Bug #50472
closed
orchestrator doesn't provide a way to remove an entire cluster
Added by Paul Cuzner about 3 years ago.
Updated almost 3 years ago.
Description
In prior toolchains like ceph-ansible, purging a cluster and returning a set of hosts to their original state was possible. With the removal of ceph-ansible, this is no longer an option.
The ceph-ansible purge-cluster playbook removed all daemons and optionally zapped each OSD, in a simple and controlled manner
How is this to be done under orch/cephadm?
A few problems:
- cephadm rm-cluster only removes the cluster on the local host
- mgr/cephadm cannot remove the cluster, as it is intended to be super save and cannot remove itself.
Workaround is to run cephadm rm-cluster on all hosts using an ansible playbook or using ceph-salt
Sebastian Wagner wrote:
A few problems:
- cephadm rm-cluster only removes the cluster on the local host
- mgr/cephadm cannot remove the cluster, as it is intended to be super save and cannot remove itself.
Workaround is to run cephadm rm-cluster on all hosts using an ansible playbook or using ceph-salt
IMO, uur answer can NOT be the user still needs to invest in ansible or salt and maintain two separate configuration management solutions for ceph. rm-cluster is part of a cluster's lifecycle - and we should be providing that capability natively.
- Priority changed from High to Low
Paul Cuzner wrote:
Sebastian Wagner wrote:
A few problems:
- cephadm rm-cluster only removes the cluster on the local host
- mgr/cephadm cannot remove the cluster, as it is intended to be super save and cannot remove itself.
Workaround is to run cephadm rm-cluster on all hosts using an ansible playbook or using ceph-salt
IMO, uur answer can NOT be the user still needs to invest in ansible or salt and maintain two separate configuration management solutions for ceph. rm-cluster is part of a cluster's lifecycle - and we should be providing that capability natively.
I'm a bit reluctant introducing SSH capabilities to the cephadm binary just for this use case here. So many downsides:
- there are much better tools for this job, like Salt or Ansible
- without the ability to have proper python dependencies, we will end up duplicating python SSH libraries.
- we're breaking the assumption that the cephadm binary is for managing the local host only.
- Users need to maintain two separate ssh setups anyway.
- Related to Feature #50529: cephadm rm-cluster is also not resetting any disks that were used as osds added
- Status changed from New to Resolved
Also available in: Atom
PDF