Data inconsistency might happen if the disk is accessed just after disk
zapping because direct I/O is not synchronous by itself.
Signed-off-by: Satoru Takeuchi <satoru.takeuchi@gmail.com>
To simplify the deployment of Rook, both the operator and cluster crd are now created in the
same namespace in the examples. By default the operator will only manage a cluster crd
in the same namespace instead of watching for clusters in all namespaces. To manage a cluster
in another namespace, the roles in latter half of operator-common.yaml need to point to the new
namespace and the operator env var ROOK_CURRENT_NAMESPACE_ONLY should be set to false.
Signed-off-by: travisn <tnielsen@redhat.com>
The command `kubectl delete -n rook-ceph pool replicapool` used to
work, but currently returns the error `error: the server doesn't
have a resource type "pool"`. The actual resource type is
`cephblockpools`. The resource type `cluster` was also recently
changed to `cephcluster`.
Since the function commands end with `|| true`, the errors are
silently masked and the resources are not deleted.
This PR adds delete commands for `cephblockpools` and `cephcluster`,
so that regardless of whether you're testing old or new versions
of Rook the delete commands will work.
Signed-off-by: Earl C. Ruby III <earlruby@users.noreply.github.com>
Recent change to support multiple backends changed the docker image name
but missed the change in the test script. This patch addresses that
issue.
Signed-off-by: Ganesh Maharaj Mahalingam <ganesh.mahalingam@intel.com>
This commit adds instructions to develop on a multi-node deployment,
which is somehow similar to a production setup.
It also proposes a development workflow approach to
write/build/test/ship
code to Rook.
Signed-off-by: Sébastien Han <seb@redhat.com>
Satoru Takeuchi
Sébastien Han
travisn
Earl C. Ruby III
Alexander Trost
Ganesh Maharaj Mahalingam
Dan Mick