Filter out AI zones when discovering zones in the region. AI zones
do not have quota for general compute resources, so we should not provision
nodes there by default.
** While the regional support is valid, we will not be using this in openshift. Regional support
requires that each api have its own endpoint. Only one api is associated with an endpoint, and managing
this access will be difficult and unnessary at this time.
Control plane machines were intermittently being created in different
availability zones than specified in their machine specs. This occurred
because the zone list returned from FilterZonesBasedOnInstanceType used
a set's UnsortedList() func, which has a non-deterministic order.
When CAPI and MAPI manifest generation independently called this func,
they could receive zones in different orders, causing a mismatch in
machine zone placements between CAPI and MAPI manifests.
This commit ensures that we sort the zone slices before further
processing.
** Ensure that the feature is backwards compatible for original XPN cases. The new field
firewallRulesManagement is an explicit setting of whether the user has the firewall rules
or not. In old versions this did not exist but XPN installs did not need firewall rules.
Now we will default to checking permissions when no field value is provided. If the rules
do not exist, the rules management is set to unmanaged.
A change to CAPZ[0], creates an SSH rule if one is not specified in
the cluster spec. Prior to this commit, we had been creating the
SSH rule with installer SDK hooks, which is still somewhat necessary
to add the inbound NAT rules, because we are not yet using CAPZ
to provision a public load balancer.
But we can use CAPZ to just create the rule, which will stop CAPZ
from preventing a redundant SSH rule which we were leaking during
bootstrap destroy.
This change will also result in creating an SSH rule for private clusters
which is fine, and something we do on other providers.
0: https://github.com/kubernetes-sigs/cluster-api-provider-azure/pull/5525
hive does not know the rhcos image value and will pass an empty
string for azure machines. In this case, we can leave the image
empty so that it will use the default from MAPI.
Remove the interactive disconnected workflow type. The appliance
embeds registries.conf and CA certificates directly in the system ignition
for bootstrap, eliminating the need for a separate workflow type.
Key changes:
Command-line interface:
- Remove --interactive flag from unconfigured-ignition command
- Remove ContextWrapper and workflow context management
Workflow handling:
- Simplify to always use AgentWorkflowTypeInstall for unconfigured ignition
- Remove AgentWorkflowTypeInstallInteractiveDisconnected constant
- Remove workflow type switching logic in UnconfiguredIgnition
- Remove workflow dependency from UnconfiguredIgnition asset
Mirror configuration:
- Remove RegistriesConf and CaBundle dependencies from UnconfiguredIgnition
- Remove addMirrorData() call (appliance provides this)
- Remove early returns for interactive workflow in mirror assets
Testing:
- Remove interactive-disconnected-workflow test case
- Remove with-mirror-configs test case from unconfigured ignition tests
- Update default dependencies in test helpers
Rationale:
The OVE appliance provides a more robust solution for disconnected
installations by embedding all necessary configuration (registries,
certificates, UI) directly in the appliance image. This approach:
- Eliminates workflow type complexity
- Decouples installer and appliance repositories
- Simplifies the codebase by removing conditional logic
- Aligns with the architecture where MCO manages post-bootstrap config
After first node reboot, the Machine Config Operator manages registry
configuration and trust bundles via IDMS/IDMT resources.
Commit message text generated by: Claude AI <noreply@anthropic.com>
Added cleanup for bootstrap VM FIP in the insatller as part of the
PostDestroyer to streamline FIP creation and deletion to be done by the
installer avoiding orphan FIPs.
Fixes the issue of bootstrap machine logs cannot be collected when installation fails because
the bootstrap VM uses the master security group, which doesn't allow SSH
access from the installation source address. This prevents gathering
diagnostic information needed to troubleshoot failed installations.
- Created dedicated bootstrap security group with SSH access from anywhere
- Tagged with `openshiftRole=bootstrap` for lifecycle management
- Enables SSH access to bootstrap VM for log collection on failure
- Implement PostDestroyer for openstack bootstrap vm
Set the default type EBS volume for machine pools:
- Controlplane, arbiter and worker pool default to gp3 volume.
- Edge pool default to gp2 volume.
The default decision is taken from existing code [0]. This commit just
makes the defaulting earlier.
This also loosens the validation to allow 0 for volume size and iops as
they are int type (i.e. not pointer).
Reference:
[0] fd5a518e49/pkg/asset/machines/worker.go (L102-L117)
GP3 volumes have the ability to configure throughput from 125 MiB/s to
2000 MiB/s. This allows the ability to set this at install time in the
install-config.
https://issues.redhat.com/browse/CORS-4212
Added a preflight check for the destroy code (AWS) to check if owned
subnets is safe to destroy: subnets are not shared by any other clusters.
An owned subnet should not be shared by other clusters as this is a
misconfiguration, but in practice, it happened [0].
References
[0] https://issues.redhat.com//browse/OCPBUGS-60071
The AWS IAM role and instance profile waiters had a 2 minute timeout
with custom delay options (1-5 seconds). This timeout was insufficient
in CI environment where IAM calls can be throttled.
Increased the timeout to 15 minutes and removed the custom delay
options to use the AWS SDK defaults (min 1s and max 120s).
This commit updates default value handling when loading the
install config to set values in machine pools based on the
defaultMachinePlatform.
By populating the values directly in the install config, we can
avoid repetitive checks throughout the codebase to ensure the
default machine platform is applied to the relevant machine pool.
Marketplace images do not support confidential VMs or trusted launch,
so when machinesets use confidential VMs the installer will still
create an image gallery compatible with the security settings.
Now that we can use marketplace images, we only need to create
managed images for OKD or AzureStack. We can also safely remove the
environment variable that allowed skipping image upload and the
corresponding validation.
Defaults the Rhcos image asset to the non-paid Azure marketplace image.
Fallback to Azure disk and image creation if the marketplqace image
is not available, which is the case for OKD.
Refactors the installconfig metadata to store the VM capabilities
and hyperVGeneration, so that we do not need to repeatedly call the
azure API to get the same information.
GP3 volumes have the ability to configure throughput from 125 MiB/s to
2000 MiB/s. This allows the ability to set this at install time in the
install-config.
https://issues.redhat.com/browse/CORS-4212
