Clusterpedia v0.2.0 Release

Tuesday, April 12, 2022

5 minute read

Use helm to install

Users can already use Helm to install Clusterpedia:

$ helm install clusterpedia . \
  --namespace clusterpedia-system \
  --create-namespace \
  --set persistenceMatchNode={{ LOCAL_PV_NODE }} \
  # --set installCRDs=true

Lean More

Use the Kube Config to import a cluster

For v0.1.0, users need to Configure the address for the imported cluster and the authentication information.

apiVersion: cluster.clusterpedia.io/v1alpha2
kind: PediaCluster
metadata:
  name: cluster-example
spec:
  apiserver: "https://10.30.43.43:6443"
  caData:
  tokenData:
  certData:
  keyData:
  syncResources: []

In v0.2.0, the PediaCluster added the spec.kubeconfig field so that users can use kube config to import the cluster directly.

First you need to base64 encode the kube config for the imported cluster.

$ base64 ./kubeconfig.yaml

Set the content after the base64 to PediaCluster spec.kubeconfig, in addition spec.apiserver and other authentication fields don’t need to set.

apiVersion: cluster.clusterpedia.io/v1alpha2
kind: PediaCluster
metadata:
  name: cluster-example
spec:
  kubeconfig: **base64 kubeconfig**
  syncResources: []

However, since the cluster address is configured in kube config, the APIServer is empty when you use kubectl get pediacluster.

$ kubectl get pediacluster
NAME              APISERVER   VERSION   STATUS
cluster-example               v1.22.2   Healthy

Mutating addmission webhooks will be added in the future to automatically set spec.apiserver, currently if you want to show the cluster apiserver address when kubectl get pediacluster, then you need to manually configure the spec.apiserver field additionally.

New Search Feature

Search by creation time interval

Description	Search Label Key	URL Query
Since	search.clusterpedia.io/since	since
Before	search.clusterpedia.io/before	before

The creation time interval used for the search is left closed and right open, since <= creation time < before.

There are four formats for creation time:

Unix Timestamp for ease of use will distinguish between units of s or ms based on the length of the timestamp. The 10-bit timestamp is in seconds, the 13-bit timestamp is in milliseconds.
RFC3339 2006-01-02T15:04:05Z or 2006-01-02T15:04:05+08:00
UTC Date 2006-01-02
UTC Datetime 2006-01-02 15:04:05

Because of the limitation of the kube label selector, the search label only supports Unix Timestamp and UTC Date.

All formats are available using the url query method.

Look at what resources are under the default namespace

$ kubectl --cluster clusterpedia get pods
CLUSTER           NAME                                                 READY   STATUS      RESTARTS       AGE
cluster-example   quickstart-ingress-nginx-admission-create--1-kxlnn   0/1     Completed   0              171d
cluster-example   fake-pod-698dfbbd5b-wvtvw                            1/1     Running     0              8d
cluster-example   fake-pod-698dfbbd5b-74cjx                            1/1     Running     0              21d
cluster-example   fake-pod-698dfbbd5b-tmcw7                            1/1     Running     0              8d

We use the creation time to filter the resources.

$ kubectl --cluster clusterpedia get pods -l "search.clusterpedia.io/since=2022-03-20"
CLUSTER           NAME                        READY   STATUS    RESTARTS   AGE
cluster-example   fake-pod-698dfbbd5b-wvtvw   1/1     Running   0          8d
cluster-example   fake-pod-698dfbbd5b-tmcw7   1/1     Running   0          8d


$ kubectl --cluster clusterpedia get pods -l "search.clusterpedia.io/before=2022-03-20"
CLUSTER           NAME                                                 READY   STATUS      RESTARTS       AGE
cluster-example   quickstart-ingress-nginx-admission-create--1-kxlnn   0/1     Completed   0              171d
cluster-example   fake-pod-698dfbbd5b-74cjx                            1/1     Running     0              21d

Search by Owner Name

As of v0.1.0, we can specify ancestor or parent Owner UID to query resources, but Owner UID is not convenient to use, after all, you still need to know the UID of the Owner resource in advance.

In v0.2.0, we support querying directly with Owner Name, and the Owner query has been moved from experimental to released functionality, the prefix of Search Label has been upgraded from internalstorage.c lusterpedia.io to *search.clusterpedia.io *, and URL Query is provided.

Role	search label key	url query
Specified Owner UID	search.clusterpedia.io/owner-uid	ownerUID
Specified Owner Name	search.clusterpedia.io/owner-name	ownerName
SPecified Owner Group Resource	search.clusterpedia.io/owner-gr	ownerGR
Specified Owner Seniority	`internalstorage.clusterpedia.io/owner-seniority`	`ownerSeniority`

Note that when specifying Owner UID, Owner Name and Owner Group Resource will be ignored.

$ kubectl --cluster cluster-example get pods -l \
    "search.clusterpedia.io/owner-name=fake-pod, \
     search.clusterpedia.io/owner-seniority=1"
CLUSTER           NAME                        READY   STATUS    RESTARTS   AGE
cluster-example   fake-pod-698dfbbd5b-wvtvw   1/1     Running   0          8d
cluster-example   fake-pod-698dfbbd5b-74cjx   1/1     Running   0          21d
cluster-example   fake-pod-698dfbbd5b-tmcw7   1/1     Running   0          8d

In addition, to avoid multiple types of owner resources in some cases, we can use the Owner Group Resource to restrict the type of owner.

$ kubectl --cluster cluster-example get pods -l \
    "search.clusterpedia.io/owner-name=fake-pod,\
     search.clusterpedia.io/owner-gr=deployments.apps,\
     search.clusterpedia.io/owner-seniority=1"
... some output

Fuzzy Search base on resource names

Since fuzzy search needs to be discussed further, it is temporarily provided as an experimental feature.

$ kubectl --cluster clusterpedia get deployments -l "internalstorage.clusterpedia.io/fuzzy-name=fake"
CLUSTER           NAME       READY   UP-TO-DATE   AVAILABLE   AGE
cluster-example   fake-pod   3/3     3            3           113d

You can use the in operator to pass multiple fuzzy arguments, so that you can filter out resources that have all strings in their names.

Other Features

In v0.1.0, searching the resources allow the number of remaining resources to be returned so that the user can calculate the total number of resources.

This feature has been enhanced in v0.2.0. When offset is too large, remainingItemCount may be negative, ensuring that the total number of resources can always be calculated.

Lean More

Release Notes

Support of using Helm Charts for installation (#53, #125, @calvin0327, @wzshiming)
PediaCluster supports for importing a cluster using the kubeconfig (#115, @wzshiming)

APIServer

Support for filtering resources by a period of creation (#113, @cleverhu)
Support for searching for resources by an Owner name. Now, the feature of Search by Owner is officially released. (#91, @Iceber)

Default Storage Layer

Support for fuzzy search by a resource name (#117, @cleverhu)
RemainingItemCount can be a negative number. We can still use offset + len(items) + remainingItemCount to calculate the total amount of resources if the Offset is too large. (#123, @cleverhu)

Bug Fixes

Fixed unnecessary json.Unmarshal and improved performance when searching (#89, #92, @Iceber)

Deprecation

Search by Owner has been released as an official feature. internalstorage.clusterpedia.io/owner-name and internalstorage.clusterpedia.io/owner-seniority will be removed in the next release. (#91, @Iceber)

Other

golangci-lint is used as a static checking tool (#86, #88, @Iceber)
Added CI Workloads such as static checking and unit testing for code #87, @Iceber)

←Previous