[adding ceph-devel] On Wed, 25 Jun 2014, Federico Simoncelli wrote: > New version of the patch: > > http://github.com/simon3z/ceph/commit/ef42184 > > sadly not fully addressing the cluster problem (it is now configurable > but it is unique). It is clear that systemd doesn't support multiple > instance identifiers %i (we could have used one for the mon/osd/mds id > and one for the cluster) and it doesn't support parsing the single %i > (to be split in two). That is unfortunate, but so it goes. Any idea if this is something that systemd may get later? > Instances are not even that common in systemd, I think the way fedora would > integrate ceph is using the /etc/sysconfig/ceph file to configure the ids: > > CLUSTER="ceph" > MON_ID="ceph-node01" > OSD_ID="0" > MDS_ID="ceph-node01" > > and then just run: > > # systemctl start ceph-mon ceph-osd ceph-mds > > So I have the feeling that we're already giving an extra degree of freedom > by using instances to run multiple mon/osd/mds (even though I am not sure > how much that's useful if they have to be part of the same cluster). > > Actually, do you feel that having multiple mon/osd/mds is useless if they > have to be of the same cluster? I am talking about the regular deployments, > not testing/developing. Having multiple instances of daemons is definitely required. The vast majority of deployments have many OSDs per node. It's less common for the mon and mds daemons, but I think it is simplest to make them match. That being the case, I think your original suggestion of having the cluster specified somewhere in /etc/ceph or /etc/sysconfig/ceph is the right interim solution. Is there a way to set up meta-jobs that capture all instances of each daemon type, or all ceph daemons? Thanks! sage -- To unsubscribe from this list: send the line "unsubscribe ceph-devel" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
