java/org/apache/drill/exec/physical/impl/aggregate/HashAggTemplate.java
---
@@ -733,28 +780,32 @@ private void restoreReservedMemory() {
* @param records
*/
private void allocateOutgoing(int records) {
- // Skip the keys and only allocate for outputting the workspace values
- // (keys will be output through splitAndTransfer)
- Iterator<VectorWrapper<?>> outgoingIter = outContainer.iterator();
- for (int i = 0; i < numGroupByOutFields; i++) {
- outgoingIter.next();
- }
-
// try to preempt an OOM by using the reserved memory
useReservedOutgoingMemory();
long allocatedBefore = allocator.getAllocatedMemory();
- while (outgoingIter.hasNext()) {
+ for (int columnIndex = numGroupByOutFields; columnIndex <
outContainer.getNumberOfColumns(); columnIndex++) {
+ final VectorWrapper wrapper =
outContainer.getValueVector(columnIndex);
@SuppressWarnings("resource")
- ValueVector vv = outgoingIter.next().getValueVector();
+ final ValueVector vv = wrapper.getValueVector();
- AllocationHelper.allocatePrecomputedChildCount(vv, records,
maxColumnWidth, 0);
+ final RecordBatchSizer.ColumnSize columnSizer = new
RecordBatchSizer.ColumnSize(wrapper.getValueVector());
+ int columnSize;
+
+ if (columnSizer.hasKnownSize()) {
+ // For fixed width vectors we know the size of each record
+ columnSize = columnSizer.getKnownSize();
+ } else {
+ // For var chars we need to use the input estimate
+ columnSize = varcharValueSizes.get(columnIndex);
+ }
+
+ AllocationHelper.allocatePrecomputedChildCount(vv, records,
columnSize, 0);
--- End diff --
I think we should also get elementCount from sizer and use that instead of
passing 0.
---
