Thread: Major Memory Leak in PostgreSQL JDBC Driver

Major Memory Leak in PostgreSQL JDBC Driver

From
"Gerald Gutierrez"
Date:
At least the 7.1 RC3 driver has a memory leak in that the
.../jdbc2/PreparedStatement.java class uses Java's ThreadLocal object for
"optimization" reasons. However, Sun's J2SE 1.3 leaks memory whenever this
class is used. Hence, repeated use of the JDBC driver's PreparedStatement
class will cause a major memory leak.

Our application running on the Orion application server using PostgreSQL
will crash with an OutOfMemoryError in less than an hour while stress
testing.

Our workaround is to disable the ThreadLocal object by patching the driver
before using it. The patch is simple and small; instead of using the
ThreadLocal object to retrieve a thread-local SimpleDateFormat instance so
that it doesn't have to be instantiated each time (this is the
optimization), just instantiate the SimpleDateFormat class each time. We've
noticed no performance loss and our application no longer has any memory
leaks.

Please consider fixing this problem before releasing 7.1.

Thanks.


Gerald.

Re: Major Memory Leak in PostgreSQL JDBC Driver

From
Peter T Mount
Date:
Quoting Gerald Gutierrez <gutz@kalador.com>:

>
> At least the 7.1 RC3 driver has a memory leak in that the
> .../jdbc2/PreparedStatement.java class uses Java's ThreadLocal object
> for
> "optimization" reasons. However, Sun's J2SE 1.3 leaks memory whenever
> this
> class is used. Hence, repeated use of the JDBC driver's
> PreparedStatement
> class will cause a major memory leak.

This has already been reported (and there now exists a bug registered with Sun).

As soon as I can I'm going to remove the need for ThreadLocal from the driver.

> that it doesn't have to be instantiated each time (this is the
> optimization), just instantiate the SimpleDateFormat class each time.
> We've
> noticed no performance loss and our application no longer has any
> memory
> leaks.

There was some discussion about this some time ago, and the concensus then was
to limit the number of instanciated SimpleDateFormat's to a minimum.

> Please consider fixing this problem before releasing 7.1.

It should be (assuming I get some spare time between now and then - ie why I'm
a week behind on emails).

Peter

--
Peter Mount peter@retep.org.uk
PostgreSQL JDBC Driver: http://www.retep.org.uk/postgres/
RetepPDF PDF library for Java: http://www.retep.org.uk/pdf/