jprdonnelly/impala
1
0
Fork 0
mirror of https://github.com/apache/impala.git synced 2025-12-19 18:12:08 -05:00
Code Issues Projects Releases Wiki Activity

IMPALA-8299: Fix crash in GroupingAggregator on uninited hash table

Browse Source 
If HashTable::Init() fails, in some cases we may try to call
HashTable::Close() on the table in GroupingAggregator::Close(), which
can cause a crash.

The solution is to set the hash table to a nullptr if Init() fails.

I also verified that the other use of HashTable, PhjBuilder, handles
this case correctly.

Testing:
- Manually verified that the crash no longer occurs if HashTable::Init
  fails.

Change-Id: I381657c43f93eb4871b93d54532cb886198fd0b2
Reviewed-on: http://gerrit.cloudera.org:8080/12746
Reviewed-by: Impala Public Jenkins <impala-public-jenkins@cloudera.com>
Tested-by: Impala Public Jenkins <impala-public-jenkins@cloudera.com>
This commit is contained in:
Thomas Tauber-Marshall
2019-03-13 20:38:48 +00:00
committed by Gabor Kaszab
parent e8c6adcc2b
commit 030c445ac9
2 changed files with 8 additions and 1 deletions
Download Patch File Download Diff File Expand all files Collapse all files

7
be/src/exec/grouping-aggregator-partition.cc
View File

@@ -89,7 +89,12 @@ Status GroupingAggregator::Partition::InitHashTable(bool* got_memory) {
1L << (32 - NUM_PARTITIONING_BITS), PAGG_DEFAULT_HASH_TABLE_SZ));
// Please update the error message in CreateHashPartitions() if initial size of
// hash table changes.
return hash_tbl->Init(got_memory);
Status status = hash_tbl->Init(got_memory);
if (!status.ok() || !(*got_memory)) {
hash_tbl->Close();
hash_tbl.reset();
}
return status;
}
Status GroupingAggregator::Partition::SerializeStreamForSpilling() {

2
be/src/exec/grouping-aggregator.h
View File

@@ -346,6 +346,8 @@ class GroupingAggregator : public Aggregator {
/// Initializes the hash table. 'aggregated_row_stream' must be non-NULL.
/// Sets 'got_memory' to true if the hash table was initialised or false on OOM.
/// After returning, 'hash_tbl' will be non-null iff 'got_memory' is true and the
/// returned status is OK.
Status InitHashTable(bool* got_memory) WARN_UNUSED_RESULT;
/// Called in case we need to serialize aggregated rows. This step effectively does