Merge pull request 'v26.05.26.4 — hotfix: migration 0022 pre-DELETE colliding ImageProvenance before UPDATE' (#26) from dev into main

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

.forgejo/workflows/ci.yml

@@ -24,22 +24,17 @@ jobs:
     steps:
       - uses: actions/checkout@v4
 
-      - name: Cache pip wheels
+      # Cache step removed 2026-05-26: act_runner's cache backend has been
-        # continue-on-error: act_runner's cache backend has been broken on
+      # broken on this homelab runner since 2026-05-15 (first as request-
-        # this homelab setup since 2026-05-15 — bolt.db + cache/ dir exist
+      # timeout warnings, then as hard "Cannot find module .../dist/restore/
-        # but the action's JS bundle now fails to load at all
+      # index.js" failures that tank the whole job). The cache step targeted
-        # ("Cannot find module .../dist/restore/index.js"), hard-failing
+      # ~/.cache/pip but the install below uses `uv pip install` primarily,
-        # the whole job. The install step that follows handles cold caches
+      # whose own cache lives at ~/.cache/uv — so the cache step's real
-        # natively (uv pip / pip install both work without it), so net
+      # benefit was marginal even when working. Cost of removal: ~30s of
-        # cost of disabling the cache here is ~30s of wheel downloads per
+      # wheel downloads per job. Future re-enable: mount ~/.cache/uv as a
-        # job. Re-enable once the runner-side cache backend is fixed.
+      # docker volume at the runner level (skips actions/cache entirely),
-        continue-on-error: true
+      # or fix the runner-side cache backend (clear /var/run/act/actions/*,
-        uses: actions/cache@v4
+      # pin act_runner version, etc.).
-        with:
-          path: ~/.cache/pip
-          key: pip-${{ runner.os }}-py314-${{ hashFiles('requirements.txt') }}
-          restore-keys: |
-            pip-${{ runner.os }}-py314-
 
       - name: Install Python deps
         # ruff is pre-installed in the ci-python image (see CI-Runner/CI-python/
@@ -133,22 +128,6 @@ jobs:
           --health-retries 10
     steps:
       - uses: actions/checkout@v4
-      - name: Cache pip wheels
-        # continue-on-error: act_runner's cache backend has been broken on
-        # this homelab setup since 2026-05-15 — bolt.db + cache/ dir exist
-        # but the action's JS bundle now fails to load at all
-        # ("Cannot find module .../dist/restore/index.js"), hard-failing
-        # the whole job. The install step that follows handles cold caches
-        # natively (uv pip / pip install both work without it), so net
-        # cost of disabling the cache here is ~30s of wheel downloads per
-        # job. Re-enable once the runner-side cache backend is fixed.
-        continue-on-error: true
-        uses: actions/cache@v4
-        with:
-          path: ~/.cache/pip
-          key: pip-${{ runner.os }}-py314-${{ hashFiles('requirements.txt') }}
-          restore-keys: |
-            pip-${{ runner.os }}-py314-
       - name: API integration shard (resolve service IPs, migrate, test)
         run: |
           set -eux
@@ -207,22 +186,6 @@ jobs:
           --health-retries 10
     steps:
       - uses: actions/checkout@v4
-      - name: Cache pip wheels
-        # continue-on-error: act_runner's cache backend has been broken on
-        # this homelab setup since 2026-05-15 — bolt.db + cache/ dir exist
-        # but the action's JS bundle now fails to load at all
-        # ("Cannot find module .../dist/restore/index.js"), hard-failing
-        # the whole job. The install step that follows handles cold caches
-        # natively (uv pip / pip install both work without it), so net
-        # cost of disabling the cache here is ~30s of wheel downloads per
-        # job. Re-enable once the runner-side cache backend is fixed.
-        continue-on-error: true
-        uses: actions/cache@v4
-        with:
-          path: ~/.cache/pip
-          key: pip-${{ runner.os }}-py314-${{ hashFiles('requirements.txt') }}
-          restore-keys: |
-            pip-${{ runner.os }}-py314-
       - name: Importer integration shard (resolve service IPs, migrate, test)
         run: |
           set -eux
@@ -281,22 +244,6 @@ jobs:
           --health-retries 10
     steps:
       - uses: actions/checkout@v4
-      - name: Cache pip wheels
-        # continue-on-error: act_runner's cache backend has been broken on
-        # this homelab setup since 2026-05-15 — bolt.db + cache/ dir exist
-        # but the action's JS bundle now fails to load at all
-        # ("Cannot find module .../dist/restore/index.js"), hard-failing
-        # the whole job. The install step that follows handles cold caches
-        # natively (uv pip / pip install both work without it), so net
-        # cost of disabling the cache here is ~30s of wheel downloads per
-        # job. Re-enable once the runner-side cache backend is fixed.
-        continue-on-error: true
-        uses: actions/cache@v4
-        with:
-          path: ~/.cache/pip
-          key: pip-${{ runner.os }}-py314-${{ hashFiles('requirements.txt') }}
-          restore-keys: |
-            pip-${{ runner.os }}-py314-
       - name: Core integration shard (everything not api / importer / migration / phash / sidecar / scan / archive / backfill)
         run: |
           set -eux

alembic/versions/0022_source_per_artist_platform.py

@@ -83,55 +83,80 @@ def upgrade() -> None:
         if not other_ids:
             continue
 
-        # STEP 2: PRE-merge colliding Posts BEFORE the bulk reparent.
+        # STEP 2: PRE-merge ALL Posts with duplicate external_post_id
-        # Find pairs (keep, drop) where:
+        # across the entire (canonical + others) group, BEFORE the bulk
-        #   keep  = a Post under the canonical source
+        # reparent. Two cases must both be handled:
-        #   drop  = a Post under one of the non-canonical sources whose
+        #   (A) canonical has Post X with epid=N; an "other" source has
-        #           external_post_id equals keep.external_post_id
+        #       Post Y with epid=N → after bulk UPDATE, (canonical, N)
-        # If we let the bulk UPDATE try to repoint drop onto canonical,
+        #       collides with itself.
-        # uq_post_source_external_id fires row-by-row and aborts the
+        #   (B) two different "other" sources each have a Post with
-        # whole migration.
+        #       epid=N; canonical has none → after bulk UPDATE, both
-        colliding = conn.execute(
+        #       are repointed to (canonical, N) and the second collides.
+        # The earlier version of this migration only handled (A); the
+        # operator's deploy 2026-05-26 tripped (B) at line 139.
+        # Fix: group ALL Posts in the (artist, platform) by epid; for
+        # any group with count>1, pick the keep (prefer one already
+        # under canonical; else lowest id) and merge the rest into it.
+        all_posts = conn.execute(
             text("""
-                SELECT keep.id AS keep_id, drop_.id AS drop_id
+                SELECT external_post_id, id, source_id
-                FROM post AS keep
+                FROM post
-                JOIN post AS drop_
+                WHERE source_id = :canonical OR source_id = ANY(:others)
-                  ON drop_.external_post_id = keep.external_post_id
+                ORDER BY external_post_id, id
-                 AND drop_.id != keep.id
-                WHERE keep.source_id = :canonical
-                  AND drop_.source_id = ANY(:others)
             """),
             {"canonical": canonical_id, "others": other_ids},
         ).fetchall()
-        for keep_id, drop_id in colliding:
+        by_epid: dict = {}
-            conn.execute(
+        for epid, post_id, src_id in all_posts:
-                text("""
+            by_epid.setdefault(epid, []).append((post_id, src_id))
-                    UPDATE image_provenance SET post_id = :keep
+        for _epid, posts in by_epid.items():
-                    WHERE post_id = :drop_
+            if len(posts) <= 1:
-                """),
+                continue
-                {"keep": keep_id, "drop_": drop_id},
+            # Prefer a Post already under canonical as the keep.
-            )
+            canonical_posts = [p for p in posts if p[1] == canonical_id]
-            conn.execute(
+            if canonical_posts:
-                text("""
+                keep_id = canonical_posts[0][0]
-                    UPDATE image_record SET primary_post_id = :keep
+            else:
-                    WHERE primary_post_id = :drop_
+                keep_id = posts[0][0]  # already sorted by id ASC
-                """),
+            drop_ids = [p[0] for p in posts if p[0] != keep_id]
-                {"keep": keep_id, "drop_": drop_id},
+            for drop_id in drop_ids:
-            )
+                # Pre-delete image_provenance rows under drop_ whose
-            # Repointed provenance may now collide on
+                # image_record_id ALREADY has a provenance under keep —
-            # uq_image_provenance_image_post (alembic 0021). Dedupe:
+                # the UPDATE below would otherwise repoint them and
-            # keep min(id) per (image_record_id, post_id).
+                # trip uq_image_provenance_image_post (alembic 0021)
-            conn.execute(text("""
+                # row-by-row before any after-the-fact dedupe could
-                DELETE FROM image_provenance ip1
+                # run. Operator's v26.05.26.3 deploy 2026-05-26 tripped
-                USING image_provenance ip2
+                # this at line 123.
-                WHERE ip1.image_record_id = ip2.image_record_id
+                conn.execute(
-                  AND ip1.post_id = ip2.post_id
+                    text("""
-                  AND ip1.id > ip2.id
+                        DELETE FROM image_provenance
-            """))
+                        WHERE post_id = :drop_
-            conn.execute(
+                          AND image_record_id IN (
-                text("DELETE FROM post WHERE id = :drop_"),
+                              SELECT image_record_id FROM image_provenance
-                {"drop_": drop_id},
+                              WHERE post_id = :keep
-            )
+                          )
+                    """),
+                    {"keep": keep_id, "drop_": drop_id},
+                )
+                # Now safe to repoint the survivors.
+                conn.execute(
+                    text("""
+                        UPDATE image_provenance SET post_id = :keep
+                        WHERE post_id = :drop_
+                    """),
+                    {"keep": keep_id, "drop_": drop_id},
+                )
+                conn.execute(
+                    text("""
+                        UPDATE image_record SET primary_post_id = :keep
+                        WHERE primary_post_id = :drop_
+                    """),
+                    {"keep": keep_id, "drop_": drop_id},
+                )
+                conn.execute(
+                    text("DELETE FROM post WHERE id = :drop_"),
+                    {"drop_": drop_id},
+                )
 
         # STEP 3: Bulk reparent the remaining Posts off the other
         # Sources. After step 2, no collisions on