From: Iannaccone, G. <gia...@in...> - 2009-07-14 22:50:02
|
Hi Oana, thanks for the detailed report. It appears that we are not handling correctly the calls to glob() leading to a large memory leak for directories with many files. I've attached a patch that should fix this problem. Unfortunately I cannot test whether it works right now (don't have the right setup here with me). cheers, gianluca > -----Original Message----- > From: Oana Goga [mailto:oan...@en...] > Sent: Tuesday, July 14, 2009 10:46 AM > To: com...@li... > Subject: [como-devel] Memory leak problem > > Hello, > > Lately I have been testing CoMo and I observed a memory leak problem > when this is running with flowtools sniffer. > > I am using CoMo v1.5 because the v2 is not working correctly with > flowtools sniffer (is only reading files till the current moment, and > after, when flowtools is generating new files they are not read by > CoMo). > > I observed that the problem is starting after CoMo has finished to read > all the flowtools files and is just waiting for new ones, and > eventually reads them when they appear. > I tested CoMo with valgrind, and this, indeed, showed that there are > memory leaks: > > LEAK SUMMARY: > ==25325== definitely lost: 8,391,809 bytes in 2,325 blocks. > ==25325== indirectly lost: 67,726,144 bytes in 1,026,184 blocks. > ==25325== possibly lost: 13,728 bytes in 14 blocks. > ==25325== still reachable: 202,062 bytes in 1,721 blocks. > ==25325== suppressed: 0 bytes in 0 blocks. > > > I attached to the email a png image that is showing the evolution of > the memory and also the log of valgrind. > > My concern is that after a few hours of running CoMo the system crashes > because of insufficient memory. > > Thank you, > Oana > > > > > |