Add metrics to track how the rate limiter is affecting requests (sleep/reject)

changelog.d/13534.misc

@@ -0,0 +1 @@
+Add metrics to track how the rate limiter is affecting requests (sleep/reject).

synapse/util/ratelimitutils.py

@@ -18,6 +18,8 @@
 import typing
 from typing import Any, DefaultDict, Iterator, List, Set
 
+from prometheus_client.core import Counter
+
 from twisted.internet import defer
 
 from synapse.api.errors import LimitExceededError
@@ -35,6 +37,11 @@
 logger = logging.getLogger(__name__)
 
 
+rate_limit_sleep_counter = Counter("synapse_rate_limit_sleep", "", ["host"])
+
+rate_limit_reject_counter = Counter("synapse_rate_limit_reject", "", ["host"])
+
+
 class FederationRateLimiter:
     def __init__(self, clock: Clock, config: FederationRatelimitSettings):
         def new_limiter() -> "_PerHostRatelimiter":
@@ -59,7 +66,7 @@ def ratelimit(self, host: str) -> "_GeneratorContextManager[defer.Deferred[None]
         Returns:
             context manager which returns a deferred.
         """
-        return self.ratelimiters[host].ratelimit()
+        return self.ratelimiters[host].ratelimit(host)
 
 
 class _PerHostRatelimiter:
@@ -94,12 +101,14 @@ def __init__(self, clock: Clock, config: FederationRatelimitSettings):
         self.request_times: List[int] = []
 
     @contextlib.contextmanager
-    def ratelimit(self) -> "Iterator[defer.Deferred[None]]":
+    def ratelimit(self, host: str) -> "Iterator[defer.Deferred[None]]":
         # `contextlib.contextmanager` takes a generator and turns it into a
         # context manager. The generator should only yield once with a value
         # to be returned by manager.
         # Exceptions will be reraised at the yield.
 
+        self.host = host
+
         request_id = object()
         ret = self._on_enter(request_id)
         try:
@@ -119,6 +128,7 @@ def _on_enter(self, request_id: object) -> "defer.Deferred[None]":
         # sleeping or in the ready queue).
         queue_size = len(self.ready_request_queue) + len(self.sleeping_requests)
         if queue_size > self.reject_limit:
+            rate_limit_reject_counter.labels(self.host).inc()
             raise LimitExceededError(
                 retry_after_ms=int(self.window_size / self.sleep_limit)
             )
@@ -146,6 +156,7 @@ def queue_request() -> "defer.Deferred[None]":
 
         if len(self.request_times) > self.sleep_limit:
             logger.debug("Ratelimiter: sleeping request for %f sec", self.sleep_sec)
+            rate_limit_sleep_counter.labels(self.host).inc()
             ret_defer = run_in_background(self.clock.sleep, self.sleep_sec)
 
             self.sleeping_requests.add(request_id)