Adrian Nistor created HIVE-23841: ------------------------------------ Summary: Field writers is an HashSet, i.e., not thread-safe. Field writers is typically protected by synchronization on lock, but not in 1 location. Key: HIVE-23841 URL: https://issues.apache.org/jira/browse/HIVE-23841 Project: Hive Issue Type: Bug Environment: Any environment Reporter: Adrian Nistor
Field {{writers}} is a {{HashSet}} ([line 70|https://github.com/apache/hive/blob/c93d7797329103d6c509bada68b6da7f907b3dee/ql/src/java/org/apache/hadoop/hive/llap/LlapOutputFormatService.java#L70]), i.e., not thread-safe. Accesses to field {{writers}} are protected by synchronization on {{lock}}, e.g., at lines: [141-144|https://github.com/apache/hive/blob/c93d7797329103d6c509bada68b6da7f907b3dee/ql/src/java/org/apache/hadoop/hive/llap/LlapOutputFormatService.java#L141-L144], [212-213|https://github.com/apache/hive/blob/c93d7797329103d6c509bada68b6da7f907b3dee/ql/src/java/org/apache/hadoop/hive/llap/LlapOutputFormatService.java#L212-L213], and [212-215|https://github.com/apache/hive/blob/c93d7797329103d6c509bada68b6da7f907b3dee/ql/src/java/org/apache/hadoop/hive/llap/LlapOutputFormatService.java#L212-L215]. However, the {{writers.remove()}} at [line 249|https://github.com/apache/hive/blob/c93d7797329103d6c509bada68b6da7f907b3dee/ql/src/java/org/apache/hadoop/hive/llap/LlapOutputFormatService.java#L249] is protected by synchronization on {{INSTANCE}}, *not* on {{lock}}. Synchronizing on 2 different objects does not ensure mutual exclusion. This is because 2 threads synchronizing on different objects can still execute in parallel at the same time. Note that lines [215|https://github.com/apache/hive/blob/c93d7797329103d6c509bada68b6da7f907b3dee/ql/src/java/org/apache/hadoop/hive/llap/LlapOutputFormatService.java#L215] and [249|https://github.com/apache/hive/blob/master/ql/src/java/org/apache/hadoop/hive/llap/LlapOutputFormatService.java#L249] are modifying {{writers}} with {{put()}} and {{remove()}}, respectively. h1. The Code for This Fix This fix is very simple: just change {{synchronized (INSTANCE)}} to {{synchronized (lock)}}, just like the methods containing the other lines listed above. -- This message was sent by Atlassian Jira (v8.3.4#803005)