karmada

Changes since v1.2.1

Bug Fixes

• karmadactl: Fixed a panic issue when retrieving resources from an unknown cluster(karmadactl get xxx --cluster=not-exist). (#2171, #2201, @my-git9)
• karmadactl: Fix Kubernetes v1.24 can not be joined issue. (#1972, @zgfh)
• karmada-controller-manager: Fixed Argo CD can not assess Deployment health status issue. (#2256, @xuqianjins)
• karmada-controller-manager: Fixed Argo CD can not assess StatefulSet/DaemonSet health status issue. (#2264, @RainbowMango)
• karmada-search: Fixed can not query a single namespace issue. (#2274, @luoMonkeyKing)
• karmada-search: Fixed panic issue when dumps error info. (#2333, @AllenZMC)
• Helm Chart: Fixed misconfigured MutatingWebhookConfiguration and added missing APIService configuration for karmada-aggregated-apiserver. (#2420, @zhixian82)

Changes since v1.1.3

Bug Fixes

• karmadactl: Fix Kubernetes v1.24 can not be joined issue. (#2199, @zgfh)
• Helm Chart: Fixed misconfigured MutatingWebhookConfiguration and added missing APIService configuration for karmada-aggregated-apiserver. (#2421, @zhixian82)

Changes since v1.0.4

Bug Fixes

• karmadactl: Fixed Kubernetes v1.24 can not be joined issue. (#2200, @zgfh)
• Helm Chart: Fixed misconfigured MutatingWebhookConfiguration and added missing APIService configuration for karmada-aggregated-apiserver. (#2422, @zhixian82)

Changes since v1.2.0

Bug Fixes

• karmada-search: Fixed panic when the resource annotation is nil. (#1939, @XiShanYongYe-Chang )
• karmada-search: Fixed a panic issue comparing uncomparable type cache.ResourceEventHandlerFuncs. (#1971, @liys87x )
• karmadactl: fixed promoting failed if a resource with another kind using the same name has been promoted before. (#1983, @wuyingjun-lucky )
• karmadactl: Removed --dry-run flag from describe, exec and log commands. (#2036, @wlp1153468871 )
• karmada-controller-manager: Fixed Job status might be incorrectly marked as Completed issue. (#2007, @Garrybest )
• karmada-controller-manager/karmada-agent : fixed panic issue when dumps error info. (#2127, @AllenZMC )

Changes since v1.1.2

Bug Fixes

• karmadactl: Fixed namespace can not be customized issue. (#1827, @likakuli )
• karmadactl: fixed karmadactl can not taint while karmada control plane config is not located on default path. (#1838, @wuyingjun-lucky )
• karmada-controller-manager: Fixed Job status might be incorrectly marked as Completed issue. (#2010, @Garrybest )
• karmada-controller-manager/karmada-agent : fixed panic issue when dumps error info. (#2126, @AllenZMC )

Changes since v1.0.3

Bug Fixes

• karmadactl: Fixed namespace can not be customized issue. (#1826, @likakuli)
• karmadactl: fixed can not taint while karmada control plane config is not located on default path. (#1837, @wuyingjun-lucky)
• karmada-controller-manager: Fixed Job status might be incorrectly marked as Completed issue. (#2011, @Garrybest)
• karmada-controller-manager/karmada-agent : fixed panic issue when dumps error info. (#2133, @AllenZMC)

What's New

Significant improvement on scheduling capability and scalability

1. Karmada Descheduler

A new component karmada-descheduler was introduced, for rebalancing the scheduling decisions over time.
One example use case is: it helps evict pending replicas (Pods) from resource-starved clusters so that karmada-scheduler can "reschedule" these replicas (Pods) to a cluster with sufficient resources.
For more details please refer to Descheduler user guide.

(Feature contributor: @Garrybest)

2. Multi region HA support

By leveraging the newly added spread-by-region constraint, users are now able to deploy workloads aross regions, e.g. people may want their workloads always running on different regions for HA purposes.
We also introduced two plugins to karmada-scheduler, which add to accurate scheduling.

• ClusterLocality is a scoring plugin that favors clusters already assigned.
• SpreadConstraint is a filter plugin that filters clusters as per spread constraints.

(Feature contributors: @huone1, @gf457832386)

We are also in the progress of enhancing the multi-cluster failover mechanism. Part of the work has been included in this release.
For example:

• A new flag(--cluster-failure-threshold) has been added to both karmada-controller-manager and karmada-agent, which specifies the cluster failure threshold (defaults to 30s). A cluster will be considered not-ready only when it stays unhealthy longer than supposed.
• A new flag(--failover-eviction-timeout) has been added to karmada-controller-manager, which specifies the grace period of eviction (defaults to 5 minutes). If a cluster stays not-ready longer than supposed, the controller taints the cluster. (Note: The taint is essentially the eviction order and the implementation is planned for the next release.)

(Feature contributors: @Garrybest, @dddddai)

Fully adopted aggregated API

The Aggregated API was initially introduced in Release 1.0, which allows users to access clusters through Karmada by a single aggregated API endpoint. By leveraging this feature, we introduced a lot of interesting features to karmadactl and kubectl-karmada.

1. The get sub-command now supports clusters both in push and pull mode.

# karmadactl get deployment -n default
NAME      CLUSTER   READY   UP-TO-DATE   AVAILABLE   AGE     ADOPTION
nginx     member1   2/2     2            2           33h     N
nginx     member2   1/1     1            1           4m38s   Y
podinfo   member3   2/2     2            2           27h     N

2. The newly added logs command prints the container logs in a specific cluster.

# ./karmadactl logs nginx-6799fc88d8-9mpxn -c nginx  -C member1
/docker-entrypoint.sh: /docker-entrypoint.d/ is not empty, will attempt to perform configuration
/docker-entrypoint.sh: Looking for shell scripts in /docker-entrypoint.d/
/docker-entrypoint.sh: Launching /docker-entrypoint.d/10-listen-on-ipv6-by-default.sh
10-listen-on-ipv6-by-default.sh: info: Getting the checksum of /etc/nginx/conf.d/default.conf
...

3. We also added watch and exec commands to karmadactl, in addition to get and logs. They all use the aggregated API.

(Feature contributor: @lonelyCZ)

Distributed search and analytics engine for Kubernetes resources (alpha)

The newly introduced karmada-search caches resources in clusters and allows users to search for resources without directly touching real clusters.

# kubectl get --raw /apis/search.karmada.io/v1alpha1/search/cache/apis/apps/v1/deployments
{
	"apiVersion": "v1",
	"kind": "List",
	"metadata": {},
	"items": [{
		"apiVersion": "apps/v1",
		"kind": "Deployment",
		"metadata": {
			"annotations": {
				"cluster.karmada.io/name": "member1",
			},
		}
	},
	]
}

The karmada-search also supports syncing cached resources to backend stores like Elasticsearch or OpenSearch. By leveraging the search engine, you can perform full-text searches with all desired features, by field, and by indice; rank results by score, sort results by field, and aggregate results.

(Feature contributors: @huntsman-li, @liys87x)

Resource Interpreter Webhook enhancement

Introduced InterpretStatus for the Resource Interpreter Webhook framework, which enables customized resource status collection.
Karmada can thereby learn how to collect status for your resources, especially custom resources. For example, a custom resource may have many status fields and only Karmada can collect only those you want.

Refer to [Customizing Resource Interpreter][https://github.com/karmada-io/karmada/blob/master/docs/userguide/customizing-resource-interpreter.md] for more details.

(Feature contributor: @XiShanYongYe-Chang)

Integrating verification with the ecosystem

Benefiting from the Kubernetes native APIs, Karmada can easily integrate the Kubernetes ecosystem. The following components are verified by the Karmada community:

• Kyverno: policy engine. Refer to working with kyverno for more details.
• Gatekeeper: another policy engine. Refer to working with gatekeeper for more details.
• fluxcd: GitOps tooling for helm chart. Refer to working with fluxcd for more details.

(Feature contributors: @Poor12, @learner0810)

Other Notable Changes

Bug Fixes

• karmadactl: Fixed the cluster joining failures in the case of legacy secrets. (@zgfh, #1306)
• karmadactl: Fixed the issue that you cannot use the '-v 6' log level. (@zgfh, #1426)
• karmadactl: Fixed the issue that the --namespace flag of init command did not work. (@sayaoailun, #1416)
• karmadactl: Allowed namespaces to be customized. (@sayaoailun, #1449)
• karmadactl: Fixed the init failure due to data path not clean. (@prodanlabs, #1455)
• karmadactl: Fixed the init failure to read the KUBECONFIG environment variable. (@lonelyCZ, #1437)
• karmadactl: Fixed the init command failure to select the default release version. (@prodanlabs, #1456)
• karmadactl: Fixed the issue that the karmada-system namespace already exists when deploying karmada-agent. (@hanweisen, #1604)
• karmadactl: Fixed the issue that the karmada-controller-manager args did not honor customized namespaces.` (@prodanlabs, #1683)
• karmadactl: Fixed a panic due to nil annotation when promoting resources to Karmada.` (@duanmengkk, #1759)
• karmadactl: Fixed the promote command failure to migrate cluster-scoped resources. (@duanmengkk, #1766)
• karmadactl: fixed the karmadactl taint failure while the karmada control plane config is not located in the default path. (@wuyingjun-lucky, #1825)
• helm-chart: Fixed the karmada-agent installation failure due to the lack of permission. (@AllenZMC, #1457)
• helm-chart: Fixed the issue that version constraints skip pre-releases. (@pigletfly, #1444)
• karmada-controller-manager: Fixed the issue that ResourceBinding may hinder en-queue in the case of schedule failures. (@mrlihanbo, #1499)
• karmada-controller-manager: Fixed the panic when the interpreter webhook returns nil patch. (@CharlesQQ, #1584)
• karmada-controller-manager: Fixed the RB/CRB controller failure to aggregate status in the case of work condition update. (@mrlihanbo, #1513)
• karmada-aggregate-apiserver: Fixed timeout issue when requesting cluster/proxy with options -w or logs -f from karmadactl get. (@XiShanYongYe-Chang, #1620)
• karmada-aggregate-apiserver: Fixed exec failed: error: unable to upgrade connection: you must specify at least 1 of stdin, stdout, stderr. (@pangsq, #1632)

Features & Enhancements

• karmada-controller-manager: Introduced several flags to specify controller's concurrent capacities(--rate-limiter-base-delay, --rate-limiter-max-delay, --rate-limiter-qps, --rate-limiter-bucket-size). (@pigletfly, #1399)
• karmada-controller-manager: The klog flags now have been grouped for better readability. (@RainbowMango, #1468)
• karmada-controller-manager: Fixed the FullyApplied condition of ResourceBinding/ClusterResourceBinding mislabeling issue in the case of non-scheduling. (@huone1, #1512)
• karmada-controller-manager: Added default AggregateStatus webhook for DaemonSet and StatefulSet. (@Poor12, #1586)
• karmada-controller-manager: OverridePolicy with empty ResourceSelector will be considered to match all resources just like nil. (@likakuli, #1706)
• karmada-controller-manager: Introduced --failover-eviction-timeout to specify the grace period of eviction. Tants(cluster.karmada.io/not-ready or cluster.karmada.io/unreachable) will be set on unhealthy clusters after the period. (@Garrybest, #1781)
• karmada-controller-manager/karmada-agent: Introduced --cluster-failure-threshold flag to specify cluster failure threshold. (@dddddai, #1829)
• karmada-scheduler: Workloads can now be rescheduled after the cluster is unregistered. (@huone1, #1383)
• karmada-scheduler: The klog flags now have been grouped for better readability. (@jameszhangyukun, #1491)
• karmada-scheduler: Added a scoring plugin ClusterLocality to favor clusters already requested. (@huone1, #1334)
• karmada-scheduler: Introduced filter plugin SpreadConstraint to filter clusters that do not meet the spread constraints. (@gf457832386, #1570)
• karmada-scheduler: Supported spread constraints by region strategy. (@huone1, #1646)
• karmada-webhook: Introduced --tls-cert-file-name and --tls-private-key-file-name flags to specify the server certificate and private key. (@mrlihanbo, #1464)
• karmada-agent: The klog flags now have been grouped for better readability. (@lonelyCZ, #1389)
• • karmada-agent: Introduced several flags to specify controller's concurrent capacities(--rate-limiter-base-delay, --rate-limiter-max-delay, --rate-limiter-qps, --rate-limiter-bucket-size). (@dddddai, #1505)
• karmada-scheduler-estimator: The klog flags now have been grouped for better readability. (@AllenZMC, #1493)
• karmadactl: Introduced --context flag to specify the context name to use. (@lonelyCZ, #1748)
• karmadactl: Introduced --kube-image-mirror-country and --kube-image-registry flags to init subcommand for Chinese mainland users. (@wuyingjun-lucky, #1764)
• karmadactl: Introduced deinit sub-command to uninstall Karmada. (@prodanlabs, #1337)
• Introduced Swagger docs for Karmada API. (@lonelyCZ, #1401)

Other (Dependencies)

• The base image alpine has been promoted to v3.15.1. (@RainbowMango, #1519)

Deprecation

• karmada-controller-manager: The hpa controller is disabled by default now. (@Poor12, #1580)
• karmada-aggregated-apiserver: The deprecated flags --karmada-config and --master in v1.1 have been removed from the codebase. (@AllenZMC, #1834)

Contributors

Thank you to everyone who contributed to this release!

Users whose commits are in this release (alphabetically by user name)

• @AllenZMC
• @anu491
• @carlory
• @CharlesQQ
• @chaunceyjiang
• @chinmaym07
• @CuiDengdeng
• @dddddai
• @duanmeng
• @duanmengkk
• @ErikJiang
• @fanzhihai0215
• @fleeto
• @Garrybest
• @gf457832386
• @gy95
• @hanweisen
• @huiwq1990
• @huntsman-li
• @huone1
• @ikaven1024
• @jameszhangyukun
• @kerthcet
• @learner0810
• @lfbear
• @likakuli
• @liys87x
• @lonelyCZ
• @lvyanru8200
• @mikeshng
• @mrlihanbo
• @my-git9
• @pangsq
• @pigletfly
• @Poor12
• @prodanlabs
• @RainbowMango
• @sayaoailun
• @snowplayfire
• @stingshen
• @Tingtal
• @wuyingjun-lucky
• @wwwnay
• @XiShanYongYe-Chang
• @xyz2277
• @YueHonghui
• @zgfh
• @zirain

Changes since v1.1.1

Bug Fixes

• karmadactl: fixed the karmada-systemnamespace already existing issue when deployingkarmada-agent` issue. (#1608, @hanweisen)
• karmadactl: Fixed karmada-controller-manager args not honor customized namespace issue. (#1689, @prodanlabs)
• karmada-controller-manager: Fixed ResourceBinding maybe prevents en-queue in case of schedule failure. (#1507, @mrlihanbo)
• karmada-controller-manager: Fixed the FullyApplied condition of ResourceBinding/ClusterResourceBinding mislabeling issue in case of non-scheduled. (#1517, @huone1)
• karmada-controller-manager: Fixed RB/CRB controller can't aggregate status in case of work condition update issue. (#1523, @mrlihanbo)
• karmada-controller-manager: Fixed panic in case of interpreter webhook returns nil patch. (#1592, @CharlesQQ)
• karmada-aggregate-apiserver: Fixed timeout issue when request cluster/proxy with options -w or logs -f from karmadactl get. (#1630, @XiShanYongYe-Chang)
• karmada-aggregate-apiserver: Fixed exec failed: error: unable to upgrade connection: you must specify at least 1 of stdin, stdout, stderr. (#1642, @XiShanYongYe-Chang)

Other

• The base image alpine has been promoted to v3.15.1. (#1583, @RainbowMango)

Changes since v1.0.2

Bug Fixes

• karmadactl: fixed the karmada-systemnamespace already existing issue when deployingkarmada-agent` issue. (#1609, @hanweisen)
• karmadactl: Fixed karmada-controller-manager args not honor customized namespace issue. (#1690, @prodanlabs)
• karmada-controller-manager: Fixed the FullyApplied condition of ResourceBinding/ClusterResourceBinding mislabeling issue in case of non-scheduled. (#1518, @huone1)
• karmada-controller-manager: Fixed RB/CRB controller can't aggregate status in case of work condition update issue. (#1524, @mrlihanbo)
• karmada-controller-manager: Fixed panic in case of interpreter webhook returns nil patch. (#1591, @CharlesQQ)
• karmada-aggregate-apiserver: Fixed timeout issue when request cluster/proxy with options -w or logs -f from karmadactl get. (#1631, @XiShanYongYe-Chang)
• karmada-aggregate-apiserver: Fixed exec failed: error: unable to upgrade connection: you must specify at least 1 of stdin, stdout, stderr. (#1641, @pangsq)

Other

• The base image alpine has been promoted to v3.15.1. (#1582, @RainbowMango)

Changes since v1.1.0

Bug Fixes

• karmadactl: Fixed --namespace flag of init command not work issue. (#1452, @sayaoailun)
• karmadactl: Fixed init failure due to data path not clean issue. (#1473, @prodanlabs)
• karmadactl: Fixed init can not read KUBECONFIG environment variable issue. (#1482, @lonelyCZ)
• karmadactl: Fixed init can not select default release version issue. (#1496, @prodanlabs)
• helm chart: Fixed version constraints skip the pre-releases issue. (#1466, @pigletfly)

Changes since v1.0.1

Bug Fixes

• karmadactl: Fixed --namespace flag of init command not work issue. (#1452, @sayaoailun)
• karmadactl: Fixed init failure due to data path not clean issue. (#1473, @prodanlabs)
• karmadactl: Fixed init can not select default release version issue. (#1495, @prodanlabs)
• karmadactl: Fixed init can not read KUBECONFIG environment variable issue. (#1482, @lonelyCZ)
• helm chart: Fixed version constraints skip the pre-releases issue. (#1466, @pigletfly)
• karmada-controller-manager: Fixed a bug where resource binding is not created occasionally. (#1384, @dddddai)

What's New

Multi-Cluster Ingress

The newly introduced MultiClusterIngress API exposes HTTP and HTTPS routes that target multi-cluster services within the Karmada control plane. The specification of MultiClusterIngress is compatible with Kubernetes Ingress.

Traffic routing is controlled by rules defined on the MultiClusterIngress resource, an MultiClusterIngress controller is responsible for fulfilling the ingress. The Multi-Cluster-Nginx Ingress Controller is one of the MultiClusterIngress controller implementations maintained by the community.

(Feature contributors: @GitHubxsy @XiShanYongYe-Chang)

Federated ResourceQuota

The newly introduced FederatedResourceQuota provides constraints that limit total resource consumption per namespace across all clusters. It can limit the number of objects that can be created in a namespace by type, as well as the total amount of compute resources that may be consumed by resources in that namespace.

(Feature contributors: @RainbowMango @XiShanYongYe-Chang)

Configurability improvement for performance tuning

The default number of reconciling workers has been enlarged and configurable. A larger number of workers means higher responsiveness but heavier CPU and network load. The number of concurrent workers could be configured by the flags introduced to karmada-controller-manager and karmada-agent.

Flags introduced to karmada-controller-manager:

• --concurrent-work-syncs
• --concurrent-namespace-syncs
• --concurrent-resource-template-syncs
• --concurrent-cluster-syncs
• --concurrent-clusterresourcebinding-syncs
• --concurrent-resourcebinding-syncs

Flags introduced to karmada-agent:

• --concurrent-work-syncs
• --concurrent-cluster-syncs

(Feature contributor: @pigletfly)

Resource Interpreter Webhook Enhancement

Introduced AggregateStatus support for the Resource Interpreter Webhook framework, which enables customized resource status aggregating.

Introduced InterpreterOperationInterpretDependency support for the Resource Interpreter Webhook framework, which enables propagating workload's dependencies automatically.

Refer to Customizing Resource Interpreter for more details.

(Feature contributors: @iawia002 @mrlihanbo)

Other Notable Changes

Bug Fixes

• karmadactl and kubectl-karmada: Fixed that init cannot update the APIService. (@prodanlabs, #1207)
• karmada-controller-manager: Fixed ApplyPolicySucceed event type mistake (should be Normal but not Warning). (@Garrybest, #1267)
• karmada-controller-manager and karmada-agent: Fixed that resync slows down reconciliation. (@Garrybest, #1265)
• karmada-controller-manager/karmada-agent: Fixed continually updating cluster status due to unordered apiEnablements. (@pigletfly, #1304)
• karmada-controller-manager: Fixed that Replicas set by OverridePolicy will be reset by the ReviseReplica interpreterhook. (@likakuli, #1352)
• karmada-controller-manager: Fixed that ResourceBinding couldn't be created in a corner case. (@dddddai, #1368)
• karmada-scheduler: Fixed inaccuracy in requested resources in the case that pod limits are specified but requests are not. (@Garrybest, #1225)
• karmada-scheduler: Fixed spreadconstraints[i].MaxGroups is invalidated in some scenarios. (@huone1, #1324)

Features & Enhancements

• karmadactl: Introduced --tls-min-version flag to specify the minimum TLS version. (@carlory, #1278)
• karmadactl: Improved the get command to show more useful information. (@lonelyCZ, #1270)
• karmada-controller-manager/karmada-agent: Introduced --resync-period flag to specify reflector resync period (defaults to 0, meaning no resync). (@Garrybest, #1261)
• karmada-controller-manager: Introduced --metrics-bind-address flag to specify the customized address for metrics. (@pigletfly, #1341)
• karmada-webhook: Introduced --metrics-bind-address and --health-probe-bind-address flags. (@mrlihanbo, #1346)

Instrumentation (Metrics and Events)

• karmada-controller-manager: Fixed ApplyPolicySucceed event type mistake (should be Normal but not Warning). (@Garrybest, #1267)

Deprecation

• OverridePolicy/ClusterOverridePolicy: The .spec.targetCluster and spec.overriders have been deprecated in favor of spec.overrideRules. (@RainbowMango #1238)
• karmada-aggregate-apiserver: Deprecated --master and --karmada-config flags. Please use --kubeconfig instead. (@carlory, #1336)

Contributors

Thank you to everyone who contributed to this release!

Users whose commits are in this release (alphabetically by user name)

• @AllenZMC
• @ashley-rongfang
• @carlory
• @CuiDengdeng
• @dddddai
• @EvaDD
• @Fish-pro
• @Garrybest
• @helen-frank
• @huone1
• @iawia002
• @jinglina
• @kerthcet
• @liangyongzhenya
• @likakuli
• @lonelyCZ
• @mrlihanbo
• @pigletfly
• @prodanlabs
• @RainbowMango
• @RishiKumarRay
• @Tingtal
• @viniciuspietscher
• @weilaaa
• @wlp1153468871
• @XiShanYongYe-Chang
• @zach593
• @zgfh

Changes since v1.0.0

Bug Fixes

• karmadactl and kubectl-karmada: Fixed init can not update the APIService issue. (#1207, @prodanlabs )
• karmada-controller-manager: Fixed ApplyPolicySucceed event type mistake(should be Normal but Warning). (#1267, @Garrybest )
• karmada-controller-manager and karmada-agent: Fixed resync slow down reconcile issue. (#1265, @Garrybest )

What's New

Aggregated Kubernetes API Endpoint

The newly introduced karmada-aggregated-apiserver component aggregates all registered clusters and allows users to access member clusters through Karmada by the proxy endpoint, e.g.

- Retrieve `Node` from `member1`:  /apis/cluster.karmada.io/v1alpha1/clusters/member1/proxy/api/v1/nodes
- Retrieve `Pod` from `member2`: /apis/cluster.karmada.io/v1alpha1/clusters/member2/proxy/api/v1/namespaces/default/pods

Please refer to user guide for more details.

(Feature contributor: @kevin-wangzefeng @GitHubxsy @XiShanYongYe-Chang @mrlihanbo @jrkeen @prodanlabs @carlory @RainbowMango)

Promoting Workloads from Legacy Clusters to Karmada

Legacy workloads running in Kubernetes now can be promoted to Karmada smoothly without container restart.
In favor of promote commands added to Karmada CLI, any kind of Kubernetes resources can be promoted to Karmada easily, e.g.

# Promote deployment(default/nginx) from cluster1 to Karmada
kubectl karmada promote deployment nginx -n default -c cluster1

(Feature contributor: @lonelyCZ @iawia002 @dddddai)

Verified Integration with Ecosystem

Benefiting from the Kubernetes native API support, Karmada can easily integrate the single cluster ecosystem for multi-cluster, multi-cloud purpose. The following components have been verified by the Karmada community:

• argo-cd: refer to working with argo-cd
• Flux: refer to propagating helm charts with flux
• Istio: refer to working with Istio
• Filebeat: refer to working with Filebeat
• Submariner: refer to working with Submariner
• Velero: refer to working with Velero
• Prometheus: refer to working with Prometheus

(Feature contributor: @lfbear @learner0810 @zirain @Rains6 @gy95 @XiShanYongYe-Chang )

OverridePolicy Improvements

By leverage of the new-introduced RuleWithCluster fields to OverridePolicy and ClusterOverridePolicy, users are now able to define override policies with a single policy for specified workloads.

(Feature contributor: @iawia002 @lfbear @RainbowMango @lonelyCZ @jameszhangyukun )

Karmada Installation Improvements

Introduced init command to Karmada CLI. Users are now able to install Karmada by a single command.

Please refer to Installing Karmada for more details.

(Feature contributor: @prodanlabs @lonelyCZ @jrkeen )

Configuring Karmada Controllers

Now all controllers provided by Karmada work as plug-ins. Users are now able to turn off any of them from the default enabled list.
See --controllers flag of karmada-controller-manager and karmada-agent for more details.

(Feature contributor: @snowplayfire @iawia002 @jameszhangyukun )

Resource Interpreter Webhook Enhancement

Introduced ReviseReplica support for the Resource Interpreter Webhook framework, which enables scheduling all customized workloads just like Kubernetes native ones.

Refer to Resource Interpreter Webhook Proposal for more design details.

(Feature contributor: @iawia002)

Other Notable Changes

Bug Fixes

• karmada-controller-manager: Fixed the issue that the annotation of resource template cannot be updated. (@mrlihanbo #1012)
• karmada-controller-manager: Fixed the issue of generating binding reference key. (@JarHMJ #1003)
• karmada-controller-manager: Fixed the inefficiency of en-queue failed task issue. (@Garrybest #1068)

Features & Enhancements

• Karmada CLI: Introduced --cluster-provider flag to join command to specify provider of joining cluster. (@2hangchen #1025)
• Karmada CLI: Introduced taint command to set taints for clusters. (@lonelyCZ #889)
• Karmada CLI: The Applied condition of Work and Scheduled/FullyApplied of ResourceBinding are available for kubectl get. (@lonelyCZ #1110)
• karmada-controller-manager: The cluster discovery feature now supports v1beta1 of cluster-api. (@iawia002 #1029)
• karmada-controller-manager: The Job's startTime and completionTime now available at resource template. (@Garrybest #1034)
• karmada-controller-manager: introduced --controllers flag to enable or disable controllers. (@snowplayfire #1083)
• karmada-controller-manager: Support retain ownerReference from observed objects. (@snowplayfire #1116)
• karmada-controller-manager and karmada-agent: Introduced cluster-cache-sync-timeout flag to specify the time waiting for cache sync. (@snowplayfire #1112)

Instrumentation (Metrics and Events)

• karmada-scheduler-estimator: Introduced /metrics endpoint to emit metrics. (@Garrybest #1030)
• Introduced ApplyPolicy and ScheduleBinding events for resource template. (@mrlihanbo #1070)

Deprecation

• The ReplicaSchedulingPolicy API deprecated at v0.9.0 now has been removed in favor of ReplicaScheduling of PropagationPolicy. (@iawia002 #1161)

Contributors

Thank you to everyone who contributed to this release!

Users whose commits are in this release (alphabetically by user name)

• @2hangchen
• @aven-ai
• @BDXGD
• @carlory
• @dddddai
• @eightzero
• @fanzhihai0215
• @feeltimeQ
• @fleeto
• @Garrybest
• @ghl116
• @gy95
• @haiker2011
• @Haleygo
• @iawia002
• @imroc
• @JackZxj
• @jameszhangyukun
• @JarHMJ
• @jrkeen
• @kevin-wangzefeng
• @leonharetd
• @lfbear
• @lonelyCZ
• @mrlihanbo
• @Phil-sun
• @pigletfly
• @prodanlabs
• @RainbowMango
• @Rains6
• @Shike-Ada
• @snowplayfire
• @wawa0210
• @XiShanYongYe-Chang
• @zirain

Changes since v0.10.0

Bug Fixes

• karmada-controller-manager: fixed the issue of generating binding reference key. (#1003, @JarHMJ)
• karmada-controller-manager: fixed resource template annotations can not update issue. (#1012, @mrlihanbo)

What's New

Resource Interpreter Webhook

The newly introduced Resource Interpreter Webhook framework allows users to implement their own CRD plugins that will be consulted at all parts of propagation process. With this feature, CRDs and CRs will be propagated just like Kubernetes native resources, which means all scheduling primitives also support custom resources. An example as well as some helpful utilities are provided to help users better understand how this framework works.

Refer to Proposal for more details.

(Feature contributor: @RainbowMango, @XiShanYongYe-Chang, @gy95)

Significant Scheduling Enhancement

1. Introduced dynamicWeight primitive to PropagationPolicy and ClusterPropagationPolicy. With this feature, replicas could be divided by a dynamic weight list, and the weight of each cluster will be calculated based on the available replicas during scheduling.
   This feature can balance the cluster's utilization significantly. (#841)

2. Introduced Job schedule (divide) support. A Job that desires many replicas now could be divided into many clusters just like Deployment.
   This feature makes it possible to run huge Jobs across small clusters. (#898)

(Feature contributor: @Garrybest )

Workloads Observation from Karmada Control Plane

After workloads (e.g. Deployments) are propagated to member clusters, users may also want to get the overall workload status across many clusters, especially the status of each pod. In this release, a get subcommand was introduced to the kubectl-karmada. With this command, user are now able get all kinds of resources deployed in member clusters from the Karmada control plane.

For example (get deployment and pods across clusters):

$ kubectl karmada get deployment
NAME    CLUSTER   READY   UP-TO-DATE   AVAILABLE   AGE   ADOPTION
nginx   member2   1/1     1            1           19m   Y
nginx   member1   1/1     1            1           19m   Y
$ kubectl karmada get pods
NAME                     CLUSTER   READY   STATUS    RESTARTS   AGE
nginx-6799fc88d8-vzdvt   member1   1/1     Running   0          31m
nginx-6799fc88d8-l55kk   member2   1/1     Running   0          31m

(Feature contributor: @lfbear @QAQ-rookie)

Other Notable Changes

• karmada-scheduler-estimator: The number of pods becomes an important reference when calculating available replicas for the cluster. (@Garrybest, #777)
• The labels (resourcebinding.karmada.io/namespace, resourcebinding.karmada.io/name, clusterresourcebinding.karmada.io/name) which were previously added on the Work object now have been moved to annotations. (@XiShanYongYe-Chang, #752)
• Bugfix: Fixed the impact of cluster unjoining on resource status aggregation. (@dddddai, #817)
• Instrumentation: Introduced events (SyncFailed and SyncSucceed) to the Work object. (@wawa0210, #800)
• Instrumentation: Introduced condition (Scheduled) to the ResourceBinding and ClusterResourceBinding. (@dddddai, #823)
• Instrumentation: Introduced events (CreateExecutionNamespaceFailed and RemoveExecutionNamespaceFailed) to the Cluster object. (@pigletfly, #749)
• Instrumentation: Introduced several metrics (workqueue_adds_total, workqueue_depth, workqueue_longest_running_processor_seconds, workqueue_queue_duration_seconds_bucket) for karmada-agent and karmada-controller-manager. (@Garrybest, #831)
• Instrumentation: Introduced condition (FullyApplied) to the ResourceBinding and ClusterResourceBinding. (@lonelyCZ, #825)
• karmada-scheduler: Introduced feature gates. (@iawia002, #805)
• karmada-controller-manager: Deleted resources from member clusters that use "Background" as the default delete option. (@RainbowMango, #970)

Contributors

Thank you to everyone who contributed to this release!

Users whose commits are in this release (alphabetically by user name)

• @2hangchen
• @algebra2k
• @benjaminhuo
• @Charlie17Li
• @ctripcloud
• @dddddai
• @duguhaotian
• @fleeto
• @Garrybest
• @gf457832386
• @gy95
• @hyschumi
• @iawia002
• @jameszhangyukun
• @kerthcet
• @kevin-wangzefeng
• @learner0810
• @lfbear
• @lonelyCZ
• @mrlihanbo
• @penghuima
• @Phil-sun
• @pigletfly
• @QAQ-rookie
• @RainbowMango
• @snowplayfire
• @TeodoraBoros
• @wawa0210
• @wzshiming
• @XiShanYongYe-Chang
• @youhonglian
• @yvoilee

What's New

Upgrading support
Users are now able to upgrade from the previous version smoothly. With the multiple version feature of CRD, objects with different schemas can be automatically converted between versions. Karmada uses the semantic versioning and will provide workarounds for inevitable breaking changes.

In this release, the ResourceBining and ClusterResourceBinding promote to v1alpha2 and the previous v1alpha1 version is still available for one more release. With the upgrading instruction, the previous version of Karmada can promote smoothly.

(Feature contributor: @RainbowMango )

Introduced karmada-scheduler-estimator to facilitate end-to-end scheduling accuracy
Karmada scheduler aims to assign workload to clusters according to constraints and available resources of each member cluster. The kube-scheduler working on each cluster takes the responsibility to assign Pods to Nodes.
Even though Karmada has the capacity to reschedule failure workload between member clusters, but the community still commits lots of effort to improve the accuracy of the end-to-end scheduling.

The karmada-scheduler-estimator is the effective assistant of karmada-scheduler, it provides prediction-based scheduling decisions that can significantly improve the scheduling efficiency and avoid the wave of rescheduling among clusters. Note that this feature is implemented as a pluggable add-on. For the instructions please refer to scheduler estimator guideline.

(Feature contributor: @Garrybest )

Maintainability improvements
A bunch of significant maintainability improvements were added to this release, including:

• Simplified Karmada installation with helm chart.
  (Feature contributor: @algebra2k @jrkeen )

• Provided metrics to observe scheduler status, the metrics API now served at /metrics of karmada-scheduler.
  With these metrics, users are now able to evaluate the scheduler's performance and identify the bottlenecks.
  (Feature contributor: @qianjun1993 )

• Provided events to Karmada API objects as supplemental information to debug problems.
  (Feature contributor: @pigletfly )

Other Notable Changes

• karmada-controller-manager: The ResourceBinding/ClusterResourceBinding won't be deleted after associate PropagationPolicy/ClusterPropagationPolicy is removed and is still available until resource template is removed.(@qianjun1993, #601)
• Introduced --leader-elect-resource-namespace which is used to specify the namespace of election object to components karmada-controller-manager/karmada-scheduler/karmada-agent`. (@XiShanYongYe-Chang #698)
• Deprecation: The API ReplicaSchedulingPolicy has been deprecated and will be removed from the following release. The feature now has been integrated into ReplicaScheduling.
• Introduced kubectl-karmada commands as the extensions for kubectl. (@XiShanYongYe-Chang #686)
• karmada-controller-manager introduced a version command to represent version information. (@RainbowMango #717 )
• karmada-scheduler/karmada-webhook/karmada-agent/karmada-scheduler-estimator introduced a version command to represent version information. (@lonelyCZ #719 )
• Provided instructions about how to use the Submariner to connect the network between member clusters. (@XiShanYongYe-Chang #737 )
• Added four metrics to the karmada-scheduler to monitor scheduler performance. (@qianjun1993 #747)

Contributors

Thank you to everyone who contributed to this release!

Users whose commits are in this release (alphabetically by user name)

• @2hangchen
• @CKchen0726
• @dddddai
• @ESonata
• @fleeto
• @Garrybest
• @Hrishikesh156
• @iawia002
• @jrkeen
• @just1900
• @kerthcet
• @lfbear
• @lonelyCZ
• @mrlihanbo
• @MukulKolpe
• @Onyinye91-ctrl
• @phantooom
• @pigletfly
• @qianjun1993
• @RainbowMango
• @saiyan86
• @smartding
• @XiShanYongYe-Chang
• @zhuyaguang

What's New

Automatic cluster discovery with cluster-api
For users who are using cluster-api (sigs.k8s.io/cluster-api), Karmada is now able to automatically discover & join clusters when provisioned, and unjoin them in case of destroyed.

Note that this features is implemented as a built-in plugin. To enbale it, simply indicate the following to flags in karmada-controller-manager config:

--cluster-api-kubeconfig string        Path to the cluster-api management cluster kubeconfig file.
--cluster-api-context string           Name of the cluster context in cluster-api management cluster kubeconfig file.

(Feature contributor: @XiShanYongYe-Chang )

Introduced CommandOverrider and ArgsOverrider to simplify commands customization per cluster
For multi-cluster applications, it's quite common to set different arguments when running on different clusters or environments.
In this release, two overrider plugins: CommandOverrider and ArgsOverrider are introduced, based on industry best practices. These two handy tools allow users to declare complex declarations and avoid configuration mistakes.

Workload types supported now are: Deployment, ReplicaSet, DaemonSet, StatefulSet and Pod, more types including CRDs will be supported in later releases.

(Feature contributor: @lfbear @betaincao )

Better integration support with Kubernetes ecosystem
The Kubernetes native APIs support and patterns to run cloud-native applications of Karmada make it quite easy to quickly integrate with other projects in the Kubernetes ecosystem.

In release, several useful features that will help Karmada work seamlessly with other systems.

• ResourceBinding and ClusterResourceBinding now supports present the applied status. (@pigletfly #595)
• More types of resources now support aggregating status to the resource template, inlcuding Job, Service, and Ingress. (@mrlihanbo #609)
• argo-cd is also verified to run full featured with Karmada to achieve multi-cluster GitOps.

Other Notable Changes

• karmadactl: introduced cordon and uncordon commands to mark a cluster schedulable and un-schedulable. (#464, @algebra2k )
• karmada-controller-manager: introduced --skipped-propagating-namespaces flag to skip resources in certain namespaces from propagating. (#533, @pigletfly )
• karmada-controller-manager/karmada-agent/karmada-scheduler: Introduced flags to config the QPS and burst which are used to control the client traffic interacting with Karmada or member cluster's kube-apiserver. (#611, @Garrybest )
  • --cluster-api-qps QPS to use while talking with cluster kube-apiserver.
  • --cluster-api-burst Burst to use while talking with cluster kube-apiserver.
  • --kube-api-qps QPS to use while talking with karmada-apiserver.
  • --kube-api-burst Burst to use while talking with karmada-apiserver.
• Karmada quick-start scripts now support running on Mac OS. (#538, @lfbear )

Contributors

Thank you to everyone who contributed to this release!

Users whose commits are in this release (alphabetically by user name)

• @algebra2k
• @betaincao
• @garfcat
• @Garrybest
• @gy95
• @Iceber
• @lfbear
• @lushenle
• @mrlihanbo
• @phantooom
• @pigletfly
• @qianjun1993
• @RainbowMango
• @wawa0210
• @weilaaa
• @XiShanYongYe-Chang

What's New

Support multi-cluster service discovery
In many cases, a Kubernetes user may want to split their deployments across multiple clusters, but still retain mutual dependencies between workloads running in those clusters.

Users are now able to export and import services between clusters with Multi-Cluster Service API (MCS-API). (@XiShanYongYe-Chang)

Support more precise cluster status management
Besides reporting cluster status, the cluster status controller now also renews the lease. The newly introduced cluster monitor monitors the lease and will mark cluster ready status to unknown in case of cluster status controller not working. (@Garrybest)

Support replica scheduling based on cluster resources
In some scenarios, users want to divide the replicas in a deployment to multiple clusters if a single cluster doesn't have sufficient resources.
Users now able to declare the replica scheduling preference by the new field ReplicaDivisionPreference in PropagationPolicy and ClusterPropagationPolicy. (@qianjun1993)

Support more convenient APIs to divide replicas by weight list
Users now able to declare cluster weight by ReplicaDivisionPreference in PropagationPolicy and ClusterPropagationPolicy, with the preference Weighted, the scheduler will divide replicas according to the WeightPreference. (@qianjun1993)

This feature is designed to replace the standalone ReplicaSchedulingPolicy API in the future.

Other Notable Changes

• karmada-agent: Introduced --karmada-context flag to indicate the cluster context in karmada kubeconfig file. (#415, @mrlihanbo)
• karmada-agent and karmada-controller-manager: Introduced --cluster-lease-duration and --cluster-lease-renew-interval-fraction flags to specify the lease expiration period and renew interval fraction. (#421, @pigletfly)
• karmada-scheduler: Added a filter plugin to prevent the cluster from scheduling if the required API is not installed. (#470, @vincent-pli)
• karmada-controller-manager: Introduced --skipped-propagating-apis flag to skip the resources from propagating. (#345, @pigletfly)
• Installation: Now the hack/deploy-karmada.sh and hack/deploy-karmada-agent.sh scripts support install Karmada components on both Kind clusters and standalone clusters. (#458, @lfbear)
• In the case of resources already in member clusters, in order to avoid conflict karmada will refuse to propagate and adopt the resource by default. (#471, @mrlihanbo)

Contributors

Thank you to everyone who contributed to this release!

Users whose commits are in this release (alphabetically by user name)

• @garfcat
• @Garrybest
• @gy95
• @huiwq1990
• @likakuli
• @lfbear
• @mrlihanbo
• @pigletfly
• @qianjun1993
• @RainbowMango
• @RhnSharma
• @shinytang6
• @vincent-pli
• @XiShanYongYe-Chang

What's New

Support syncing with member cluster behind proxy
In some scenarios where certain clusters may not be directly connected from the Internet, such as:

• The member clusters are behind a NAT gateway from the Karmada control plane
• The member clusters are in an on-prem Intranet while Karmada runs in the cloud

By setting proxy-url in the kubeconfig when registering member clusters, Karmada will talk to member clusters through indicated proxy. (#307, @liufen90)

Introduced ImageOverrider for simplifying image replacement
In most scenarios where clusters are running in different cloud or data centers, the workload requires a different image registry. ImageOverrider is a handy tool to override images for a workload before they are propagated to clusters. (#370, @XiShanYongYe-Chang)

Support scheduling based on cluster taint toleration
Karmada-scheduler now reflects taints on member clusters and tolerations defined in PropagationPolicy and ClusterPropagationPolicy
when scheduling resources. (#320, @mrlihanbo)

Support scheduling based on cluster topology
Karmada-scheduler now supports scheduling resources according to the topology information(cluster/provider/region/zone)
defined in cluster objects. (#357, @mrlihanbo)

Other Notable Changes

• Installation: introduced hack/remote-up-karmada.sh to install Karmada on a specified Kubernetes as host. (#367, @lfbear)
• karmadactl: introduced the version command to show the version it is built from. Try it on by command: # karmadactl version. (#285, @algebra2k)
• API: added short name for most APIs. (#376, @pigletfly)
• The resource templates now match PropagationPolicy or ClusterPropagationPolicy in alphabetical order
  when there are multiple policies that match. (#306, @XiShanYongYe-Chang)
• Always generates ResourceBinding objects for namespace-scoped resource template. (#315, @vincent-pli)
• karmada-controller-manager: introduced the leader-elect command line flag to enable or disable leadership election. (#321, @pigletfly)
• The Work objects name now consist of the resource template's .metada.name, .metada.kind and .metadata.namespace. (#359, @Garrybest)

Contributors

Thank you to everyone who contributed to this release!

Users whose commits are in this release (alphabetically by user name)

• @algebra2k
• @anirudhramnath
• @daixiang0
• @futuretea
• @Garrybest
• @gy95
• @hantmac
• @huiwq1990
• @Iceber
• @kevin-wangzefeng
• @leofang94
• @LeoLiuYan
• @liufen90
• @lfbear
• @mrlihanbo
• @pigletfly
• @RainbowMango
• @vincent-pli
• @XiShanYongYe-Chang
• @yangcheng-icbc