Clarification needed on "Managing Power Events" shutdown order
Affects | Status | Importance | Assigned to | Milestone | |
---|---|---|---|---|---|
OpenStack Charm Guide |
Triaged
|
High
|
Unassigned |
Bug Description
Hi,
The shutdown order specified on the Managing Power Events page isn't clearly justified as far as I can tell [1]. Two observations:
- Radosgw should be shut down before ceph osds/mons as it is a ceph client
- Glance and Cinder both consume ceph for storage, but are listed to be shut down after ceph. Why is this?
It makes sense to make a distinction between the data plane (DP) and control plane (CP) services, since after compute goes down there won't be many attempts to use e.g. Glance. That said, it's also reasonable to come at this from a "shut down dependents before dependencies" angle, which isn't called out in the doc; there's definitely space for confusion here.
Additionally, there's no guidance given for additional services that a user may have deployed in their environment, especially those with their own dependencies (Gnocchi on memcached, etc). While getting this order right isn't critical for operation of the cloud I think it's still worth providing some hint for the diligent: juju relations are a reasonable proxy for what talks to what.
[1] https:/
Yes, I agree. Ceph is a statefull app and should be shutdown after all the consumers have been shutdown/paused.