Travis Nielsen

It's not a common scenario to have no pod network available, so seems like we would need a new setting in the CephCluster CR such as: `network.requireHostNetwork: true`

> Need to be updated manually when the cephcsi image is updated in Rook Since the daemonset spec is updated with a new image, but the policy isn't to update...

> > Webhooks are very challenging and will introduce a lot of complexity. We struggled with this for a long time in the admission controller and finally decided it wasn't...

@BlaineEXE Since #13890 was merged, shall we close this? Or anything else blocking for v1.14 related to this?

A complication is that the exporter is running on each node in the external cluster. An endpoint is needed on each node. Can the servicemonitor connect to multiple endpoints? And...

As discussed separately, we need a design review for the approach. Scraping metrics from a list of nodes not even inside K8s doesn't seem correct.

If it's v18.2.1, check if it's the same issue as #13527.

@rkachach Did you mean to remove all OSD management from the rook mgr module? If so, sounds good.

> @travisn I think the fix for this to be already included in `Ceph 18.2.4` Great to know it's in the latest release!

I do see in the operator log several times that the reconcile started again immediately. For example, this snippet shows one reconcile completes, and then 0.04 seconds later the next...