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

4 files changed, 81 insertions, 21 deletions

diff --git a/changelog.d/13717.misc b/changelog.d/13717.misc
new file mode 100644
index 0000000000..07ace50b12
--- /dev/null
+++ b/changelog.d/13717.misc
@@ -0,0 +1 @@
+Add experimental configuration option to allow disabling legacy Prometheus metric names.
\ No newline at end of file
diff --git a/synapse/metrics/_legacy_exposition.py b/synapse/metrics/_legacy_exposition.py
index 133f1603dd..563d8cc2c6 100644
--- a/synapse/metrics/_legacy_exposition.py
+++ b/synapse/metrics/_legacy_exposition.py
@@ -34,8 +34,6 @@ from prometheus_client.core import Sample
 from twisted.web.resource import Resource
 from twisted.web.server import Request
 
-from synapse.util import caches
-
 CONTENT_TYPE_LATEST = "text/plain; version=0.0.4; charset=utf-8"
 
 
@@ -107,11 +105,6 @@ def generate_latest(registry: CollectorRegistry, emit_help: bool = False) -> byt
     by prometheus-client.
     """
 
-    # Trigger the cache metrics to be rescraped, which updates the common
-    # metrics but do not produce metrics themselves
-    for collector in caches.collectors_by_name.values():
-        collector.collect()
-
     output = []
 
     for metric in registry.collect():
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
 
 
diff --git a/synapse/util/metrics.py b/synapse/util/metrics.py
index bc3b4938ea..9687120ebf 100644
--- a/synapse/util/metrics.py
+++ b/synapse/util/metrics.py
@@ -15,9 +15,9 @@
 import logging
 from functools import wraps
 from types import TracebackType
-from typing import Awaitable, Callable, Optional, Type, TypeVar
+from typing import Awaitable, Callable, Generator, List, Optional, Type, TypeVar
 
-from prometheus_client import Counter
+from prometheus_client import CollectorRegistry, Counter, Metric
 from typing_extensions import Concatenate, ParamSpec, Protocol
 
 from synapse.logging.context import (
@@ -208,3 +208,33 @@ class Measure:
         metrics.real_time_sum += duration
 
         # TODO: Add other in flight metrics.
+
+
+class DynamicCollectorRegistry(CollectorRegistry):
+    """
+    Custom Prometheus Collector registry that calls a hook first, allowing you
+    to update metrics on-demand.
+
+    Don't forget to register this registry with the main registry!
+    """
+
+    def __init__(self) -> None:
+        super().__init__()
+        self._pre_update_hooks: List[Callable[[], None]] = []
+
+    def collect(self) -> Generator[Metric, None, None]:
+        """
+        Collects metrics, calling pre-update hooks first.
+        """
+
+        for pre_update_hook in self._pre_update_hooks:
+            pre_update_hook()
+
+        yield from super().collect()
+
+    def register_hook(self, hook: Callable[[], None]) -> None:
+        """
+        Registers a hook that is called before metric collection.
+        """
+
+        self._pre_update_hooks.append(hook)