Problem:
-------
For rbd nbd userspace mounter backends, after a restart of the nodeplugin
all the mounts will start seeing IO errors. This is because, for rbd-nbd
backends there will be a userspace mount daemon running per volume, post
restart of the nodeplugin pod, there is no way to restore the daemons
back to life.
Solution:
--------
The volume healer is a one-time activity that is triggered at the startup
time of the rbd nodeplugin. It navigates through the list of volume
attachments on the node and acts accordingly.
For now, it is limited to nbd type storage only, but it is flexible and
can be extended in the future for other backend types as needed.
From a few feets above:
This solves a severe problem for nbd backed csi volumes. The healer while
going through the list of volume attachments on the node, if finds the
volume is in attached state and is of type nbd, then it will attempt to
fix the rbd-nbd volumes by sending a NodeStageVolume request with the
required volume attributes like secrets, device name, image attributes,
and etc.. which will finally help start the required rbd-nbd daemons in
the nodeplugin csi-rbdplugin container. This will allow reattaching the
backend images with the right nbd device, thus allowing the applications
to perform IO without any interruptions even after a nodeplugin restart.
Signed-off-by: Prasanna Kumar Kalever <prasanna.kalever@redhat.com>
Nodeplugin needs below cluster roles:
persistentvolumes: get
volumeattachments: list, get
These additional permissions are needed by the volume healer. Volume healer
aims at fixing the volume health issues at the very startup time of the
nodeplugin. As part of its operations, volume healer has to run through
the list of volume attachments and understand details about each
persistentvolume.
The later commits will use these additional cluster roles.
Signed-off-by: Prasanna Kumar Kalever <prasanna.kalever@redhat.com>
As part of stage transaction if the mounter is of type nbd, then capture
device path after a successful rbd-nbd map.
Signed-off-by: Prasanna Kumar Kalever <prasanna.kalever@redhat.com>
clone controller make sure there is no delete operation happens
on the source PVC which has been referred as the datasource of
clone PVC, we are safe to operate without looking at delete
operation lock in this case.
Subjected code in the controller:
...
if claim.Spec.DataSource != nil && rc.clone {
err = p.setCloneFinalizer(ctx, claim)
...
}
if !checkFinalizer(claim, pvcCloneFinalizer) {
claim.Finalizers = append(claim.Finalizers, pvcCloneFinalizer)
_, err := p.client.CoreV1().PersistentVolumeClaims(claim.Namespace).Update(..claim..)
}
Signed-off-by: Humble Chirammal <hchiramm@redhat.com>
Considering kubelet make sure the stage and publish operations
are serialized, we dont need any extra locking in nodePublish
Signed-off-by: Humble Chirammal <hchiramm@redhat.com>
Considering kubelet make sure the stage and publish operations
are serialized, we dont need any extra locking in nodePublish
Signed-off-by: Humble Chirammal <hchiramm@redhat.com>
Considering kubelet make sure the unstage and unpublish operations
are serialized, we dont need any extra locking in nodeUnpublish
Signed-off-by: Humble Chirammal <hchiramm@redhat.com>
Considering kubelet make sure the unstage and unpublish operations
are serialized, we dont need any extra locking in nodeUnpublish
Signed-off-by: Humble Chirammal <hchiramm@redhat.com>
Minikube has bumped it's support for latest Kubernetes version to
1.22.0-beta.0. This might improve our CI jobs with Kubernetes 1.22 too.
See-also: https://github.com/kubernetes/minikube/releases/tag/v1.22.0
Signed-off-by: Niels de Vos <ndevos@redhat.com>
When an error occurs, the pvc object is overwritten in the
PollImmediate() loop. Re-using the pvc.Namespace results in error
messages like
Error getting pvc in namespace: '': an empty namespace may not be set when a resource name is provided
and prevents the retry by PollImmediate() to never succeed. Storing the
namespace in a local variable prevents this from happening.
Reported-by: Rakshith R <rar@redhat.com>
Signed-off-by: Niels de Vos <ndevos@redhat.com>
There are regular CI failures where etcdserver times out. These errors
seem not to get caught by any of the existing error comparing. Matching
the error by string should prevent temporary etcdserver issues now too.
Updates: #2218Closes: #1969
Signed-off-by: Niels de Vos <ndevos@redhat.com>
parseTenantConfig() only allowed configuring a defined set of options,
and KMSs were not able to re-use the implementation. Now, the function
parses the ConfigMap from the Tenants Namespace and returns a map with
options that the KMS supports.
The map that parseTenantConfig() returns can be inspected by the KMS,
and applied to the vaultTenantConnection type by calling parseConfig().
Signed-off-by: Niels de Vos <ndevos@redhat.com>
Kubelet sometimes reports the following error:
failed to "StartContainer" for "vault-init-job" with CreateContainerConfigError: container has runAsNonRoot and image will run as root
Setting securityContext.runAsUser resolves this.
Signed-off-by: Niels de Vos <ndevos@redhat.com>
The ServiceAccount "ceph-csi-vault-sa" is expected to be placed in the
Namespace "tenant" so that the provisioner and node-plugin fetch the
ServiceAccount from a Namespace where Ceph-CSI is not deployed.
Signed-off-by: Niels de Vos <ndevos@redhat.com>
The provisioner and node-plugin have the capability to connect to
Hashicorp Vault with a ServiceAccount from the Namespace where the PVC
is created. This requires permissions to read the contents of the
ServiceAccount from an other Namespace than where Ceph-CSI is deployed.
Signed-off-by: Niels de Vos <ndevos@redhat.com>
This new KMS uses a Kubernetes ServiceAccount from a Tenant (Namespace)
to connect to Hashicorp Vault. The provisioner and node-plugin will
check for the configured ServiceAccount and use the token that is
located in one of the linked Secrets. Subsequently the Vault connection
is configured to use the Kubernetes token from the Tenant.
Signed-off-by: Niels de Vos <ndevos@redhat.com>
This makes the Tenant configuration for Hashicorp Vault KMS connections
more modular. Additional KMS implementations that use Hashicorp Vault
with per-Tenant options can re-use the new vaultTenantConnection.
Signed-off-by: Niels de Vos <ndevos@redhat.com>
A new KMS that supports Hashicorp Vault with the Kubernetes Auth backend
and ServiceAccounts per Tenant (Kubernetes Namespace).
Updates: #2222
Signed-off-by: Niels de Vos <ndevos@redhat.com>
This commit resolves parallel test issues
and also excludes internal/util/conn_pool_test.go
as those test can't run in parallel.
Updates: #1586
Signed-off-by: Yati Padia <ypadia@redhat.com>
This commit resolves godot linter issue
which says "Comment should end in a period (godot)".
Updates: #1586
Signed-off-by: Yati Padia <ypadia@redhat.com>
This adds a new `kmsConfig` interface that can be used to validate
different KMS services and setting. It makes checking for the available
support easier, and fetching the passphrase simpler.
The basicKMS mirrors the current validation of the KMS implementations
that use secrets and metadata. vaultKMS can be used to validate the
passphrase stored in a Vault service.
Signed-off-by: Niels de Vos <ndevos@redhat.com>
This commit adds t.Helper() to the test helper
function. With this call go test prints correct
lines of code for failed tests. Otherwise,
printed lines will be inside helpers functions.
For more details check: https://github.com/kulti/thelper
Updates: #1586
Signed-off-by: Yati Padia <ypadia@redhat.com>
* Make kernel version parsing to support more (valid) version strings
* Put version string parsing into a separate, testable function
* Fixes#2248 (Kernel Subversion Parsing Failure)
Signed-off-by: Jonas Zeiger <jonas.zeiger@talpidae.net>
This commit adds e2e for user secret based metadata encryption,
adds user-secret.yaml and makes required changes in kms-connection-details,
kms-config yamls.
Signed-off-by: Rakshith R <rar@redhat.com>
This commit adds capability to `metadata` encryption
to be able to fetch `encryptionPassphrase` from user
specified secret name and namespace(if not specified,
will default to namespace where PVC was created).
This behavior is followed if `secretName` key is found
in the encryption configuration else defaults to fetching
`encryptionPassphrase` from storageclass secrets.
Closes: 2107
Signed-off-by: Rakshith R <rar@redhat.com>
This commit updates the helm chart documentations
with the configurations available while deploying
these helm charts.
Signed-off-by: Yati Padia <ypadia@redhat.com>
setting metadata in isThickProvisioned method
helps us to avoid checking thick metakey and
deprecated metakey for both thick and thin
provisioned images and also this will easily
help us to migrated the deprecated key to new key.
Signed-off-by: Madhu Rajanna <madhupr007@gmail.com>
isThickProvisioned is already method of the rbdImage
to keep similar thick provisioner related functions
common making isThickProvisioned as method of rbdImage.
Signed-off-by: Madhu Rajanna <madhupr007@gmail.com>
instead of checking the parent is thick provisioned
or not we can decide based on the rbdVol generated
from the request. If the request is to create a Thick
Image. set metadata without checking the parent.
Signed-off-by: Madhu Rajanna <madhupr007@gmail.com>
As image metadata key starting with '.rbd' will not
be copied when we do clone or mirroring, deprecating
the old key for the same reason use
'csi.ceph.com/thick-provisioned' to set image metadata.
Signed-off-by: Madhu Rajanna <madhupr007@gmail.com>
Mirror-snapshots can also be automatically created on a
periodic basis if mirror-snapshot schedules are defined.
The mirror-snapshot can be scheduled globally, per-pool,
or per-image levels. Multiple mirror-snapshot schedules
can be defined at any level.
To create a mirror-snapshot schedule with rbd, specify
the mirror snapshot schedule add command along with an
optional pool or image name; interval; and optional start time:
The interval can be specified in days, hours, or minutes
using d, h, m suffix respectively. The optional start-time
can be specified using the ISO 8601 time format. For example:
```
$ rbd --cluster site-a mirror snapshot schedule
add --pool image-pool --image image1 24h 14:00:00-05:00
```
Signed-off-by: Madhu Rajanna <madhupr007@gmail.com>
New go-ceph admin package api's expects to
pass the rados connection as argument.
added new method called GetRBDAdmin
to get admin connection to administrate
rbd volumes.
Signed-off-by: Madhu Rajanna <madhupr007@gmail.com>
added go-ceph rbd admin package to vendor. keeping
this as a separate commit which helps in review.
Signed-off-by: Madhu Rajanna <madhupr007@gmail.com>
