Merge pull request #2844 from matrix-org/rav/evicted_metrics

montoring metrics for number of cache evictions

synapse/metrics/metric.py

@@ -193,7 +193,9 @@ class DistributionMetric(object):
 
 
 class CacheMetric(object):
-    __slots__ = ("name", "cache_name", "hits", "misses", "size_callback")
+    __slots__ = (
+        "name", "cache_name", "hits", "misses", "evicted_size", "size_callback",
+    )
 
     def __init__(self, name, size_callback, cache_name):
         self.name = name
@@ -201,6 +203,7 @@ class CacheMetric(object):
 
         self.hits = 0
         self.misses = 0
+        self.evicted_size = 0
 
         self.size_callback = size_callback
 
@@ -210,6 +213,9 @@ class CacheMetric(object):
     def inc_misses(self):
         self.misses += 1
 
+    def inc_evictions(self, size=1):
+        self.evicted_size += size
+
     def render(self):
         size = self.size_callback()
         hits = self.hits
@@ -219,6 +225,9 @@ class CacheMetric(object):
             """%s:hits{name="%s"} %d""" % (self.name, self.cache_name, hits),
             """%s:total{name="%s"} %d""" % (self.name, self.cache_name, total),
             """%s:size{name="%s"} %d""" % (self.name, self.cache_name, size),
+            """%s:evicted_size{name="%s"} %d""" % (
+                self.name, self.cache_name, self.evicted_size
+            ),
         ]
 
 

synapse/util/caches/descriptors.py

@@ -75,6 +75,7 @@ class Cache(object):
         self.cache = LruCache(
             max_size=max_entries, keylen=keylen, cache_type=cache_type,
             size_callback=(lambda d: len(d)) if iterable else None,
+            evicted_callback=self._on_evicted,
         )
 
         self.name = name
@@ -83,6 +84,9 @@ class Cache(object):
         self.thread = None
         self.metrics = register_cache(name, self.cache)
 
+    def _on_evicted(self, evicted_count):
+        self.metrics.inc_evictions(evicted_count)
+
     def check_thread(self):
         expected_thread = self.thread
         if expected_thread is None:

synapse/util/caches/expiringcache.py

@@ -79,7 +79,11 @@ class ExpiringCache(object):
         while self._max_len and len(self) > self._max_len:
             _key, value = self._cache.popitem(last=False)
             if self.iterable:
-                self._size_estimate -= len(value.value)
+                removed_len = len(value.value)
+                self.metrics.inc_evictions(removed_len)
+                self._size_estimate -= removed_len
+            else:
+                self.metrics.inc_evictions()
 
     def __getitem__(self, key):
         try:

synapse/util/caches/lrucache.py

@@ -49,7 +49,24 @@ class LruCache(object):
     Can also set callbacks on objects when getting/setting which are fired
     when that key gets invalidated/evicted.
     """
-    def __init__(self, max_size, keylen=1, cache_type=dict, size_callback=None):
+    def __init__(self, max_size, keylen=1, cache_type=dict, size_callback=None,
+                 evicted_callback=None):
+        """
+        Args:
+            max_size (int):
+
+            keylen (int):
+
+            cache_type (type):
+                type of underlying cache to be used. Typically one of dict
+                or TreeCache.
+
+            size_callback (func(V) -> int | None):
+
+            evicted_callback (func(int)|None):
+                if not None, called on eviction with the size of the evicted
+                entry
+        """
         cache = cache_type()
         self.cache = cache  # Used for introspection.
         list_root = _Node(None, None, None, None)
@@ -61,8 +78,10 @@ class LruCache(object):
         def evict():
             while cache_len() > max_size:
                 todelete = list_root.prev_node
-                delete_node(todelete)
+                evicted_len = delete_node(todelete)
                 cache.pop(todelete.key, None)
+                if evicted_callback:
+                    evicted_callback(evicted_len)
 
         def synchronized(f):
             @wraps(f)
@@ -111,12 +130,15 @@ class LruCache(object):
             prev_node.next_node = next_node
             next_node.prev_node = prev_node
 
+            deleted_len = 1
             if size_callback:
-                cached_cache_len[0] -= size_callback(node.value)
+                deleted_len = size_callback(node.value)
+                cached_cache_len[0] -= deleted_len
 
             for cb in node.callbacks:
                 cb()
             node.callbacks.clear()
+            return deleted_len
 
         @synchronized
         def cache_get(key, default=None, callbacks=[]):

tests/metrics/test_metric.py

@@ -141,6 +141,7 @@ class CacheMetricTestCase(unittest.TestCase):
             'cache:hits{name="cache_name"} 0',
             'cache:total{name="cache_name"} 0',
             'cache:size{name="cache_name"} 0',
+            'cache:evicted_size{name="cache_name"} 0',
         ])
 
         metric.inc_misses()
@@ -150,6 +151,7 @@ class CacheMetricTestCase(unittest.TestCase):
             'cache:hits{name="cache_name"} 0',
             'cache:total{name="cache_name"} 1',
             'cache:size{name="cache_name"} 1',
+            'cache:evicted_size{name="cache_name"} 0',
         ])
 
         metric.inc_hits()
@@ -158,4 +160,14 @@ class CacheMetricTestCase(unittest.TestCase):
             'cache:hits{name="cache_name"} 1',
             'cache:total{name="cache_name"} 2',
             'cache:size{name="cache_name"} 1',
+            'cache:evicted_size{name="cache_name"} 0',
+        ])
+
+        metric.inc_evictions(2)
+
+        self.assertEquals(metric.render(), [
+            'cache:hits{name="cache_name"} 1',
+            'cache:total{name="cache_name"} 2',
+            'cache:size{name="cache_name"} 1',
+            'cache:evicted_size{name="cache_name"} 2',
         ])