[Prangshuman Das (t-prda)]

Diagnostic warm-up experiment - DO NOT MERGE / DO NOT QUEUE

Related work: AB#646383 (link only; this experiment does not resolve the work item).

Draft for online review. Based on #12447 (features/646383-sql-reset-pilot) so Files changed isolates the warm-up treatment from the existing SQL-reset instrumentation. Not intended to merge into that pilot or the production stack.

Treatment

• Freeze the pristine tenant template before running the existing first-app warm-up from Warm up container before parallel test dispatch #10204 on default.
• For each clean batch: reset all three workers, externally probe each worker's companies API, then dispatch AL tests.
• Preserve existing password-based authentication, reused SQL database names, pinned PR2 source/packages, and NST 30.0.55665.0. No AL-source or assertion changes.
• Five W1 and five DE trials; maximum five concurrent trials and 120 minutes per trial. Manual dispatch only, no automatic replacement runs.
• Exact ordered 21-codeunit prefix through Expense Users, Capabilities and Activity Log. This is not the full API suite; expected W1 baseline is 234 passed and 19 skipped, including 19 non-skipped Expense target methods.
• No AL-test, first-app warm-up, scheduler or CI retries. Current branch permits companies-probe retries only: at most three attempts, two-second delay, 20-second connection/read timeouts. Only HTTP 500/502/503/504 and connection/timeout failures are eligible; authentication/permanent errors fail immediately.
• Record sanitized probe attempts, warm-up separately from clean results, provenance, reset timeline, JUnit, event logs and ownership-checked disposable-container cleanup. No local/shared NST operations.

Running experiment versus current review head

Follow Actions run 37608087700.

That run is pinned to fcc1776c6b165199dd66e4227675b1cc31da7a8f: app warm-up plus a single companies probe, without probe retries. It cannot acquire later branch changes.

The current review head 8a7ebc7d665c0a812c2b3b4902c63df570959954 additionally includes bounded companies-probe retries and improved collection of nested .buildartifacts test results. Those later changes have not been dispatched. Creating this draft does not launch another matrix; normal full-country PR initialization is excluded for this branch.

Validation and limits

• Dispatched candidate: 123 local PowerShell tests passed.
• Probe-retry candidate: 141 tests passed; final artifact-collection update: 56 diagnostic tests passed, with 87 scheduler tests previously passing.
• Live outcomes are pending. A green run would be non-reproduction/mitigation evidence, not proof of root cause or production reliability; earlier controls also passed.
• This combines first-app warm-up, companies probes and timing/concurrency changes, so it cannot isolate their individual effects.
• Shared failing Expense methods/runtime match the unrelated Keep the draft unit of measure and variant code when creating the purchase line #12333 investigation, but the complete package cohort is pinned PR2, not identical to Keep the draft unit of measure and variant code when creating the purchase line #12333.

Protocol: build/scripts/SqlResetPilot/README.md.

[github-actions]

⚠️ Stale Status Check Deleted

The Pull Request Build workflow run for this PR was older than 72 hours and has been deleted.

📋 Why was it deleted?

Status checks that are too old may no longer reflect the current state of the target branch. To ensure this PR is validated against the latest code and passes up-to-date checks, a fresh build is required.

---

🔄 How to trigger a new status check:

1. 📤 Push a new commit to the PR branch, or
2. 🔁 Close and reopen the PR

This will automatically trigger a new Pull Request Build workflow run.