perf: accelerate complete wildcard and literal candidate scans

app.py

@@ -61,12 +61,15 @@ FULLTEXT_MATCH_CACHE_TTL_SECONDS = 300
 FULLTEXT_MATCH_CACHE_MAX_ENTRIES = 16
 FULLTEXT_MATCH_CACHE_MAX_TOTAL_IDS = 250000
 WILDCARD_FTS_SCAN_MIN_CANDIDATES = 5000
+METADATA_MATCH_CACHE_MAX_ENTRIES = 128
 PREVIEW_CACHE_MAX_ENTRIES = 64
 PREVIEW_CACHE_MAX_ENTRY_BYTES = 2 * 1024 * 1024
 api_response_cache: dict[tuple, tuple[float, object]] = {}
 api_response_cache_lock = threading.Lock()
 search_response_cache: dict[tuple, tuple[float, object]] = {}
 search_response_cache_lock = threading.Lock()
+metadata_match_cache: dict[tuple, frozenset[int]] = {}
+metadata_match_cache_lock = threading.Lock()
 preview_text_cache: dict[str, str] = {}
 preview_text_cache_lock = threading.Lock()
 cache_key_locks: dict[tuple, threading.Lock] = {}
@@ -247,6 +250,8 @@ class FulltextDatabases:
         self.file_signatures[source] = signature
         self.generation += 1
         self.match_cache.clear()
+        with metadata_match_cache_lock:
+            metadata_match_cache.clear()
         if old_connection is not None:
             if self.connection_readers.get(old_connection, 0):
                 self.retired_connections.add(old_connection)
@@ -287,6 +292,8 @@ class FulltextDatabases:
             with self.lock:
                 self.closed = True
                 self.match_cache.clear()
+                with metadata_match_cache_lock:
+                    metadata_match_cache.clear()
                 for event in self.match_inflight.values():
                     event.set()
                 self.match_inflight.clear()
@@ -778,6 +785,8 @@ def load_json_gz(path: Path):
 
 def build_indexes() -> None:
     global word_index, latin_all_index, latin_file_index, latin_vocabulary_all, source_records_map, extension_counts, source_extension_counts, name_order, size_order, source_name_order, source_size_order
+    with metadata_match_cache_lock:
+        metadata_match_cache.clear()
     word_index = {}
     latin_all_index = {}
     latin_file_index = {}
@@ -1041,7 +1050,7 @@ def metadata_prefilter(terms: list[str], search_paths: bool) -> set[int] | None:
             break
     return candidates
 
-def metadata_matches(q: str, exact: bool, search_paths: bool) -> set[int]:
+def _metadata_matches_uncached(q: str, exact: bool, search_paths: bool) -> set[int]:
     field = "_search_text" if search_paths else "_file_search_text"
     if exact:
         terms = query_terms(q)
@@ -1077,6 +1086,20 @@ def metadata_matches(q: str, exact: bool, search_paths: bool) -> set[int]:
         if all(term in records[idx][field] for term in terms)
     }
 
+
+def metadata_matches(q: str, exact: bool, search_paths: bool) -> set[int]:
+    key = (q, exact, search_paths)
+    with metadata_match_cache_lock:
+        cached = metadata_match_cache.get(key)
+        if cached is not None:
+            return set(cached)
+    result = _metadata_matches_uncached(q, exact, search_paths)
+    with metadata_match_cache_lock:
+        metadata_match_cache[key] = frozenset(result)
+        while len(metadata_match_cache) > METADATA_MATCH_CACHE_MAX_ENTRIES:
+            metadata_match_cache.pop(next(iter(metadata_match_cache)), None)
+    return set(result)
+
 def search(q="", sources_filter=None, folders=None, min_size=None, max_size=None, page=1, page_size=100, sort="relevance", exact=False, search_paths=True, timings=None):
     started = time.perf_counter()
     q = q.strip()