of-dev/v1.7.1

Ddsionovvm controller: Clear RestartRequired when no restart is needed

51399e81创建于 2月4日历史提交

文件	最后提交记录	最后更新时间
clone	fix non-constant format string error hit by make go-test https://prow.ci.kubevirt.io/view/gs/kubevirt-prow/pr-logs/pull/kubevirt_kubevirt/15784/pull-kubevirt-unit-test-s390x/1972990835410604032 Signed-off-by: Brian Carey <bcarey@redhat.com>	8 个月前
common	Merge pull request #14183 from aqilbeig/VMPoolUpdateStrategy Adding MaxUnavailable to VMPool to ensure smoother update	1 年前
descheduler	Derive eviction-in-progress annotation from VMI eviction status A VM is considered marked for eviction when Status.EvacuationNodeName is set. When this happens, the eviction manager attempts to evict it based on its eviction strategy. For the LiveMigration strategy, this means creating a VMIM object. Migrations can succeed or fail; in case of failure, the VM remains marked for eviction and the eviction manager retries until eviction succeeds (potentially forever). Currently, there is no supported way to cancel an accepted eviction request. The annotation descheduler.alpha.kubernetes.io/eviction-in-progress is set on the virt-launcher pod to signal to the descheduler that the eviction request was accepted and is proceeding in the background (e.g. with a live migration that can take time). Currently, the annotation’s lifecycle is tied to the status of a specific migration attempt. This is misleading: removing the annotation after a failed migration reports that the eviction was completed, when in fact the migration attempt failed but the eviction manager will retry the eviction with another migration attempt. Such incorrect signaling can cause the descheduler to enqueue other VMs for eviction. If migrations are temporarily failing (e.g., due to a migration network outage), the descheduler could accumulate many eviction requests, potentially bypassing its configured concurrency limits, while the eviction manager loops over new VMIMs for each marked VM. This double retry behavior risks destabilizing the cluster. This change updates the logic so that the eviction-in-progress annotation reflects only vmi.IsMarkedForEviction value, ignoring the outcome of individual migration attempts. In case of an active migration, the annotation should only be added to the source pod to prevent a double counting on the descheduler side. Note: If migrations are stuck and evictions are never canceled (KubeVirt currently has no such mechanism), and the descheduler is configured with a limit on concurrent evictions, the descheduler may stall until some evictions complete or are manually canceled. Signed-off-by: Simone Tiraboschi <stirabos@redhat.com>	9 个月前
dra	build: make generate Signed-off-by: Brian Carey <bcarey@redhat.com>	8 个月前
drain	evacuation: always create migrations The evacuation controller has its own mechanism of delaying the migration creation if there are no spot available. Since the evacuation migrations will have the priority over the others, we should not delay the migration creation if the priorityQueue FG is enabled. Signed-off-by: fossedihelm <ffossemo@redhat.com>	7 个月前
migration	Fix flakiness in migration priority queue unit tests Signed-off-by: Jed Lejosne <jed@redhat.com>	5 个月前
node	Merge pull request #15680 from fossedihelm/do-not-use-only-label-for-pod refactoring(virt-ctrl): Do not fallback using labels for getting owner ref	8 个月前
pool	Move v1alpha1 to v1beta1 version Signed-off-by: svarnam <svarnam@nvidia.com>	7 个月前
replicaset	hack: Run bazel-race Signed-off-by: Lee Yarwood <lyarwood@redhat.com>	8 个月前
testing	consistent punctuation of Copyright notices under pkg/ Signed-off-by: Dan Kenigsberg <danken@redhat.com>	1 年前
topology	hack: Run bazel-race Signed-off-by: Lee Yarwood <lyarwood@redhat.com>	8 个月前
util	add test case to check VMExport is no longer stuck in Pending when using a source Pod that is Completed Signed-off-by: dsanatar <dsanatar@redhat.com>	6 个月前
vm	vm controller: Clear RestartRequired when no restart is needed The RestartRequired condition is set when non-live-updatable fields in the VM spec differ from the spec used to start the VMI. However, the condition was never cleared when the specs become equal again. This caused issues where VMs with a previously set RestartRequired condition would keep it indefinitely, blocking hotplug operations for CPU, memory, and volumes, even when no restart was actually needed. This commit refactors addRestartRequiredIfNeeded into syncRestartRequired, which now both adds and removes the RestartRequired condition as needed. The condition is cleared when specs match after neutralizing live-updatable fields. Signed-off-by: Daniel Sionov <dsionov@redhat.com> Assisted-By: Claude <noreply@anthropic.com>	4 个月前
vmi	add new eventhandler in vmi controller to reenqueuevmis when the owner vm's volumes update. this will be removed once HotplugVolumes feature gate is removed Signed-off-by: dsanatar <dsanatar@redhat.com>	5 个月前
volume-migration	Added unit tests for the GenerateReceiverMigratedVolumes Signed-off-by: Alexander Wels <awels@redhat.com>	4 个月前
vsock	hack: Run bazel-race Signed-off-by: Lee Yarwood <lyarwood@redhat.com>	8 个月前
workload-updater	workload updater: set priority to the migrations - system upgrade will have the system-critical priority - hotplug and volume upgrade will have user-triggered Signed-off-by: fossedihelm <ffossemo@redhat.com>	7 个月前
BUILD.bazel	Move v1alpha1 to v1beta1 version Signed-off-by: svarnam <svarnam@nvidia.com>	7 个月前
application.go	Move v1alpha1 to v1beta1 version Signed-off-by: svarnam <svarnam@nvidia.com>	7 个月前
application_test.go	Merge pull request #15663 from oshoval/deviceinfo virt-controller, migration: Add missing network-info	7 个月前
watch_suite_test.go	Consistent copyright notice under pkg Red Hat is willing to mark these files as copyrighted by the KubeVirt Authors. If accepted, a similar sed PR would follow to the reset of the code based. sed -i 's/\(20.., \)*20.. Red Hat, Inc./The KubeVirt Authors./' git grep -l '20.. Red Hat, Inc.' Signed-off-by: Dan Kenigsberg <danken@redhat.com>	1 年前