Files
KArtSell.Aegis/docs/CURRENT/PHASE-1_REQUEUE_READINESS.md
T
kjh2064 fbff7cfbda
ci / backend (push) Failing after 1s
ci / static (push) Failing after 11s
ci / publish (push) Has been cancelled
ci / frontend (push) Has been cancelled
Build & Test with Secrets / build (push) Failing after 1s
Build & Test with Secrets / security-scan (push) Failing after 6s
deploy / deploy (push) Successful in 2m59s
Build & Test with Secrets / frontend (push) Successful in 3m57s
deploy / notify (push) Successful in 1s
Build & Test with Secrets / notification (push) Failing after 1s
PHASE-1-SHADOW-RUN: assign readiness owner and deadline
2026-08-06 14:28:00 +09:00

68 lines
4.3 KiB
Markdown

# Phase 1 Shadow Run Requeue Readiness
## Traceability
- WBS: `PHASE-1-SHADOW-RUN`
- Slice: requeue readiness and approval package
- Source: `docs/CURRENT/WBS_EXECUTION_PROCEDURES.md`, `docs/CURRENT/PHASE-1_SHADOW_RUN_STATUS_CORRECTION.md`, `docs/CURRENT/AEG-X-004_DBUP_EVIDENCE.md`, `src/KArtSell.Host/Features/ShadowRun/API_CONTRACT.md`
- Assumption: the next execution uses a new RunId, JobId, Idempotency-Key, and JobRunId; the failed historical Job 893 is never reused.
- Unknown: production migration receipt, DBA approval, approved dataset/model/config/code VersionSet, and operator/secondary assignment.
- Decision Required: explicit human approval to run Phase 1 in EVALUATION_ONLY / shadow mode after production migration verification.
## Current evidence boundary
- `0032_shadow_run_queued_status_contract.sql` is append-only and accepts the existing application `Queued` state.
- Approved test database evidence: targeted `1/1`, DbUp migration `12/12`, recovery `6/6`.
- Job 893 was never started; no historical run may be resumed or mutated.
- Automatic order, KIS submission, model promotion, rollback automation, and threshold mutation remain disabled.
## Unsafe legacy script
`scripts/EXECUTE_PHASE_1_NOW.ps1` is not an approved execution path and must not be run. Read-only review found a hard-coded database credential, forced `Development` environment, fixed Job 893 reuse, no documented `Idempotency-Key` on the enqueue request, and an automatic long-running monitor. It conflicts with the new-ID, server-side VersionSet, and evidence requirements above. Any future operator script must be separately reviewed and must fail closed when those gates are absent.
## Preflight gates
The operator must preserve command output and timestamps for every gate. A failed gate stops the procedure.
1. Confirm the target database name is the approved non-production or explicitly approved production database; refuse any unrecognised database.
2. Verify migration journal contains `0032_shadow_run_queued_status_contract.sql` and the `check_status` constraint includes `Queued`.
3. Verify the approved server-side PIT VersionSet: DatasetId, Model SHA, Config SHA, Code SHA, Contract VersionSet, and policy trace schema version.
4. Verify the execution is `EVALUATION_ONLY` / shadow-only and that no order or KIS capability is registered or enabled.
5. Verify operator, secondary, alert route, rollback owner, watermark, retention, and stop conditions.
6. Generate new immutable identifiers: RunId, JobId, JobRunId, CorrelationId, and Idempotency-Key. Do not reuse Job 893.
7. Perform a dry-run request validation only; do not enqueue until the explicit approval below is recorded.
## Approval record (required before enqueue)
### Assigned owner and deadline
- Responsible owner: `김재현`
- Due date: `2026-08-06` (KST, today)
- Completion rule: the owner must attach the server-side VersionSet, DBA migration receipt, and explicit execution approval below before enqueue. Owner assignment alone does not constitute those evidences.
```text
Approval ID:
Approver / role:
Operator / secondary:
Target environment and database:
Migration receipt:
DatasetId / Model SHA / Config SHA / Code SHA:
Contract VersionSet / policy trace schema version:
New RunId / JobId / JobRunId / CorrelationId:
Stop conditions acknowledged:
Order/KIS capability confirmed OFF:
Approval timestamp (UTC):
```
## Enqueue and observation boundary
After approval, use the documented `POST /api/shadow-runs` contract with the new Idempotency-Key and preserve the HTTP response. Poll only the new RunId. Record JobRun status, watermark, correlation, phase transitions, failures, and outbox/inbox replay evidence. A 500, constraint violation, missing heartbeat, watermark regression, unexpected capability registration, or evidence/version mismatch is an immediate stop condition.
## Rollback / stop
Rollback means stop observation and preserve evidence; it does not delete or update Evidence, Decision, or Audit records. Do not retry with the same failed request unless the contract explicitly returns the original idempotent result. Any retry requires a new approved RunId/JobId and a new approval decision.
## Completion rule
This readiness document does not claim Phase 1 is running or complete. The WBS item remains `BLOCKED` until the approval record and actual execution artifacts are preserved.