Fix cache metrics not being updated when not using the legacy exposition module. (#13717)

1 files changed, 48 insertions, 12 deletions

diff --git a/synapse/util/caches/__init__.py b/synapse/util/caches/__init__.py
index d4a2b77c29..35c0be08b0 100644
--- a/synapse/util/caches/__init__.py
+++ b/synapse/util/caches/__init__.py
@@ -20,9 +20,11 @@ from sys import intern
 from typing import Any, Callable, Dict, List, Optional, Sized, TypeVar
 
 import attr
+from prometheus_client import REGISTRY
 from prometheus_client.core import Gauge
 
 from synapse.config.cache import add_resizable_cache
+from synapse.util.metrics import DynamicCollectorRegistry
 
 logger = logging.getLogger(__name__)
 
@@ -30,27 +32,62 @@ logger = logging.getLogger(__name__)
 # Whether to track estimated memory usage of the LruCaches.
 TRACK_MEMORY_USAGE = False
 
+# We track cache metrics in a special registry that lets us update the metrics
+# just before they are returned from the scrape endpoint.
+CACHE_METRIC_REGISTRY = DynamicCollectorRegistry()
 
 caches_by_name: Dict[str, Sized] = {}
-collectors_by_name: Dict[str, "CacheMetric"] = {}
 
-cache_size = Gauge("synapse_util_caches_cache_size", "", ["name"])
-cache_hits = Gauge("synapse_util_caches_cache_hits", "", ["name"])
-cache_evicted = Gauge("synapse_util_caches_cache_evicted_size", "", ["name", "reason"])
-cache_total = Gauge("synapse_util_caches_cache", "", ["name"])
-cache_max_size = Gauge("synapse_util_caches_cache_max_size", "", ["name"])
+cache_size = Gauge(
+    "synapse_util_caches_cache_size", "", ["name"], registry=CACHE_METRIC_REGISTRY
+)
+cache_hits = Gauge(
+    "synapse_util_caches_cache_hits", "", ["name"], registry=CACHE_METRIC_REGISTRY
+)
+cache_evicted = Gauge(
+    "synapse_util_caches_cache_evicted_size",
+    "",
+    ["name", "reason"],
+    registry=CACHE_METRIC_REGISTRY,
+)
+cache_total = Gauge(
+    "synapse_util_caches_cache", "", ["name"], registry=CACHE_METRIC_REGISTRY
+)
+cache_max_size = Gauge(
+    "synapse_util_caches_cache_max_size", "", ["name"], registry=CACHE_METRIC_REGISTRY
+)
 cache_memory_usage = Gauge(
     "synapse_util_caches_cache_size_bytes",
     "Estimated memory usage of the caches",
     ["name"],
+    registry=CACHE_METRIC_REGISTRY,
 )
 
-response_cache_size = Gauge("synapse_util_caches_response_cache_size", "", ["name"])
-response_cache_hits = Gauge("synapse_util_caches_response_cache_hits", "", ["name"])
+response_cache_size = Gauge(
+    "synapse_util_caches_response_cache_size",
+    "",
+    ["name"],
+    registry=CACHE_METRIC_REGISTRY,
+)
+response_cache_hits = Gauge(
+    "synapse_util_caches_response_cache_hits",
+    "",
+    ["name"],
+    registry=CACHE_METRIC_REGISTRY,
+)
 response_cache_evicted = Gauge(
-    "synapse_util_caches_response_cache_evicted_size", "", ["name", "reason"]
+    "synapse_util_caches_response_cache_evicted_size",
+    "",
+    ["name", "reason"],
+    registry=CACHE_METRIC_REGISTRY,
 )
-response_cache_total = Gauge("synapse_util_caches_response_cache", "", ["name"])
+response_cache_total = Gauge(
+    "synapse_util_caches_response_cache", "", ["name"], registry=CACHE_METRIC_REGISTRY
+)
+
+
+# Register our custom cache metrics registry with the global registry
+REGISTRY.register(CACHE_METRIC_REGISTRY)
 
 
 class EvictionReason(Enum):
@@ -168,9 +205,8 @@ def register_cache(
         add_resizable_cache(cache_name, resize_callback)
 
     metric = CacheMetric(cache, cache_type, cache_name, collect_callback)
-    metric_name = "cache_%s_%s" % (cache_type, cache_name)
     caches_by_name[cache_name] = cache
-    collectors_by_name[metric_name] = metric
+    CACHE_METRIC_REGISTRY.register_hook(metric.collect)
     return metric