On Tue, Feb 19, 2008 at 08:17:27AM -0700, Brad Nicholes wrote: > > All of the other memory leak fixes in 3.1.0 were specific to that code > base. Although there might be something similar going on in 3.0.x. > The other memory leak fixes dealt with the XDR functions that create > and free the XDR data. There were instances in some of the new code > that I wrote where XDR data structures were being created but not > freed. There could be similar instances in the 3.0.x code base. We > would just have to take a closer look at the code path that begins > from process_udp_recv_channel() when a metric packet is being read and > stored by other gmond nodes.
I still haven't figured out where we should be freeing this memory, or how we're dropping the pointers on the floor (mostly due to still figuring out how gmond works, and how XDR works). It is trivially reproducible. Just inject any metric you want with the spoof "-S" argument. Both of the strings will be leaked. E.g., gmetric -n "bleh" -v 5 -t uint8 -u "goobers" -S 10.0.0.1:myhost valgrind will report two blocks for a total of 16 bytes as being lost. "10.0.0.1\0" and "myhost\0" would be the blocks, I believe. mh ------------------------------------------------------------------------- This SF.net email is sponsored by: Microsoft Defy all challenges. Microsoft(R) Visual Studio 2008. http://clk.atdmt.com/MRT/go/vse0120000070mrt/direct/01/ _______________________________________________ Ganglia-developers mailing list [email protected] https://lists.sourceforge.net/lists/listinfo/ganglia-developers
