Merge pull request #4008 from matrix-org/erikj/log_looping_exceptions

Log looping call exceptions

3 files changed, 36 insertions, 13 deletions

diff --git a/changelog.d/4008.misc b/changelog.d/4008.misc
new file mode 100644
index 0000000000..5730210054
--- /dev/null
+++ b/changelog.d/4008.misc
@@ -0,0 +1 @@
+Log exceptions in looping calls
diff --git a/synapse/handlers/appservice.py b/synapse/handlers/appservice.py
index f0f89af7dc..17eedf4dbf 100644
--- a/synapse/handlers/appservice.py
+++ b/synapse/handlers/appservice.py
@@ -28,6 +28,7 @@ from synapse.metrics import (
     event_processing_loop_room_count,
 )
 from synapse.metrics.background_process_metrics import run_as_background_process
+from synapse.util import log_failure
 from synapse.util.logcontext import make_deferred_yieldable, run_in_background
 from synapse.util.metrics import Measure
 
@@ -36,17 +37,6 @@ logger = logging.getLogger(__name__)
 events_processed_counter = Counter("synapse_handlers_appservice_events_processed", "")
 
 
-def log_failure(failure):
-    logger.error(
-        "Application Services Failure",
-        exc_info=(
-            failure.type,
-            failure.value,
-            failure.getTracebackObject()
-        )
-    )
-
-
 class ApplicationServicesHandler(object):
 
     def __init__(self, hs):
@@ -112,7 +102,10 @@ class ApplicationServicesHandler(object):
 
                         if not self.started_scheduler:
                             def start_scheduler():
-                                return self.scheduler.start().addErrback(log_failure)
+                                return self.scheduler.start().addErrback(
+                                    log_failure, "Application Services Failure",
+                                )
+
                             run_as_background_process("as_scheduler", start_scheduler)
                             self.started_scheduler = True
 
diff --git a/synapse/util/__init__.py b/synapse/util/__init__.py
index 680ea928c7..9a8fae0497 100644
--- a/synapse/util/__init__.py
+++ b/synapse/util/__init__.py
@@ -68,7 +68,10 @@ class Clock(object):
         """
         call = task.LoopingCall(f)
         call.clock = self._reactor
-        call.start(msec / 1000.0, now=False)
+        d = call.start(msec / 1000.0, now=False)
+        d.addErrback(
+            log_failure, "Looping call died", consumeErrors=False,
+        )
         return call
 
     def call_later(self, delay, callback, *args, **kwargs):
@@ -109,3 +112,29 @@ def batch_iter(iterable, size):
     sourceiter = iter(iterable)
     # call islice until it returns an empty tuple
     return iter(lambda: tuple(islice(sourceiter, size)), ())
+
+
+def log_failure(failure, msg, consumeErrors=True):
+    """Creates a function suitable for passing to `Deferred.addErrback` that
+    logs any failures that occur.
+
+    Args:
+        msg (str): Message to log
+        consumeErrors (bool): If true consumes the failure, otherwise passes
+            on down the callback chain
+
+    Returns:
+        func(Failure)
+    """
+
+    logger.error(
+        msg,
+        exc_info=(
+            failure.type,
+            failure.value,
+            failure.getTracebackObject()
+        )
+    )
+
+    if not consumeErrors:
+        return failure