Merge pull request #2027 from matrix-org/rav/logcontext_leaks

A few fixes to logcontext things

11 files changed, 86 insertions, 62 deletions

diff --git a/synapse/app/appservice.py b/synapse/app/appservice.py
index 83ee3e3ce3..a6f1e7594e 100644
--- a/synapse/app/appservice.py
+++ b/synapse/app/appservice.py
@@ -29,7 +29,7 @@ from synapse.replication.slave.storage.registration import SlavedRegistrationSto
 from synapse.storage.engines import create_engine
 from synapse.util.async import sleep
 from synapse.util.httpresourcetree import create_resource_tree
-from synapse.util.logcontext import LoggingContext
+from synapse.util.logcontext import LoggingContext, PreserveLoggingContext
 from synapse.util.manhole import manhole
 from synapse.util.rlimit import change_resource_limit
 from synapse.util.versionstring import get_version_string
@@ -187,7 +187,11 @@ def start(config_options):
     ps.start_listening(config.worker_listeners)
 
     def run():
-        with LoggingContext("run"):
+        # make sure that we run the reactor with the sentinel log context,
+        # otherwise other PreserveLoggingContext instances will get confused
+        # and complain when they see the logcontext arbitrarily swapping
+        # between the sentinel and `run` logcontexts.
+        with PreserveLoggingContext():
             logger.info("Running")
             change_resource_limit(config.soft_file_limit)
             if config.gc_thresholds:
diff --git a/synapse/app/client_reader.py b/synapse/app/client_reader.py
index 7ed0de4117..a821a6ce62 100644
--- a/synapse/app/client_reader.py
+++ b/synapse/app/client_reader.py
@@ -35,7 +35,7 @@ from synapse.storage.client_ips import ClientIpStore
 from synapse.storage.engines import create_engine
 from synapse.util.async import sleep
 from synapse.util.httpresourcetree import create_resource_tree
-from synapse.util.logcontext import LoggingContext
+from synapse.util.logcontext import LoggingContext, PreserveLoggingContext
 from synapse.util.manhole import manhole
 from synapse.util.rlimit import change_resource_limit
 from synapse.util.versionstring import get_version_string
@@ -193,7 +193,11 @@ def start(config_options):
     ss.start_listening(config.worker_listeners)
 
     def run():
-        with LoggingContext("run"):
+        # make sure that we run the reactor with the sentinel log context,
+        # otherwise other PreserveLoggingContext instances will get confused
+        # and complain when they see the logcontext arbitrarily swapping
+        # between the sentinel and `run` logcontexts.
+        with PreserveLoggingContext():
             logger.info("Running")
             change_resource_limit(config.soft_file_limit)
             if config.gc_thresholds:
diff --git a/synapse/app/federation_reader.py b/synapse/app/federation_reader.py
index ca742de6b2..e52b0f240d 100644
--- a/synapse/app/federation_reader.py
+++ b/synapse/app/federation_reader.py
@@ -31,7 +31,7 @@ from synapse.server import HomeServer
 from synapse.storage.engines import create_engine
 from synapse.util.async import sleep
 from synapse.util.httpresourcetree import create_resource_tree
-from synapse.util.logcontext import LoggingContext
+from synapse.util.logcontext import LoggingContext, PreserveLoggingContext
 from synapse.util.manhole import manhole
 from synapse.util.rlimit import change_resource_limit
 from synapse.util.versionstring import get_version_string
@@ -184,7 +184,11 @@ def start(config_options):
     ss.start_listening(config.worker_listeners)
 
     def run():
-        with LoggingContext("run"):
+        # make sure that we run the reactor with the sentinel log context,
+        # otherwise other PreserveLoggingContext instances will get confused
+        # and complain when they see the logcontext arbitrarily swapping
+        # between the sentinel and `run` logcontexts.
+        with PreserveLoggingContext():
             logger.info("Running")
             change_resource_limit(config.soft_file_limit)
             if config.gc_thresholds:
diff --git a/synapse/app/federation_sender.py b/synapse/app/federation_sender.py
index 0cf5b196e6..76c4cc54d1 100644
--- a/synapse/app/federation_sender.py
+++ b/synapse/app/federation_sender.py
@@ -35,7 +35,7 @@ from synapse.storage.engines import create_engine
 from synapse.storage.presence import UserPresenceState
 from synapse.util.async import sleep
 from synapse.util.httpresourcetree import create_resource_tree
-from synapse.util.logcontext import LoggingContext
+from synapse.util.logcontext import LoggingContext, PreserveLoggingContext
 from synapse.util.manhole import manhole
 from synapse.util.rlimit import change_resource_limit
 from synapse.util.versionstring import get_version_string
@@ -193,7 +193,11 @@ def start(config_options):
     ps.start_listening(config.worker_listeners)
 
     def run():
-        with LoggingContext("run"):
+        # make sure that we run the reactor with the sentinel log context,
+        # otherwise other PreserveLoggingContext instances will get confused
+        # and complain when they see the logcontext arbitrarily swapping
+        # between the sentinel and `run` logcontexts.
+        with PreserveLoggingContext():
             logger.info("Running")
             change_resource_limit(config.soft_file_limit)
             if config.gc_thresholds:
diff --git a/synapse/app/homeserver.py b/synapse/app/homeserver.py
index 0b9d78c13c..2cdd2d39ff 100755
--- a/synapse/app/homeserver.py
+++ b/synapse/app/homeserver.py
@@ -52,7 +52,7 @@ from synapse.api.urls import (
 )
 from synapse.config.homeserver import HomeServerConfig
 from synapse.crypto import context_factory
-from synapse.util.logcontext import LoggingContext
+from synapse.util.logcontext import LoggingContext, PreserveLoggingContext
 from synapse.metrics import register_memory_metrics, get_metrics_for
 from synapse.metrics.resource import MetricsResource, METRICS_PREFIX
 from synapse.replication.resource import ReplicationResource, REPLICATION_PREFIX
@@ -456,7 +456,12 @@ def run(hs):
     def in_thread():
         # Uncomment to enable tracing of log context changes.
         # sys.settrace(logcontext_tracer)
-        with LoggingContext("run"):
+
+        # make sure that we run the reactor with the sentinel log context,
+        # otherwise other PreserveLoggingContext instances will get confused
+        # and complain when they see the logcontext arbitrarily swapping
+        # between the sentinel and `run` logcontexts.
+        with PreserveLoggingContext():
             change_resource_limit(hs.config.soft_file_limit)
             if hs.config.gc_thresholds:
                 gc.set_threshold(*hs.config.gc_thresholds)
diff --git a/synapse/app/media_repository.py b/synapse/app/media_repository.py
index c5579d9e38..3c7984a237 100644
--- a/synapse/app/media_repository.py
+++ b/synapse/app/media_repository.py
@@ -32,7 +32,7 @@ from synapse.storage.engines import create_engine
 from synapse.storage.media_repository import MediaRepositoryStore
 from synapse.util.async import sleep
 from synapse.util.httpresourcetree import create_resource_tree
-from synapse.util.logcontext import LoggingContext
+from synapse.util.logcontext import LoggingContext, PreserveLoggingContext
 from synapse.util.manhole import manhole
 from synapse.util.rlimit import change_resource_limit
 from synapse.util.versionstring import get_version_string
@@ -190,7 +190,11 @@ def start(config_options):
     ss.start_listening(config.worker_listeners)
 
     def run():
-        with LoggingContext("run"):
+        # make sure that we run the reactor with the sentinel log context,
+        # otherwise other PreserveLoggingContext instances will get confused
+        # and complain when they see the logcontext arbitrarily swapping
+        # between the sentinel and `run` logcontexts.
+        with PreserveLoggingContext():
             logger.info("Running")
             change_resource_limit(config.soft_file_limit)
             if config.gc_thresholds:
diff --git a/synapse/app/pusher.py b/synapse/app/pusher.py
index b025db54d4..ab682e52ec 100644
--- a/synapse/app/pusher.py
+++ b/synapse/app/pusher.py
@@ -31,7 +31,8 @@ from synapse.storage.engines import create_engine
 from synapse.storage import DataStore
 from synapse.util.async import sleep
 from synapse.util.httpresourcetree import create_resource_tree
-from synapse.util.logcontext import LoggingContext, preserve_fn
+from synapse.util.logcontext import LoggingContext, preserve_fn, \
+    PreserveLoggingContext
 from synapse.util.manhole import manhole
 from synapse.util.rlimit import change_resource_limit
 from synapse.util.versionstring import get_version_string
@@ -275,7 +276,11 @@ def start(config_options):
     ps.start_listening(config.worker_listeners)
 
     def run():
-        with LoggingContext("run"):
+        # make sure that we run the reactor with the sentinel log context,
+        # otherwise other PreserveLoggingContext instances will get confused
+        # and complain when they see the logcontext arbitrarily swapping
+        # between the sentinel and `run` logcontexts.
+        with PreserveLoggingContext():
             logger.info("Running")
             change_resource_limit(config.soft_file_limit)
             if config.gc_thresholds:
diff --git a/synapse/app/synchrotron.py b/synapse/app/synchrotron.py
index 449fac771b..a68bb873fe 100644
--- a/synapse/app/synchrotron.py
+++ b/synapse/app/synchrotron.py
@@ -48,7 +48,8 @@ from synapse.storage.presence import PresenceStore, UserPresenceState
 from synapse.storage.roommember import RoomMemberStore
 from synapse.util.async import sleep
 from synapse.util.httpresourcetree import create_resource_tree
-from synapse.util.logcontext import LoggingContext, preserve_fn
+from synapse.util.logcontext import LoggingContext, preserve_fn, \
+    PreserveLoggingContext
 from synapse.util.manhole import manhole
 from synapse.util.rlimit import change_resource_limit
 from synapse.util.stringutils import random_string
@@ -496,7 +497,11 @@ def start(config_options):
     ss.start_listening(config.worker_listeners)
 
     def run():
-        with LoggingContext("run"):
+        # make sure that we run the reactor with the sentinel log context,
+        # otherwise other PreserveLoggingContext instances will get confused
+        # and complain when they see the logcontext arbitrarily swapping
+        # between the sentinel and `run` logcontexts.
+        with PreserveLoggingContext():
             logger.info("Running")
             change_resource_limit(config.soft_file_limit)
             if config.gc_thresholds:
diff --git a/synapse/handlers/federation.py b/synapse/handlers/federation.py
index 10b2325b27..888dd01240 100644
--- a/synapse/handlers/federation.py
+++ b/synapse/handlers/federation.py
@@ -933,8 +933,9 @@ class FederationHandler(BaseHandler):
             # lots of requests for missing prev_events which we do actually
             # have. Hence we fire off the deferred, but don't wait for it.
 
-            synapse.util.logcontext.reset_context_after_deferred(
-                self._handle_queued_pdus(room_queue))
+            synapse.util.logcontext.preserve_fn(self._handle_queued_pdus)(
+                room_queue
+            )
 
         defer.returnValue(True)
 
diff --git a/synapse/storage/background_updates.py b/synapse/storage/background_updates.py
index 94b2bcc54a..813ad59e56 100644
--- a/synapse/storage/background_updates.py
+++ b/synapse/storage/background_updates.py
@@ -12,6 +12,7 @@
 # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 # See the License for the specific language governing permissions and
 # limitations under the License.
+import synapse.util.async
 
 from ._base import SQLBaseStore
 from . import engines
@@ -84,24 +85,14 @@ class BackgroundUpdateStore(SQLBaseStore):
         self._background_update_performance = {}
         self._background_update_queue = []
         self._background_update_handlers = {}
-        self._background_update_timer = None
 
     @defer.inlineCallbacks
     def start_doing_background_updates(self):
-        assert self._background_update_timer is None, \
-            "background updates already running"
-
         logger.info("Starting background schema updates")
 
         while True:
-            sleep = defer.Deferred()
-            self._background_update_timer = self._clock.call_later(
-                self.BACKGROUND_UPDATE_INTERVAL_MS / 1000., sleep.callback, None
-            )
-            try:
-                yield sleep
-            finally:
-                self._background_update_timer = None
+            yield synapse.util.async.sleep(
+                self.BACKGROUND_UPDATE_INTERVAL_MS / 1000.)
 
             try:
                 result = yield self.do_next_background_update(
diff --git a/synapse/util/logcontext.py b/synapse/util/logcontext.py
index d73670f9f2..7cbe390b15 100644
--- a/synapse/util/logcontext.py
+++ b/synapse/util/logcontext.py
@@ -308,47 +308,44 @@ def preserve_context_over_deferred(deferred, context=None):
     return d
 
 
-def reset_context_after_deferred(deferred):
-    """If the deferred is incomplete, add a callback which will reset the
-    context.
-
-    This is useful when you want to fire off a deferred, but don't want to
-    wait for it to complete. (The deferred will restore the current log context
-    when it completes, so if you don't do anything, it will leak log context.)
-
-    (If this feels asymmetric, consider it this way: we are effectively forking
-    a new thread of execution. We are probably currently within a
-    ``with LoggingContext()`` block, which is supposed to have a single entry
-    and exit point. But by spawning off another deferred, we are effectively
-    adding a new exit point.)
+def preserve_fn(f):
+    """Wraps a function, to ensure that the current context is restored after
+    return from the function, and that the sentinel context is set once the
+    deferred returned by the funtion completes.
 
-    Args:
-        deferred (defer.Deferred): deferred
+    Useful for wrapping functions that return a deferred which you don't yield
+    on.
     """
     def reset_context(result):
         LoggingContext.set_current_context(LoggingContext.sentinel)
         return result
 
-    if not deferred.called:
-        deferred.addBoth(reset_context)
-
-
-def preserve_fn(f):
-    """Ensures that function is called with correct context and that context is
-    restored after return. Useful for wrapping functions that return a deferred
-    which you don't yield on.
-    """
+    # XXX: why is this here rather than inside g? surely we want to preserve
+    # the context from the time the function was called, not when it was
+    # wrapped?
     current = LoggingContext.current_context()
 
     def g(*args, **kwargs):
-        with PreserveLoggingContext(current):
-            res = f(*args, **kwargs)
-            if isinstance(res, defer.Deferred):
-                return preserve_context_over_deferred(
-                    res, context=LoggingContext.sentinel
-                )
-            else:
-                return res
+        res = f(*args, **kwargs)
+        if isinstance(res, defer.Deferred) and not res.called:
+            # The function will have reset the context before returning, so
+            # we need to restore it now.
+            LoggingContext.set_current_context(current)
+
+            # The original context will be restored when the deferred
+            # completes, but there is nothing waiting for it, so it will
+            # get leaked into the reactor or some other function which
+            # wasn't expecting it. We therefore need to reset the context
+            # here.
+            #
+            # (If this feels asymmetric, consider it this way: we are
+            # effectively forking a new thread of execution. We are
+            # probably currently within a ``with LoggingContext()`` block,
+            # which is supposed to have a single entry and exit point. But
+            # by spawning off another deferred, we are effectively
+            # adding a new exit point.)
+            res.addBoth(reset_context)
+        return res
     return g