I found there is an option `mds_health_summarize_threshold` so it could show the clients that are lagging.
I increased the default value. I ran `ceph daemon perf dump` to make sure `inodes < inodes_max`. The problem still persists. I'll try looking into the code for clues. Thanks On Fri, Nov 11, 2016 at 1:57 PM Goncalo Borges <[email protected]> wrote: > Doesn't the mds log tell you which clients ids are with problems? > > Does you mds has enough RAM so that you can increase the default value > 100000 of the mds cache size > ? > > Cheers > G. > > ------------------------------ > *From:* Yutian Li [[email protected]] > *Sent:* 11 November 2016 14:03 > *To:* Goncalo Borges; [email protected] > *Subject:* Re: [ceph-users] Locating CephFS clients in warn message > > As for now, when I run `dump_ops_in_flight`, `ops` in empty and `num_ops` > is 0. > But when I run `ceph status`, I still get 15 clients failing to respond to > cache pressure. > > Where should I start solving this problem? > > On Thu, Nov 10, 2016 at 6:16 PM Goncalo Borges < > [email protected]> wrote: > > Hi > > "ceph daemon mds.<id> session ls", executed in your mds server, should > give you hostname and client id of all your cephfs clients. > > "ceph daemon mds.<id> dump_ops_in_flight" should give you operations not > completed or pending to complete for certain clients ids. In case of > problems, that those problematic clients will probably appear there. > > Cheers > Goncalo > > > ________________________________________ > From: ceph-users [[email protected]] on behalf of Yutian > Li [[email protected]] > Sent: 10 November 2016 15:21 > To: [email protected] > Subject: [ceph-users] Locating CephFS clients in warn message > > I get a HEALTH_WARN when I run `ceph status`. It says > > health HEALTH_WARN > mds0: Many clients (17) failing to respond to cache pressure > > I have 50 OSDs, 3 MONs, and 1 MDS. I just use CephFS and attach it to 20 ~ > 30 clients using kernel mount option. > > I wonder how to locate those "many" clients that are failing to respond. I > don't even see an ID of the lagging clients anywhere. > > Thanks! > >
_______________________________________________ ceph-users mailing list [email protected] http://lists.ceph.com/listinfo.cgi/ceph-users-ceph.com
