baigd created HIVE-11917:
----------------------------
Summary: Report a possible race issue
Key: HIVE-11917
URL: https://issues.apache.org/jira/browse/HIVE-11917
Project: Hive
Issue Type: Bug
Reporter: baigd
Priority: Critical
Hi, Developers of apache/hive,
I am writing to report a race issue on use of ConcurrentHashMap. The issue is
reported by our tool in an automatic way. Although manually confirmed, it would
be a false positive, given we do not know the specification of the program. We
would very appreciate if you could check below for details and confirm with us
whether it is a real problem. For more information, please refer to our
website: http://sav.sutd.edu.sg/?page_id=2845
File:
"apache/hive/ql/src/java/org/apache/hadoop/hive/ql/exec/spark/SmallTableCache.java"
Location: Line (61,45/46/49）
Description:
The check-get-clear operations (line 45,46 and 49) in method "initialize()" is
guarded by lock "tableContainerMap". I assume the intention is to guarantee the
atomicity of the check-get-clear operation. However the put operation in line
61 is not properly locked, therefore it may break the atomicity of the
check-get-clear operation in method "initialize()". Relying on the
ConcurrentHashMap to ensure exclusive access is dangerous since
ConcurrentHashMap has no guarantee of exclusive access.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
