If an OSD is down and needs to be removed from the cluster,
a job can be run that will remove the OSD deployment
and purge the OSD from ceph. If the OSD is still up,
the purge will be rejected.
To purge multiple OSDs at the same time, the OSD IDs can be
specified as a comma-separated list.
Signed-off-by: Servesha Dudhgaonkar <sdudhgao@redhat.com>
When the cluster is external there are no running pods so setting a
Selector on the Service Spec does not make sense, even worse, it makes
Kubernetes reconciling and overriding the configured endpoint.
Signed-off-by: Sébastien Han <seb@redhat.com>
Earlier, we were returning only when the object was not the config
override configmap, we want the opposite.
Also, this was blocking all subsequent conditions and add the correct
check to avoid cm to reconcile.
Signed-off-by: Sébastien Han <seb@redhat.com>
When the cluster is external the service monitor port name should match
with the name of the external metrics one.
Signed-off-by: Sébastien Han <seb@redhat.com>
It's messy to answer `y` everytime in `tests/script/kubeadm.sh clean.`
Since the cluster only exists for testing, it's safe to reset without prompting.
Signed-off-by: Satoru Takeuchi <satoru.takeuchi@gmail.com>
Add a note that NFS service names created by efscli need to comply with
k8 DNS naming restrictions. Failure to do so will result in an invalid NFS
object manifest name.
Signed-off-by: Obleskar <benjamindeuson@gmail.com>
Add documentation for the setting AllowUninstallWithVolumes to the
following files:
Documentation/ceph-cluster-crd.md
cluster/examples/kubernetes/ceph/cluster.yaml
Co-authored-by: Nitin Goyal <nigoyal@redhat.com>
Signed-off-by: Raghavendra Talur <raghavendra.talur@gmail.com>
If AllowUninstallWithVolumes is set in the cleanup policy then we don't
check and wait for the deletion of existing volumes from the
cephCluster.
Co-authored-by: Nitin Goyal <nigoyal@redhat.com>
Signed-off-by: Raghavendra Talur <raghavendra.talur@gmail.com>
This adds an uninstall mode to the ceph cluster. If
allowUninstallWithVolumes in the CleanupPolicySpec is set to true then
the cluster deletion does not wait for the PVs to be deleted.
Co-authored-by: Nitin Goyal <nigoyal@redhat.com>
Signed-off-by: Raghavendra Talur <raghavendra.talur@gmail.com>
The upgrade tests can safely use the rbac.authorization.k8s.io/v1 instead of
rbac.authorization.k8s.io/v1beta1. In K8s 1.19, the upgrade test was failing
because v1beta1 has been deprecated.
Signed-off-by: Travis Nielsen <tnielsen@redhat.com>
With the pending release of k8s 1.19 we need to start running
the integration tests on 1.19. In master we remove support for
1.14 to keep the last five releases running in master.
Signed-off-by: Travis Nielsen <tnielsen@redhat.com>
Trying to execute a binary from the host within the container can result
in mismatched libraries. The host version is compiled with paths of
libraries on the host where the libs inside the container can differ.
Because the path is known and we only need to make sure whether the bin
exists or not then simply doing a lookup when we fallback on /rootfs is
sufficient.
Closes: https://github.com/rook/rook/issues/6078
Signed-off-by: Sébastien Han <seb@redhat.com>
By addding ".svc.cluster.local" to the endpoint of the object store, the
OBC requests won't go through a proxy if any is configured.
Signed-off-by: Sébastien Han <seb@redhat.com>
node-hangs-after-reboot problem can't be fixed with ceph-csi:2.0.1
if Rook is v1.2.x. It's because this version doesn't have a bind
mount that is needed to fix this problem.
Signed-off-by: Satoru Takeuchi <satoru.takeuchi@gmail.com>
There was a broken link in the cluster CR documentation that was pointing
to /Documentation instead of the .md in the same dir.
Signed-off-by: Travis Nielsen <tnielsen@redhat.com>
This allows users to specify additional labels to be added to the components
created by the Rook Ceph operator.
Additionally the rbd mirrors are now getting their annotations and
labels set.
Signed-off-by: Alexander Trost <galexrt@googlemail.com>
This change is the first part to improve the watch OSD nodes logic in
the Rook Ceph operator.
This commit changes the "is node valid check" to use the OSD placement
instead of the all placement. The OSD placement is "merged" with the
all placement already.
Second step will be to skip the first "iteration" of the "nodes added"
events and then start the actual watch logic.
Partially resolves#6111.
Signed-off-by: Alexander Trost <galexrt@googlemail.com>
Sébastien Han
Servesha Dudhgaonkar
Satoru Takeuchi
Satoru Takeuchi
Obleskar
Raghavendra TalurandNitin Goyal
Gregor Krmelj
Travis Nielsen
Alexander Trost