Report publish transform failures honestly and keep healthy pieces processing #100

New Issue

2026-10-05T12:50:13Z

eleith commented

2026-10-05 12:50:13 +00:00

Goal

Do not convert failed transforms into successful empty output. Preserve healthy-piece continuation and report failed pieces explicitly.

The owner explicitly dropped the larger complete-piece/unchanged-published-version goal as not worth pursuing. Files and metadata already changed for a failed piece may remain changed; this is an accepted limitation.

Agreed scope

Transform and image-preparation errors must propagate rather than become empty records/jobs.
Legitimate empty/not-applicable outputs remain successful and may clear old records normally.
Do not persist asset records if production for that piece fails.
Catch failures at the piece-processing boundary, collect file path/message, and continue healthy pieces.
Carry failed-piece details into the publish result and show an honest partial-failure outcome rather than an unqualified Published successfully label.
Keep the existing publish/CDN/cache phases; no file staging, rollback, historical versions, database migration or new retry policy.

Review slices

Worker failure propagation, per-piece batch handling and regression tests.
Publish result/types and web reporting/tests.

Do not land the worker slice alone while final reporting still claims all-success. Interruption/retry recovery remains a separate investigation in #102.

## Goal Do not convert failed transforms into successful empty output. Preserve healthy-piece continuation and report failed pieces explicitly. The owner explicitly dropped the larger complete-piece/unchanged-published-version goal as not worth pursuing. Files and metadata already changed for a failed piece may remain changed; this is an accepted limitation. ## Agreed scope - Transform and image-preparation errors must propagate rather than become empty records/jobs. - Legitimate empty/not-applicable outputs remain successful and may clear old records normally. - Do not persist asset records if production for that piece fails. - Catch failures at the piece-processing boundary, collect file path/message, and continue healthy pieces. - Carry failed-piece details into the publish result and show an honest partial-failure outcome rather than an unqualified Published successfully label. - Keep the existing publish/CDN/cache phases; no file staging, rollback, historical versions, database migration or new retry policy. ## Review slices 1. Worker failure propagation, per-piece batch handling and regression tests. 2. Publish result/types and web reporting/tests. Do not land the worker slice alone while final reporting still claims all-success. Interruption/retry recovery remains a separate investigation in #102.

eleith added the bug label 2026-10-05 12:50:13 +00:00

eleith changed title from ~~Preserve last successful derivatives when a transform fails~~ to Report transform failures in the best-effort publish summary

2026-10-05 23:03:22 +00:00

eleith changed title from ~~Report transform failures in the best-effort publish summary~~ to Publish each Markdown independently and leave failed updates unchanged

2026-10-05 23:43:43 +00:00

eleith changed title from ~~Publish each Markdown independently and leave failed updates unchanged~~ to Report publish transform failures honestly and keep healthy pieces processing

2026-10-06 14:55:04 +00:00

eleith referenced a pull request that will close this issue

2026-10-06 19:33:37 +00:00

Report publish transform failures and continue healthy pieces #109

eleith closed this issue

2026-10-06 19:37:17 +00:00

Sign in to join this conversation.

1 Participants

Notifications

Due Date

No due date set.

Dependencies

No dependencies set.

Reference: eleith/luzzle#100