[umbrella] Add Implementation for Service discovery with native Kubernetes naming and resolution

[jwcesign]

What would you like to be added:
In #4287, we propose a way to implement service discovery with native Kubernetes naming and resolution.

The specific work items are as follows:

release-1.8(deadline: 1130)

• Add proposal for service discovery with native Kubernetes naming and resolution - doc: add proposal for mcs with native service name #4287, @jwcesign
• Add corresponding API - api: define API for MultiClusterService #4290, @jwcesign
• For mcs-controller, List&watch mcs and service, reconcile the work in execution namespace, feat: List&watch mcs and service, reconcile the work in execution namespace #4323, @Rains6
• For mcs-eps-controller, List&watch mcs, collect the corresponding EndpointSlice from ServerLocations and create/update the corresponding Work. feat: collect endpoint slice from provision clusters and sync them to consumption clusters #4329, @jwcesign
• e2e test case, e2e: add mcs e2e with CrossCluster type #4333, @jwcesign
• Add mutation webhook to add permanent-id to MCS(helm/karmadactl/operator). feat: add webhook for multiclusterservice to add the permanentid #4335, @jwcesign
• Add feature gate of MultiClusterService, Add MultiClusterService FeatureGate #4347, @XiShanYongYe-Chang
• Documentation of the usage, @jwcesign doc: add user guide and tutorial doc of mcs website#548
• Documentation of Configure Controllers, doc: update controller configuration website#520

release-1.9 iteration-1 (deadline: 1207)

• For mcs-controller, List&watch cluster creation/deletion, reconcile the work in the corresponding cluster execution namespace. feat: sync mcs work when cluster joined or unjoin #4360, @Rains6
• For mcs-eps-controller, List&watch cluster creation/deletion, reconcile the EndpointSlice's work in the corresponding cluster execution namespace. feat: dispatch eps to the newly joined consumption clusters #4356, @jwcesign
• If a cluster gets unhealth, mcs-eps-controller should delete the EndpointSlice from all the cluster execution namespace. feat: sync mcs work when cluster joined or unjoin #4360, @Rains6

iteration-2 (deadline: 1223)

• Fix the following scenario problem: We've changed our work naming style from the previous one, which has resulted in an issue where deleted resources in the member cluster cannot be recreated. feat: fix resource(work/resource in member clusters) conflicting between PP and MCS. #4414
• The work name is not sufficient to show the resource information, for example, I want to know the corresponding original resource name/kind, and I need to get the yaml to check. It's better to add additional columns to show these information. Add additionalPrinterColumns WORKLOAD-KIND for Work CRD #2066

Legacy Issues:
Basic:

• Currently, we have MCS, Service, Provider EndpointSlice work, and Consumer EndpointSlice work existing independently from each other. In the future, we consider implementing interaction through MCS status to optimize process control.
• Sort out abnormal scenarios, and provide comprehensive things about traffic anomaly events.
• Organize performance points, provide comprehensive metrics, and facilitate performance observation.
• Construct a comprehensive mcs e2e test.
• Construct a comprehensive mcs unit test.

Conflict:

• Now, after dispatch the endpointslice to consumption cluster, label resourcetemplate.karmada.io/uid is empty, We should check if it's OK.
• Conflict issue with MCI/LoadBalancer MCS: Both MCI and LoadBalancer MCS will synchronize EndpointSlice to the management side as Work. At this time, there will be a Work management conflict with CrossCluster MCS that needs to be resolved.
• Conflicts with PP/CPP, control the RB together.

Performance:

• Improving MCS Performance: In the case of 20 clusters, where ProviderCluster and ConsumerCluster are empty, a single Service will generate 20*19=380 Works. According to actual tests, Work management synchronization is relatively slow (on the minute level).

Reliability:

• Reliability issues: Consider the following abnormal scenarios, how MCS avoids service access anomalies.
  - Karmada control plane anomaly, member cluster's EndpointSlice invalidation issue.
  - East-west network anomaly between clusters, service access anomaly.
  - Intra-cluster network anomaly, but the cluster health check passes, leading to cross-cluster access anomalies.
  - Cluster apiserver failure and normal network cause EndpointSlice synchronization deletion anomalies.

Experience Precipitation:

• Based on MCS experience, precipitate the implementation standards for components, anomalies/events/performance/monitoring metrics; at the same time, define how to solve conflicts with previous resources through scenario-based Policy.

Why is this needed:

Ref proposal #4287

Help

Anyone who wants to participate in one of the tasks can come and assign the tasks. Feel free to let me know in the comments and I will mark it after the corresponding task item, thanks~

/help

[karmada-bot]

@jwcesign:
This request has been marked as needing help from a contributor.

Please ensure the request meets the requirements listed here.

If this request no longer meets these requirements, the label can be removed
by commenting with the /remove-help command.

In response to this:

What would you like to be added:
In #4287, we propose a way to implement service discovery with native Kubernetes naming and resolution.

The specific work items are as follows:

release-1.8(1130)

• Add proposal for service discovery with native Kubernetes naming and resolution - doc: add proposal for mcs with native service name #4287
• Add corresponding API - api: define API for MultiClusterService #4290
• For mcs-controller, List&watch mcs and service, reconcile the work in execution namespace @Rains6
• For mcs-eps-controller, List&watch mcs, collect the corresponding EndpointSlice from ServerLocations and create/update the corresponding Work. @jwcesign

release-1.9(1215)

• For mcs-controller, List&watch cluster creation/deletion, reconcile the work in corresponding cluster execution namespace.
• For mcs-eps-controller, List&watch cluster creation/deletion, reconcile the EndpointSlice's work in corresponding cluster execution namespace.
• If cluster gets unhealth, mcs-eps-controller should delete the EndpointSlice from all the cluster execution namespace.

Why is this needed:

Ref proposal #4287

Help

Anyone who wants to participate in one of the tasks can come and assign the tasks. Feel free to let me know in the comments and I will mark it after the corresponding task item, thanks~

/help

Instructions for interacting with me using PR comments are available here. If you have questions or suggestions related to my behavior, please file an issue against the kubernetes/test-infra repository.

[RainbowMango]

/assign @jwcesign @Rains6

[karmada-bot]

@RainbowMango: GitHub didn't allow me to assign the following users: Rains6.

Note that only karmada-io members, repo collaborators and people who have commented on this issue/PR can be assigned. Additionally, issues/PRs can only have 10 assignees at the same time.
For more information please see the contributor guide

In response to this:

/assign @jwcesign @Rains6

Instructions for interacting with me using PR comments are available here. If you have questions or suggestions related to my behavior, please file an issue against the kubernetes/test-infra repository.

[jwcesign]

Now, the EPS synchronization logic is as follows:

1.List&watch mcs, build informer to list&watch the eps in ServiceProvisionClusters
2.List&watch eps in ServiceProvisionClusters
3.create the corresponding work
4.Sync the endpointslice's work in step 3 to ServiceConsumptionClusters, the endpoint slice name will be changed to {source-cluster-name}-{endpointslcie-name}
5.The work will trigger to create the endpointslice int ServiceConsumptionClusters.

[RainbowMango]

We are going to introduce 3 controllers for this feature, proposing the controller name here:

• multiclusterservice : focus on dispatching services to member clusters
• endpointsliceCollect: focus on collecting endpointslices from member clusters
• endpointsliceDispatch: focus on dispatch endpointslices to member clusters

[Affan-7]

Hi @jwcesign, can you please assign me something to work on?