[Diamond (diamondpowell)]

Summary

• Preserve default collector operation names, row schema, and append behavior.
• Add explicit --legacy-output support for existing internal ADX consumers, retaining canonical operation data alongside legacy aliases.
• Accept pipeline context through CLI arguments or environment variables; preserve numeric counts and JSON-encoded node-pool data.
• Validate legacy input before atomic output publication, preserving prior results on failures.
• Update focused tests and collector documentation.

Integration

Companion internal migration: https://msazure.visualstudio.com/CloudNativeCompute/_git/telescope/pullrequest/16193735

The internal wrapper opts into legacy output. Existing open-source callers do not need to change. No ADX table or dashboard connection changes are included. Merge this package change before the internal migration consumes open-source main.

Validation

• Full local Python suite: 836 passed, 1 skipped; 83.46% coverage.
• Pylint: 10.00/10; no Pylance diagnostics; whitespace checks passed.
• Installed-wheel compatibility checks: small VMSS, small VirtualMachines, and large VMSS; actual internal collector script and seven CLI dispatch signatures (cloud clients mocked).
• Confirmed canonical default output/append behavior, legacy fields and types, selected-cluster context, failure metadata, unchanged source files, and legacy rerun idempotency.
• Live internal installed-package validation: pending; temporary configuration will remain on a disposable internal validation branch.

Kept as draft until live validation evidence is available.

[github-actions]

For reviewers only: reply /run-tf-integration to trigger the terraform integration pipeline before approving the PR.

[Diamond (diamondpowell)]

Live installed-package validation has been queued: internal build 188117777.

• Open-source package commit: 11b8f69914dd8143db6c6e9e59b76f6c0693b37c.
• Internal production PR companion commit: c13adf8d49e1a53b1dc5ff7e9592cc17af135598 (PR 16193735).
• Disposable validation commit: 9ca42e8e94d6ff2b727b390305e3e6680fff4c3e, branch dipowell/crud-collector-validation-20261028.
• Scope: westcentralus, small VMSS and VirtualMachines cases, all three workload types, exact package-commit provenance check, and seven-row legacy result-contract assertions.

The temporary pipeline/configuration changes are not on the production PR branch. Run is queued/running, not yet reported as passed. Production ADX table names and upload routing are unchanged.

[Diamond (diamondpowell)]

Build 188117777 passed the VMSS lifecycle and all seven legacy-result assertions. VirtualMachines failed before collection because the CLI helper returned a boolean while shared provisioning instrumentation expected (request_started_at, retry_occurred). Both resource-group cleanup steps succeeded. Fixed in 35f440e with real-instrumentation create/direct/progressive scale and failure regressions: 201 targeted tests passed, 93.18% CLI-helper coverage, Pylint 10/10. Replacement internal build 188195239 is queued on disposable validation commit 20425a0ea91c38d46894628e82eb44c89f2ec866: https://msazure.visualstudio.com/CloudNativeCompute/_build/results?buildId=188195239 . Live rerun has not yet passed; no production table/routing changes.

[Diamond (diamondpowell)]

Audited live validation: build 188195239

Verified the evidence beyond the green pipeline status:

• Internal source: 20425a0ea91c38d46894628e82eb44c89f2ec866; package source: 35f440e88a60f4cb31972c5c86c8db8b854c4ab3 (current PR head); shared templates: 1889c34df737aed96181e121b9ff1ecbdeca99f2.
• Both jobs built/reinstalled the feature-branch wheel and imported the collector from Python 3.10 site-packages, not a source-tree PYTHONPATH.
• Both small VMSS and VirtualMachines jobs executed create, scale up, Deployment, StatefulSet, Job, scale down, delete, collect, upload, and resource-group cleanup successfully.
• Independently downloaded both persisted result blobs and verified seven distinct operations each, actual ARM pool types (not just matrix labels), scale 1 -> 2 -> 1, all requested workloads (1 Deployment x 10 replicas, 3 StatefulSets x 1 replica, 20 Jobs x 1 completion), successful metadata, timing values, run IDs, and legacy field types.
• Existing production ADX column schema and JSON mapping match the output envelope. Read-only KQL evaluation of the actual uploaded rows passes the existing scale-operation name, Azure cloud, succeeded, command timing, and node-count predicates for both pool types.
• Current GitHub Python Validation on this head: 843 passed, 1 skipped; 83.66% coverage; Pylint 10/10.
• Compared disposable validation branch against production internal PR source c13adf8d: differences are test YAML, package provenance checks, and post-collection assertions; no substituted CRUD implementation or changed production command sequence.

Scope and remaining gates

This is a valid one-region, small-size installed-package lifecycle and output-contract pass. It is not proof of production ingestion/dashboard visibility: blobs were uploaded under the validation-branch path, no matching rows were found in existing CRUD ADX tables, and no production ingestion was forced. Medium/large production scales and other regions were not exercised.

After this package merges, validate the resulting open-source main package before internal cutover, and verify production ingestion/consumer visibility during the approved rollout. The internal default tracks mutable main, so this run proves the recorded commits, not arbitrary future open-source changes.

Run: https://msazure.visualstudio.com/CloudNativeCompute/_build/results?buildId=188195239

[Diamond (diamondpowell)]

Added timeout follow-up in commit 96aa4e0.

• The configured AKS client timeout now reaches node-pool create, direct scale, and progressive scale for both SDK and VirtualMachines CLI provisioning. Low-level helpers retain their 1800-second default.
• Regression tests cover a 2400-second allowance surviving beyond the old 1800-second limit, exact configured/default polling limits, timeout propagation, and publication of failed legacy operation records.
• Local validation: 205 targeted CRUD/client/helper tests passed; all changed Python files passed Pylint at 10/10. Pylance signature checks passed for the SDK helper and both callable factories.
• An installed-wheel smoke with synthetic failed-operation data confirmed failure fields are retained, repeat collection is idempotent, and malformed input does not replace previously published output.

A new live cloud run has not been performed for this revision. This comment does not claim end-to-end ADX or dashboard validation.

[Diamond (diamondpowell)]

Backward-compatibility follow-up published in 9c2b133.

Corrections:

• Client ARM provisioning now keeps the historical 1800-second allowance when readiness settings are shorter; longer settings still extend it. The default 600-second readiness budget is unchanged. Regression coverage includes an actual mocked SDK operation finishing after 630 seconds and 32 pool/configuration/action combinations across create, direct scale, and progressive scale.
• Default collection again preserves source enumeration order. Deterministic sorting is limited to opt-in legacy mode.

Validation:

• Full local suite: 850 passed, 1 skipped; 83.69% coverage; changed Python files Pylint 10/10.
• Differential execution of the pre-PR default collector versus this revision produced exactly matching 30 output rows for Azure/GPU/AWS/Machine input shapes, including failed records, source order, field values, and append behavior.
• Fresh wheel build/install passed. Source and installed imports resolved the real Azure SDK namespace; all nine CRUD module CLI help paths and both telescope-crud/telescope-iperf3 console entries worked.

No new live cloud provisioning run was performed for this revision. These checks do not claim deployed ADX/dashboard validation.

[Diamond (diamondpowell)]

Validation of exact source commit 3dfd548 is complete:

• GitHub python-validation, yaml-validation, and CLA checks all passed on this head.
• Both small live Azure node-pool cases (VirtualMachineScaleSets and VirtualMachines) passed all seven CRUD/workload operations, scaling 1 → 2 → 1.
• Independently downloaded and validated all 14 uploaded legacy-compatible JSONL rows: actual ARM pool types, successful workloads (one deployment with ten replicas, three StatefulSets, twenty Jobs), readiness targets, timing fields, aliases, and typed context.
• Installed package source-byte checks and Azure SDK import checks passed in both jobs; teardown completed with no remaining run resource groups.

This does not certify 51/201-node scale or automatic production ADX/dashboard ingestion. Neither PR was merged; no production ingestion settings were changed.