Bluestore is assumed as the default backend and string StoreType is left empty.
Propogating that assumption to device partitioning.
Resolves: #1748
Signed-off-by: Ganesh Maharaj Mahalingam <ganesh.mahalingam@intel.com>
To make it more applicable to what is actually running when using the
operator the used test mon names have been replaced by the actual format
used in normal operations: `rook-ceph-mon[0-9]` instead of `mon[0-9]`.
This also removes some duplicated test code and used the already
available function for it.
Signed-off-by: Alexander Trost <galexrt@googlemail.com>
This introduces a MonSpec to the Ceph Cluster CRD which contains settings for
mons, current options are the count and if more than one mon is allowed
to run per node.
Fixes#1386
Signed-off-by: Alexander Trost <galexrt@googlemail.com>
This patch attempts to fix PV delete issues after an operator service
restart.
Resolves: #1721
Signed-off-by: Ganesh Maharaj Mahalingam <ganesh.mahalingam@intel.com>
Multiple file system support is still considered experimental in Ceph.
Disabling that as default in rook until such time it is production
ready.
A small code cleanup in filesystem file
Resolves#1720
Signed-off-by: Ganesh Maharaj Mahalingam <ganesh.mahalingam@intel.com>
run "rook discover" on storage nodes and discover devices on each node. The discovered disks are saved in a per node configmap, local-device-nodename.
Device information consits of name and persistent names, uuid, partition, filesystem, rotational, readonly, size, etc.
Signed-off-by: Huamin Chen <hchen@redhat.com>
This allows users to scrape their ceph-mgr deployment without being
forced to use the Prometheus operator
Signed-off-by: Alvaro Aleman <alv2412@googlemail.com>
The stopCh value was only present in struct cluster, whose instance was
local to the onAdd function; save it to ClusterController.
Second, it was not initialized.
Third, ClusterController's "devicesInUse" needs to be set false on
cluster delete.
With this fix, Ceph clusters can be deleted and recreated without
restarting the Rook operator.
Signed-off-by: Dan Mick <dan.mick@redhat.com>
If watching the osd orchestration status map fails, ensure the loop is broken out of so a new watcher can be created.
Log the osd device/dir mapping structures in a more friendly way
Signed-off-by: Jared Watts <jbw976@gmail.com>
- Daemonsets deployed by rook currently use the "delete" update
controller. Which means the pods have to be manually deleted
before being upgraded.
- Using the update controller will make daemonset behave like
deployments when it comes to upgrade.
Partially fixes https://github.com/rook/rook/issues/997
Signed-off-by: Steve Leon <kokhang@gmail.com>
Jared Watts
travisn
Ganesh Maharaj Mahalingam
Alexander Trost
Bence Szikora
Huamin Chen
rohan47
Alvaro Aleman
Tony Allen
Illya Chekrygin
Dan Mick
David González Ruiz
Travis Nielsen
John Spray
Omri Shiv
Steve Leon
Christian Hüning