Github user kevinjmh commented on a diff in the pull request:
https://github.com/apache/carbondata/pull/2900#discussion_r230972236
--- Diff:
hadoop/src/main/java/org/apache/carbondata/hadoop/api/CarbonTableInputFormat.java
---
@@ -575,6 +576,8 @@ private BitSet setMatchedPartitions(String
partitionIds, Expression filter,
*/
public BlockMappingVO getBlockRowCount(Job job, CarbonTable table,
List<PartitionSpec> partitions) throws IOException {
+ // no useful information for count star query without filter, so
disable explain collector
+ ExplainCollector.remove();
--- End diff --
You are right.
Normal query flow goes to `CarbonInputFormat#getPrunedBlocklets` and
initialize the pruning info for table we queried. But count star query without
filter use a different query plan, it does not go into that method, so no
pruning info does not init. When it calls default data map to prune(with a null
filter), exception will occur during settingg pruning info.
One solution is to init the pruning info for this type of query in mrthod
`getBlockRowCount`. But considering
no useful information about block/blocklet pruning for such query(actually
no pruning), I choose to disable the expalin collector instead.
---
