Thread: refcount leak warnings

I have a PL/Java user that performs some lengthy operations. Eventually, 
he get warnings like:

WARNING: buffer refcount leak: [779] (rel=1663/16440/52989, blockNum=3, 
flags=0x27, refcount=1 2)

I traced this to the function PrintBufferLeakWarning. AFAICS, it's only 
called from the function ResourceOwnerReleaseInternal under the 
following comment:

* During a commit, there shouldn't be any remaining pins --- that
* would indicate failure to clean up the executor correctly --- so
* issue warnings.    In the abort case, just clean up quietly.

I have no idea where to go from here. What should I look for when trying 
to find the cause of such warnings?

Regards,
Thomas Hallgren

On Tue, Jun 27, 2006 at 03:55:06PM +0200, Thomas Hallgren wrote:
> I have a PL/Java user that performs some lengthy operations. Eventually,
> he get warnings like:
>
> WARNING: buffer refcount leak: [779] (rel=1663/16440/52989, blockNum=3,
> flags=0x27, refcount=1 2)

I think the comment about failing to clean up is correct. AIUI, if
you've done a heap_open without a heap_close, or an index_open without
an index_close, you'll get warnings like this. Maybe using SPI without
clearing everything.

Obviously something didn't get cleaned up somewhere, but what? Maybe
identifying the rel and looking at that block might help identify the
issue.

Have a ncie day,
--
Martijn van Oosterhout   <kleptog@svana.org>   http://svana.org/kleptog/
> From each according to his ability. To each according to his ability to litigate.

Thomas Hallgren <thomas@tada.se> writes:
> I have a PL/Java user that performs some lengthy operations. Eventually, 
> he get warnings like:

> WARNING: buffer refcount leak: [779] (rel=1663/16440/52989, blockNum=3, 
> flags=0x27, refcount=1 2)

Look for ReadBuffer calls not matched by ReleaseBuffer.
        regards, tom lane