Hmmmm. I have somewhat similar issues, and I'm not entirely happy with
what I've got, but let me fill you in.
Ceph supports NFS by launching instances of Ganesha-nfs. If you're using
managed services, this will be run out of the master Ceph container
image and the name of this container is rather long and ugly. You could
set up a systemd dependency on this name, but it's not a pretty thing to do.
I'm not sure how much putting an "nfs" tag on a server does other than
simply labelling it. On the other hand, you can configure NFS servers on
multiple ceph nodes and those nodes will always be available as NFS
servers . It's not a Highlander situation where only one can be active
at a time. I keep 2 going myself.
The documentation makes reference to the idea that keepalived could
failover NFS, but there are cautions. It's worth trying, though.
Tim
On 11/8/24 17:56, Alex Buie wrote:
Hello all,
Facing an issue with my first ceph with cephfs deployment.
I have 2 admin (mgr) hosts, 3 mon/mds hosts, and 3 OSD hosts with 3 OSDs
each.
I also have been using the OSDs to run the nfs service that my hypervisor
connects to.
I have these service definitions I've applied:
service_type: nfs
service_id: nfs
service_name: nfs.nfs
placement:
count: 1
label: nfs
spec:
port: 2049
virtual_ip: 172.19.19.165
service_type: ingress
service_id: nfs.nfs
service_name: ingress.nfs.nfs
placement:
count: 1
label: nfs
spec:
backend_service: nfs.nfs
monitor_port: 9049
virtual_ip: 172.19.19.165
keepalive_only: true
and I have tagged the 3 OSD nodes with the `nfs` label.
Generally, both the ingress and the nfs service start on the same node
(osd01, for example). But sometimes, errors occur with the nfs service and
it gets rescheduled and ends up on a different node (osd03, for example) -
and because the ingress service isn't running on this node, it breaks my
NFS mount and the hypervisor gets very unhappy.
Is there a way I can specify some kind of scheduling dependency between the
nfs.nfs and ingress.nfs.nfs services so that they will get scheduled to run
on the same node so that the virtual IP is present for nfs to bind to? Or,
am I doing something wrong here?
Thanks a bunch!
*Alex*
_______________________________________________
ceph-users mailing list -- ceph-users@xxxxxxx
To unsubscribe send an email to ceph-users-leave@xxxxxxx
_______________________________________________
ceph-users mailing list -- ceph-users@xxxxxxx
To unsubscribe send an email to ceph-users-leave@xxxxxxx
