Merge pull request #3257 from lonvia/slow-queries

Performance tweaks for search

nominatim/api/logging.py

@@ -235,6 +235,10 @@ def __init__(self) -> None:
         self.buffer = io.StringIO()
 
 
+    def _timestamp(self) -> None:
+        self._write(f'[{dt.datetime.now()}]\n')
+
+
     def get_buffer(self) -> str:
         return self.buffer.getvalue()
 
@@ -247,6 +251,7 @@ def function(self, func: str, **kwargs: Any) -> None:
 
 
     def section(self, heading: str) -> None:
+        self._timestamp()
         self._write(f"\n# {heading}\n\n")
 
 
@@ -283,6 +288,7 @@ def table_dump(self, heading: str, rows: Iterator[Optional[List[Any]]]) -> None:
 
 
     def result_dump(self, heading: str, results: Iterator[Tuple[Any, Any]]) -> None:
+        self._timestamp()
         self._write(f'{heading}:\n')
         total = 0
         for rank, res in results:
@@ -298,6 +304,7 @@ def result_dump(self, heading: str, results: Iterator[Tuple[Any, Any]]) -> None:
 
     def sql(self, conn: AsyncConnection, statement: 'sa.Executable',
             params: Union[Mapping[str, Any], Sequence[Mapping[str, Any]], None]) -> None:
+        self._timestamp()
         sqlstr = '\n| '.join(textwrap.wrap(self.format_sql(conn, statement, params), width=78))
         self._write(f"| {sqlstr}\n\n")
 

nominatim/api/search/db_search_builder.py

@@ -208,7 +208,7 @@ def yield_lookups(self, name: TokenRange, address: List[TokenRange])\
                            and all(t.is_indexed for t in addr_partials)
         exp_count = min(t.count for t in name_partials) / (2**(len(name_partials) - 1))
 
-        if (len(name_partials) > 3 or exp_count < 3000) and partials_indexed:
+        if (len(name_partials) > 3 or exp_count < 8000) and partials_indexed:
             yield penalty, exp_count, dbf.lookup_by_names(name_tokens, addr_tokens)
             return
 

nominatim/api/search/db_searches.py

@@ -24,6 +24,13 @@
 #pylint: disable=singleton-comparison,not-callable
 #pylint: disable=too-many-branches,too-many-arguments,too-many-locals,too-many-statements
 
+def no_index(expr: SaColumn) -> SaColumn:
+    """ Wrap the given expression, so that the query planner will
+        refrain from using the expression for index lookup.
+    """
+    return sa.func.coalesce(sa.null(), expr) # pylint: disable=not-callable
+
+
 def _details_to_bind_params(details: SearchDetails) -> Dict[str, Any]:
     """ Create a dictionary from search parameters that can be used
         as bind parameter for SQL execute.
@@ -107,14 +114,14 @@ def _make_interpolation_subquery(table: SaFromClause, inner: SaFromClause,
 def _filter_by_layer(table: SaFromClause, layers: DataLayer) -> SaColumn:
     orexpr: List[SaExpression] = []
     if layers & DataLayer.ADDRESS and layers & DataLayer.POI:
-        orexpr.append(table.c.rank_address.between(1, 30))
+        orexpr.append(no_index(table.c.rank_address).between(1, 30))
     elif layers & DataLayer.ADDRESS:
-        orexpr.append(table.c.rank_address.between(1, 29))
-        orexpr.append(sa.and_(table.c.rank_address == 30,
+        orexpr.append(no_index(table.c.rank_address).between(1, 29))
+        orexpr.append(sa.and_(no_index(table.c.rank_address) == 30,
                               sa.or_(table.c.housenumber != None,
                                      table.c.address.has_key('addr:housename'))))
     elif layers & DataLayer.POI:
-        orexpr.append(sa.and_(table.c.rank_address == 30,
+        orexpr.append(sa.and_(no_index(table.c.rank_address) == 30,
                               table.c.class_.not_in(('place', 'building'))))
 
     if layers & DataLayer.MANMADE:
@@ -124,15 +131,15 @@ def _filter_by_layer(table: SaFromClause, layers: DataLayer) -> SaColumn:
         if not layers & DataLayer.NATURAL:
             exclude.extend(('natural', 'water', 'waterway'))
         orexpr.append(sa.and_(table.c.class_.not_in(tuple(exclude)),
-                              table.c.rank_address == 0))
+                              no_index(table.c.rank_address) == 0))
     else:
         include = []
         if layers & DataLayer.RAILWAY:
             include.append('railway')
         if layers & DataLayer.NATURAL:
             include.extend(('natural', 'water', 'waterway'))
         orexpr.append(sa.and_(table.c.class_.in_(tuple(include)),
-                              table.c.rank_address == 0))
+                              no_index(table.c.rank_address) == 0))
 
     if len(orexpr) == 1:
         return orexpr[0]
@@ -295,7 +302,7 @@ async def lookup_category(self, results: nres.SearchResults,
                                        else_ = tgeom.c.centroid.ST_Expand(0.05))))\
                      .order_by(tgeom.c.centroid.ST_Distance(table.c.centroid))
 
-        sql = sql.where(t.c.rank_address.between(MIN_RANK_PARAM, MAX_RANK_PARAM))
+        sql = sql.where(no_index(t.c.rank_address).between(MIN_RANK_PARAM, MAX_RANK_PARAM))
         if details.countries:
             sql = sql.where(t.c.country_code.in_(COUNTRIES_PARAM))
         if details.excluded:

nominatim/api/search/icu_tokenizer.py

@@ -101,10 +101,16 @@ def from_db_row(row: SaRow) -> 'ICUToken':
         penalty = 0.0
         if row.type == 'w':
             penalty = 0.3
+        elif row.type == 'W':
+            if len(row.word_token) == 1 and row.word_token == row.word:
+                penalty = 0.2 if row.word.isdigit() else 0.3
         elif row.type == 'H':
             penalty = sum(0.1 for c in row.word_token if c != ' ' and not c.isdigit())
             if all(not c.isdigit() for c in row.word_token):
                 penalty += 0.2 * (len(row.word_token) - 1)
+        elif row.type == 'C':
+            if len(row.word_token) == 1:
+                penalty = 0.3
 
         if row.info is None:
             lookup_word = row.word

nominatim/api/types.py

@@ -538,7 +538,9 @@ def is_impossible(self) -> bool:
                 or (self.bounded_viewbox
                     and self.viewbox is not None and self.near is not None
                     and self.viewbox.contains(self.near))
-                or self.layers is not None and not self.layers)
+                or (self.layers is not None and not self.layers)
+                or (self.max_rank <= 4 and
+                    self.layers is not None and not self.layers & DataLayer.ADDRESS))
 
 
     def layer_enabled(self, layer: DataLayer) -> bool: