[
http://issues.apache.org/jira/browse/DERBY-1025?page=comments#action_12369590 ]
Kathey Marsden commented on DERBY-1025:
---------------------------------------
So on XAResource.start() we need to flowcommit if
1) We are in autocommit mode.
conn_.autocommit_ == true
2) We are in a an active local Transaction.
conn_getXAState() == XA_T0_NOT_ASSOCIATED
conn_.inUnitOfWork_ == true
For my first shot at fixing this bug I added the folling code to
NetXAResource.start()
// DERBY-1025
// autocommit the local transaction before starting the global
transaction
// if autocommit is set
if (conn_.autoCommit_ && conn_.inActiveLocalTransaction())
{
conn_.flowAutoCommit();
}
Where in NetXAConnection we have
public boolean inActiveLocalTransaction() {
return (isInUnitOfWork_() &&
(getXAState() == XA_T0_NOT_ASSOCIATED));
}
And in Connection.java
/**
* @return Returns the inUnitOfWork_.
*/
public boolean isInUnitOfWork_() {
return inUnitOfWork_;
}
With my change, putting this code in I found some very strange behaviour in
this case in XATest
1) I got a commit on start when no local transaction was active after a
suspend
2) I seemed to lose a totally different suspended transaction in the process.
See comments for behaviour:
private static void interleavingTransactions(XADataSource xads) {
System.out.println("interleavingTransactions");
try {
XAConnection xac = xads.getXAConnection("sku", "testxa");
XAResource xar = xac.getXAResource();
Xid xid1 = XATestUtil.getXid(1, 93, 18);
Xid xid2 = XATestUtil.getXid(2, 45, 77);
xar.start(xid1, XAResource.TMNOFLAGS);
Connection conn = xac.getConnection();
Statement s = conn.createStatement();
s.executeUpdate("insert into APP.foo values (1)");
xar.end(xid1, XAResource.TMSUSPEND);
// commit occurred here because inUnitOfWork_ was true
// when it should be false.
xar.start(xid2, XAResource.TMNOFLAGS);
s.executeUpdate("insert into APP.foo values (2)");
xar.end(xid2, XAResource.TMSUSPEND);
xar.start(xid1, XAResource.TMRESUME);
s.executeUpdate("insert into APP.foo values (3)");
// XAER_NOTA on suspend of xid1.
// very strange because the resume worked ok.
// where did it go?
xar.end(xid1, XAResource.TMSUSPEND);
I am trying to understand:
1) When and how inUnitOfWork_ is set and its relation to XA
2) Why the autocommit after the suspension of xid2 caused xid1 to go away
after it resumed just fine.
All very strange.
> [xa] client XAResource.start() does not commit an active local transaction
> when auto commit is true
> ---------------------------------------------------------------------------------------------------
>
> Key: DERBY-1025
> URL: http://issues.apache.org/jira/browse/DERBY-1025
> Project: Derby
> Type: Bug
> Components: Network Client
> Reporter: Daniel John Debrunner
> Assignee: Kathey Marsden
>
> Embedded XAResource.start() implementation commits the active local
> transaction on the Connection associated with the XAResource if the
> connection is auto-commit mode.
> Client incorrectly throws an XAException with the XAER_RMFAIL error code (see
> DERBY-1024)
> XATest contains a work-around for client (calling commit) with a comment with
> this bug number.
--
This message is automatically generated by JIRA.
-
If you think it was sent incorrectly contact one of the administrators:
http://issues.apache.org/jira/secure/Administrators.jspa
-
For more information on JIRA, see:
http://www.atlassian.com/software/jira
