Delete api-tracking.md
no longer need this file since the content is now in the KEP and the user-facing documentation
This commit is contained in:
Michael Michael
GitHub
parent
6dda1ac0cd
commit
458a23054e
|
|
@ -1,76 +0,0 @@
|
|||
# Windows & Kubernetes APIs
|
||||
|
||||
This document will grow into an API by API list of work that needs to be done to clarify Windows & Linux differences. This will be used to help clarify what needs to be eventually implemented (need a tracking issue), or not implemented (need a doc note).
|
||||
|
||||
|
||||
## Volumes
|
||||
|
||||
`V1.Pod.Volumes`
|
||||
|
||||
Out of the various volume types, these should all be possible on Windows but tests are lacking:
|
||||
|
||||
- EmptyDirVolumeSource
|
||||
- Secret
|
||||
- hostPath
|
||||
|
||||
The main gaps in Windows Server 2016 & 1709 are that symlinks are pretty much broken. The only ones that work are SMB/CIFS mount points. Workarounds need to be investigated.
|
||||
|
||||
`V1.Container.volumeMounts`
|
||||
Mounting volumes across some (but not all) containers will need changes to Windows. Not ready in Windows Server 2016/1709.
|
||||
|
||||
### Links
|
||||
|
||||
- [FlexVolume does not work on Windows node](https://github.com/kubernetes/kubernetes/issues/56875)
|
||||
- [feature proposal add SMB(cifs) volume plugin](https://github.com/kubernetes/kubernetes/issues/56005)
|
||||
- [add NFS volume support for Windows](https://github.com/kubernetes/kubernetes/issues/56188)
|
||||
|
||||
## V1.Pod.Resources & V1.Container.ResourceRequirements
|
||||
|
||||
`V1.Container.ResourceRequirements.limits.cpu`
|
||||
`V1.Container.ResourceRequirements.limits.memory`
|
||||
|
||||
Windows schedules CPU based on CPU count & percentage of cores. We need this represented because it can help optimize app performance. CPU count is immutable once set but you can change % of core allocations.
|
||||
|
||||
`V1.Container.ResourceRequirements.requests.cpu`
|
||||
`V1.Container.ResourceRequirements.requests.memory`
|
||||
|
||||
Also of note, requests aren't supported. Will pod eviction policies in the kubelet ensure reserves are met by not overprovisioning the node?
|
||||
|
||||
Windows can either expose a NUMA topology matching the host (best performance) or fake it to be 1 big NUMA node (suboptimal). We should think of a way to turn this on/off later - probably q2 2018
|
||||
|
||||
### Links
|
||||
[Kubernetes Container Runtime Interface (CRI) doesn't support WindowsContainerConfig and WindowsContainerResources](https://github.com/kubernetes/kubernetes/issues/56734)
|
||||
|
||||
|
||||
|
||||
## Networking features
|
||||
|
||||
`V1.Pod.dnsPolicy` - I think only ClusterFirst is implemented
|
||||
|
||||
`V1.Pod.hostNetwork` - Not feasible on Windows Server 2016 / 1709
|
||||
|
||||
## IPC & Pid
|
||||
|
||||
`V1.Pod.hostIPC`, `v1.pod.hostpid`
|
||||
|
||||
How important are these? They're not implemented in Windows Server 2016 / 1709, and I'm not too sure if they'd be helpful or not.
|
||||
|
||||
For cases where a pod/container need to talk to the host docker / containerd daemon we could map a named pipe as a volume which would offer the same functionality as the unix socket to the Linux daemons. It works in moby but isn't hooked up in the kubelet yet.
|
||||
|
||||
## Security
|
||||
|
||||
- `V1.Container.SecurityContext.Capabilities`
|
||||
- `V1.Container.SecurityContext.seLinuxOptions`
|
||||
|
||||
These don't have Windows equivalents since the permissions model is substantially different
|
||||
|
||||
`V1.Container.SecurityContext.readOnlyRootFilesystem`
|
||||
|
||||
This is probably doable if needed but not possible in Windows Server 2016 / 1709.
|
||||
|
||||
### User Mapping
|
||||
|
||||
There are a few fields that refer to uid/gid. These probably need to be supplemented with a Windows SID (string) and username (string)
|
||||
|
||||
`V1.podSecurityContext.runAsUser` provides a UID
|
||||
`V1.podSecurityContext.supplementalGroups` provides GID
|
||||
Loading…
Reference in New Issue