github/kubernetes
1
0
Fork 0
mirror of https://github.com/k3s-io/kubernetes.git synced 2026-07-21 00:33:15 +00:00
Code Issues Projects Releases Wiki Activity

DRA kubeletplugin: some doc updates

Browse Source 
The TODO was resolved and the "seamless update" PR merged already in 1.33.
The updated guidance in NodeUnprepareResources is meant to emphasize
existing practice.
This commit is contained in:
Patrick Ohly
2025-05-23 13:05:42 +02:00
parent 6da56bd4b7
commit 7215de79b3
2 changed files with 9 additions and 7 deletions
Download Patch File Download Diff File Expand all files Collapse all files

5
staging/src/k8s.io/dynamic-resource-allocation/kubeletplugin/doc.go
View File

@@ -57,8 +57,9 @@ limitations under the License.
// the node the DaemonSet runs on supports that. Trying
// to do a rolling update with a kubelet which does not support it yet
// will fail because shutting down the old Pod unregisters the driver
// even though the new Pod is running. See https://github.com/kubernetes/kubernetes/pull/129832
// for details (TODO: link to doc after merging instead).
// even though the new Pod is running. See
// https://github.com/kubernetes/kubernetes/blob/v1.33.0/pkg/kubelet/pluginmanager/pluginwatcher/README.md#seamless-upgrade
// for details.
//
// A DRA driver can either require 1.33 as minimal Kubernetes version or
// provide two variants of its DaemonSet deployment. In the variant with

11
staging/src/k8s.io/dynamic-resource-allocation/kubeletplugin/draplugin.go
View File

@@ -100,6 +100,12 @@ type DRAPlugin interface {
// It is the responsibility of the DRA driver to cache whatever additional
// information it might need about prepared resources.
//
// The DRA driver cannot assume that the matching PrepareResourceClaims
// call was handled by the same process:
// - The driver might have been restarted in the meantime.
// - [RollingUpdate], if enabled, can lead to PrepareResourceClaims being
// called in one driver instance and UnprepareResourceClaims in another.
//
// This call must be idempotent because the kubelet might have to ask
// for un-preparation multiple times, for example if it gets restarted.
// Therefore it is not an error if this gets called for a ResourceClaim
@@ -626,11 +632,6 @@ func (d *Helper) PublishResources(_ context.Context, resources resourceslice.Dri
// our background context, not the one passed into this
// function, and thus is connected to the lifecycle of the
// plugin.
//
// TODO: don't delete ResourceSlices, not even on a clean shutdown.
// We either support rolling updates and want to hand over seamlessly
// or don't and then perhaps restart the pod quickly enough that
// the kubelet hasn't deleted ResourceSlices yet.
controllerCtx := d.backgroundCtx
controllerLogger := klog.FromContext(controllerCtx)
controllerLogger = klog.LoggerWithName(controllerLogger, "ResourceSlice controller")