deletion of cluster is stuck when cluster is not configured. The PR skips the cleanup and removes the finalizers if the cluster was not configured, that is, if clusterInfo could not be loaded.
Signed-off-by: Santosh Pillai <sapillai@redhat.com>
It's better to validate ownerReferences when setting them. In addition, we should use
controllerrutil.Set{Controller,Owner}Reference, that have such validation, as possible.
Signed-off-by: Satoru Takeuchi <satoru.takeuchi@gmail.com>
If one of the suite fails, it is sometimes handy to ssh into the
instance to precisely understand what's going on.
So by applying a debug label so as "debug-smoke-suite" you will be able
to ssh into the runner to debug. By default it looks
like the runner will be cancelled automatically after 6 hours as per the
official doc: https://docs.github.com/en/actions/reference/workflow-syntax-for-github-actions#jobsjob_idtimeout-minutes
I've had a different experience, and somehow 45min seems to be the limit
but I might be wrong.
Once the label is applied, go into the check and the once you are
interested in and search the step "setup tmate session for debugging".
It should output lines with the connection details.
To get the list of all labels see: https://github.com/rook/rook/labels?q=debug-
Example run is available here: https://github.com/rook/rook/pull/7416/checks?check_run_id=2120103989
Signed-off-by: Sébastien Han <seb@redhat.com>
Finally! We can now stop editing manually our crds definition. Simply
run `make crds-gen`. These two files:
* `cluster/examples/kubernetes/ceph/crds.yaml`
* `cluster/charts/rook-ceph/templates/resources.yaml`
will be autogenerated for us.
We rely on the controller-gen tool, it reads our API definitions from
`pkg/apis/` and produces the CRD files accordingly.
It uses "markers" to add extra yaml fields to the CRD, for instance we
have a lot fields with:
```
nullable: true
x-kubernetes-preserve-unknown-fields: true
```
The corresponding API type markers are:
```
// +kubebuilder:pruning:PreserveUnknownFields
// +nullable
```
For more API convention see
https://github.com/kubernetes/community/blob/master/contributors/devel/sig-architecture/api-conventions.md#optional-vs-required
and for the markers see: https://book.kubebuilder.io/reference/markers/crd-validation.html
Signed-off-by: Sébastien Han <seb@redhat.com>
This is a follow up to:
https://github.com/rook/rook/pull/7264
which added pool quotas. It was stated that it would be nice to
be able to set pool max_bytes quotas using standard k8s quantity
formats rather than an integer representing the number of bytes.
I was too busy at the time to make the changes. This PR adds the
functionality.
Signed-off-by: Frank Ritchie <12985912+fritchie@users.noreply.github.com>
Fix an apt-get update issue by bumping to the latest (fixed) version of
manusa/actions-setup-minikube (v2.3.1).
Signed-off-by: Blaine Gardner <blaine.gardner@redhat.com>
As a recommendation from the core team we should turn on PG auto repair
when running on Bluestore OSDs. The Quincy release will enable this by
default but in the meantime let's enable it here.
Signed-off-by: Sébastien Han <seb@redhat.com>
Sometimes `radosgw-admin` succeeds after showing logs to stderr. We should
skip non-json strings if parsing output as json.
Here is an example.
```
2021-02-26 04:10:44.190418 I | op-bucket-prov: creating Ceph user "ceph-user-aSzNqgE7"
E0226 04:11:37.901310 8 controller.go:199] error syncing 'logging/loki-bucket': error provisioning bucket: Provision: can't create ceph user: error creating ceph user "ceph-user-aSzNqgE7": failed to unmarshal json. 2021-02-26T04:11:21.425+0000 7f6714be4980 1 robust_notify: If at first you don't succeed: (110) Connection timed out
2021-02-26T04:11:21.426+0000 7f6714be4980 0 ERROR: failed to distribute cache for ceph-hdd-object-store.rgw.meta:users.uid:ceph-user-aSzNqgE7
2021-02-26T04:11:32.168+0000 7f6714be4980 1 robust_notify: If at first you don't succeed: (110) Connection timed out
2021-02-26T04:11:32.168+0000 7f6714be4980 0 ERROR: failed to distribute cache for ceph-hdd-object-store.rgw.meta:users.keys:23Z8GUEXR0TJDO86PSJR
{
"user_id": "ceph-user-aSzNqgE7",
"display_name": "ceph-user-aSzNqgE7",
...
"mfa_ids": []
}: invalid character '-' after top-level value: failed to unmarshal json.
```
In this case, some logs like "robust_notify:..." was shown in stderr.
Unmarsharing was failed due to tried to parse these logs as json.
Signed-off-by: Satoru Takeuchi <satoru.takeuchi@gmail.com>
If the count of mgr daemons is reduced, we remove the extra one.
The check is made during each reconcile.
Signed-off-by: Travis Nielsen <tnielsen@redhat.com>
In Pacific the simple command to retrieve the active mgr is with
ceph mgr stat instead of retrieving the full ceph mgr dump.
Signed-off-by: Travis Nielsen <tnielsen@redhat.com>
If there are multiple mgrs running, the operator will automatically
add pod antiaffinity across hosts by default. For stretch clusters,
antiaffinity across the stretch failure domain (e.g. zones)
will be added. Required antiaffinity will be created unless the
test setting allowMultiplePerNode is set to true.
Signed-off-by: Travis Nielsen <tnielsen@redhat.com>
If there are two mgr daemons, they should be running on different
hosts or zones, depending on the failure domain.
Signed-off-by: Travis Nielsen <tnielsen@redhat.com>
The mgr daemon may be failed over by ceph if the active mgr is not
responding and the standby mgr is available. If the active mgr changes
the services for the dashboard and metrics will be updated with a
label selector for the new active mgr. The services cannot direct
traffic to the standby mgr or else they will be incorrectly redirected
to the active mgr.
Signed-off-by: Travis Nielsen <tnielsen@redhat.com>
The active mgr should match the labels on the services that
are available for prometheus and the dashboard. The selector
labels must be updated whenever there is a new active mgr.
Signed-off-by: Travis Nielsen <tnielsen@redhat.com>
The GRPC metrics exposed by both the provisioner pod
and the node plugin pod on some port. Provisioner pod
is running on the pod network and the daemonset pods
run on the host network. sometimes starting the GRPC
metrics by default can lead to node plugin pods
crashloopback state this is due to the port conflict.
Moreover, the GRPC metrics are not for the user it's
for the one which will help to debug the time taken
by cephcsi to serve each GRPC call. Enabling it by
default won't be a good idea. So the plan is to
disable it by default, If someone faces any issue it
can be enabled later at some point in time.
closes#7378
Signed-off-by: Madhu Rajanna <madhupr007@gmail.com>
We now create 3 disks which will give us 2 OSDs and potentially help us
to catch more errors in our scenarios, especially on iterations.
Signed-off-by: Sébastien Han <seb@redhat.com>
When c-v runs after the end of the prepare sequence, it will look for
lvm and raw-based OSDs. The raw mode detects LVM OSDs so we must trim
the list of OSD, only to report the LVM ones if their ID exists.
Closes: https://github.com/rook/rook/issues/7359
Signed-off-by: Sébastien Han <seb@redhat.com>
The integration tests will now read the manifests from the same source
that is used in the examples for installation. No longer is there a need
for a copy of the CRDs in the integration test code.
The tests will read the manifests directly from disk under a path
relative to the rook root. The upgrade test will read the manifests from
a github url where the base version has been published.
Signed-off-by: Travis Nielsen <tnielsen@redhat.com>
Santosh Pillai
Satoru Takeuchi
Anmol Sachan
Sébastien Han
subhamkrai
Travis Nielsen
Frank Ritchie
Blaine Gardner
Blaine Gardner
ushen
Madhu Rajanna