There are two issues at odds with the cephfs driver. The
first is that older kernels less than 4.17 do not support
quotas so PVC sizes will not be enforced with the kernel
driver in that case. If the FUSE client is used instead, there
are issues during upgrade because the client mount will be
disrupted and the client pods will have to restart.
To prevent client pods from breaking during upgrade, we enable
the kernel driver by default. If the quotas are critical,
the admin will need to take steps during upgrade to prevent
breaking the applications.
Signed-off-by: Travis Nielsen <tnielsen@redhat.com>
In ceph-csi v1.2.2 we have add a flag `forcecephkernelclient`
to force use Ceph Kernel clients on kernel < 4.17 which
support quotas for Cephfs
Signed-off-by: Madhu Rajanna <madhupr007@gmail.com>
Removed some obsolete instructions from the 1.1 upgrade
and made a few minor fixes to make the upgrade to 1.2 smoother.
With the addition of the CephClient CRD, the CRD needs to be
created before the upgrade or else the operator log will be
filled with errors until it is created. The CSI driver also
had an error when registering after the upgrade due to an RBAC
change.
Signed-off-by: Travis Nielsen <tnielsen@redhat.com>
gp2 devices present themselves as SSDs,
but have much more variable performance.
Ceph attempts to adjust some settings based on the drive type,
HDD/SSD
In this case client I/O can be impacted too much by assuming gp2 devices are like other SSDs
(an example is https://bugzilla.redhat.com/show_bug.cgi?id=1765923#c6)
In order to prevent this, rook can set the following for OSDs
if the new CR setting `tuneSlowDeviceClass` is set to `true`:
```
osd_recovery_sleep = 0.1
osd_snap_trim_sleep = 2
osd_delete_sleep = 2
```
Closes: https://github.com/rook/rook/issues/4298 and https://bugzilla.redhat.com/show_bug.cgi?id=1771047
Signed-off-by: Sébastien Han <seb@redhat.com>
An owner reference cannot be set with BlockOwnerDeletion: true
if the operator does not have privs to add a finalizer to the parent.
The discover resources don't need to block deletion based on the
parent so we remove that property from the owner reference.
Signed-off-by: Travis Nielsen <tnielsen@redhat.com>
Removal of OSDs is a sensitive area. If you get it wrong you could have data
loss. Now a help topic is provided for manual steps to remove OSDs when
absolutely necessary.
Signed-off-by: Travis Nielsen <tnielsen@redhat.com>
Removing an OSD unintentionally can result in data loss,
which is to be prevented at all costs. Rook's functionality
to remove OSDs automatically when a node was removed from the
cluster CR has led on several occasions to users having OSDs
unintentionally removed from their cluster.
If the admin decides they really need to remove OSDs they should
do it as a one-time action that they intentionally run
rather than the operator attempting to manage the deletion
with desired state. The risk of misinterpreting the desired
state, even with a typo, is too great.
Signed-off-by: Travis Nielsen <tnielsen@redhat.com>
It's not 100% safe since before Nautilus osd safe-to-destroy command
didn't have any json output. So added a version check.
Signed-off-by: rohan47 <rohgupta@redhat.com>
Add owner reference to the discover config map so that when the
discover pod goes away we also remove its configmap.
Signed-off-by: Sébastien Han <seb@redhat.com>
Attach owner reference to the discover daemon set so that when the
opetator goes away, the discover pod gets removed too.
Signed-off-by: Sébastien Han <seb@redhat.com>
As of 14.2.5, the crash collector pod will use the client.crash user
instead of the client.admin keyring to post crash dumps.
Closes: https://github.com/rook/rook/issues/4388
Signed-off-by: Sébastien Han <seb@redhat.com>
Disable the dashboard in the upgrade test because upgrading Ceph from
Mimic to Nautilu with Rook v1.0 can stall for a LONG time when updating
the mgr module. We are pretty sure there is very little risk that the
upgrade test will expose problems with the dashboard module that aren't
caught by other integration tests.
Signed-off-by: Blaine Gardner <blaine.gardner@suse.com>
The file-test pod still hangs sometimes when exiting. Instead of
trapping SIGTERM as before, use tini's `-g` option to kill the process
group. According to further reading on bash signal handling [1], Ctrl-C
sends a kill signal to the entire process group in order to kill `sleep`
commands.
[1] Link also noted in code comments:
http://mywiki.wooledge.org/SignalTrap#When_is_the_signal_handled.3F
Signed-off-by: Blaine Gardner <blaine.gardner@suse.com>
Reset the integration test k8s helper's `RetryLoop` to its original
value, and instead only wait an extra long time to allow the mgr module
updates to take a long time after Ceph is updated from Mimic to Nautilus
as part of Ceph's upgrade integration test.
Updating mgr modules can hang for quite a while, which causes the tests
to time out waiting for the OSDs to be updated. Allow this to take a
long time so the tests aren't as flaky.
Signed-off-by: Blaine Gardner <blaine.gardner@suse.com>
Both --id and --name do the same thing, so let's use one only: --id
which is shared accross all the daemons already.
Signed-off-by: Sébastien Han <seb@redhat.com>
rgw needs the fronted on its cli line to select the fronted as well the
port. This option is not supported in the mon config store.
Closes: https://github.com/rook/rook/issues/4471
Signed-off-by: Sébastien Han <seb@redhat.com>
As performed with mon, gateways from same ceph object store
must not run on the same host if hostNetwork is set to true
Signed-off-by: n.fraison <n.fraison@criteo.com>
If editing the cluster CR we must update the spec to reflect its new
content. Now the spec is updated so we can update with an unsupported
image as well as setting allowUnsupported: true.
It was not possible before we had to restart the operator, now we can do
it as an update.
Signed-off-by: Sébastien Han <seb@redhat.com>