If the spec.mon.count is 5 or more, then create
mon pdbs with maxUnavailable count as 2. We can
afford to have more than 1 mons down while
draining the nodes in this case.
Signed-off-by: sp98 <sapillai@redhat.com>
(cherry picked from commit a4866b4b06)
retry other cephfs/rbd pvc if there
are no active clients found on ceph cluster.
Signed-off-by: Madhu Rajanna <madhupr007@gmail.com>
(cherry picked from commit ca21bd9715)
to keep the code consistent and to propogate
more details about error rook uses error wraping.
updating current code whereever its required to
wrap the details.
Signed-off-by: Madhu Rajanna <madhupr007@gmail.com>
(cherry picked from commit eb97390df2)
The cephfs PVC might exist on the kubernetes
node object but due to some timing issues the
ip might not be visible on the ceph cluster or
the client might already got evicted or disconnected
from ceph cluster. In this case we will not be
able to get IP details for the subvolume and we dont
have any check for empty ip's in the code and rook
tries to create NetworkFence CR with empty Ip's
and the NetworkFence will get moved to the Failed state.
This PR adds the necessary check and logging to prevent
this one.
Signed-off-by: Madhu Rajanna <madhupr007@gmail.com>
(cherry picked from commit 0e115c44bf)
since networkFence is a cluster-based resource so that we don't need
the namespace and ownerReferences as it cause garbage-collector errors.
Also, now we create the networkFence with clusteUID label so when doing
cleanup we match the cephCluster uid and networkFence label clusterUID.
Signed-off-by: subhamkrai <srai@redhat.com>
(cherry picked from commit 5bff860380)
Mgr sidecar was using a privileged security context because it needed
to create the /var/lib/rook directory, generate the config on it, and
then copy it to the /etc/ceph directory. This change creates empty
directories on /var/lib/rook and /etc/ceph so the mgr sidecar can use
them, removing the need for using a privileged security context.
Fixes: https://github.com/rook/rook/issues/13719
Signed-off-by: Redouane Kachach <rkachach@redhat.com>
(cherry picked from commit db4b3322fa)
Fixes: #13692
Since the introduction of the "host" network provider, the legacy
"hostNetwork" setting is intended to be used only in combination with
the default network provider (""), but the code did not enforce this.
This change adds the required validation checks to throw errors
in invalid constellations.
These checks are added both in the operator's input validation code
and as kubernetes x-validation admission policies in the Cepcluster CRD.
Signed-off-by: Michael Adam <obnox@samba.org>
(cherry picked from commit 25aa15e6c1)
Updating the cephcsi image to 3.10.2
which is the latest released one.
Signed-off-by: Madhu Rajanna <madhupr007@gmail.com>
(cherry picked from commit c22597a32c)
Currently, when rook provisions OSDs(in the OSD prepare job), rook effectively run a
c-v command such as the following.
```console
ceph-volume lvm batch --prepare <deviceA> <deviceB> <deviceC> --db-devices <metadataDevice>
```
but c-v lvm batch only supports disk and lvm, instead of disk partitions.
We can resort to `ceph-volume lvm prepare` to implement it.
Signed-off-by: Liang Zheng <zhengliang0901@gmail.com>
(cherry picked from commit ee05e82cea)
since networkFence is a cluster-based resource so that we don't need
the namespace and ownerReferences as it cause garbage-collector errors.
Signed-off-by: subhamkrai <srai@redhat.com>
(cherry picked from commit 2ac418a3b2)
When CPU requests and limits are assigned to a pod,
the pod will be guaranteed the requests, up to the limits.
Even if there are spare CPU cycles, the pod cannot use
them. Thus, pods can be unnecessarily denied compute
when they need to burst if the limits are set.
Therefore, it is not recommended to set CPU limits
since the CPU requests are already guaranteeing that
no pod will be starved at least for its requests.
Signed-off-by: travisn <tnielsen@redhat.com>
(cherry picked from commit 27265ff279)
The validation logic of checking the number of devices is
wrong when `metadataDevice` is set and `osdsPerDevice` > 1.
`len(cvReports)` is the expected number of OSDs and is
the number of specified data devices multiplied
by `osdsPerDevice`.
Closes: https://github.com/rook/rook/issues/13637
Signed-off-by: Satoru Takeuchi <satoru.takeuchi@gmail.com>
(cherry picked from commit 753bbfe1c4)
When `managePodBudgets` is enabled, the Rook operator sets a blocking
PDB by considering the failure domains of the OSDs. This functionality
is implemented by `reconcilePDBsForOSDs`, and it sets the PDB only after
unhealthy PGs appear. However, there are no unhealthy PGs when an OSD
with no PGs becomes down. In this case, the PDB is never enabled.
This PR makes the operator configure the blocking PDB without waiting
for the unhealthy PGs to appear. This PR solves the above problem
because the blocking PDB is always enabled when a down OSD is detected.
Signed-off-by: Ryotaro Banno <ryotaro.banno@gmail.com>
(cherry picked from commit 6def9c8e30)
If the configmap rook-config-override is empty,
there is no need to trigger the reconcile to update
the ceph daemons. This configmap update is causing
unnecessary reconciles periodically in some clusters
even when it is empty.
Signed-off-by: travisn <tnielsen@redhat.com>
(cherry picked from commit 0240ee8563)
this commit updates the name of network fence CR to
node name + driver name, which will allow the creation of
multiple network fence per drivers on the same node, and
modifies the unit test and deletion procedure for the
same.
Signed-off-by: Riya Singhal <rsinghal@redhat.com>
(cherry picked from commit c2e3cf367c)
The name for cephcosi driver should be ceph-cosi-driver and namespace
should same as the rook operator. Otherwise request need to be failed.
Fixes: #13025
Signed-off-by: Jiffin Tony Thottan <thottanjiffin@gmail.com>
(cherry picked from commit 5b5e1c0292)
this kernal version is greater than 5.11 let's
use mounter kernal instead of fuse.
Signed-off-by: subhamkrai <srai@redhat.com>
(cherry picked from commit abc272c91c)
This PR fixes the failure while running multicluster mirroring CI tests
Signed-off-by: sp98 <sapillai@redhat.com>
(cherry picked from commit 45adae1931)
The 'extra' block device attached to GH actions runners has changed size
twice in 3 months. The previous strategy of detecting the disk by size
is becoming harder to maintain. Additionally, the block size with recent
changes (75G) is now the same as the boot device (also 75G), making the
method inexact.
The method can now be summarized as, "find the boot disk and choose the
disk that isn't the boot disk to be the 'extra' one used."
Prior to this, we used a one-liner based on `lsblk`. While we could
still make this a one-liner, the method is now updated to 2 effective
lines, plus debug text output to stderr to help if we need to debug
further in the future.
Of note, the 'extra' disk has a mount point of "/mnt", but it is unclear
whether this is a reliable heuristic for detecting the extra disk. For
years now, GH action runners have had only 2 disks. Therefore, it seems
slightly more likely that a heuristic to "choose the non-boot disk" will
be a more robust long-term solution.
If this strategy proves to be unreliable in the future, it may be wise
to consider whether "the device with a partition mounted to '/mnt'"
would be a good alternative.
Signed-off-by: Blaine Gardner <blaine.gardner@ibm.com>
(cherry picked from commit feacb6424e)
the disk size in the github action machine has
increased from 64G to 75G. Now, we detech the version
automatically not fetching hard coded value.
Co-authored-by: Jan Klippel <jan.klippel@uhurutec.com>
Signed-off-by: subhamkrai <srai@redhat.com>
(cherry picked from commit 6b0deb32d9)
This is needed because not all contributors may be using gomft to format
source code, and this could result in inconsistent usage over time.
Signed-off-by: Blaine Gardner <blaine.gardner@ibm.com>
(cherry picked from commit c98ce0a56b)
Corrected the NetNamespaceFilePath generation in the
subvolumegroup & radosnamespace controller by replacing
the cephCluster.Name with cephCluster.Namespace.
Signed-off-by: Praveen M <m.praveen@ibm.com>
(cherry picked from commit 67ff924b57)
The exporter service and service monitor should not
be deleted if an individual daemon is being removed
from a node. As long as the exporter is enabled with
v18 and newer, the exporter service and service monitor
should exist. Clusters were seeing their metrics stop
collection when the exporter service and service monitor
would disappear, especially when multiple clusters are
configured.
Signed-off-by: travisn <tnielsen@redhat.com>
(cherry picked from commit a1fc70f2fc)
sp98
Travis Nielsen
Tim Olow
mergify[bot]
Madhu Rajanna
subhamkrai
Redouane Kachach
Michael Adam
Liang Zheng
Satoru Takeuchi
Joshua Hoblitt
Ryotaro Banno
Riya Singhal
Jiffin Tony Thottan
subham rai
Blaine Gardner
Praveen M