Ceph » Orchestrator

When a cluster specification is passed like:

cephadm bootstrap --apply-spec <spec>

And spec looks like:

---
# END mds-000 ANSIBLE MANAGED BLOCK
# BEGIN mds-001 ANSIBLE MANAGED BLOCK
service_type: host
hostname: mds-001
addr: 173.255.224.209
labels:
  - mdss
---
# END mds-001 ANSIBLE MANAGED BLOCK
...
# BEGIN mon spec ANSIBLE MANAGED BLOCK
service_type: mon
placement:
  label: mons
---
# END mon spec ANSIBLE MANAGED BLOCK
# BEGIN mgr spec ANSIBLE MANAGED BLOCK
service_type: mgr
placement:
  label: mgrs
---
# END mgr spec ANSIBLE MANAGED BLOCK
# BEGIN osd spec ANSIBLE MANAGED BLOCK
service_id: default
service_type: osd
placement:
  label: osds
data_devices:
  all: true
---
# END osd spec ANSIBLE MANAGED BLOCK
# BEGIN mds spec ANSIBLE MANAGED BLOCK
service_id: all
service_type: mds
placement:
  label: mdss
---
# END mds spec ANSIBLE MANAGED BLOCK
...

If one host is (transiently) down, the spec file processing stops. Make it possible for cephadm to continue processing the spec file.

(Ideally, I think cephadm would still add the host to the cluster even if it cannot be contacted but maybe that's not easily possible.)