On Wed, Feb 11, 2015 at 5:30 AM, Dennis Kramer (DT) <dennis@xxxxxxxxx> wrote: > After setting the debug level to 2, I can see: > 2015-02-11 13:36:31.922262 7f0b38294700 2 mds.0.cache check_memory_usage > total 58516068, rss 57508660, heap 32676, malloc 1227560 mmap 0, baseline > 39848, buffers 0, max 67108864, 8656261 / 9999931 inodes have caps, 10367318 > caps, 1.03674 caps per inode > > It doesn't look like it has serious memory problems, unless my > interpretation is wrong of the output. The MDS currently requests trimming based on simple dentry counts rather than actual amount of memory in use. That's a configurable; I believe mds_max_cache_size? It defaults to 100,000. Looks like you've already increased it to more like 10 million. You can go to the clients and run the "status" and "dump_cache" commands on their admin sockets and see if the kernel is holding references to their inodes, preventing cap releases. > It looks like I have the same symptoms as: > http://tracker.ceph.com/issues/10151 > > I'm running 0.87 on all my nodes. That bug is just about whether the health warnings for it show up, and is resolved in v0.89, so you seeing it is expected. -Greg _______________________________________________ ceph-users mailing list ceph-users@xxxxxxxxxxxxxx http://lists.ceph.com/listinfo.cgi/ceph-users-ceph.com
 |