You are viewing a plain text version of this content. The canonical link for it is here.
Posted to dev@openjpa.apache.org by "Pawel Veselov (Jira)" <ji...@apache.org> on 2023/03/08 19:40:00 UTC
[jira] [Created] (OPENJPA-2904) OpenJPAEntityManager.getConnection() may lead to sudden connection closure
Pawel Veselov created OPENJPA-2904:
--------------------------------------
Summary: OpenJPAEntityManager.getConnection() may lead to sudden connection closure
Key: OPENJPA-2904
URL: https://issues.apache.org/jira/browse/OPENJPA-2904
Project: OpenJPA
Issue Type: Bug
Affects Versions: 3.2.2, 3.2.1, 3.2.0
Reporter: Pawel Veselov
(looking at commit 05069dfee, but this code hasn't changed in ages).
There is a number of methods that chain to {{JDBCStore.getClientConnection()}}
* {{StoreManager.getClientConnection()}}
* {{StoreContext.getConnection()}}
* {{OpenJPAEntityManager.getConnection()}}
* {{EntityManagerImpl.unwrap()}}
Callers of these methods typically use the retrieved connection object for an operation, and then let go of it, without any cleanup (as they are supposed to).
However, {{JDBCStore.getClientConnection()}} implementation creates an instance of {{JDBCStoreManager.ClientConnection}} class every time it's called. These instances, considering how they are used, are therefore short-lived
However, the implementation {{JDBCStoreManager.ClientConnection}} closes the underlying connection object on finalization.
This has a relatively high chance of having the underlying connection closed before it's actually properly released.
I don't necessarily know what the right fix for this is, but locally I will just make ClientConnection a field of RefCountConnection, it's a lightweight class so I can create it right away, then I'll have {{JDBCStore.getClientConnection()}} just return that. There may be some gotchas there, but I just need a quick solution because this suddenly became a persistent problem.
I've overridden the pooled connection implementation to try to trace the connection random connection closure that we are experiencing under some random scenarios, and I attached an exception to the connection object when it's closed, and I see when an attempt to use a closed connection happens - it's because the GC closed it through {{JDBCStoreManager.ClientConnection}}.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)