Repository navigation
Conversation
…or 1.2.1 The imageset pinned OCP 4.22.8 while the deployed installer is 4.22.17, and trustee-operator 1.2.0 while values-baremetal pins CSV v1.2.1. The mirror must contain the exact release the installer consumes (K12) and the operator bundles the pattern resolves. Version audit: specs/002-multi-tranche-testing/version-audit.md (coco-gsd)
…crets.sh The script only uncommented registryCaCert and bootstrap_secrets. The branch-pair contract for disconnected testing also requires tdxCollateral enabled (offline DCAP collateral) and intelPcsApiKey disabled (manual PCK provisioning via dcap-pck.py in airgap; the connected key file must not be required by load-secrets). - Wrap the tdxCollateral block in AIRGAP-TDX-COLLATERAL markers in values-secret.yaml.template (gen-secrets copies the template verbatim, so markers flow through to the generated file) - Uncomment both marker blocks; comment out the intelPcsApiKey entry - Idempotent; verified against a fresh template copy (YAML-valid output, airgap secrets active, intelPcsApiKey disabled) Contract: specs/002-multi-tranche-testing/contracts/branch-pair.md (coco-gsd)
… in imageset oc-mirror v2 fails catalog filtering when the package's catalog default channel is filtered out: openshift-gitops-operator defaults to 'latest' (we pin gitops-1.22) and gpu-operator-certified defaults to v26.7 (we pin v26.3). The defaultChannel field re-points the default to the pinned channel. Known-failure class recorded as K13 in the T1 mirror checklist (coco-gsd specs/002-multi-tranche-testing/mirror-known-failures.md).
…t is a hard mirror error
registry.redhat.io rejects :latest for container-native-virtualization/virtio-win
('must be pulled with an explicit version'), which fails oc-mirror on every run.
The three guest images are documented-optional (~2 GB each) and unconsumed:
Phase 26 sets enableCommonBootImageImport: false, so CNV never imports them.
Commenting them out also removes the large-blob timeout flakiness (rhel8/rhel9
guest images hit context-deadline-exceeded on upload).
T1 mirror run evidence: coco-gsd specs/002-multi-tranche-testing/t1-progress.md
…disabled subscription The gpu-operator subscription is commented out in values-baremetal.yaml (NVIDIA bundle mirror issues); nothing in T1 consumes the bundle. The nvcr.io related images (dcgm, driver) hit repeated copy timeouts across two resumable retries, blocking mirror certification for content the disconnected deployment never pulls. Tranche 2 re-enables the GPU operator at v26.7 (pawelpros cherry-pick), which mandates its own re-mirror + re-certification per the T022 trigger logic — at that point the bundle returns to the imageset.
The mirror registry is now identified by its DNS name everywhere: - Phase 0-0: canonical MIRROR_REGISTRY=user-jump.int.apac-tech-lab.net:8443, with the resolved IP recorded as a generation-time fact only - Phase 0-1b: cert CN/SAN carry the DNS name (IP SAN kept for transition) - Phase 0-1h: certs.d trust entries for both name and IP - oc-mirror targets the DNS name so generated cluster-resources (IDMS/ITMS/ CatalogSources) reference it — no IP enters the ignition chain - itms-manual-mirrors placeholders substituted from MIRROR_REGISTRY at apply time (was applied verbatim with the literal placeholder) - git-server URLs and repoURL hints use the DNS name Lab IPs are not stable (the jump host moved subnets on 2026-09-15, breaking the T013 install through the reserved-VLAN routing); DNS is maintained through moves and the node's ignition resolver uses it. Spec: FR-017 (specs/002-multi-tranche-testing, R6)
…ries ansible-lint's yaml profile (comments.require-starting-space) flags the commented-out guest-image entries; add the space after the marker.
butler54
force-pushed
the
feat/t1-connected-testing
branch
from
October 28, 2026 03:11
335561f to
a3b9366
Compare
Collaborator
Author
|
Closing in favor of #164, which is a superset of this PR — it contains all the same commits (imageset alignment, DNS identity, airgap secrets toggle, trustee SNP fix) plus the additional fixes from the Quay consolidation work (vault pin update, pattern-ui-catalog removal, Makefile cleanup, runbook rewrite). No changes are lost. |
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
Airgap/deployment robustness fixes from the Tranche 1 testing lifecycle — validated on bare-metal Intel TDX (SNO, OCP 4.22.17) through a full connected pattern deployment plus mirror creation and content certification.
What's in here
fix(airgap): align imageset with installer 4.22.17 and trustee-operator 1.2.1feat(secrets): complete the airgap secrets toggle in enable-airgap-secrets.shfix(airgap): add defaultChannel overrides for gitops and gpu-operator in imagesetlatest, gpu-operator to v26.7) —defaultChannelre-points the default to the pinned channelfix(airgap): disable CNV guest images in imageset:latestfor virtio-win ("must be pulled with an explicit version") — a mirror error on every run. The guest images are documented-optional (~2 GB each) and unconsumed (enableCommonBootImageImport: false)fix(airgap): disable gpu-operator-certified in T1 imagesetfeat(airgap): DNS-based mirror registry identityuser-jump.int.apac-tech-lab.net:8443) everywhere: cert CN/SAN (DNS + IP), certs.d trust entries for both, oc-mirror target (so generated cluster-resources carry no IPs — nothing IP-based enters the ignition chain), git-server URL reporting, and the itms-manual-mirrors placeholders (previously applied verbatim with the literalMIRROR_REGISTRY_HOSTplaceholder — a latent bug) are now substituted fromMIRROR_REGISTRYat apply time. Lab host IPs are not stable long-term; DNS is maintained through movesValidation
POST /kbs/v0/attest200), resource delivery 200s, all CoCo workloads Running under kata-cc, zero-touch PCK registration via cache reuseDisconnected deployment status (for transparency)
The full disconnected install on our lab is currently blocked by a lab-network infrastructure fault on the path between the cluster node and the registry host (rate-dependent sustained-TCP collapse; both endpoints validated clean; escalated to the lab owners with instrumentation). These changes are the connected-side + mirror-side output of that work; the disconnected run completes once the lab network issue is resolved. Happy to share the diagnosis details if useful.
Testing
bash -non both modified scripts; YAML validated on all modified manifestspattern.sh make show) and a direct clustergroup 0.9.58 chart render with the full values chain pass; the airgap overlay selection (global.airgap→ sharedValueFiles) verified in the rendered output