Initial multi-stream enhancement #1692
Conversation
Copied from https://hackmd.io/q1Txm0twTYatSvSf1vNFeA?view where there's some comments which warrant understanding.
openshift-ci
bot
added
the
do-not-merge/work-in-progress
| At install time the **cluster creator** will either specify the desired os | ||
| for `ControlPlane` and `Compute` or not, if they provide no value the installer | ||
| is to render the current default stream into relevant resources. |
There was a problem hiding this comment.
Is there strong value in exposing this via install config versus just supporting it as "day 0" machineconfig?
There was a problem hiding this comment.
I guess it doesn't have to be, for all platforms we already allow overriding the boot image and you could patch MCP manifests as well.
| Whenever compute resources aren't elastic should we support a special mode where the host | ||
| OS is reinstalled across versions specifically NOT preserving any data / config? | ||
|
|
||
| #### Single-node Deployments or MicroShift |
There was a problem hiding this comment.
Today MicroShift is pretty different in that the user chooses the base OS version already.
| When the installer is built for OCP valid `osStream` must start with "rhcos" | ||
| and match the name of a file in data/data/coreos/ | ||
|
|
||
| Somewhere in MCO's templates/ add streams/{rhcos-9,rhcos-10} anything outside |
There was a problem hiding this comment.
What I suspect we may need to add to the MCO is a conditional like this for user provided machineconfig.
There was a problem hiding this comment.
If we hold that stream is configured at MCP level and that value is effectively immutable, wouldn't they just provide MachineConfig that matches the labels for each pool?
Need to vet the viability of stream being immutable with how we would potentially handle in-place major OS upgrades. My thinking for now is that entails moving a node from one pool to another rather than reconfiguring a pool for a different stream.
| recorded in /etc/os-release or other available facilities. There's probably | ||
| some systemd magic here or something. This probably also pushes more static |
There was a problem hiding this comment.
Yes, we I think we need to make it easy dispatch on the OS major/minor; this gets into a lot of details around our use of VERSION_ID and how that looks...and the fact that today our versions include the OCP version and the OS version...
|
There's intersections with OLM here right? Does it already exist as a way for operators to declare their compatible OS versions? |
|
[APPROVALNOTIFIER] This PR is NOT APPROVED This pull-request has been approved by: The full list of commands accepted by this bot can be found here.
Needs approval from an approver in each of these files:
Approvers can indicate their approval by writing |
@cgwalters dd32891 adds my opinion but I've put it as an open question because I think it warrants additional discussion. I had previously asked if we had a way to determine which operators requested RBAC necessary to run privileged containers and we may or may not be able to assess that. I'm hoping to hear back on that soon so we can perform some more targeted analysis. However I assume that those who truly become part of the OS will have motivation not to ignore RHEL10, such as GPU management operators who embed drivers. They may skimp on the openshift operator side but it seems like they won't forego RHEL10 drivers all together and hopefully the additional operator work is minimal. |
| * As an OpenShift admin adding newer hardware to an existing cluster I want the | ||
| new hardware to boot, run, and update from a specific OS stream. | ||
|
|
||
| * As an OpenShift admin wishing to migrate existing hosts to a newer stream I |
There was a problem hiding this comment.
We debated this live, but I want to write things down here for consistency. I have no opposition to making it obvious/easy for admins to use CAPI to spawn separate RHEL-$next hosts for testing, etc.
But I am pretty confident in stating that we can support seamless in-place upgrades from 9 to 10 for the majority of e.g. cloud-deployed clusters. Take our own Prow clusters for example. I bet we can just flip the flag to run those on rhel10 and watch it roll out in place by default and it would Just Work.
Copied from https://hackmd.io/q1Txm0twTYatSvSf1vNFeA?view where there's some comments which warrant understanding.