chore(deps): update aks/aks-gpu-cuda docker tag to v595

[renovate]

This PR contains the following updates:

Package	Update	Change
aks/aks-gpu-cuda	major	580.126.09-20260126030251 → 595.71.05-20260623180420

---

Configuration

📅 Schedule: (UTC)

• Branch creation
  • At any time (no schedule defined)
• Automerge
  • At any time (no schedule defined)

🚦 Automerge: Enabled.

♻ Rebasing: Whenever PR becomes conflicted, or you tick the rebase/retry checkbox.

🔕 Ignore: Close this PR and you won't be reminded about this update again.

---

• If you want to rebase/retry this PR, check this box

---

This PR was generated by Mend Renovate. View the repository job log.

[Copilot]

Copilot encountered an error and was unable to review this pull request. You can try again by re-requesting a review.

[Copilot]

Copilot encountered an error and was unable to review this pull request. You can try again by re-requesting a review.

[Copilot]

Copilot encountered an error and was unable to review this pull request. You can try again by re-requesting a review.

[Copilot]

Copilot encountered an error and was unable to review this pull request. You can try again by re-requesting a review.

[aks-node-assistant]

AgentBaker Linux PR gate — Single NetworkIsolated localdns-exporter assertion failure (test-assertion issue, not PR-caused)

• Run: 167348479 (failed)
• Failed task: Run AgentBaker E2E
• Test summary: DONE 457 tests, 95 skipped, 2 failures in 1837.481s (notably NO fwupd hits in this run)
• Failing scenario: Test_Ubuntu2204_ArtifactStreaming_NetworkIsolatedCluster/default

Exact failure signature:

validation.go:82: 🔴 FAIL:
  validate_localdns_exporter_metrics.go:71
  Error: Not equal: expected: "0"  actual: "1"
  Test:   Test_Ubuntu2204_ArtifactStreaming_NetworkIsolatedCluster/default
  Messages: localdns exporter metrics validation failed

Node provisions cleanly, kubelet ready, CSE OK, ethtool/waagent/node-exporter all pass; the failing assertion is inside validate_localdns_exporter_metrics.go:71 comparing an expected metric value of "0" to an actual "1" after the test sends ~1000 DNS queries through localdns. The localdns exporter on port 9353 is up, the metrics endpoint responds with a populated payload (localdns_memory_usage_bytes, localdns_cpu_usage_seconds_total, localdns_vnetdns_forward_info{...} 1, etc.), so the issue is the assertion's expected value, not a missing/unreachable exporter.

Three-level analysis:

1. L1: assertion expects metric 0, gets 1, on a NetworkIsolated 22.04 ArtifactStreaming scenario.
2. L2 corroboration: all other validators in this scenario pass; the rest of the run is green (only 2 failures out of 457); the other failure is a single companion sub-scenario of the same test. No fwupd, no CSE error, no infra timeout. PR chore(deps): update aks/aks-gpu-cuda docker tag to v595 #8439 changes nvidia-gpu-cuda to v595 — touches GPU driver path only, not localdns exporter, not NetworkIsolated, not 22.04 base.
3. L3 challenge: alternatives — (a) PR-caused: refuted, nvidia-gpu-cuda has zero overlap with localdns exporter assertions on a non-GPU 22.04 scenario; (b) infra flake: weaker than usual — node and exporter healthy; this looks like a test-assertion drift (expected value was probably tightened recently and doesn't match real-world counters after the load-priming step generates legitimate traffic). Strongest alternative is "recently-tightened assertion in validate_localdns_exporter_metrics.go" — i.e. test-code regression in main, not in this PR.

Build-vs-test: test-code/assertion (not PR-caused, not product regression).
Confidence: HIGH that PR #8439 is not the cause; MEDIUM-HIGH that this is a recently-introduced assertion mismatch in e2e/validate_localdns_exporter_metrics.go:71.

Recommended next action / owner: NodeSIG-dev / E2E owner of validate_localdns_exporter_metrics.go — review the line 71 assertion (which counter is expected to be exactly 0 after sending 1000 priming queries?). The exporter is reporting localdns_kubedns_forward_info{...} 1 and localdns_vnetdns_forward_info{...} 1 for several blocks; either the assertion should expect ≥0 / >0, or the priming step should not target the metric under test. PR author: do NOT block merge on this; this is the only failing scenario and it's outside your PR's scope.

Posted by Clawpilot AgentBaker gate detective.