MyBatis Hanging?

[Rob Ogilvie]

List,

I seem to have encountered an odd situation in which MyBatis is hanging.  It's working fine for select statements, but the update statement I'm calling is hanging.  I set the timeout to 2000 on the update to ensure it wasn't the database hanging.

Is there a good place to start troubleshooting this?  I walked through the code in debug mode and it hits the mapper.updateInfo(updatedObject) and sits there forever; the servlet handler timed out 28 minutes, 33 seconds later and returned nothing to the client.

Rob

[Rob Ogilvie]

I appear to have stumbled upon a bug, but would like confirmation before I report it to a bug tracker.

The cause of the problem is I had a different session locking the table I was trying to update.  I believe the timeout of 2000 should have caused a failure in the event of a lock that doesn't clear within the timeout... shouldn't it?

Rob

[Rob Ogilvie]

....Oracle, BTW.

[Clinton Begin]

Timeout may not work for update statements on many RDBMS.  A lot of databases support putting the timeout directly in the SQL though.

Is there any way you can confirm this behavior with a select statement?

Clinton

[Rob Ogilvie]

I can't think of a way to lock a resource against SELECT statements in Oracle.

Ideally, MyBatis should implement its own timeout, shouldn't it?

Rob

[Rob Ogilvie]

If an update locks some resources, they're locked for writing.  A select on those resources will show you the pre-update version (until you commit, at which point it will show you the post-update version) of the data.

Rob

On Mon, Jul 12, 2010 at 01:50, Giovanni Cuccu <giovann...@gmail.com> wrote:

Hi,
   correct me if I'm wrong but a select for update locks resources,
and if I remerember correctly even a select on a Oracle RAC system may
lock some resources.
I'm using Oracle and the statement timeout property works as expect
even when there are waitings related to database locks.
ciao,
   Giovanni

[Larry Meadors]

No, "select for update" in Oracle is a different animal - it locks the
rows until the next rollback or commit.

Larry

[Giovanni Cuccu]

Yes, and it's not an Oracle specific feature, see
http://www.postgresql.org/docs/8.1/static/sql-select.html#SQL-FOR-UPDATE-SHARE
for an example.
Giovanni

--------------------------------------------------------------------
"You don't know the power of dark side" - Darth Vader

[Rob Ogilvie]

Is that what you would like tested?  A SELECT FOR UPDATE doesn't lock it for SELECT statements... only for writing actions.  As far as I know, there's no way to lock a table for read operations in Oracle.

Rob

[Giovanni Cuccu]

yes, you're right, there is no way to prevent a lock for read other
than select for update.
in one of your email you said

The cause of the problem is I had a different session locking the
table I was trying to update. I believe the timeout of 2000 should
have caused a failure in the event of a lock that doesn't clear within
the timeout... shouldn't it?

What I understood is that you were expecting to statementTimeout to
work properly and raise an exception when different locks on the table
were preventing the update within n seconds.
If so I can say that with version 2 of MyBatis the I experienced a
timeout in the situation you described when using the statementTimeout
property.
I have configured it in productions systems ad sometimes we see an
oracle error reporting the cancel of the operation caused by the
timeout.
If I did understand your problem the cause does not seems a Mybatis
bug, so you should create a simple test case and someone on the list
can try to reproduce it and check if it's a bug or not.
ciao,
Giovanni

--------------------------------------------------------------------
"You don't know the power of dark side" - Darth Vader