On Monday 29 March 2010 wrote Krzysztof Strasburger: > On Fri, Mar 26, 2010 at 03:40:27PM +0100, Amon Ott wrote: > > > Now we should check, whether fuse_forget() is called at all during the > > > test. I would not assume blindly that it is. > > > > Added a counter to fuse_forget(): > > [2010-03-26 15:34:31] N [fuse-bridge.c:3203:fuse_thread_proc] fuse: > > unmounting /home/user after 42732 forgets > > Great. BTW, is the number of forgets consistent with the number of touched > files? If the latter would be significantly larger, it could mean that the > kernel decided not to forget some inodes, for whatever reason. Hmmm. The > calls to forget could be also made after the umount request. The solution > would be to log each forget() with its time stamp. I know, this means > flooding the logfile, but for a test, we could live with it. But maybe not > needed. I remember running glusterfs under valgrind and there were lots of > still allocated memory before exiting. So fuse_forget() does not forget? 40161 forgets for non-dirs: [2010-03-29 15:22:14] N [fuse-bridge.c:3207:fuse_thread_proc] fuse: unmounting /home/user after 2839 dir and 40161 other forgets But 56473 files there backed up: find /home/user/ -type f| wc -l 56473 And 5324 dirs: find /home/user/ -type d| wc -l 5324 So either I get something wrong or fuse does not deliver all forgets. Still, some memory should be freed after a while, but memory usage only grows. If the dentry and inode structs in the gluster code follow the idea in the Linux kernel, it could be that the dentry cache is not cleaned up with forget. So repeated lookups find the existing dentry in the cache and the glusterfs process does not grow with repeated accesses, but memory does not get freed. Memory does not even seem to be freed when I delete complete directory trees, so I guess that something with the reference counters of dentries is wrong. The gluster code for inodes and dentries is too complicated for me to understand within some minutes, so I leave it to others who know it better. Amon Ott -- Dr. Amon Ott - m-privacy GmbH Am K?llnischen Park 1, 10179 Berlin Tel: +49 30 24342334 Fax: +49 30 24342336 Web: http://www.m-privacy.de Handelsregister: Amtsgericht Charlottenburg HRB 84946 Gesch?ftsf?hrer: Dipl.-Kfm. Holger Maczkowsky, Roman Maczkowsky GnuPG-Key-ID: EA898571
