Fix stack overflow in _PerHostRatelimiter due to synchronous reques…

…ts (#14812)

When there are many synchronous requests waiting on a
`_PerHostRatelimiter`, each request will be started recursively just
after the previous request has completed. Under the right conditions,
this leads to stack exhaustion.

A common way for requests to become synchronous is when the remote
client disconnects early, because the homeserver is overloaded and slow
to respond.

Avoid stack exhaustion under these conditions by deferring subsequent
requests until the next reactor tick.

Fixes #14480.

Signed-off-by: Sean Quah <seanq@matrix.org>

changelog.d/14812.bugfix

@@ -0,0 +1 @@
+Fix a long-standing bug where Synapse would exhaust the stack when processing many federation requests where the remote homeserver has disconencted early.

synapse/rest/client/register.py

@@ -310,6 +310,7 @@ def __init__(self, hs: "HomeServer"):
         self.hs = hs
         self.registration_handler = hs.get_registration_handler()
         self.ratelimiter = FederationRateLimiter(
+            hs.get_reactor(),
             hs.get_clock(),
             FederationRatelimitSettings(
                 # Time window of 2s

synapse/server.py

@@ -768,6 +768,7 @@ def get_replication_streams(self) -> Dict[str, Stream]:
     @cache_in_self
     def get_federation_ratelimiter(self) -> FederationRateLimiter:
         return FederationRateLimiter(
+            self.get_reactor(),
             self.get_clock(),
             config=self.config.ratelimiting.rc_federation,
             metrics_name="federation_servlets",

synapse/util/ratelimitutils.py

@@ -34,6 +34,7 @@
 from typing_extensions import ContextManager
 
 from twisted.internet import defer
+from twisted.internet.interfaces import IReactorTime
 
 from synapse.api.errors import LimitExceededError
 from synapse.config.ratelimiting import FederationRatelimitSettings
@@ -146,12 +147,14 @@ class FederationRateLimiter:
 
     def __init__(
         self,
+        reactor: IReactorTime,
         clock: Clock,
         config: FederationRatelimitSettings,
         metrics_name: Optional[str] = None,
     ):
         """
         Args:
+            reactor
             clock
             config
             metrics_name: The name of the rate limiter so we can differentiate it
@@ -163,7 +166,7 @@ def __init__(
 
         def new_limiter() -> "_PerHostRatelimiter":
             return _PerHostRatelimiter(
-                clock=clock, config=config, metrics_name=metrics_name
+                reactor=reactor, clock=clock, config=config, metrics_name=metrics_name
             )
 
         self.ratelimiters: DefaultDict[
@@ -194,19 +197,22 @@ def ratelimit(self, host: str) -> "_GeneratorContextManager[defer.Deferred[None]
 class _PerHostRatelimiter:
     def __init__(
         self,
+        reactor: IReactorTime,
         clock: Clock,
         config: FederationRatelimitSettings,
         metrics_name: Optional[str] = None,
     ):
         """
         Args:
+            reactor
             clock
             config
             metrics_name: The name of the rate limiter so we can differentiate it
                 from the rest in the metrics. If `None`, we don't track metrics
                 for this rate limiter.
                 from the rest in the metrics
         """
+        self.reactor = reactor
         self.clock = clock
         self.metrics_name = metrics_name
 
@@ -364,12 +370,22 @@ def on_both(r: object) -> object:
 
     def _on_exit(self, request_id: object) -> None:
         logger.debug("Ratelimit(%s) [%s]: Processed req", self.host, id(request_id))
-        self.current_processing.discard(request_id)
-        try:
-            # start processing the next item on the queue.
-            _, deferred = self.ready_request_queue.popitem(last=False)
 
-            with PreserveLoggingContext():
-                deferred.callback(None)
-        except KeyError:
-            pass
+        # When requests complete synchronously, we will recursively start the next
+        # request in the queue. To avoid stack exhaustion, we defer starting the next
+        # request until the next reactor tick.
+
+        def start_next_request() -> None:
+            # We only remove the completed request from the list when we're about to
+            # start the next one, otherwise we can allow extra requests through.
+            self.current_processing.discard(request_id)
+            try:
+                # start processing the next item on the queue.
+                _, deferred = self.ready_request_queue.popitem(last=False)
+
+                with PreserveLoggingContext():
+                    deferred.callback(None)
+            except KeyError:
+                pass
+
+        self.reactor.callLater(0.0, start_next_request)

tests/util/test_ratelimitutils.py

@@ -13,6 +13,7 @@
 # limitations under the License.
 from typing import Optional
 
+from twisted.internet import defer
 from twisted.internet.defer import Deferred
 
 from synapse.config.homeserver import HomeServerConfig
@@ -29,7 +30,7 @@ def test_ratelimit(self) -> None:
         """A simple test with the default values"""
         reactor, clock = get_clock()
         rc_config = build_rc_config()
-        ratelimiter = FederationRateLimiter(clock, rc_config)
+        ratelimiter = FederationRateLimiter(reactor, clock, rc_config)
 
         with ratelimiter.ratelimit("testhost") as d1:
             # shouldn't block
@@ -39,7 +40,7 @@ def test_concurrent_limit(self) -> None:
         """Test what happens when we hit the concurrent limit"""
         reactor, clock = get_clock()
         rc_config = build_rc_config({"rc_federation": {"concurrent": 2}})
-        ratelimiter = FederationRateLimiter(clock, rc_config)
+        ratelimiter = FederationRateLimiter(reactor, clock, rc_config)
 
         with ratelimiter.ratelimit("testhost") as d1:
             # shouldn't block
@@ -57,6 +58,7 @@ def test_concurrent_limit(self) -> None:
 
             # ... until we complete an earlier request
             cm2.__exit__(None, None, None)
+            reactor.advance(0.0)
             self.successResultOf(d3)
 
     def test_sleep_limit(self) -> None:
@@ -65,7 +67,7 @@ def test_sleep_limit(self) -> None:
         rc_config = build_rc_config(
             {"rc_federation": {"sleep_limit": 2, "sleep_delay": 500}}
         )
-        ratelimiter = FederationRateLimiter(clock, rc_config)
+        ratelimiter = FederationRateLimiter(reactor, clock, rc_config)
 
         with ratelimiter.ratelimit("testhost") as d1:
             # shouldn't block
@@ -81,6 +83,43 @@ def test_sleep_limit(self) -> None:
             sleep_time = _await_resolution(reactor, d3)
             self.assertAlmostEqual(sleep_time, 500, places=3)
 
+    def test_lots_of_queued_things(self) -> None:
+        """Tests lots of synchronous things queued up behind a slow thing.
+
+        The stack should *not* explode when the slow thing completes.
+        """
+        reactor, clock = get_clock()
+        rc_config = build_rc_config(
+            {
+                "rc_federation": {
+                    "sleep_limit": 1000000000,  # never sleep
+                    "reject_limit": 1000000000,  # never reject requests
+                    "concurrent": 1,
+                }
+            }
+        )
+        ratelimiter = FederationRateLimiter(reactor, clock, rc_config)
+
+        with ratelimiter.ratelimit("testhost") as d:
+            # shouldn't block
+            self.successResultOf(d)
+
+            async def task() -> None:
+                with ratelimiter.ratelimit("testhost") as d:
+                    await d
+
+            for _ in range(1, 100):
+                defer.ensureDeferred(task())
+
+            last_task = defer.ensureDeferred(task())
+
+            # Upon exiting the context manager, all the synchronous things will resume.
+            # If a stack overflow occurs, the final task will not complete.
+
+        # Wait for all the things to complete.
+        reactor.advance(0.0)
+        self.successResultOf(last_task)
+
 
 def _await_resolution(reactor: ThreadedMemoryReactorClock, d: Deferred) -> float:
     """advance the clock until the deferred completes.