Merge pull request #3722 from matrix-org/erikj/bg_process_iteration

LaterGauge needs to call thread safe functions

changelog.d/3722.bugfix

@@ -0,0 +1 @@
+Fix error collecting prometheus metrics when run on dedicated thread due to threading concurrency issues

synapse/http/request_metrics.py

@@ -15,6 +15,7 @@
 # limitations under the License.
 
 import logging
+import threading
 
 from prometheus_client.core import Counter, Histogram
 
@@ -111,6 +112,9 @@ in_flight_requests_db_sched_duration = Counter(
 # The set of all in flight requests, set[RequestMetrics]
 _in_flight_requests = set()
 
+# Protects the _in_flight_requests set from concurrent accesss
+_in_flight_requests_lock = threading.Lock()
+
 
 def _get_in_flight_counts():
     """Returns a count of all in flight requests by (method, server_name)
@@ -120,7 +124,8 @@ def _get_in_flight_counts():
     """
     # Cast to a list to prevent it changing while the Prometheus
     # thread is collecting metrics
-    reqs = list(_in_flight_requests)
+    with _in_flight_requests_lock:
+        reqs = list(_in_flight_requests)
 
     for rm in reqs:
         rm.update_metrics()
@@ -154,10 +159,12 @@ class RequestMetrics(object):
         # to the "in flight" metrics.
         self._request_stats = self.start_context.get_resource_usage()
 
-        _in_flight_requests.add(self)
+        with _in_flight_requests_lock:
+            _in_flight_requests.add(self)
 
     def stop(self, time_sec, request):
-        _in_flight_requests.discard(self)
+        with _in_flight_requests_lock:
+            _in_flight_requests.discard(self)
 
         context = LoggingContext.current_context()
 

synapse/metrics/background_process_metrics.py

@@ -13,6 +13,8 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 
+import threading
+
 import six
 
 from prometheus_client.core import REGISTRY, Counter, GaugeMetricFamily
@@ -78,6 +80,9 @@ _background_process_counts = dict()  # type: dict[str, int]
 # of process descriptions that no longer have any active processes.
 _background_processes = dict()  # type: dict[str, set[_BackgroundProcess]]
 
+# A lock that covers the above dicts
+_bg_metrics_lock = threading.Lock()
+
 
 class _Collector(object):
     """A custom metrics collector for the background process metrics.
@@ -92,7 +97,11 @@ class _Collector(object):
             labels=["name"],
         )
 
-        for desc, processes in six.iteritems(_background_processes):
+        # We copy the dict so that it doesn't change from underneath us
+        with _bg_metrics_lock:
+            _background_processes_copy = dict(_background_processes)
+
+        for desc, processes in six.iteritems(_background_processes_copy):
             background_process_in_flight_count.add_metric(
                 (desc,), len(processes),
             )
@@ -167,19 +176,26 @@ def run_as_background_process(desc, func, *args, **kwargs):
     """
     @defer.inlineCallbacks
     def run():
-        count = _background_process_counts.get(desc, 0)
+        with _bg_metrics_lock:
-        _background_process_counts[desc] = count + 1
+            count = _background_process_counts.get(desc, 0)
+            _background_process_counts[desc] = count + 1
+
         _background_process_start_count.labels(desc).inc()
 
         with LoggingContext(desc) as context:
             context.request = "%s-%i" % (desc, count)
             proc = _BackgroundProcess(desc, context)
-            _background_processes.setdefault(desc, set()).add(proc)
+
+            with _bg_metrics_lock:
+                _background_processes.setdefault(desc, set()).add(proc)
+
             try:
                 yield func(*args, **kwargs)
             finally:
                 proc.update_metrics()
-                _background_processes[desc].remove(proc)
+
+                with _bg_metrics_lock:
+                    _background_processes[desc].remove(proc)
 
     with PreserveLoggingContext():
         return run()