pingcap / advanced-statefulset Goto Github PK

This is useful for projects which use a different version of k8s code.

branches:

• client-1.16 => track code with k8s 1.16
• client-1.17 => track code with k8s 1.17
• ...

tags:

• client/v1.16.x => for k8s 1.16
• client/v1.17.x => for k8s 1.17

https://github.com/go-modules-by-example/submodules

not necessary anymore

• remove v1alpha1 API version
• update client to use 1.17.0 k8s code
• use ./client go submodule
• push client-1.17 branch
• tag client/v1.17.0

based on #13

• scale in/out with delete slots [podManagementPolicy=OrderedReady]
• scale in/out with delete slots [podManagementPolicy=Parallel]
• scale up/down with delete slots [updateStrategy=RollingUpdate]
• scale up/down with delete slots [updateStrategy=OnDelete]

Currently, we don't have any mechanism to know the status of delete slots. For example:

When replicas 3, delete slots[1] changes to replicas 3, delete slots[2]. We have no way to know when the scaling operation is done. Because we can't simply compare spec.replicas with status.replicas.

The workaround solution is listing the pods from the APIServer.

In this CRD version of StatefulSet, some validations cannot be implemented via CRD OpenAPI Schema. For safety, we can run validation logic in the controller and emit errors via events to notify the user to fix them.

We can leverage the builtin ControllerRevision type.

previous:

• #44

PRs:

• kubernetes/kubernetes#87299 (this is incompatibility with Kubernetes client-go prior to 1.18, we should generate a client for k8s prior to 1.18)
• kubernetes/kubernetes#86801
• kubernetes/kubernetes#84984
• kubernetes/kubernetes#87706
• kubernetes/kubernetes#90898

no bugfixes or any improvements, only some cleanup

• kubernetes/kubernetes#79129
• kubernetes/kubernetes#82858
• kubernetes/kubernetes#84836

https://onsi.github.io/ginkgo/#parallel-specs

write up a doc to summarize it

ref: pingcap/tidb-operator#2859

Consider the following scenario:
I have a asts with .spec.replicas=6, the pod {0,2,4} are in zoneA, the pod {1,3,5} are in zoneB, say zoneA completely down at some time, so I try delete all pods in zoneA while scale down replicas to 3 since I don't want another 3 pods up in zoneB, use the command

kubectl patch asts --type merge -p='{"metadata":{"annotations":{"delete-slots":"[0,2,4]"}},"spec":{"replicas":3}}'

now asts only treates [0,2] as deleted and finally we got 3 alive pods {1,3,4} but the expect is {1,3,5}

pod 4 not recognized by asts controller for deleting because 4 is out range of [0, .spec.replicas=3) as the code shows:

pingcap/tidb-operator#2859 (comment)

I wanna watch advanced-statefulset resource in k8s, but i can't import the client package. a error here is:

when comment advanced-statefulset go.mod, then run go mod tidy, the same error

then, users can configure timezone via TZ env

xref: https://golang.org/src/time/zoneinfo_unix.go (how time pkg handles timezone)