Repository navigation
fix(acl): wait for Node registration before starting TAA - #9960
bfjelds (bfjelds) wants to merge 4 commits into
Conversation
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com> Copilot-Session: 84294fd5-1096-4b5a-a33e-b5d4730445e3
Windows Unit Test Results 3 files 18 suites 1m 19s ⏱️ Results for commit 59bd8ce. ♻️ This comment has been updated with latest results. |
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com> Copilot-Session: 84294fd5-1096-4b5a-a33e-b5d4730445e3
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com> Copilot-Session: 84294fd5-1096-4b5a-a33e-b5d4730445e3
|
Cameron Meissner (@cameronmeissner) / Adams Anaglo (@adamsanaglo) ... I believe this will address the trident-acl-agent missing-k8s-node issue breaking other e2e tests. I started https://msazure.visualstudio.com/CloudNativeCompute/_build/results?buildId=188664133&view=results and will monitor. |
There was a problem hiding this comment.
🟡 Changes recommended
The Node lookup uses an unnormalized hostname, so mixed-case hostnames cause every probe to fail and prevent TAA from starting even when the Node is registered.
1 open issue
What changed in this PR
Adds a Node-registration preflight to prevent trident-acl-agent from starting before its Kubernetes Node exists, addressing the reported startup race.
Changes:
- Requires two consecutive successful Node checks, one second apart.
- Limits preflight to 60 checks with a five-second request timeout and fails if registration remains unavailable.
- Sets the startup timeout to 400 seconds while preserving existing diagnostics and restart policy.
| File | Description |
|---|---|
parts/linux/cloud-init/artifacts/acl/trident-acl-agent-override.conf |
Adds the Node-registration gate and extends the service startup timeout. |
Review effort: Balanced
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com> Copilot-Session: 84294fd5-1096-4b5a-a33e-b5d4730445e3
There was a problem hiding this comment.
🔵 Needs a closer look
Cloud E2E validation remains pending on AMD64 and ARM64 to confirm startup and recovery avoid service/path start limits in the actual Trident 0.28 environment.
No open issues 🎉
1 resolved since last review
Review effort: Balanced
|

Change
Trident 0.28 exits successfully when kubeconfig exists but the Node has not registered. The path trigger then leaves the service and path units start-limited.
Build 188602915 failed six ACL scenarios with this sequence. In scenario-logs-1793393947, ACL/journalctl records the NodeGone exit at 21:05:18, registration at 21:05:18.190457, and start-limit failures at 21:05:22.
Validation