On Friday 26 March 2010 wrote Krzysztof Strasburger: > On Fri, Mar 26, 2010 at 12:08:48PM +0100, Amon Ott wrote: > > I can confirm this problem on our test system. glusterfs goes up to the > > memory limit set for performance/io-cache (1.2 GB here), logs out of > > memory errors and further access fails with "Stale NFS file handle". The > > above command did not help, only remounting did. > > > > So I assume some memory leak in the io-cache filter. > > No, this should happen without any filters. What happens, if you disable > io-cache? Disabled both io-cache and read-ahead. Now it only grew to 739 MB. Again, drop_caches did not reduce that size, although it reduced the fuse_inode slabs significantly. No out of memory now. My test is a backup run of some linux kernel source trees, so many small files and dirs. Data is on four servers, distributed over two mirrors of two nodes each. We would like to use glusterfs for distributed storage in a cluster of application servers - but this bug is a show stopper, because it leads to denial of service. Please note that I can live with 1GB of memory for the glusterfs process, if it needs that. It must not block accesses, this is my concern. 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
