Support YurtAppDaemon to deploy workload to different NodePools
A YurtAppDaemon ensures that all (or some) NodePools run a copy of a Deployment or StatefulSet. As nodepools are added to the cluster, Deployment or StatefulSet are added to them. As nodepools are removed from the cluster, those Deployments or StatefulSet are garbage collected. The behavior of YurtAppDaemon is similar to that of DaemonSet, except that YurtAppDaemon creates workloads from a node pool. More information can be found at: (#422, yurt-app-manager, @kadisi)
Using YurtIngress to unify service across NodePools
YurtIngress acts as a unified interface for services access request from outside the NodePool, it abstracts and simplifies service access logic to users, it also reduces the complexity of NodePool services management. More information can be found at: (#373, #645, yurt-app-manager, @zzguang, @gnunu, @LindaYu17)
Improve the user experience of OpenYurt
- OpenYurt Experience Center
New users who want to try out OpenYurt's capabilities do not need to install an OpenYurt cluster from scratch. They can apply for a test account on the OpenYurt Experience Center and immediately have an OpenYurt cluster available. More information can be found at: (OpenYurt Experience Center Introduction, @luc99hen, @qclc, @Peeknut)
- YurtCluster
This YurtCluster Operator is to translate a vanilla Kubernetes cluster into an OpenYurt cluster, through a simple API (YurtCluster CRD). And we recommend that you do the cluster conversion based on the declarative API of YurtCluster Operator. More information can be found at: (#389, #518, yurtcluster-operator, @SataQiu, @gnunu)
- Yurtctl init/join
In order to improve efficiency of creating OpenYurt cluster, a new tool named sealer has
been integrated into yurtctl init
command. and OpenYurt cluster image(based on Kubernetes v1.19.7 version) has been prepared.
Users can use yurtctl init
command to create OpenYurt cluster control-plane, and use yurtctl join
to add worker nodes(including
cloud nodes and edge nodes). More information can be found at: (#704, #697, @Peeknut, @rambohe-ch, @adamzhoul)
Update docs of OpenYurt
The docs of OpenYurt installation, core concepts, user manuals, developer manuals etc. have been updated, and all of them are located at OpenYurt Docs. Thanks to all contributors for maintaining docs for OpenYurt. (@huangyuqi, @kadisi, @luc99hen, @SataQiu, @mowangdk, @rambohe-ch, @zyjhtangtang, @qclc, @Peeknut, @Congrool, @zzguang, @adamzhoul, @windydayc, @villanel)
- Proposal: enhance cluster networking capabilities (#637, @DrmagicE, @BSWANG)
- add node-servant (#516, adamzhoul)
- improve yurt-tunnel-server to automatically update server certificates when service address changed (#525, @YRXING)
- automatically clean dummy interface and iptables rule when yurthub is stopped by k8s (#530, @Congrool)
- enhancement: add openyurt.io/skip-discard annotation verify for discardcloudservice filter (#524, @rambohe-ch)
- inject working_mode (#552, @ngau66)
- Yurtctl revert adds the function of deleting yurt app manager (#555, @yanyhui)
- Add edge device demo in the README.md (#553, #554, @Fei-Guo, @qclc, @lwmqwer)
- Refactor: separate the creation of informers from tunnel server component (#585, @YRXING)
- add make push for pushing images generated during make release (#601, @gnunu)
- add trafficforward that contains two diversion modes: DNAT and DNS (#606, @JcJinChen)
- yurthub verify bootstrap ca on start (#631, @gnunu)
- deprecate kubelet certificate management mode (#639, @qclc)
- remove k8s.io/kubernetes dependency from OpenYurt (#650, #664, #681, #697, #704, @rambohe-ch, @qclc, @Peeknut, @Rachel-Shao)
- add unit tests for yurthub data filtering framework (#670, @windydayc)
- enable yurthub to handle upgrade request (#673, @Congrool)
- Yurtctl: add precheck for reducing convert failure (#675, @Peeknut)
- ingress: add nodepool endpoints filtering for nginx ingress controller (#696, @zzguang)
- fix some bugs when local up openyurt (#517, @Congrool)
- reject delete pod request by yurthub when cloud-edge network disconnected (#593, @rambohe-ch)
- service topology filter can not work when hub agent work on cloud mode (#607, @rambohe-ch)
- fix transport race conditions in yurthub (#683, @rambohe-ch, @DrmagicE)
Manage EdgeX Foundry system in OpenYurt in a cloud-native, non-intrusive way
-
yurt-edgex-manager
Yurt-edgex-manager enable OpenYurt to be able to manage the EdgeX lifecycle. Each EdgeX CR (Custom Resource) stands for an EdgeX instance. Users can deploy/update/delete EdgeX in OpenYurt cluster by operate the EdgeX CR directly. (yurt-edgex-manager, @yixingjia, @lwmqwer)
-
yurt-device-controller
Yurt-device-controller aims to provider device management functionalities to OpenYurt cluster by integrating with edge computing/IOT platforms, like EdgeX in a cloud native way. It will automatically synchronize the device status to device CR (custom resource) in the cloud and any update to the device will pass through to the edge side seamlessly. More information can be found at: (yurt-device-controller, @charleszheng44, @qclc, @Peeknut, @rambohe-ch, @yixingjia)
Yurt-tunnel support more flexible settings for forwarding requests from cloud to edge
-
Support forward https request from cloud to edge for components(like prometheus) on the cloud nodes can access the https service(like node-exporter) on the edge node. Please refer to the details: (#442, @rambohe-ch, @Fei-Guo, @DrmagicE, @SataQiu)
-
support forward cloud requests to edge node's localhost endpoint for components(like prometheus) on cloud nodes collect edge components(like yurthub) metrics(http://127.0.0.1:10267). Please refer to the details: (#443, @rambohe-ch, @Fei-Guo)
- Proposal: YurtAppDaemon (#422, @kadisi, @zzguang, @gnunu, @Fei-Guo, @rambohe-ch)
- update yurtcluster operator proposal (#429, @SataQiu)
- yurthub use original bearer token to forward requests for inclusterconfig pods (#437, @rambohe-ch)
- yurthub support working on cloud nodes (#483 and #495, @DrmagicE)
- support discard cloud service(like LoadBalancer service) on edge side (#440, @rambohe-ch, @Fei-Guo)
- improve list/watch node pool resource in serviceTopology filter (#454, @neo502721)
- add configmap to configure user agents for specify response cache. (#466, @rambohe-ch)
- improve the usage of certificates (#475, @ke-jobs)
- improve unit tests for yurt-tunnel. (#470, @YRXING)
- stop renew node lease when kubelet's heartbeat is stopped (#482, @SataQiu)
- add check-license-header script to github action (#487, @lonelyCZ)
- Add tunnel server address for converting (#494, adamzhoul)
- fix incomplete data copy of resource filter (#452, @SataQiu)
- remove excess chan that will block the program (#446, @zc2638)
- use buffered channel for signal notifications (#471, @SataQiu)
- add create to yurt-tunnel-server ClusterRole (#500, adamzhoul)
Join or Reset node in one step
In order to enable users to use OpenYurt clusters quickly and reduce the cost of users learning OpenYurt, yurtctl has provided the subcommands convert
and revert
, to implement the conversion between OpenYurt cluster and Kubernetes cluster. However, It still has some shortcomings:
- No support for new nodes to join the cluster directly;
- Users are required to pre-built a kubernetes cluster, and then do conversion, which has a relatively high learning cost for beginners.
So, we need to add subcommands init
, join
, reset
for yurtctl. and join
and reset
subcommands can be used in v0.4.1, and init
subcommand will comes in next version.
Please refer to the proposal doc and usage doc for details. (#387, #402, @zyjhtangtang)
Support Pods use InClusterConfig
access kube-apiserver through yurthub
Many users in OpenYurt community have requested that support InClusterConfig for pods to access kube-apiserver through yurthub on edge nodes. so pods on cloud can move to edge cluster smoothly. so we add the following features.
- yurthub supports https serve.
- env
KUBERNETES_SERVICE_HOST
andKUBERNETES_SERVICE_PORT
should be the address that yurthub listening, like169.254.2.1:10261
Please refer to the issue #372 for details. (#386, #394, @luckymrwang, @rambohe-ch)
Support filter cloud response data on edge side
In the cloud-edge scenario, In response to requests from edge components (such as kube-proxy) or user pods to the cloud, it is hoped that some customized processing can be performed on the data returned by the cloud, we consider providing a generic data filtering framework in the yurthub component, which can customize the data returned from the cloud without being aware of the edge components and user pods, so as to meet business needs simply and conveniently. And provides two specific filtering handlers.
- support endpointslice filter for keeping service traffic in-bound of nodePool
- support master service mutation for pod use InClusterConfig access kube-apiserver
Please refer to the Proposal doc for details. (#388, #394, @rambohe-ch, @Fei-Guo)
- Yurtctl modify kube-controllersetting to close the nodelifecycle-controller (#399, @Peeknut)
- Proposal: EdgeX integration with OpenYurt (#357, @yixingjia, @lwmqwer)
- Proposal: add ingress feature support to nodepool (#373, @zzguang, @wenjun93)
- Proposal: OpenYurt Convertor Operator for converting K8S to OpenYurt (#389, @gnunu)
- add traffic(from cloud to edge) collector metrics for yurthub (#398, @rambohe-ch)
- Add sync.Pool to cache *bufio.Reader in tunnel server (#381, @DrmagicE)
- improve tunnel availability (#375, @aholic)
- yurtctl adds parameter enable app manager to control automatic deployment of yurtappmanager. (#352, @yanhui)
- fix tunnel-agent/tunnel-server crashes when the local certificate can not be loaded correctly (#378, @SataQiu)
- fix the error when cert-mgr-mode set to kubelet (#359, @qclc)
- fix the same prefix key lock error (#396, @rambohe-ch)
Node Resource Manager Released
Node resource manager is released in this version, which provides local node resources management of OpenYurt cluster in a unified manner. It currently supports LVM, QuotaPath and Pmem, and create or update the compute, storage resources based on local devices. It works as daemonset spread on each edge node, and manages local resources with a predefined spec stored in configmap. Please refer to the usage doc for details. (#1, @mowangdk, @wenjun93)
Add Cloud Native IOT Device Management API definition
Inspiring by the Unix philosophy, "Do one thing and do it well", we believe that Kubernetes should focus on managing computing resources while edge devices management can be done by adopting existing edge computing platforms. Therefore, we define several generic custom resource definitions(CRD) that act as the mediator between OpenYurt and the edge platform. Any existing edge platforms can be integrated into the OpenYurt by implementing custom controllers for these CRDs. In addition, these CRDs allow users to manage edge devices in a declarative way, which provides users with a Kubernetes-native experience.(#233, #236, @Fei-Guo, @yixingjia, @charleszheng44)
Kubernetes V1.18 is supported
OpenYurt officially supports version v1.18 of Kubernetes. Now, OpenYurt users are able to convert v1.18 Kubernetes cluster to OpenYurt cluster or deploy components of OpenYurt on v1.18 Kubernetes cluster manually. the main work for supporting v1.18 Kubernetes as following:
- refactor serializer of cache manager to adapt ClientNegotiator in k8s.io/apimachinery v0.18
- add context parameter in api for calling client-go v0.18
and based on Kubernetes compatibility, v1.16 Kubernetes is still supported. (#288, @rambohe-ch)
UnitedDeployment support patch for pool
UnitedDeployment controller provides a new way to manage pods in multi-pool by using multiple workloads. Each workload managed by UnitedDeployment is called a pool and user can only configure the workload replicas in the pool. Based on the patch feature, besides the workload replicas configuration, user can easily configure other fields(like images and resoures) of workloads in the pool.(#242, #12, @kadisi)
Support caching CRD resources by yurthub
Because resources in the resourceToKindMap can be cached by yurt-hub component, when network between cloud and edge disconnected, if any pod(eg: calico) on the edge node that used some resources(like crd) not in the above map want to run continuously, that is to say, the pod can not restarted successfully because resources(like crd) are not cached by yurt-hub. This PR can solve this limitation. Now yurt-hub is able to cache all kubernetes resources, including crd resource that defined by user.(#162, #231, #225, #265, @qclc, @rambohe-ch)
Prometheus and Yurt-Tunnel-Server cross-node deployment is supported via DNS
In the edge computing scenario, the IP addresses of the edge nodes are likely to be the same. So we can not rely on the node IP to forward the request but should use the node hostname(unique in one cluster). This PR provides the ability for the yurt-tunnel-server to handle requests in the form of scheme://[hostname]:[port]/[req_path].(#270, #284, @SataQiu, @rambohe-ch)
Support kind cluster and node level conversion by yurtctl
OpenYurt supports the conversion between the OpenYurt cluster and the Kubernetes cluster created by minikube, kubeadm, and ack. Now OpenYurt supports the conversion between kind cluster and OpenYurt cluster. (#230, #206, #220, #234, @Peeknut)
- add edge-pod-network doc (#302, @wenjun93)
- add resource and system requirements (#315, @wawlian)
- feature: add dummy network interface for yurthub (#289, @rambohe-ch)
- refactor: divide the yurthubServer into hubServer and proxyServert (#237, @rambohe-ch)
- using lease for cluster remote server healthz checker. (#249, @zyjhtangtang)
- yurtctl cluster-info subcommand that list edge/cloud nodes (#208, @neo502721)
- Feature: Support specified kubeadm conf for join cluster (#210, @liangyuanpeng)
- Add --feature-gate flag to yurt-controller-manager (#222, @DrmagicE)
- feature: add promtheus metrics for yurthub (#238, @rambohe-ch)
- Update the Manual document (#228, @yixingjia)
- feature: add meta server for handling prometheus metrics and pprof by yurttunnel (#253, @rambohe-ch)
- feature: add 'yurthub-healthcheck-timeout' flag for 'yurtctl convert' command (#290, @SataQiu)
- fix list runtimeclass and csidriver from cache error when cloud-edge network disconnected (#258, @rambohe-ch)
- fix the error of cluster status duration statistics (#295, @zyjhtangtang)
- fix bug when ServeHTTP panic (#198, @aholic)
- solve ips repeated question from addr.go (#209, @luhaopei)
- fix t.Fatalf from a non-test goroutine (#269, @contrun)
- Uniform label for installation of yurt-tunnel-agent openyurt.io/is-edge-worker=true (#275, @yanhui)
- fix the bug that dns controller updates dns records incorrectly (#283, @SataQiu)
- It solves the problem that the cloud node configuration taint (#299, @yanhui)
- Fixed incorrect representation in code comments. (#296, @felix0080)
- fix systemctl restart in manually-setup tutorial (#205, @DrmagicE)
- Add new component Yurt App Manager that runs on cloud nodes
- Add new provider=kubeadm for yurtctl
- Add hubself certificate mode for yurthub
- Support log flush for yurt-tunnel
- New tutorials for Yurt App Manager
- Implement NodePool CRD that provides a convenient management experience for a pool of nodes within the same region or site
- Implement UnitedDeployment CRD by defining a new edge application management methodology of using per node pool workload
- Add tutorials to use Yurt App Manager
- Add hubself certificate mode for generating and rotating certificate that used to connect with kube-apiserver as default mode
- Add timeout mechanism for proxying watch request
- Optimize the response when cache data is not found
- Add integration test
- Support log flush request from kube-apiserver
- Optimize tunnel interceptor for separating context dailer and proxy request
- Optimize the usage of sharedIndexInformer
- Add new provider=kubeadm that kubernetes cluster installed by kubeadm can be converted to openyurt cluster
- Adapt new certificate mode of yurthub when convert edge node
- Fix image pull policy from
Always
toIfNotPresent
for all components deployment setting
- Support Kubernetes 1.16 dependency for all components
- Support multi-arch binaries and images (arm/arm64/amd64)
- Add e2e test framework and tests for node autonomy
- New tutorials (e2e test and yurt-tunnel)
- Implement yurt-tunnel-server and yurt-tunnel-agent based on Kubernetes apiserver network proxy framework
- Implement cert-manager to manage yurt-tunnel certificates
- Add timeout mechanism for yurt-tunnel
- Add global lock to prevent multiple yurtctl invocations concurrently
- Add timeout for acquiring global lock
- Allow user to set the label prefix used to identify edge nodes
- Deploy yurt-tunnel using convert option
- Remove kubelet config bootstrap args during manual setup
- Avoid evicting Pods from nodes that have been marked as
autonomy
nodes
- Use Kubelet certificate to communicate with APIServer
- Implement a http proxy for all Kubelet to APIServer requests
- Cache the responses of Kubelet to APIServer requests in local storage
- Monitor network connectivity and switch to offline mode if health check fails
- In offline mode, response Kubelet to APIServer requests based on the cached states
- Resync and clean up the states once node is online again
- Support to proxy for other node daemons
- Support install/uninstall all OpenYurt components in a native Kubernetes cluster
- Pre-installation validation check