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

photo_pipeline/services/library.py

@@ -15,10 +15,9 @@ from __future__ import annotations
 
 import json
 import re
-from collections import Counter
 from pathlib import Path
 
-from sqlalchemy import and_, func, or_, select
+from sqlalchemy import String, and_, case, cast, func, or_, select, text
 from sqlalchemy.orm import sessionmaker
 
 from photo_pipeline.models import AnalysisResult, Asset
@@ -71,38 +70,75 @@ class LibraryService:
         return {"rows": rows, "total": total, "offset": offset, "limit": limit}
 
     def stats(self) -> dict:
+        """Library-wide totals, aggregated in SQL (US07-06).
+
+        This page used to load every analysis row — object, tags, and all — to count
+        them in Python, which cost half a second at 100k assets and grew from there.
+        Only the album breakdown still walks rows, and only their path and status:
+        SQLite has no ``dirname``, and two short strings per asset is cheap.
+        """
         with self._session_factory() as session:
             status = dict(
                 session.execute(
                     select(AnalysisResult.status, func.count()).group_by(AnalysisResult.status)
                 ).all()
             )
-            rows = list(
-                session.execute(
-                    select(AnalysisResult, Asset.current_path).join(
-                        Asset, Asset.id == AnalysisResult.asset_id
-                    )
-                )
-            )
             albums: dict[str, dict] = {}
-            tag_counts: Counter = Counter()
-            year_counts: Counter = Counter()
-            people: Counter = Counter()
-            errors = []
-            for result, path in rows:
+            for path, row_status in session.execute(
+                select(Asset.current_path, AnalysisResult.status).join(
+                    Asset, Asset.id == AnalysisResult.asset_id
+                )
+            ):
                 album = _album_of(path)
                 bucket = albums.setdefault(album, {"album": album, "done": 0, "total": 0})
                 bucket["total"] += 1
-                if result.status in DONE:
+                if row_status in DONE:
                     bucket["done"] += 1
-                for tag in _tags(result.tags):
-                    tag_counts[tag] += 1
-                if result.approx_year is not None:
-                    year_counts[result.approx_year] += 1
-                if result.people_count is not None:
-                    people["3+" if result.people_count >= 3 else str(result.people_count)] += 1
-                if result.status == "error":
-                    errors.append({"path": path, "error": result.error_message})
+            year_counts = dict(
+                session.execute(
+                    select(AnalysisResult.approx_year, func.count())
+                    .where(AnalysisResult.approx_year.is_not(None))
+                    .group_by(AnalysisResult.approx_year)
+                ).all()
+            )
+            people = dict(
+                session.execute(
+                    select(
+                        case(
+                            (AnalysisResult.people_count >= 3, "3+"),
+                            else_=cast(AnalysisResult.people_count, String),
+                        ),
+                        func.count(),
+                    )
+                    .where(AnalysisResult.people_count.is_not(None))
+                    .group_by(
+                        case(
+                            (AnalysisResult.people_count >= 3, "3+"),
+                            else_=cast(AnalysisResult.people_count, String),
+                        )
+                    )
+                ).all()
+            )
+            # SQLite's JSON1 counts the tag arrays where they are: parsing 50k JSON
+            # strings in Python to keep the top 40 is the definition of doing work
+            # the database already does. Malformed tags are skipped, not fatal.
+            tag_counts = session.execute(
+                text(
+                    "SELECT tag.value AS value, count(*) AS total "
+                    "FROM analysis_results, json_each(analysis_results.tags) AS tag "
+                    "WHERE analysis_results.tags IS NOT NULL "
+                    "AND json_valid(analysis_results.tags) "
+                    "GROUP BY tag.value ORDER BY total DESC, value LIMIT 40"
+                )
+            ).all()
+            errors = [
+                {"path": path, "error": message}
+                for path, message in session.execute(
+                    select(Asset.current_path, AnalysisResult.error_message)
+                    .join(Asset, Asset.id == AnalysisResult.asset_id)
+                    .where(AnalysisResult.status == "error")
+                )
+            ]
         return {
             "total": sum(status.values()),
             "status": status,
@@ -111,7 +147,7 @@ class LibraryService:
             "season": self._facet("season"),
             "people": [{"value": v, "count": n} for v, n in sorted(people.items())],
             "years": [{"value": y, "count": year_counts[y]} for y in sorted(year_counts)],
-            "top_tags": [{"value": t, "count": n} for t, n in tag_counts.most_common(40)],
+            "top_tags": [{"value": t, "count": n} for t, n in tag_counts],
             "albums": sorted(albums.values(), key=lambda d: d["album"]),
             "errors": sorted(errors, key=lambda e: e["path"] or ""),
         }