mirror of
https://github.com/k3s-io/k3s.git
synced 2024-06-07 19:41:36 +00:00
e204d863a5
* Update Kubernetes to v1.22.1 * Update dependent modules to track with upstream Signed-off-by: Brad Davidson <brad.davidson@rancher.com>
258 lines
11 KiB
Go
258 lines
11 KiB
Go
/*
|
|
Copyright 2016 The Kubernetes Authors.
|
|
|
|
Licensed under the Apache License, Version 2.0 (the "License");
|
|
you may not use this file except in compliance with the License.
|
|
You may obtain a copy of the License at
|
|
|
|
http://www.apache.org/licenses/LICENSE-2.0
|
|
|
|
Unless required by applicable law or agreed to in writing, software
|
|
distributed under the License is distributed on an "AS IS" BASIS,
|
|
WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
|
|
See the License for the specific language governing permissions and
|
|
limitations under the License.
|
|
*/
|
|
|
|
package deployment
|
|
|
|
import (
|
|
"context"
|
|
"sort"
|
|
"strconv"
|
|
|
|
appsv1 "k8s.io/api/apps/v1"
|
|
corev1 "k8s.io/api/core/v1"
|
|
apiequality "k8s.io/apimachinery/pkg/api/equality"
|
|
"k8s.io/apimachinery/pkg/api/meta"
|
|
metav1 "k8s.io/apimachinery/pkg/apis/meta/v1"
|
|
"k8s.io/apimachinery/pkg/runtime"
|
|
intstrutil "k8s.io/apimachinery/pkg/util/intstr"
|
|
runtimeresource "k8s.io/cli-runtime/pkg/resource"
|
|
appsclient "k8s.io/client-go/kubernetes/typed/apps/v1"
|
|
)
|
|
|
|
const (
|
|
// RevisionAnnotation is the revision annotation of a deployment's replica sets which records its rollout sequence
|
|
RevisionAnnotation = "deployment.kubernetes.io/revision"
|
|
// RevisionHistoryAnnotation maintains the history of all old revisions that a replica set has served for a deployment.
|
|
RevisionHistoryAnnotation = "deployment.kubernetes.io/revision-history"
|
|
// DesiredReplicasAnnotation is the desired replicas for a deployment recorded as an annotation
|
|
// in its replica sets. Helps in separating scaling events from the rollout process and for
|
|
// determining if the new replica set for a deployment is really saturated.
|
|
DesiredReplicasAnnotation = "deployment.kubernetes.io/desired-replicas"
|
|
// MaxReplicasAnnotation is the maximum replicas a deployment can have at a given point, which
|
|
// is deployment.spec.replicas + maxSurge. Used by the underlying replica sets to estimate their
|
|
// proportions in case the deployment has surge replicas.
|
|
MaxReplicasAnnotation = "deployment.kubernetes.io/max-replicas"
|
|
// RollbackRevisionNotFound is not found rollback event reason
|
|
RollbackRevisionNotFound = "DeploymentRollbackRevisionNotFound"
|
|
// RollbackTemplateUnchanged is the template unchanged rollback event reason
|
|
RollbackTemplateUnchanged = "DeploymentRollbackTemplateUnchanged"
|
|
// RollbackDone is the done rollback event reason
|
|
RollbackDone = "DeploymentRollback"
|
|
// TimedOutReason is added in a deployment when its newest replica set fails to show any progress
|
|
// within the given deadline (progressDeadlineSeconds).
|
|
TimedOutReason = "ProgressDeadlineExceeded"
|
|
)
|
|
|
|
// GetDeploymentCondition returns the condition with the provided type.
|
|
func GetDeploymentCondition(status appsv1.DeploymentStatus, condType appsv1.DeploymentConditionType) *appsv1.DeploymentCondition {
|
|
for i := range status.Conditions {
|
|
c := status.Conditions[i]
|
|
if c.Type == condType {
|
|
return &c
|
|
}
|
|
}
|
|
return nil
|
|
}
|
|
|
|
// Revision returns the revision number of the input object.
|
|
func Revision(obj runtime.Object) (int64, error) {
|
|
acc, err := meta.Accessor(obj)
|
|
if err != nil {
|
|
return 0, err
|
|
}
|
|
v, ok := acc.GetAnnotations()[RevisionAnnotation]
|
|
if !ok {
|
|
return 0, nil
|
|
}
|
|
return strconv.ParseInt(v, 10, 64)
|
|
}
|
|
|
|
// GetAllReplicaSets returns the old and new replica sets targeted by the given Deployment. It gets PodList and
|
|
// ReplicaSetList from client interface. Note that the first set of old replica sets doesn't include the ones
|
|
// with no pods, and the second set of old replica sets include all old replica sets. The third returned value
|
|
// is the new replica set, and it may be nil if it doesn't exist yet.
|
|
func GetAllReplicaSets(deployment *appsv1.Deployment, c appsclient.AppsV1Interface) ([]*appsv1.ReplicaSet, []*appsv1.ReplicaSet, *appsv1.ReplicaSet, error) {
|
|
rsList, err := listReplicaSets(deployment, rsListFromClient(c), nil)
|
|
if err != nil {
|
|
return nil, nil, nil, err
|
|
}
|
|
newRS := findNewReplicaSet(deployment, rsList)
|
|
oldRSes, allOldRSes := findOldReplicaSets(deployment, rsList, newRS)
|
|
return oldRSes, allOldRSes, newRS, nil
|
|
}
|
|
|
|
// GetAllReplicaSetsInChunks is the same as GetAllReplicaSets, but accepts a chunk size argument.
|
|
// It returns the old and new replica sets targeted by the given Deployment. It gets PodList and
|
|
// ReplicaSetList from client interface. Note that the first set of old replica sets doesn't include the ones
|
|
// with no pods, and the second set of old replica sets include all old replica sets. The third returned value
|
|
// is the new replica set, and it may be nil if it doesn't exist yet.
|
|
func GetAllReplicaSetsInChunks(deployment *appsv1.Deployment, c appsclient.AppsV1Interface, chunkSize int64) ([]*appsv1.ReplicaSet, []*appsv1.ReplicaSet, *appsv1.ReplicaSet, error) {
|
|
rsList, err := listReplicaSets(deployment, rsListFromClient(c), &chunkSize)
|
|
if err != nil {
|
|
return nil, nil, nil, err
|
|
}
|
|
newRS := findNewReplicaSet(deployment, rsList)
|
|
oldRSes, allOldRSes := findOldReplicaSets(deployment, rsList, newRS)
|
|
return oldRSes, allOldRSes, newRS, nil
|
|
}
|
|
|
|
// RsListFromClient returns an rsListFunc that wraps the given client.
|
|
func rsListFromClient(c appsclient.AppsV1Interface) rsListFunc {
|
|
return func(namespace string, initialOpts metav1.ListOptions) ([]*appsv1.ReplicaSet, error) {
|
|
rsList := &appsv1.ReplicaSetList{}
|
|
err := runtimeresource.FollowContinue(&initialOpts,
|
|
func(opts metav1.ListOptions) (runtime.Object, error) {
|
|
newRs, err := c.ReplicaSets(namespace).List(context.TODO(), opts)
|
|
if err != nil {
|
|
return nil, runtimeresource.EnhanceListError(err, opts, "replicasets")
|
|
}
|
|
rsList.Items = append(rsList.Items, newRs.Items...)
|
|
return newRs, nil
|
|
})
|
|
if err != nil {
|
|
return nil, err
|
|
}
|
|
var ret []*appsv1.ReplicaSet
|
|
for i := range rsList.Items {
|
|
ret = append(ret, &rsList.Items[i])
|
|
}
|
|
return ret, err
|
|
}
|
|
}
|
|
|
|
// TODO: switch this to full namespacers
|
|
type rsListFunc func(string, metav1.ListOptions) ([]*appsv1.ReplicaSet, error)
|
|
|
|
// listReplicaSets returns a slice of RSes the given deployment targets.
|
|
// Note that this does NOT attempt to reconcile ControllerRef (adopt/orphan),
|
|
// because only the controller itself should do that.
|
|
// However, it does filter out anything whose ControllerRef doesn't match.
|
|
func listReplicaSets(deployment *appsv1.Deployment, getRSList rsListFunc, chunkSize *int64) ([]*appsv1.ReplicaSet, error) {
|
|
// TODO: Right now we list replica sets by their labels. We should list them by selector, i.e. the replica set's selector
|
|
// should be a superset of the deployment's selector, see https://github.com/kubernetes/kubernetes/issues/19830.
|
|
namespace := deployment.Namespace
|
|
selector, err := metav1.LabelSelectorAsSelector(deployment.Spec.Selector)
|
|
if err != nil {
|
|
return nil, err
|
|
}
|
|
options := metav1.ListOptions{LabelSelector: selector.String()}
|
|
if chunkSize != nil {
|
|
options.Limit = *chunkSize
|
|
}
|
|
all, err := getRSList(namespace, options)
|
|
if err != nil {
|
|
return nil, err
|
|
}
|
|
// Only include those whose ControllerRef matches the Deployment.
|
|
owned := make([]*appsv1.ReplicaSet, 0, len(all))
|
|
for _, rs := range all {
|
|
if metav1.IsControlledBy(rs, deployment) {
|
|
owned = append(owned, rs)
|
|
}
|
|
}
|
|
return owned, nil
|
|
}
|
|
|
|
// EqualIgnoreHash returns true if two given podTemplateSpec are equal, ignoring the diff in value of Labels[pod-template-hash]
|
|
// We ignore pod-template-hash because:
|
|
// 1. The hash result would be different upon podTemplateSpec API changes
|
|
// (e.g. the addition of a new field will cause the hash code to change)
|
|
// 2. The deployment template won't have hash labels
|
|
func equalIgnoreHash(template1, template2 *corev1.PodTemplateSpec) bool {
|
|
t1Copy := template1.DeepCopy()
|
|
t2Copy := template2.DeepCopy()
|
|
// Remove hash labels from template.Labels before comparing
|
|
delete(t1Copy.Labels, appsv1.DefaultDeploymentUniqueLabelKey)
|
|
delete(t2Copy.Labels, appsv1.DefaultDeploymentUniqueLabelKey)
|
|
return apiequality.Semantic.DeepEqual(t1Copy, t2Copy)
|
|
}
|
|
|
|
// FindNewReplicaSet returns the new RS this given deployment targets (the one with the same pod template).
|
|
func findNewReplicaSet(deployment *appsv1.Deployment, rsList []*appsv1.ReplicaSet) *appsv1.ReplicaSet {
|
|
sort.Sort(replicaSetsByCreationTimestamp(rsList))
|
|
for i := range rsList {
|
|
if equalIgnoreHash(&rsList[i].Spec.Template, &deployment.Spec.Template) {
|
|
// In rare cases, such as after cluster upgrades, Deployment may end up with
|
|
// having more than one new ReplicaSets that have the same template as its template,
|
|
// see https://github.com/kubernetes/kubernetes/issues/40415
|
|
// We deterministically choose the oldest new ReplicaSet.
|
|
return rsList[i]
|
|
}
|
|
}
|
|
// new ReplicaSet does not exist.
|
|
return nil
|
|
}
|
|
|
|
// replicaSetsByCreationTimestamp sorts a list of ReplicaSet by creation timestamp, using their names as a tie breaker.
|
|
type replicaSetsByCreationTimestamp []*appsv1.ReplicaSet
|
|
|
|
func (o replicaSetsByCreationTimestamp) Len() int { return len(o) }
|
|
func (o replicaSetsByCreationTimestamp) Swap(i, j int) { o[i], o[j] = o[j], o[i] }
|
|
func (o replicaSetsByCreationTimestamp) Less(i, j int) bool {
|
|
if o[i].CreationTimestamp.Equal(&o[j].CreationTimestamp) {
|
|
return o[i].Name < o[j].Name
|
|
}
|
|
return o[i].CreationTimestamp.Before(&o[j].CreationTimestamp)
|
|
}
|
|
|
|
// // FindOldReplicaSets returns the old replica sets targeted by the given Deployment, with the given slice of RSes.
|
|
// // Note that the first set of old replica sets doesn't include the ones with no pods, and the second set of old replica sets include all old replica sets.
|
|
func findOldReplicaSets(deployment *appsv1.Deployment, rsList []*appsv1.ReplicaSet, newRS *appsv1.ReplicaSet) ([]*appsv1.ReplicaSet, []*appsv1.ReplicaSet) {
|
|
var requiredRSs []*appsv1.ReplicaSet
|
|
var allRSs []*appsv1.ReplicaSet
|
|
for _, rs := range rsList {
|
|
// Filter out new replica set
|
|
if newRS != nil && rs.UID == newRS.UID {
|
|
continue
|
|
}
|
|
allRSs = append(allRSs, rs)
|
|
if *(rs.Spec.Replicas) != 0 {
|
|
requiredRSs = append(requiredRSs, rs)
|
|
}
|
|
}
|
|
return requiredRSs, allRSs
|
|
}
|
|
|
|
// ResolveFenceposts resolves both maxSurge and maxUnavailable. This needs to happen in one
|
|
// step. For example:
|
|
//
|
|
// 2 desired, max unavailable 1%, surge 0% - should scale old(-1), then new(+1), then old(-1), then new(+1)
|
|
// 1 desired, max unavailable 1%, surge 0% - should scale old(-1), then new(+1)
|
|
// 2 desired, max unavailable 25%, surge 1% - should scale new(+1), then old(-1), then new(+1), then old(-1)
|
|
// 1 desired, max unavailable 25%, surge 1% - should scale new(+1), then old(-1)
|
|
// 2 desired, max unavailable 0%, surge 1% - should scale new(+1), then old(-1), then new(+1), then old(-1)
|
|
// 1 desired, max unavailable 0%, surge 1% - should scale new(+1), then old(-1)
|
|
func ResolveFenceposts(maxSurge, maxUnavailable *intstrutil.IntOrString, desired int32) (int32, int32, error) {
|
|
surge, err := intstrutil.GetScaledValueFromIntOrPercent(intstrutil.ValueOrDefault(maxSurge, intstrutil.FromInt(0)), int(desired), true)
|
|
if err != nil {
|
|
return 0, 0, err
|
|
}
|
|
unavailable, err := intstrutil.GetScaledValueFromIntOrPercent(intstrutil.ValueOrDefault(maxUnavailable, intstrutil.FromInt(0)), int(desired), false)
|
|
if err != nil {
|
|
return 0, 0, err
|
|
}
|
|
|
|
if surge == 0 && unavailable == 0 {
|
|
// Validation should never allow the user to explicitly use zero values for both maxSurge
|
|
// maxUnavailable. Due to rounding down maxUnavailable though, it may resolve to zero.
|
|
// If both fenceposts resolve to zero, then we should set maxUnavailable to 1 on the
|
|
// theory that surge might not work due to quota.
|
|
unavailable = 1
|
|
}
|
|
|
|
return int32(surge), int32(unavailable), nil
|
|
}
|