Saved Links should dedupe unresolved failures by URL and clear stale rows after successful reingest

[ahjinsolo]

Summary

Saved Links currently behaves like an attempt log, but users experience it as a list of unresolved URLs.

That mismatch creates confusing duplicates. If the same URL fails multiple times, Kompl can show multiple Saved Links rows for the same source_url. If the URL later succeeds through a new onboarding session, the old unresolved failure row may still remain because stale failure cleanup is session-scoped.

Current observed state

In one local wiki:

• 394 unresolved URL failure rows
• 267 unique unresolved URLs
• 80 duplicate URL groups
• 127 extra duplicate rows removable by keeping one row per exact source_url
• 0 active compile sessions at inspection time

So Saved Links overstates unresolved work by about 47 percent in this case.

Why this feels wrong as a user

When I open Saved Links, I expect to see links that still need attention.

Instead, I may see repeated entries for the same URL because Kompl stores each failed ingest attempt as a separate unresolved row. That makes it hard to tell:

• how many unique links still need work
• which failures are real versus repeated attempts
• whether retrying actually improved anything
• whether a successfully imported URL is still incorrectly listed as unresolved

Likely cause

Failed URL ingest inserts a new ingest_failures row.

/api/compile/retry-failed cleans stale failures only for URLs in the same failed staging session. That works for same-session retry recovery, but not for a fresh reingest of an old Saved Link.

So this flow can leave stale rows:

1. URL fails in session A.
2. Saved Links shows the unresolved URL.
3. User later reingests the same URL in session B.
4. URL succeeds and becomes a source.
5. Old session A failure row remains unresolved.

Expected behavior

Saved Links should represent unresolved URLs, not every failed attempt.

Desired behavior:

• At most one unresolved Saved Links row per canonical/exact URL.
• New failed attempts update or replace the existing unresolved row.
• If a URL imports successfully, any unresolved failure row for that URL is resolved or removed.
• Attempt history can still be preserved in activity logs or metadata.
• The Saved Links page count should match unique unresolved URLs.

Proposed implementation direction

A conservative fix could be:

1. On insertIngestFailure, check for an existing unresolved row with the same source_url.
2. If one exists, update it with the newest attempt metadata instead of inserting a second unresolved row.
3. On successful URL ingest, delete or mark resolved any unresolved ingest_failures rows for that source_url, regardless of session.
4. Regenerate the Saved Links page after either update.
5. Add a one-time cleanup/migration or admin repair command for existing duplicate rows.

Acceptance criteria

• Repeated failures for the same URL produce one visible Saved Links entry.
• A later successful import removes that URL from Saved Links.
• Retry-failed behavior remains session-scoped where it needs to be, but successful ingest performs URL-scoped reconciliation.
• Tests cover duplicate failure insert, newer failure update, and successful reingest cleanup.