feat(trip-cluster): cluster orchestration (seed derivation + coverage) + CLI

2026-06-27 17:28:27 +02:00
parent a9b443cd53
commit 5256df4117
3 changed files with 131 additions and 1 deletions
@@ -0,0 +1,61 @@
+from photoflow.core import Store
+from photoflow.core.models import Asset
+from photoflow.immich import pipeline
+from app.cluster_run import run_cluster
+
+
+def _store(tmp_path):
+    return Store(str(tmp_path / "t.db")).connect()
+
+
+def test_run_cluster_seeds_bounded_tag_not_people_tag(tmp_path):
+    s = _store(tmp_path)
+    # "Italy 2019" spans 2 days -> seeds; "Mum" spans years -> not a seed.
+    rows = [("a", "2019-06-01T10:00:00", ["Italy 2019", "Mum"]),
+            ("b", "2019-06-02T10:00:00", ["Italy 2019"]),
+            ("c", "2010-01-01T10:00:00", ["Mum"]),
+            ("d", "2022-01-01T10:00:00", ["Mum"])]
+    for i, t, tags in rows:
+        s.upsert_asset(Asset(immich_id=i, taken_at=t))
+        s.set_asset_tags(i, tags)
+    res = run_cluster(s)
+    seeded = [c for c in s.all_clusters() if c.suggested_name == "Italy 2019"]
+    assert len(seeded) == 1
+    assert sorted(x.immich_id for x, _ in s.cluster_members(seeded[0].id)) == ["a", "b"]
+    assert res["clusters"] >= 1
+    s.close()
+
+
+def test_run_cluster_excludes_processed_and_replaces(tmp_path):
+    s = _store(tmp_path)
+    s.upsert_asset(Asset(immich_id="a", taken_at="2019-06-01T10:00:00"))
+    s.upsert_asset(Asset(immich_id="p", taken_at="2019-06-01T11:00:00"))
+    s.set_asset_tags("a", [])
+    s.set_asset_tags("p", [pipeline.PROCESSED])
+    s.mark_processed("p")
+    run_cluster(s)
+    run_cluster(s)  # idempotent replace — not doubled
+    all_member_ids = [x.immich_id for c in s.all_clusters()
+                      for x, _ in s.cluster_members(c.id)]
+    assert "p" not in all_member_ids
+    assert all_member_ids.count("a") == 1
+    s.close()
+
+
+def test_run_cluster_short_span_non_dominant_tag_does_not_seed(tmp_path):
+    s = _store(tmp_path)
+    # "Lunch" is on 2 photos within a ~30-min window, but that same window is
+    # full of untagged photos -> the tag does NOT dominate its window
+    # (2/5 = 0.4 < 0.5), so it must not seed even though its span is tiny.
+    rows = [("x", "2020-03-01T12:00:00", ["Lunch"]),
+            ("y", "2020-03-01T12:30:00", ["Lunch"]),
+            ("u1", "2020-03-01T12:05:00", []),
+            ("u2", "2020-03-01T12:15:00", []),
+            ("u3", "2020-03-01T12:25:00", [])]
+    for i, t, tags in rows:
+        s.upsert_asset(Asset(immich_id=i, taken_at=t))
+        s.set_asset_tags(i, tags)
+    run_cluster(s)
+    seeded = [c for c in s.all_clusters() if c.suggested_name == "Lunch"]
+    assert seeded == []
+    s.close()