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

photo_pipeline/services/analysis.py

@@ -18,7 +18,6 @@ from __future__ import annotations
 
 import json
 import os
-import uuid
 from datetime import datetime, timezone
 from typing import Protocol
 
@@ -26,9 +25,9 @@ from sqlalchemy import func, select
 from sqlalchemy.orm import sessionmaker
 
 from photo_pipeline import path_policy
-from photo_pipeline.models import AnalysisResult, Asset, SafetyReview
+from photo_pipeline.models import AnalysisResult, Asset
 from photo_pipeline.services import exif_checkpoint
-from photo_pipeline.services.safety import SFW
+from photo_pipeline.services.safety import SFW, latest_reviews
 
 MODEL = "gemini-2.5-flash"
 PROMPT_VERSION = "1"
@@ -72,11 +71,26 @@ class AnalysisService:
 
     def _sfw_asset_ids(self, session) -> set[str]:
         """Asset ids whose latest safety decision is ``sfw`` — the ONLY assets that
-        may reach the provider."""
-        latest: dict[str, str | None] = {}
-        for review in session.scalars(select(SafetyReview).order_by(SafetyReview.created_at)):
-            latest[review.asset_id] = review.decision
-        return {aid for aid, decision in latest.items() if decision == SFW}
+        may reach the provider.
+
+        The "latest row wins" rule is applied in SQL (US07-06); loading every review
+        to fold it in Python made the gate cost grow with the review history rather
+        than with the work being gated.
+        """
+        latest = latest_reviews().subquery()
+        return set(
+            session.scalars(select(latest.c.asset_id).where(latest.c.decision == SFW))
+        )
+
+    def _sfw_count(self, session) -> int:
+        """How many assets the gate currently allows, without listing them."""
+        latest = latest_reviews().subquery()
+        return int(
+            session.scalar(
+                select(func.count()).select_from(latest).where(latest.c.decision == SFW)
+            )
+            or 0
+        )
 
     def _is_still_sfw(self, asset_id: str) -> bool:
         """Re-read the current safety decision straight from the database."""
@@ -104,9 +118,12 @@ class AnalysisService:
             )
             return [a.id for a in assets if a.id not in done]
 
-    def counts(self) -> dict[str, int]:
+    def counts(self, *, eligible: int | None = None) -> dict[str, int]:
+        """Analysis progress. ``eligible`` may be passed by a caller that has just
+        counted confirmed-SFW assets, so the workflow home does not resolve the
+        latest decision of every asset twice on one page load (US07-06)."""
         with self._session_factory() as session:
-            sfw = self._sfw_asset_ids(session)
+            eligible = self._sfw_count(session) if eligible is None else eligible
             rows = dict(
                 session.execute(
                     select(AnalysisResult.status, func.count()).group_by(AnalysisResult.status)
@@ -115,10 +132,10 @@ class AnalysisService:
             analyzed = int(rows.get("analyzed", 0))
             errored = int(rows.get("error", 0))
             return {
-                "eligible": len(sfw),
+                "eligible": eligible,
                 "analyzed": analyzed,
                 "error": errored,
-                "pending": max(len(sfw) - analyzed - errored, 0),
+                "pending": max(eligible - analyzed - errored, 0),
             }
 
     def run(self, asset_ids: list[str] | None = None) -> dict: