From 0dbbe33a65c17cdb1ad41d6109b5629029dce886 Mon Sep 17 00:00:00 2001
From: Erik Johnston <erik@matrix.org>
Date: Tue, 15 Feb 2022 14:31:04 +0000
Subject: [PATCH] Track cache invalidations (#12000)

Currently we only track evictions due to size or time constraints.
---
 changelog.d/12000.feature            | 1 +
 synapse/util/caches/__init__.py      | 1 +
 synapse/util/caches/expiringcache.py | 5 +++++
 synapse/util/caches/lrucache.py      | 4 +++-
 4 files changed, 10 insertions(+), 1 deletion(-)
 create mode 100644 changelog.d/12000.feature

diff --git a/changelog.d/12000.feature b/changelog.d/12000.feature
new file mode 100644
index 000000000..246cc87f0
--- /dev/null
+++ b/changelog.d/12000.feature
@@ -0,0 +1 @@
+Track cache invalidations in Prometheus metrics, as already happens for cache eviction based on size or time.
diff --git a/synapse/util/caches/__init__.py b/synapse/util/caches/__init__.py
index 15debd6c4..1cbc180ed 100644
--- a/synapse/util/caches/__init__.py
+++ b/synapse/util/caches/__init__.py
@@ -56,6 +56,7 @@ response_cache_total = Gauge("synapse_util_caches_response_cache:total", "", ["n
 class EvictionReason(Enum):
     size = auto()
     time = auto()
+    invalidation = auto()
 
 
 @attr.s(slots=True, auto_attribs=True)
diff --git a/synapse/util/caches/expiringcache.py b/synapse/util/caches/expiringcache.py
index 67ee4c693..c6a5d0dfc 100644
--- a/synapse/util/caches/expiringcache.py
+++ b/synapse/util/caches/expiringcache.py
@@ -133,6 +133,11 @@ class ExpiringCache(Generic[KT, VT]):
                 raise KeyError(key)
             return default
 
+        if self.iterable:
+            self.metrics.inc_evictions(EvictionReason.invalidation, len(value.value))
+        else:
+            self.metrics.inc_evictions(EvictionReason.invalidation)
+
         return value.value
 
     def __contains__(self, key: KT) -> bool:
diff --git a/synapse/util/caches/lrucache.py b/synapse/util/caches/lrucache.py
index 7548b3854..45ff0de63 100644
--- a/synapse/util/caches/lrucache.py
+++ b/synapse/util/caches/lrucache.py
@@ -560,8 +560,10 @@ class LruCache(Generic[KT, VT]):
         def cache_pop(key: KT, default: Optional[T] = None) -> Union[None, T, VT]:
             node = cache.get(key, None)
             if node:
-                delete_node(node)
+                evicted_len = delete_node(node)
                 cache.pop(node.key, None)
+                if metrics:
+                    metrics.inc_evictions(EvictionReason.invalidation, evicted_len)
                 return node.value
             else:
                 return default