On Tue, Jun 28, 2016 at 10:25 PM, João Castro <castrofjoao@xxxxxxxxx> wrote:
> Hello guys,
> From time to time I have MDS cache pressure error (Client failing to respond
> to cache pressure).
>
> If I try to increase the mds_cache_size to increase the number of inodes two
> things happen:
>
> 1) inodes keep growing until I get to the limit again
> 2) the more inodes, mds runs out of memory and gets killed by the OS, swaps
> to the failover mds.
The thing to do is to investigate why the client seems to be
misbehaving. Is it a kernel client or ceph-fuse? And what version?
John
> I have two mds, each with 12gb of ram. Running on a VM.
>
> root@mds01:~# ceph daemon mds.mds01 perf dump | head -n 30
> {
> "mds": {
> "request": 44328985,
> "reply": 44328983,
> "reply_latency": {
> "avgcount": 44328983,
> "sum": 155060.386784153
> },
> "forward": 0,
> "dir_fetch": 52149,
> "dir_commit": 7306,
> "dir_split": 0,
> "inode_max": 3000000,
> "inodes": 2884302,
> "inodes_top": 1191759,
> "inodes_bottom": 485982,
> "inodes_pin_tail": 1206561,
> "inodes_pinned": 2398701,
> "inodes_expired": 187542007,
> "inodes_with_caps": 2395109,
> "caps": 2395110,
> "subtrees": 2,
> "traverse": 44376013,
> "traverse_hit": 44305134,
> "traverse_forward": 0,
> "traverse_discover": 0,
> "traverse_dir_fetch": 22494,
> "traverse_remote_ino": 0,
> "traverse_lock": 0,
> "load_cent": 4434414393,
>
>
> Any idea how can I overcome this problem?
>
> _______________________________________________
> ceph-users mailing list
> ceph-users@xxxxxxxxxxxxxx
> http://lists.ceph.com/listinfo.cgi/ceph-users-ceph.com
_______________________________________________
ceph-users mailing list
ceph-users@xxxxxxxxxxxxxx
http://lists.ceph.com/listinfo.cgi/ceph-users-ceph.com
