Skip to content
This repository was archived by the owner on Aug 12, 2025. It is now read-only.

Conversation

@deitch
Copy link
Contributor

@deitch deitch commented May 11, 2020

Must have 0.3.0 or higher, or lots of bugs kick in.

This ensures it.

deitch and others added 30 commits May 4, 2020 14:20
It does not look like the PacketMachine needs a ProjectID. It is a value
that we get from the PacketCluster.

It is not even generated from the `make examples` file and when applied
via kubectl it fails because PachetMachine.ProjectID is a required field
in openapi.
…ctID

remove the projectID from PacketMachine crd
update the cluster template with required fields
At the end of the story it took me an hour to understand the issue and 1
minutes to fix it. I suppose.

The kubebuilder installation process is already part of the Makefile,
but it was not used in CI.

now it should work.
It is cool that we cache the go dependency as a separate layer so images
will take care of caching for us! But it does not come for free!!
The types.Machine or Cluster from v1alpha3 where not right registered to
the manager
fix manager adding custerv1 schema types
fix annotations so the CRDs are correct
proper cluster reconcile loop, remove finalizers on delete
If we attempt to remove a machine that is not running on Packet (for any
reason), we are in a loop where the reconciler returns 404 from Packet
API and the source never gets deleted.

This code should check for a 404, and it should pass forward the
finalizer trigger a garbage collection.

I presume we do not need to check for the number of `instances` anymore,
because if they are `0` we never get over the `not found error`.
From v1alpha2 to v1alpha3 the Bootstrap.Data field got deprecated in
favour of Bootstrap.DataSecretName as you can see:

https://cluster-api.sigs.k8s.io/developer/providers/v1alpha2-to-v1alpha3.html#data-generated-from-a-bootstrap-provider-is-now-stored-in-a-secret
…-present

trigger finalizer even if the machine is already deleted from packet
Gianluca Arbezzano and others added 14 commits May 7, 2020 15:36
The tags generated by the machine controller where not merged with the
one passed from the MachineSpec. This PR fixes the issue. Now we can use
those labels to filter devices via packet api.
By default we mark a machine failed when it returns a status that we do
not know. I think it is very reasonable. But we were not handled a few
status: provisioning and queued.
get ip from control plan during cluster reconciliation
add requeue timeouts for when machine cannot start because cluster or…
simplify manual install of a new cluster with make target
By default we mark a machine as failed when we receive an unrecognized
state. We were not handling the `provisioning` state.
because why not?
@deitch deitch requested a review from gianarb May 11, 2020 09:08
@deitch
Copy link
Contributor Author

deitch commented May 11, 2020

And again, wrong one!

@deitch deitch closed this May 11, 2020
@gianarb gianarb deleted the update-controller-gen branch May 11, 2020 10:40
Sign up for free to subscribe to this conversation on GitHub. Already have an account? Sign in.

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

3 participants