Merge pull request 'Unused-tag fandom fix + ML-worker logging/tuning + unified dropdown Enter' (#86) from dev into main

backend/app/services/cleanup_service.py

@@ -21,6 +21,7 @@ from sqlalchemy import func, or_, select, update
 from sqlalchemy.orm import Session
 
 from ..models import Artist, ImageRecord, LibraryAuditRun, Tag
+from ..models.series_chapter import SeriesChapter
 from ..models.series_page import SeriesPage
 from ..models.tag import image_tag
 
@@ -141,21 +142,36 @@ def count_tag_associations(session: Session, *, tag_id: int) -> int:
 def find_unused_tags(
     session: Session, *, limit: int | None = None,
 ) -> list[Tag]:
-    """Tags with no image_tag rows AND no series_page rows.
+    """Tags genuinely referenced by nothing — safe to sweep.
 
-    Sorted by name. Used by both dry-run preview and the live prune.
-    A tag is "unused" iff it has zero rows in image_tag AND zero rows
-    in series_page (so we don't accidentally prune a series tag that
-    happens to have no images yet).
+    Sorted by name. Used by both dry-run preview and the live prune. A tag is
+    "unused" iff it has zero references across ALL the ways a tag can be in use:
+      - image_tag           (applied to an image)
+      - series_page         (a series tag with ordered pages)
+      - series_chapter      (a series tag with chapters but no pages yet, e.g.
+                             all-placeholder)
+      - tag.fandom_id       (a fandom referenced by a character)
+
+    The fandom check is essential: fandom tags are NEVER applied to images — a
+    character carries its fandom via fandom_id — so without it every assigned
+    fandom looked "unused", and the FK is ondelete=SET NULL, so deleting one
+    would silently strip the fandom off all its characters (operator-flagged
+    2026-06-08).
     """
     used_via_image_tag = select(image_tag.c.tag_id).distinct()
     used_via_series = select(SeriesPage.series_tag_id).where(
         SeriesPage.series_tag_id.is_not(None)
     ).distinct()
+    used_via_chapter = select(SeriesChapter.series_tag_id).distinct()
+    used_via_fandom = select(Tag.fandom_id).where(
+        Tag.fandom_id.is_not(None)
+    ).distinct()
     stmt = (
         select(Tag)
         .where(Tag.id.not_in(used_via_image_tag))
         .where(Tag.id.not_in(used_via_series))
+        .where(Tag.id.not_in(used_via_chapter))
+        .where(Tag.id.not_in(used_via_fandom))
         .order_by(Tag.name)
     )
     if limit is not None:

backend/app/tasks/ml.py

@@ -6,8 +6,10 @@ apply_allowlist_tags sweeps which are 'maintenance' lane. Sync sessions
 (Celery workers are sync processes), same pattern as FC-2a tasks.
 """
 
+import logging
 from pathlib import Path
 
+from celery.exceptions import SoftTimeLimitExceeded
 from sqlalchemy import select
 from sqlalchemy.exc import DBAPIError, OperationalError
 
@@ -15,6 +17,8 @@ from ..celery_app import celery
 from ..models import ImageRecord, MLSettings
 from ._sync_engine import sync_session_factory as _sync_session_factory
 
+log = logging.getLogger(__name__)
+
 IMAGES_ROOT = Path("/images")
 VIDEO_EXTS = {".mp4", ".mov", ".avi", ".mkv", ".webm", ".m4v", ".wmv", ".flv"}
 
@@ -31,8 +35,8 @@ def _is_video(path: Path) -> bool:
     retry_backoff_max=60,
     retry_jitter=True,
     max_retries=3,
-    # Sized for the video branch: sample 10 frames, run tagger +
-    # embedder on each (≈20 GPU ops vs 2 for an image). A loaded
+    # Sized for the video branch: sample 6 frames, run tagger +
+    # embedder on each (≈12 GPU ops vs 2 for an image). A loaded
     # ml-worker can take 5-10 min on a long video; bumped from
     # 5min/7min on 2026-05-28 after operator-flagged image 6288 (a
     # .mp4) hit the recovery sweep at 5 min while still legitimately
@@ -46,71 +50,140 @@ def tag_and_embed(self, image_id: int) -> dict:
     then enqueue per-image allowlist application.
 
     Video: sample frames between 10% and 90% of duration (VIDEO_ML_FRAMES,
-    default 10). Max-pool tagger confidences across frames, mean-pool the
+    default 6). Max-pool tagger confidences across frames, mean-pool the
     SigLIP embeddings. On no-frames returns status='no_frames' (not an error).
     """
     import os
+    import time
 
     from ..services.ml.embedder import get_embedder
     from ..services.ml.tagger import get_tagger
 
-    SessionLocal = _sync_session_factory()
-    with SessionLocal() as session:
-        record = session.get(ImageRecord, image_id)
-        if record is None:
-            return {"status": "missing", "image_id": image_id}
-        settings = session.execute(
-            select(MLSettings).where(MLSettings.id == 1)
-        ).scalar_one()
+    # Phase + file context, so a timeout/crash names WHICH file and WHERE it
+    # died instead of a bare SoftTimeLimitExceeded() (operator-flagged 2026-06-08:
+    # the activity told them nothing about the file or why). `ctx` is enriched
+    # once the record is loaded; both feed the worker log AND the re-raised
+    # exception message (which becomes the activity's error_message).
+    started = time.monotonic()
+    phase = "open_session"
+    ctx = f"image_id={image_id}"
 
-        src = Path(record.path)
-        if not src.is_file():
-            return {"status": "file_missing", "image_id": image_id}
+    def _elapsed() -> float:
+        return time.monotonic() - started
 
-        tagger = get_tagger()
-        embedder = get_embedder()
+    try:
+        SessionLocal = _sync_session_factory()
+        with SessionLocal() as session:
+            record = session.get(ImageRecord, image_id)
+            if record is None:
+                return {"status": "missing", "image_id": image_id}
+            settings = session.execute(
+                select(MLSettings).where(MLSettings.id == 1)
+            ).scalar_one()
 
-        if _is_video(src):
-            # Layer-3 isolation: ffprobe (a separate process) validates
-            # the container before we burn ~20 GPU ops sampling frames
-            # from it. A corrupt video that would crash the frame
-            # decoder is rejected cleanly here instead of taking down
-            # the ml-worker. Operator-flagged 2026-05-28.
-            from ..utils import safe_probe
-            vprobe = safe_probe.probe_video(src)
-            if not vprobe.ok:
-                return {
-                    "status": "bad_video", "image_id": image_id,
-                    "reason": vprobe.reason,
-                }
-            frames = _sample_video_frames(
-                src, int(os.environ.get("VIDEO_ML_FRAMES", "10"))
+            src = Path(record.path)
+            is_vid = _is_video(src)
+            ctx = (
+                f"image_id={image_id} path={record.path} mime={record.mime} "
+                f"bytes={record.size_bytes} video={is_vid}"
             )
-            if not frames:
-                return {"status": "no_frames", "image_id": image_id}
-            preds = _maxpool_predictions([tagger.infer(f) for f in frames])
-            import numpy as np
+            log.info("tag_and_embed start: %s", ctx)
+            if not src.is_file():
+                log.warning("tag_and_embed file missing on disk: %s", ctx)
+                return {"status": "file_missing", "image_id": image_id}
 
-            embedding = np.mean(
-                [embedder.infer(f) for f in frames], axis=0
-            ).astype("float32")
-            for f in frames:
-                f.unlink(missing_ok=True)
-        else:
-            raw = tagger.infer(src)
-            preds = {
-                name: {"category": p.category, "confidence": p.confidence}
-                for name, p in raw.items()
-            }
-            embedding = embedder.infer(src)
+            phase = "load_models"
+            tagger = get_tagger()
+            embedder = get_embedder()
 
-        record.tagger_predictions = preds
-        record.tagger_model_version = settings.tagger_model_version
-        record.siglip_embedding = embedding.tolist()
-        record.siglip_model_version = settings.embedder_model_version
-        session.add(record)
-        session.commit()
+            if is_vid:
+                # Layer-3 isolation: ffprobe (a separate process) validates
+                # the container before we burn ~20 GPU ops sampling frames
+                # from it. A corrupt video that would crash the frame
+                # decoder is rejected cleanly here instead of taking down
+                # the ml-worker. Operator-flagged 2026-05-28.
+                phase = "video_probe"
+                from ..utils import safe_probe
+                vprobe = safe_probe.probe_video(src)
+                if not vprobe.ok:
+                    log.warning(
+                        "tag_and_embed bad video (%s): %s", vprobe.reason, ctx
+                    )
+                    return {
+                        "status": "bad_video", "image_id": image_id,
+                        "reason": vprobe.reason,
+                    }
+                phase = "video_sample_frames"
+                t0 = time.monotonic()
+                frames = _sample_video_frames(
+                    src, int(os.environ.get("VIDEO_ML_FRAMES", "6"))
+                )
+                log.info(
+                    "tag_and_embed sampled %d frame(s) in %.1fs: %s",
+                    len(frames), time.monotonic() - t0, ctx,
+                )
+                if not frames:
+                    return {"status": "no_frames", "image_id": image_id}
+                phase = "video_infer"
+                import numpy as np
 
+                preds = _maxpool_predictions([tagger.infer(f) for f in frames])
+                embedding = np.mean(
+                    [embedder.infer(f) for f in frames], axis=0
+                ).astype("float32")
+                for f in frames:
+                    f.unlink(missing_ok=True)
+            else:
+                phase = "tag"
+                t0 = time.monotonic()
+                raw = tagger.infer(src)
+                log.info(
+                    "tag_and_embed tagged in %.1fs (%d tags): %s",
+                    time.monotonic() - t0, len(raw), ctx,
+                )
+                preds = {
+                    name: {"category": p.category, "confidence": p.confidence}
+                    for name, p in raw.items()
+                }
+                phase = "embed"
+                t0 = time.monotonic()
+                embedding = embedder.infer(src)
+                log.info(
+                    "tag_and_embed embedded in %.1fs: %s",
+                    time.monotonic() - t0, ctx,
+                )
+
+            phase = "persist"
+            record.tagger_predictions = preds
+            record.tagger_model_version = settings.tagger_model_version
+            record.siglip_embedding = embedding.tolist()
+            record.siglip_model_version = settings.embedder_model_version
+            session.add(record)
+            session.commit()
+    except SoftTimeLimitExceeded:
+        log.error(
+            "tag_and_embed TIMED OUT after %.0fs in phase=%s: %s",
+            _elapsed(), phase, ctx,
+        )
+        # Re-raise as SoftTimeLimitExceeded (preserves the 'timeout' status in
+        # the task_run signal) but WITH context, so the activity error_message
+        # names the file + phase instead of being empty.
+        raise SoftTimeLimitExceeded(
+            f"timed out in phase={phase} after {_elapsed():.0f}s ({ctx})"
+        ) from None
+    except Exception:
+        # OSError/DBAPIError/OperationalError are autoretried — re-raise the
+        # ORIGINAL so the type is preserved; just make sure it's logged with
+        # context first.
+        log.exception(
+            "tag_and_embed FAILED in phase=%s after %.0fs: %s",
+            phase, _elapsed(), ctx,
+        )
+        raise
+
+    log.info(
+        "tag_and_embed ok in %.1fs (%d tags): %s", _elapsed(), len(preds), ctx
+    )
     apply_allowlist_tags.delay(image_id=image_id)
     return {"status": "ok", "image_id": image_id, "tags": len(preds)}
 

frontend/src/components/modal/AliasPickerDialog.vue

@@ -8,6 +8,7 @@
       </p>
       <v-autocomplete
         v-model="selectedId"
+        v-model:menu="menuOpen"
         :items="results"
         :item-title="(t) => t.fandom_name ? `${t.name} — ${t.fandom_name}` : t.name"
         :item-value="(t) => t.id"
@@ -15,6 +16,7 @@
         label="Canonical tag"
         no-filter clearable density="compact"
         @update:search="onSearch"
+        @keydown.enter.capture="onEnter"
       />
     </v-card-text>
     <v-card-actions>
@@ -31,9 +33,10 @@
 <script setup>
 import { ref } from 'vue'
 import { useApi } from '../../composables/useApi.js'
+import { useAcceptOnEnter } from '../../composables/useAcceptOnEnter.js'
 
 const props = defineProps({ category: { type: String, required: true } })
-defineEmits(['confirm', 'cancel'])
+const emit = defineEmits(['confirm', 'cancel'])
 
 const api = useApi()
 const results = ref([])
@@ -41,6 +44,11 @@ const loading = ref(false)
 const selectedId = ref(null)
 let debounce = null
 
+// Enter on the closed dropdown confirms the selection instead of re-opening it.
+const { menuOpen, onEnter } = useAcceptOnEnter(() => {
+  if (selectedId.value != null) emit('confirm', selectedId.value)
+})
+
 function onSearch(q) {
   if (debounce) clearTimeout(debounce)
   debounce = setTimeout(async () => {

frontend/src/components/modal/FandomPicker.vue

@@ -51,16 +51,19 @@
 <script setup>
 import { nextTick, onMounted, ref } from 'vue'
 import { useTagStore } from '../../stores/tags.js'
+import { useAcceptOnEnter } from '../../composables/useAcceptOnEnter.js'
 
 const emit = defineEmits(['confirm', 'cancel'])
 const store = useTagStore()
 
 const selectedId = ref(null)
 const newName = ref('')
-const menuOpen = ref(false)
 const fandomRef = ref(null)
 const newNameRef = ref(null)
 
+// Enter on the closed dropdown accepts the chosen fandom instead of re-opening it.
+const { menuOpen, onEnter: onSearchEnter } = useAcceptOnEnter(() => onConfirm())
+
 // Always refresh on open so the list reflects fandoms created elsewhere (#712).
 onMounted(() => store.loadFandoms())
 
@@ -92,22 +95,6 @@ function onConfirm () {
   if (f) emit('confirm', f)
 }
 
-// Enter on the search field — bound in the CAPTURE phase so this runs BEFORE
-// Vuetify's own input keydown handler (which opens the menu on Enter). When the
-// menu is open, let Vuetify pick the highlighted item (it sets selectedId +
-// closes the menu). When it's closed and a fandom is already chosen, accept it
-// and stop the event so Vuetify never (re)opens the menu — that re-open was the
-// bug: Enter popped the dropdown instead of submitting (operator-flagged
-// 2026-06-07).
-function onSearchEnter (e) {
-  if (menuOpen.value) return
-  if (selectedId.value != null) {
-    e.preventDefault()
-    e.stopPropagation()
-    onConfirm()
-  }
-}
-
 // Tab from the search field goes to the "new fandom" text field (operator-
 // specified). Shift+Tab keeps normal reverse traversal.
 function onSearchTab (e) {

frontend/src/components/modal/FandomSetDialog.vue

@@ -82,6 +82,7 @@
 <script setup>
 import { nextTick, onMounted, ref } from 'vue'
 import { useTagStore } from '../../stores/tags.js'
+import { useAcceptOnEnter } from '../../composables/useAcceptOnEnter.js'
 
 const props = defineProps({ tag: { type: Object, required: true } })
 const emit = defineEmits(['updated', 'cancel'])
@@ -93,10 +94,15 @@ const newName = ref('')
 const busy = ref(false)
 const error = ref(null)
 const collision = ref(null)
-const menuOpen = ref(false)
 const fandomRef = ref(null)
 const newNameRef = ref(null)
 
+// Enter on the closed dropdown Saves the changed fandom instead of re-opening it.
+const { menuOpen, onEnter: onSearchEnter } = useAcceptOnEnter(() => {
+  if (busy.value) return
+  if (selectedId.value !== (props.tag.fandom_id ?? null)) onSave()
+})
+
 // Always refresh on open so the list reflects fandoms created elsewhere (#712).
 onMounted(() => store.loadFandoms())
 
@@ -131,20 +137,6 @@ async function onCreate() {
   }
 }
 
-// Enter on the search field — bound in the CAPTURE phase so it runs BEFORE
-// Vuetify's own handler (which opens the menu on Enter). Menu open → let Vuetify
-// pick the highlighted item. Menu closed with a changed selection → Save and stop
-// the event so Vuetify never (re)opens the dropdown (the bug this fixes).
-function onSearchEnter(e) {
-  if (menuOpen.value) return
-  if (busy.value) return
-  if (selectedId.value !== (props.tag.fandom_id ?? null)) {
-    e.preventDefault()
-    e.stopPropagation()
-    onSave()
-  }
-}
-
 // Tab from the search field goes to the "new fandom" text field (operator-
 // specified). Shift+Tab keeps normal reverse traversal.
 function onSearchTab(e) {

frontend/src/components/posts/PostSeriesMenu.vue

@@ -32,11 +32,13 @@
           </p>
           <v-autocomplete
             v-model="picked"
+            v-model:menu="pickerMenuOpen"
             :items="hits"
             :item-title="(s) => s.name"
             :item-value="(s) => s.id"
             :loading="searching"
             label="Series" density="compact" autofocus clearable
+            @keydown.enter.capture="onEnter"
           />
         </v-card-text>
         <v-card-actions>
@@ -56,6 +58,7 @@
 import { ref } from 'vue'
 import { useRouter } from 'vue-router'
 import { useApi } from '../../composables/useApi.js'
+import { useAcceptOnEnter } from '../../composables/useAcceptOnEnter.js'
 import { toast } from '../../utils/toast.js'
 
 const props = defineProps({ post: { type: Object, required: true } })
@@ -70,6 +73,12 @@ const picked = ref(null)
 const hits = ref([])
 const searching = ref(false)
 
+// Enter on the closed dropdown adds the chosen series instead of re-opening it.
+// (aliased — `menuOpen` above is the outer "Series ▾" v-menu's state.)
+const { menuOpen: pickerMenuOpen, onEnter } = useAcceptOnEnter(() => {
+  if (picked.value != null && !busy.value) onAddExisting()
+})
+
 async function onPromote() {
   menuOpen.value = false
   busy.value = true

frontend/src/composables/useAcceptOnEnter.js

@@ -0,0 +1,35 @@
+import { ref } from 'vue'
+
+/**
+ * Unifies keyboard behavior for confirm-style `v-autocomplete` / `v-select`
+ * dropdowns — a single picker whose Enter should trigger the surrounding
+ * modal's primary action (Merge, Add, Save, Use…).
+ *
+ * The recurring bug (operator-flagged across the merge / fandom / series / alias
+ * dialogs): pressing Enter with the menu CLOSED re-opens the dropdown instead of
+ * accepting the chosen value, because Vuetify's own Enter handler opens the menu.
+ * We bind in the CAPTURE phase so this runs BEFORE Vuetify's handler:
+ *   - menu OPEN   → return; let Vuetify pick the highlighted item (it sets the
+ *                   value and closes the menu).
+ *   - menu CLOSED → preventDefault + stopPropagation so Vuetify can't re-open the
+ *                   menu, then call `accept()` — the modal's confirm. `accept`
+ *                   should itself no-op when nothing valid is selected.
+ *
+ * Usage:
+ *   const { menuOpen, onEnter } = useAcceptOnEnter(onConfirm)
+ *   <v-autocomplete v-model:menu="menuOpen" @keydown.enter.capture="onEnter" … />
+ *
+ * `menuOpen` is exposed so a component that programmatically focuses the field
+ * can force the menu shut (so the very next Enter accepts rather than being
+ * swallowed as a menu interaction).
+ */
+export function useAcceptOnEnter(accept) {
+  const menuOpen = ref(false)
+  function onEnter(e) {
+    if (menuOpen.value) return
+    e.preventDefault()
+    e.stopPropagation()
+    accept()
+  }
+  return { menuOpen, onEnter }
+}

frontend/src/views/TagsView.vue

@@ -55,11 +55,13 @@
           </p>
           <v-autocomplete
             v-model="mergeTargetId"
+            v-model:menu="mergeMenuOpen"
             :items="mergeHits" item-title="name" item-value="id"
             :loading="mergeLoading"
             density="compact" variant="outlined" hide-details
             placeholder="Search target tag…" no-filter
             @update:search="onMergeSearch"
+            @keydown.enter.capture="onMergeEnter"
           />
         </v-card-text>
         <v-card-actions>
@@ -107,6 +109,7 @@ import { useTagDirectoryStore } from '../stores/tagDirectory.js'
 import { useAdminStore } from '../stores/admin.js'
 import { useApi } from '../composables/useApi.js'
 import { useInfiniteScroll } from '../composables/useInfiniteScroll.js'
+import { useAcceptOnEnter } from '../composables/useAcceptOnEnter.js'
 import TagCard from '../components/discovery/TagCard.vue'
 import MergeConfirmDialog from '../components/discovery/MergeConfirmDialog.vue'
 import DestructiveConfirmModal from '../components/modal/DestructiveConfirmModal.vue'
@@ -181,6 +184,10 @@ const api = useApi()
 
 // Tag merge via dots-menu (separate from inline-rename collision flow above)
 const mergePickerOpen = ref(false)
+// Enter on the closed dropdown accepts the target instead of re-opening it.
+const { menuOpen: mergeMenuOpen, onEnter: onMergeEnter } = useAcceptOnEnter(
+  () => onMergeConfirm()
+)
 const mergeSource = ref(null)
 const mergeTargetId = ref(null)
 const mergeHits = ref([])

tests/test_cleanup_service.py

@@ -155,6 +155,31 @@ def test_find_unused_tags_returns_only_unreferenced(db_sync, tmp_path):
     assert names.index("aaa-unused") < names.index("zzz-unused")
 
 
+def test_find_unused_tags_excludes_fandom_referenced_by_character(db_sync):
+    # A fandom is never on an image — a character carries it via fandom_id — so
+    # an assigned fandom must NOT be flagged unused (deleting it SET-NULLs the
+    # fandom off its characters). Operator-flagged 2026-06-08.
+    fandom = _make_tag(db_sync, name="Creux", kind=TagKind.fandom)
+    char = _make_tag(db_sync, name="OcChar", kind=TagKind.character)
+    char.fandom_id = fandom.id
+    _make_tag(db_sync, name="NobodysFandom", kind=TagKind.fandom)
+    db_sync.commit()
+
+    names = [t.name for t in cleanup_service.find_unused_tags(db_sync)]
+    assert "Creux" not in names          # referenced by a character → kept
+    assert "NobodysFandom" in names      # genuinely orphaned → still swept
+
+
+def test_find_unused_tags_excludes_series_with_only_chapters(db_sync):
+    # An all-placeholder series has chapters but no pages yet — not unused.
+    series = _make_tag(db_sync, name="EmptySeries", kind=TagKind.series)
+    db_sync.add(SeriesChapter(series_tag_id=series.id, chapter_number=1))
+    db_sync.commit()
+
+    names = [t.name for t in cleanup_service.find_unused_tags(db_sync)]
+    assert "EmptySeries" not in names
+
+
 # --- unlink_image_files ---------------------------------------------