[ewlarson]

Summary

The repository dashboard exceeds the proxy's 30-second timeout because its counts repeatedly scan the resources table for each repository. Calculate harvested counts once per repository and all resource counts in one grouped query, shared by the dashboard and public repository summary endpoint.

Deduplicate tags per resource to preserve the previous array-membership behavior. Keep publication-state precedence, suppressed counts, empty repositories, and latest harvest details unchanged. No database migration or new index is required.

Validation

• Read-only production benchmark: the new query completed in 666.8 ms across roughly 321,000 resources and 48 repositories.
• Compared complete old and new query results in the same repeatable-read production snapshot: 48 rows each, zero differences.
• 7 tests passed, covering repository queries, public API responses, and dashboard HTML rendering. The PostgreSQL 15 regression test covers duplicate/shared tags, publication-state fallback and case handling, missing resources, disabled/empty repositories, and latest-run selection.
• Ruff and git diff --check passed.

The production benchmark measures database query execution, not end-to-end HTTP latency. This change has not been deployed.