[shiny-code-app]

Why

Odoo previews on open pull requests were stuck in reconcile_required and re-deployed an unchanged head about every 30 minutes. Any driver fail after a provider effect was treated as an unknown outcome, including a data schedule the provider had already reported as failed. That left CM previews with no finished state and no reason a Client could read.

What changed

• A data schedule the provider reports as failed or error is now a completed failed preview (502), through OdooPostDeployScheduleFailed and the adapter in odoo_preview_apply_execution.py. The same build and binding replays the failure instead of applying again; a new build or changed binding tries again, using the existing preview_build_failed wording.
• The Client-facing summary names only the exception type (The data schedule failed with RuntimeError.). The bounded redacted log line stays on the operation record and in the worker log.
• Template-inherited Website intent no longer sends LAUNCHPLANE_WEBSITE_BOOTSTRAP_REQUIRED on a preview, including through the import-source composition. Exact preview intent and every stable lane stay strict.
• Genuinely unknown outcomes (no terminal failed status, unreadable provider, lost lease, interrupted executor, preservation retries, schedules before the data workflow) keep the existing guarded recovery.

Depends on cbusillo/odoo-devkit#415 for images that skip inherited Website intent. This PR is safe without it: an older image raises as it does today, and now ends as a finished failed preview instead of looping.

Refs #5828. Merging does not by itself prove the finish line on the current CM previews. The issue stays open until they either serve or show their failure on deployed Launchplane.

Verification

• Classification: bug fix in preview reconcile and provider-operation recovery, a safety area. Failing first: the reconcile test failed before the fix (preview_operation_status was reconcile_required).
• Whole modules passing: tests.test_product_reconcile, tests.test_preview_reservation_recovery, tests.test_odoo_preview_apply_execution, tests.test_odoo_preview_runtime, tests.test_odoo_preview_import_source, tests.test_odoo_instance_overrides, tests.test_copy_readback_diagnostics, tests.test_odoo_post_deploy, tests.test_http_app_driver_odoo, tests.test_dokploy_schedule_trigger, tests.test_dokploy.
• ruff check, ruff format --check, mypy control_plane tests pass.
• JetBrains inspection: UNKNOWN, native_python_support_plugin_unavailable, native PyCharm, 10 files, no findings reported; tracked on Headless inspection workers: confirm project interpreter and Pydantic plugin so type findings stop being noise codex-skills#3363. Not a clean result.
• Not verified: no live provider, preview or record was touched. The real Dokploy status strings come from code and existing fixtures, not a live read.

Review

OpenAI reviewed the change, then the fix, and returned one P1 and one P2.

• P1 acted on: arbitrary exception text could reach Clients. Client-facing text now carries only the exception type.
• P2 partly acted on: the dotted-name pattern accepted non-exception names (customer.AliceSmith). It now needs an Error or Exception suffix or psycopg2.errors.*.
• P2 remainder declined: a fixed registry of exception names. A class name in the Director's own product code is not private data, and a registry is a second list to keep.
• Answered none: no unknown outcome becomes a finished failure, no stable-lane or binding weakening, and completed 502 records fit PR-close cleanup, retained-restore decisions and the retained plan chain.

Removed: the unconditional unknown-outcome raise for a schedule the provider reports as failed, and the required flag on template-inherited preview Website intent.
Net lines: +437 (105 code and docs, 332 tests). Growth is the new failure path and its tests.

[shiny-code-app]

Review result: weighed
Head: 6718b6c

Review by another model, weighed for head 6718b6c.

Reviewer: OpenAI through the model-review helper (model not reported by the CLI), full change plus a focused re-review of the fix.

• P1, acted on: arbitrary schedule exception text could reach Clients. Client text now names only the exception type; the bounded redacted line stays on the operation record and in the worker log.
• P2, acted on in part: the dotted-name pattern accepted non-exception names. It now needs an Error or Exception suffix or psycopg2.errors.*.
• P2 remainder, declined: a fixed registry of exception names adds a second list to keep, and a class name in the Director's own product code is not private data.
• Specific questions answered none: no unknown outcome becomes a finished failure, no stable-lane or preview-binding weakening from the flag change, completed 502 records fit PR-close cleanup, retained-restore decisions and the retained plan chain.

[shiny-code-app]

Ready for the merge train

Head: 6718b6c

[shiny-code-app]

Launchplane completed the merge-train step for this pull request.

• Repository: cbusillo/launchplane
• Base branch: main
• Pull request: fix: finish Odoo previews whose data schedule ran and failed #5829
• Controller action: land_batch
• Controller record: merge-train-batch-landing-plan-20261109T171609Z-68df0af995bad0a0

Launchplane landed this pull request through protected batch PR #5832.

Launchplane manages this comment and will update it as the train moves.