Hi,
On 7/21/21 9:40 AM, mabi wrote:
Hello,
I need to relocate an Octopus (15.2.13) ceph cluster of 8 nodes to another
internal network. This means that the IP address of each nodes as well as the
domain name will change. The hostname itself will stay the same.
What would be the best steps in order to achieive that from the ceph side? I
did not find any relevant documentation to such a procedure.
This cluster is installed with cephadm and runs on Ubuntu 20.04.
You need to ensure that TCP traffic is routeable between the networks
for the migration. OSD only hosts are trivial, an OSD updates its IP
information in the OSD map on boot. This should also be the case for
MDS, MGR, RGW, mirror agents and other services.
MONs are a different beast. You cannot change the IP address of an
existing mon (or with way too much effort). I would thus recommend to
remove one mon, migrate the machine to the new network, and re-add the
mon afterwards. Repeat with the other mons etc.
You will also need to adapt ceph.conf on each host (servers and clients)
to the new IP addresses or update the DNS configuration if you use SRV
entries. Running instances will be updated automatically (notification
of changes in the mon map), but newly started clients/services might
fail if they try to use the old IP addresses. This is why ceph.conf is
create by puppet in our setup......
One client with special needs is openstack cinder. The database entries
contain the mon list for volumes, and I'm not aware of a good method to
update them except manipulating the database. If you also use ceph for
nova (root disks / ephemerals), the problem might also be present there.
libvirt also has a list of mon hosts in the domain description for each
instance that is using ceph based block storage. I'm not 100% sure
whether this is updated e.g. during a live migration. Or what will
happen if you shutdown an instance and restart it after the migration
(domain xml file will not be recreated and might contain stale
information...).
Regards,
Burkhard
_______________________________________________
ceph-users mailing list -- ceph-users@ceph.io
To unsubscribe send an email to ceph-users-le...@ceph.io