Closed eugenkoenig closed 5 years ago
Could you share the content of group_vars/*.yml ?
Does the mimic deployment was already using ceph-volume (osd_scenario: lvm) ? AFAIK the gpt header is only present on device deployed by ceph-disk and ceph-disk isn't present anymore in ceph nautilus / stable-4.0.
@dsavineau it was deployed with collocated scenario, so it was using ceph-disk. But how can we use add-osd.yaml
then? Is it enough to just replace osd_scenario: collocated
with osd_scenario: lvm
?
group_vars/ceph.yaml:
ceph_mirror: http://download.ceph.com
ceph_origin: repository
ceph_repository: community
ceph_stable: true
ceph_stable_key: https://download.ceph.com/keys.release.asc
ceph_stable_release: nautilus
ceph_stable_repo: "{{ ceph_mirror }}/debian-{{ ceph_stable_release }}"
upgrade_ceph_packages: True
cluster: ceph
ceph_conf_key_directory: /etc/ceph
fetch_directory: fetch/
ntp_service_enabled: true
configure_firewall: false
mon_group_name: mons
osd_group_name: osds
mgr_group_name: mgrs
rgw_group_name: rgws
ceph_mgr_modules:
- status
- dashboard
monitor_interface: ens18
radosgw_interface: ens18
nfs_ganesha_stable: true
nfs_ganesha_stable_branch: V2.7-stable
nfs_ganesha_stable_deb_repo: "[trusted=yes] https://chacra.ceph.com/r/nfs-ganesha-stable/V2.7-stable/2356c3867730696aacc31874357b3499062fc902/ubuntu/bionic/flavors/ceph_nautilus"
nfs_file_gw: false
nfs_obj_gw: true
ceph_nfs_log_file: "/var/log/ganesha/ganesha.log"
group_vars/osds.yaml
osd_scenario: collocated
devices:
- '/dev/vdb'
- '/dev/vdc'
- '/dev/vdd'
- '/dev/vde'
osd_mkfs_type: xfs
osd_objectstore: bluestore
@andrewschoen shouldn't ceph-volume lvm batch --report
simply ignore disks when it sees GPT header instead of failing like this?
@guits true, that's a behavior I'd expect. We found, that OSDs created with ceph-disk can be easily converted to lvm (ceph-volume) by following the guide in ceph docs [1], however it's okay for small clusters, but too much effort for clusters with hundreds of OSDs. Either ignoring them or run special playbook would be better approach.
[1] http://docs.ceph.com/docs/nautilus/rados/operations/add-or-rm-osds/#replacing-an-osd
@andrewschoen shouldn't
ceph-volume lvm batch --report
simply ignore disks when it sees GPT header instead of failing like this?
We did this purposefully because batch wants to make sure all drives given to it are usable, if they are not, they are rejected. The disks would need to be zapped and/or all GPT headers removed before given to lvm batch
.
@andrewschoen IMO, upgrading from mimic with ceph-disk prepared OSDs to nautilus shouldn't require users have to manually do something regarding already deployed OSDs. As @styleart said, that could eventually be acceptable for small cluster, but what about large cluster with hundred OSDs already deployed? This doesn't ease the ceph-disk to ceph-volume transition.
This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you for your contributions.
Bug Report
What happened:
add-osd.yaml
playbook, it crashes because of GTP table error on all nodes:How to reproduce it (minimal and precise):
add-osd.yaml
playbookEnvironment:
18.04.2 LTS (Bionic Beaver)
uname -a
):Linux ceph1 4.15.0-51-generic
ansible-playbook --version
):ansible 2.8.0
git head or tag or stable branch
): stable-4.0 / masterceph -v
): ceph version 14.2.1