We now always perform ok-to-stop change when the deployment is going to
be updated. This could be for various reasons like change in the spec or
a new Ceph image.
In any case, if Ceph resources are going to be restarted this must be
done carefully.
Also trimmed the isUpgrade variable, the upcoming commit will remove the
need of that variable.
Keeping isUpgrade from the controller code since we need it to detect
whether an orchestration upgrade should be triggered or not (based on
the ceph cluster status).
Also, we keep it on the child controllers so we don't go into an
orchestration phase if the cluster image hasn't changed. The child does
not need to be updated/go into its orchestration phase because the
cluster CR changed.
Closes: https://github.com/rook/rook/issues/4379
Signed-off-by: Sébastien Han <seb@redhat.com>
In order for deployments managed outside the immediate scope of
the ClusterController to accurately detect the Ceph version of
their images, the ClusterController must store the detected
version. This change records the Ceph version as part of the
cluster status and modifies the crashcollector deployment to
consume this field to record its own Ceph version. This replaces
an earlier solution to this problem in which the Ceph version
for a given image to version mapping was stored in a global
in-memory map by the operator.
Resolves: rook#4357
Signed-off-by: Elise Gafford <egafford@redhat.com>
Multiple things:
1. We removed all the function/methods/tests that were used to
create and manage rook legacy OSDS as well as bringing support to
Bluestore OSD only.
It also fixes various go-lint issues in the respectives files.
2. use c-v inventory to detect available devices:
Now we rely on the 'ceph-volume inventory' command to tell us if a
device is available or not.
3. implement raw mode for osd on pvc
When an OSD will be bootstrap on a PVC, the new c-v raw mode will be
used. It consists of putting block, db and wal under the same device.
Here LVM is out of the picture and the raw device is used as is. The
implementation is backward compatible so existing OSD on PVC will LVM
will continue to operate.
Closes: https://github.com/rook/rook/issues/4363
Signed-off-by: Sébastien Han <seb@redhat.com>
If an image is specified in the external cluster CR, the
'rook-ceph-config' config map will be created and then child CRs like
rgw/mds/nfs can be created.
Also, calling config.GetStore() was a mistake since it's calling by the
health check if new mons are being added to clusterInfo.
Closes: https://github.com/rook/rook/issues/4686
Signed-off-by: Sébastien Han <seb@redhat.com>
Currently each time the ceph operator restart, osds also restart.
This is due to a change in the crush-location args with location not ordered
At each restart the location order can change and lead to that kind of change in the osd deployment
104c105
< "--crush-location=root=default host=hostname datacenter=PAR pod=2 rack=2",
---
> "--crush-location=root=default host=hostname pod=2 rack=2 datacenter=PAR",
Sorting the topology to ensure stability of this command arg
Signed-off-by: n.fraison <n.fraison@criteo.com>
Now, we purely rely on what the user puts in the CR and apply it.
Prior to this commit we were putting 0 if not declared, this was not the
intentional.
Closes: https://github.com/rook/rook/issues/4713
Signed-off-by: Sébastien Han <seb@redhat.com>
If all mons are down (e.g. set the replicas to 0 on all the mon deployments) and the operator is restarted, the operator waits indefinitely for a single mon to start. This PR fixes this issue by skipping the quorum check of mons if the number of ready mon pods is equal to 0.
Signed-off-by: Ashish Ranjan <ashishranjan738@gmail.com>
When the OSDs are being provisioned,
if one cluster has already provisioned the OSDs,
the new cluster will skip them since it will see they are already in use.
So we don't need this check anymore.
Coordinating the flag across multiple cluster instances is more complicated than it's worth.
Someone also might want useAllDevices: true in multiple clusters
and they would keep them on separate nodes by using the node labels and selectors.
Closes: https://github.com/rook/rook/issues/4633
Signed-off-by: Sébastien Han <seb@redhat.com>
A new CR setting has been introduced to disable the crash controller.
To disable it, add `disable: true` on the crashCollector section.
Signed-off-by: Sébastien Han <seb@redhat.com>
Added a check for --crush-location, if --crush-location is not present,
the node topology is not applied and the operator will need to determine
the value based on the node labels instead of skipping this setting.
Signed-off-by: rohan47 <rohgupta@redhat.com>
The host name of an OSD in the CRUSH map should be the real
host name for non-portable OSDs. It was incorrectly being set
to the PVC name. Now the non-portable OSDs based on PVCs will
corretly have the host name set to the node name.
Signed-off-by: Travis Nielsen <tnielsen@redhat.com>
The method to construct osd environment variables is simplified
by passing a struct param instead of individual params
Signed-off-by: Travis Nielsen <tnielsen@redhat.com>
A new CRD option `continueUpgradeAfterChecksEvenIfNotHealthy` is added.
When upgrading, Rook goes OSD by OSD and then waits for PGs to be clean
before proceeding to the next OSD. Currently, Rook waits for 5 hours but
there might be circumstances where PGs need more time to settle.
Thus setting `continueUpgradeAfterChecksEvenIfNotHealthy` to true will
pursue the upgrade process, even if PGs are not 100% active+clean.
Closes: https://bugzilla.redhat.com/show_bug.cgi?id=1786029
Signed-off-by: Sébastien Han <seb@redhat.com>
The operator kit had more utility originally when the operator
was creating and managing the TPRs and CRDs directly. Since
the CRDs are now created from a manifest and no longer by the
operators, the utility of operator kit is limited to the
controller watcher. Since we are moving to the controller runtime
we simplify the code to make the transition smoother. Now
there is only a simple WatchCR method that will need to be
replaced as we maek that transition.
Signed-off-by: Travis Nielsen <tnielsen@redhat.com>
Replaced `math/rand` with more secure `crypto/rand` which will generate CSPRNG.
Changed the length of dashboard password from 10 to 20 and included more special characters that can strengthen password.
Passwords before this change: QoMmqLkaOf,8tqTyWF1sZ
Passwords after this change: L{QZ@_c'Rl7$uvW:UHU1, fP-OjX`UVw0J0Ka/scS(
Resolves: https://github.com/rook/rook/issues/4571
Signed-off-by: Nizamudeen <nia@redhat.com>
security issues identified by TrailOfBits:
Improperly handled type assertion error could result in invalid
execution state.
If we obtain an error trying to get a Node object is better not
continue executing this function.
Resolves#4567
Signed-off-by: Juan Miguel Olmo Martínez <jolmomar@redhat.com>
"ceph-volume activate" on encrypted osds needs a ceph.conf as well as
connectivity to the mons. This commit adds that to that "activate-osd"
init container.
Closes: https://github.com/rook/rook/issues/4558
Signed-off-by: Sébastien Han <seb@redhat.com>
The OSDs on PVCs were changing location in the CRUSH
hierarchy after an operator restart due to the location
property not being re-discovered on the OSD daemon.
Since OSDs on PVCs do not run the prepare pod again
to discover this property, the operator must read it
off the existing osd deployment. The operator was
reading other properties, but now will read the location
as well.
Signed-off-by: Travis Nielsen <tnielsen@redhat.com>
The isUpgrade flag was being set every time the operator
was started. This caused the upgrade checks to be run every
time the operator restarted. Thus, if any mons were down,
you likely would get stuck on the upgrade checks since another
mon could not be stopped. We need to rely on the upgrade being
detected either by a different image name from the cluster CR
update, or from different versions of the daemons running
in case the operator was restarted in the middle of an upgrade.
Signed-off-by: Travis Nielsen <tnielsen@redhat.com>
Reduce from 5 minutes to less than 50 seconds the time needed to restart a
mgr and osds(PVCs based),rgw,mds,rbd,nfs and toolbox pod that was running
in a k8s "NotReady" Node.
A new Rook Operator setting has been added to allow the user to set the
time used in pod's toleration <node.kubernetes.io/unreachable>:
ROOK_UNREACHABLE_NODE_TOLERATION_SECONDS = <5>
New setting added in release notes
Addressed @travisn,@leseb, and @blaineEXE suggestions
Signed-off-by: Juan Miguel Olmo Martínez <jolmomar@redhat.com>
When using the "errors" package, using `%+v` (extended format),
each Frame of the error's StackTrace will be printed in detail.
Let's only print `%v` to print the error.
If the error has a Cause it will be printed recursively.
Basically `%+v` has been replaced with `%v` for all `error` type
interfaces, whether the logger is Info, Warning or Error.
Signed-off-by: Sébastien Han <seb@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>
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>
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>
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>
We must not default the selectors to the labels since labels can be
changed and selector cannot until the deployment is updated.
This way the deployment was failing with:
2019-12-10 10:48:13.414438 E | ceph-crashcollector-controller: node reconcile failed on op "unchanged": Deployment.apps "rook-ceph-crashcollector-minikube" is invalid: spec.template.metadata.labels: Invalid value: map[string]string{"app":"rook-ceph-crashcollector", "ceph-version":"15.0.0", "ceph_daemon_id":"crash", "crashcollector":"crash", "kubernetes.io/hostname":"minikube", "node_name":"minikube", "rook-version":"v1.2.0-3.g8d3b85306.dirty"}: `selector` does not match template `labels`
Signed-off-by: Sébastien Han <seb@redhat.com>
The loop wasn't using seconds for its retry so the crash controller
would simply crash since the version is used later in the pod spec.
Signed-off-by: Sébastien Han <seb@redhat.com>
Just as with the mon, mgr, mds, rgw, and rbd-mirror daemons, do not
generate a ceph.conf in an init container for filestore OSDs.
Instead use the commandline to supply all the needed arguments for
running these OSDs.
Signed-off-by: Blaine Gardner <blaine.gardner@suse.com>