ci: keep runner swap for goldendict export; revert expunge_all, preload set_ru TSV

- set swap-storage: false in free-disk-space across all release workflows —
  the parallel high-mem exporter OOMed the swapless 16GB runner (upstream
  draft_release keeps default swap and passes)
- revert local db_session.expunge_all() addition for upstream parity
- hoist read_set_ru_from_tsv() into render_data (once per export instead of
  per word) in export_dpd_ru/export_dpd_sbs, update registry local_changes

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

Author: Devamitta

.github/workflows/anki_release.yml

@@ -24,7 +24,9 @@ jobs:
         haskell: true
         large-packages: true
         docker-images: true
-        swap-storage: true
+        # keep the default swap file: the goldendict high-mem exporter needs
+        # it on 16GB runners (upstream draft_release keeps swap too)
+        swap-storage: false
 
     - name: Checkout repository
       uses: actions/checkout@v5

.github/workflows/ru_release.yml

@@ -33,7 +33,9 @@ jobs:
         haskell: true
         large-packages: true
         docker-images: true
-        swap-storage: true
+        # keep the default swap file: the goldendict high-mem exporter needs
+        # it on 16GB runners (upstream draft_release keeps swap too)
+        swap-storage: false
 
     - name: Checkout repository
       uses: actions/checkout@v5

.github/workflows/ru_release_kindle.yml

@@ -30,7 +30,9 @@ jobs:
         haskell: true
         large-packages: true
         docker-images: true
-        swap-storage: true
+        # keep the default swap file: the goldendict high-mem exporter needs
+        # it on 16GB runners (upstream draft_release keeps swap too)
+        swap-storage: false
 
     - name: Checkout repository
       uses: actions/checkout@v5

.github/workflows/ru_release_test.yml

@@ -33,7 +33,9 @@ jobs:
         haskell: true
         large-packages: true
         docker-images: true
-        swap-storage: true
+        # keep the default swap file: the goldendict high-mem exporter needs
+        # it on 16GB runners (upstream draft_release keeps swap too)
+        swap-storage: false
 
     - name: Checkout repository
       uses: actions/checkout@v5

exporter/goldendict/export_dpd.py

@@ -377,9 +377,6 @@ def _add_parts(row: DpdHeadwordDbRowItems) -> DpdHeadwordDbParts:
                     pr.counter(processed, pali_words_count, marker)
                     reported = processed
 
-            # Evict processed records to prevent memory leak
-            db_session.expunge_all()
-
     total_sizes = sum_rendered_sizes(rendered_sizes)
 
     return dpd_data_list, total_sizes

exporter/goldendict/export_dpd_ru.py

@@ -63,6 +63,7 @@ class DpdHeadwordRenderDataBase(TypedDict):
     cf_set: set[str]
     idioms_set: set[str]
     show_id: bool
+    set_ru_dict: dict[str, str]
 
 
 class DpdHeadwordRenderData(DpdHeadwordRenderDataBase):
@@ -141,7 +142,7 @@ def render_pali_word_dpd_html(
     synonyms += i.inflections_thai_list
     synonyms += i.family_set_list
 
-    set_ru_dict = read_set_ru_from_tsv()
+    set_ru_dict = rd["set_ru_dict"]
     ru_set_list = []
     for english_word in i.family_set_list:
         if english_word in set_ru_dict:
@@ -339,6 +340,7 @@ def generate_dpd_html(
         "cf_set": cf_set,
         "idioms_set": idioms_set,
         "show_id": show_id,
+        "set_ru_dict": read_set_ru_from_tsv(),
     }
 
     # Preload the family tables once instead of 3 queries per headword.
@@ -407,9 +409,6 @@ def _add_parts(row: DpdHeadwordDbRowItems) -> DpdHeadwordDbParts:
                     pr.counter(processed, pali_words_count, marker)
                     reported = processed
 
-            # Evict processed records to prevent memory leak
-            db_session.expunge_all()
-
     total_sizes = sum_rendered_sizes(rendered_sizes)
 
     return dpd_data_list, total_sizes

exporter/goldendict/export_dpd_sbs.py

@@ -69,6 +69,7 @@ class DpdHeadwordRenderDataBase(TypedDict):
     show_ru_data: bool
     show_ta_data: bool
     show_grammar: bool
+    set_ru_dict: dict[str, str]
 
 
 class DpdHeadwordRenderData(DpdHeadwordRenderDataBase):
@@ -157,7 +158,7 @@ def render_pali_word_dpd_html(
     synonyms += i.family_set_list
 
     if show_ru_data:
-        set_ru_dict = read_set_ru_from_tsv()
+        set_ru_dict = rd["set_ru_dict"]
         ru_set_list = []
         for english_word in i.family_set_list:
             if english_word in set_ru_dict:
@@ -362,6 +363,7 @@ def generate_dpd_html(
         "show_ru_data": show_ru_data,
         "show_ta_data": show_ta_data,
         "show_grammar": show_grammar,
+        "set_ru_dict": read_set_ru_from_tsv(),
     }
 
     # Preload the family tables once instead of 3 queries per headword.
@@ -429,9 +431,6 @@ def _add_parts(row: DpdHeadwordDbRowItems) -> DpdHeadwordDbParts:
                     pr.counter(processed, pali_words_count, marker)
                     reported = processed
 
-            # Evict processed records to prevent memory leak
-            db_session.expunge_all()
-
     total_sizes = sum_rendered_sizes(rendered_sizes)
 
     return dpd_data_list, total_sizes

kamma/upstream_sync/registry.json

@@ -312,7 +312,7 @@
         "`DpdHeadwordDbParts` includes `ru: Russian` model; queries filter for `Russian.id.isnot(None)`.",
         "`render_pali_word_dpd_html` uses `data_classes_dps.HeadwordData` and renders via `dpd_headword_ru.jinja`.",
         "Size tracking: `size_dict[\"dpd_summary\"]` records `data.ru_meaning` length instead of English meaning.",
-        "Synonyms include Russian set names via `tools_for_ru_exporter.read_set_ru_from_tsv()`.",
+        "`DpdHeadwordRenderDataBase` preloads Russian set names via `read_set_ru_from_tsv()` once at the start of `generate_dpd_html`; each render accesses the dict from `render_data`.",
         "`generate_dpd_html` uses `rupth: RuPaths` and `joinedload(DpdHeadword.ru)`."
       ],
       "watch_for": [
@@ -401,7 +401,7 @@
       "local_changes": [
         "`generate_dpd_html` signature adds `dpspth: DPSPaths`, `show_sbs_data`, `show_ru_data`, `show_ta_data`, and `show_grammar` flags.",
         "Eagerly loads `.rt`, `.ru`, `.ta`, and `.sbs` relationships.",
-        "Synonyms logic includes a conditional Russian set lookup if `show_ru_data` is true.",
+        "`DpdHeadwordRenderDataBase` preloads Russian set names via `read_set_ru_from_tsv()` once at the start of `generate_dpd_html`; each render accesses the dict conditionally from `render_data` if `show_ru_data` is true.",
         "`_parse_batch_top_level` uses `sbs_templates/` and passes all locale flags including `show_ta_data`.",
         "`DpdHeadwordRenderDataBase` TypedDict includes `show_ta_data: bool`."
       ],