Merge pull request #2079 from matrix-org/erikj/push_regex_cache

Cache glob to regex at a higher level for push

synapse/push/push_rule_evaluator.py

@@ -17,6 +17,7 @@ import logging
 import re
 
 from synapse.types import UserID
+from synapse.util.caches import CACHE_SIZE_FACTOR, register_cache
 from synapse.util.caches.lrucache import LruCache
 
 logger = logging.getLogger(__name__)
@@ -125,6 +126,11 @@ class PushRuleEvaluatorForEvent(object):
         return self._value_cache.get(dotted_key, None)
 
 
+# Caches (glob, word_boundary) -> regex for push. See _glob_matches
+regex_cache = LruCache(50000 * CACHE_SIZE_FACTOR)
+register_cache("regex_push_cache", regex_cache)
+
+
 def _glob_matches(glob, value, word_boundary=False):
     """Tests if value matches glob.
 
@@ -137,46 +143,63 @@ def _glob_matches(glob, value, word_boundary=False):
     Returns:
         bool
     """
+
     try:
-        if IS_GLOB.search(glob):
-            r = re.escape(glob)
-
-            r = r.replace(r'\*', '.*?')
-            r = r.replace(r'\?', '.')
-
-            # handle [abc], [a-z] and [!a-z] style ranges.
-            r = GLOB_REGEX.sub(
-                lambda x: (
-                    '[%s%s]' % (
-                        x.group(1) and '^' or '',
-                        x.group(2).replace(r'\\\-', '-')
-                    )
-                ),
-                r,
-            )
-            if word_boundary:
-                r = r"\b%s\b" % (r,)
-                r = _compile_regex(r)
-
-                return r.search(value)
-            else:
-                r = r + "$"
-                r = _compile_regex(r)
-
-                return r.match(value)
-        elif word_boundary:
-            r = re.escape(glob)
-            r = r"\b%s\b" % (r,)
-            r = _compile_regex(r)
-
-            return r.search(value)
-        else:
-            return value.lower() == glob.lower()
+        r = regex_cache.get((glob, word_boundary), None)
+        if not r:
+            r = _glob_to_re(glob, word_boundary)
+            regex_cache[(glob, word_boundary)] = r
+        return r.search(value)
     except re.error:
         logger.warn("Failed to parse glob to regex: %r", glob)
         return False
 
 
+def _glob_to_re(glob, word_boundary):
+    """Generates regex for a given glob.
+
+    Args:
+        glob (string)
+        word_boundary (bool): Whether to match against word boundaries or entire
+            string. Defaults to False.
+
+    Returns:
+        regex object
+    """
+    if IS_GLOB.search(glob):
+        r = re.escape(glob)
+
+        r = r.replace(r'\*', '.*?')
+        r = r.replace(r'\?', '.')
+
+        # handle [abc], [a-z] and [!a-z] style ranges.
+        r = GLOB_REGEX.sub(
+            lambda x: (
+                '[%s%s]' % (
+                    x.group(1) and '^' or '',
+                    x.group(2).replace(r'\\\-', '-')
+                )
+            ),
+            r,
+        )
+        if word_boundary:
+            r = r"\b%s\b" % (r,)
+
+            return re.compile(r, flags=re.IGNORECASE)
+        else:
+            r = "^" + r + "$"
+
+            return re.compile(r, flags=re.IGNORECASE)
+    elif word_boundary:
+        r = re.escape(glob)
+        r = r"\b%s\b" % (r,)
+
+        return re.compile(r, flags=re.IGNORECASE)
+    else:
+        r = "^" + re.escape(glob) + "$"
+        return re.compile(r, flags=re.IGNORECASE)
+
+
 def _flatten_dict(d, prefix=[], result={}):
     for key, value in d.items():
         if isinstance(value, basestring):
@@ -185,16 +208,3 @@ def _flatten_dict(d, prefix=[], result={}):
             _flatten_dict(value, prefix=(prefix + [key]), result=result)
 
     return result
-
-
-regex_cache = LruCache(5000)
-
-
-def _compile_regex(regex_str):
-    r = regex_cache.get(regex_str, None)
-    if r:
-        return r
-
-    r = re.compile(regex_str, flags=re.IGNORECASE)
-    regex_cache[regex_str] = r
-    return r