US07-06: Validate Performance and Resource Bounds (#93)

migrations/versions/0016_performance_indexes.py

@@ -0,0 +1,45 @@
+"""Indexes the large-library read paths need (US07-06).
+
+Revision ID: 0016_performance_indexes
+Revises: 0015_exif_projections
+Create Date: 2026-08-17
+
+Measured, not guessed. At 100k assets the workflow home and the safety queue both
+resolve "the current decision per asset" with a window function over
+``safety_reviews``; a plain ``asset_id`` index makes SQLite sort every partition by
+hand. Ordering the index by ``(asset_id, created_at DESC)`` halves that query.
+
+``duplicate_members(cluster_id, asset_id)`` serves the paged member list of a
+cluster with thousands of members, which is the other page that stopped being cheap.
+"""
+
+from alembic import op
+
+revision = "0016_performance_indexes"
+down_revision = "0015_exif_projections"
+branch_labels = None
+depends_on = None
+
+
+def upgrade() -> None:
+    op.create_index(
+        "ix_safety_reviews_asset_created",
+        "safety_reviews",
+        ["asset_id", "created_at"],
+    )
+    op.create_index(
+        "ix_duplicate_members_cluster_asset",
+        "duplicate_members",
+        ["cluster_id", "asset_id"],
+    )
+    op.create_index(
+        "ix_analysis_results_approx_year",
+        "analysis_results",
+        ["approx_year"],
+    )
+
+
+def downgrade() -> None:
+    op.drop_index("ix_analysis_results_approx_year", table_name="analysis_results")
+    op.drop_index("ix_duplicate_members_cluster_asset", table_name="duplicate_members")
+    op.drop_index("ix_safety_reviews_asset_created", table_name="safety_reviews")