From e053c86a96609382dec3a7312d9ca1012a412844 Mon Sep 17 00:00:00 2001 From: Richard van der Hoff <1389908+richvdh@users.noreply.github.com> Date: Wed, 6 May 2020 17:36:46 +0100 Subject: Make redis go faster with hiredis (#7439) For the record, the reason we need this is as follows: each RDATA command comes down the redis pipe as a subscription message. txredisapi as written needs at least three reactor ticks to read each subscription message from the tcp buffer. Hence, once the process gets loaded, it starts getting behind, and eventually redis knifes the connection. it then takes ages for the master to work its way through the backlog, before it reconnects again, during which any commands from any workers are dropped. --- synapse/python_dependencies.py | 4 +++- 1 file changed, 3 insertions(+), 1 deletion(-) (limited to 'synapse') diff --git a/synapse/python_dependencies.py b/synapse/python_dependencies.py index 733c51b758..39c99a2802 100644 --- a/synapse/python_dependencies.py +++ b/synapse/python_dependencies.py @@ -98,7 +98,9 @@ CONDITIONAL_REQUIREMENTS = { "sentry": ["sentry-sdk>=0.7.2"], "opentracing": ["jaeger-client>=4.0.0", "opentracing>=2.2.0"], "jwt": ["pyjwt>=1.6.4"], - "redis": ["txredisapi>=1.4.7"], + # hiredis is not a *strict* dependency, but it makes things much faster. + # (if it is not installed, we fall back to slow code.) + "redis": ["txredisapi>=1.4.7", "hiredis"], } ALL_OPTIONAL_REQUIREMENTS = set() # type: Set[str] -- cgit 1.4.1 From 4162c39dcf9aa9a2d393e5ab3257596c3aa717c1 Mon Sep 17 00:00:00 2001 From: Andrew Morgan <1342360+anoadragon453@users.noreply.github.com> Date: Wed, 6 May 2020 20:21:38 +0100 Subject: Port group attestation renewal slow down from matrix-org-hotfixes (#7442) --- changelog.d/7442.misc | 1 + synapse/groups/attestations.py | 6 ++---- 2 files changed, 3 insertions(+), 4 deletions(-) create mode 100644 changelog.d/7442.misc (limited to 'synapse') diff --git a/changelog.d/7442.misc b/changelog.d/7442.misc new file mode 100644 index 0000000000..a8fd5ad803 --- /dev/null +++ b/changelog.d/7442.misc @@ -0,0 +1 @@ +Run group attestation renewal in series rather than parallel for performance. \ No newline at end of file diff --git a/synapse/groups/attestations.py b/synapse/groups/attestations.py index 1eec3874b6..27b0c02655 100644 --- a/synapse/groups/attestations.py +++ b/synapse/groups/attestations.py @@ -46,7 +46,6 @@ from twisted.internet import defer from synapse.api.errors import HttpResponseException, RequestSendFailed, SynapseError from synapse.metrics.background_process_metrics import run_as_background_process from synapse.types import get_domain_from_id -from synapse.util.async_helpers import yieldable_gather_results logger = logging.getLogger(__name__) @@ -208,6 +207,5 @@ class GroupAttestionRenewer(object): "Error renewing attestation of %r in %r", user_id, group_id ) - await yieldable_gather_results( - _renew_attestation, ((row["group_id"], row["user_id"]) for row in rows) - ) + for row in rows: + await _renew_attestation((row["group_id"], row["user_id"])) -- cgit 1.4.1 From d7c2df2fa3691069cc4fdeabd5028e246882d70c Mon Sep 17 00:00:00 2001 From: Patrick Cloke Date: Wed, 6 May 2020 16:43:39 -0400 Subject: Improve per-block CPU and DB usage metrics (#7426) --- changelog.d/7426.misc | 1 + synapse/logging/context.py | 38 ++++++++++++++++++++++++++------------ 2 files changed, 27 insertions(+), 12 deletions(-) create mode 100644 changelog.d/7426.misc (limited to 'synapse') diff --git a/changelog.d/7426.misc b/changelog.d/7426.misc new file mode 100644 index 0000000000..731f4dcb52 --- /dev/null +++ b/changelog.d/7426.misc @@ -0,0 +1 @@ +Clean up some LoggingContext code. diff --git a/synapse/logging/context.py b/synapse/logging/context.py index 856534e91a..8b9c4e38bd 100644 --- a/synapse/logging/context.py +++ b/synapse/logging/context.py @@ -431,15 +431,7 @@ class LoggingContext(object): return utime_delta, stime_delta = self._get_cputime(rusage) - self._resource_usage.ru_utime += utime_delta - self._resource_usage.ru_stime += stime_delta - - # if we have a parent, pass our CPU usage stats on - if self.parent_context: - self.parent_context._resource_usage += self._resource_usage - - # reset them in case we get entered again - self._resource_usage.reset() + self.add_cputime(utime_delta, stime_delta) finally: self.usage_start = None @@ -497,30 +489,52 @@ class LoggingContext(object): return utime_delta, stime_delta + def add_cputime(self, utime_delta: float, stime_delta: float) -> None: + """Update the CPU time usage of this context (and any parents, recursively). + + Args: + utime_delta: additional user time, in seconds, spent in this context. + stime_delta: additional system time, in seconds, spent in this context. + """ + self._resource_usage.ru_utime += utime_delta + self._resource_usage.ru_stime += stime_delta + if self.parent_context: + self.parent_context.add_cputime(utime_delta, stime_delta) + def add_database_transaction(self, duration_sec: float) -> None: + """Record the use of a database transaction and the length of time it took. + + Args: + duration_sec: The number of seconds the database transaction took. + """ if duration_sec < 0: raise ValueError("DB txn time can only be non-negative") self._resource_usage.db_txn_count += 1 self._resource_usage.db_txn_duration_sec += duration_sec + if self.parent_context: + self.parent_context.add_database_transaction(duration_sec) def add_database_scheduled(self, sched_sec: float) -> None: """Record a use of the database pool Args: - sched_sec (float): number of seconds it took us to get a - connection + sched_sec: number of seconds it took us to get a connection """ if sched_sec < 0: raise ValueError("DB scheduling time can only be non-negative") self._resource_usage.db_sched_duration_sec += sched_sec + if self.parent_context: + self.parent_context.add_database_scheduled(sched_sec) def record_event_fetch(self, event_count: int) -> None: """Record a number of events being fetched from the db Args: - event_count (int): number of events being fetched + event_count: number of events being fetched """ self._resource_usage.evt_db_fetch_count += event_count + if self.parent_context: + self.parent_context.record_event_fetch(event_count) class LoggingContextFilter(logging.Filter): -- cgit 1.4.1 From d9b8d274949df7356e880a67d3aac1b25613ab1f Mon Sep 17 00:00:00 2001 From: Brendan Abolivier Date: Thu, 7 May 2020 11:35:23 +0200 Subject: Add a configuration setting for the dummy event threshold (#7422) Add dummy_events_threshold which allows configuring the number of forward extremities a room needs for Synapse to send forward extremities in it. --- changelog.d/7422.feature | 1 + docs/sample_config.yaml | 12 ++++++++++++ synapse/config/server.py | 15 +++++++++++++++ synapse/handlers/message.py | 4 +++- 4 files changed, 31 insertions(+), 1 deletion(-) create mode 100644 changelog.d/7422.feature (limited to 'synapse') diff --git a/changelog.d/7422.feature b/changelog.d/7422.feature new file mode 100644 index 0000000000..d6d5bb2169 --- /dev/null +++ b/changelog.d/7422.feature @@ -0,0 +1 @@ +Add a configuration setting to tweak the threshold for dummy events. diff --git a/docs/sample_config.yaml b/docs/sample_config.yaml index fc970986c6..98ead7dc0e 100644 --- a/docs/sample_config.yaml +++ b/docs/sample_config.yaml @@ -253,6 +253,18 @@ listeners: # bind_addresses: ['::1', '127.0.0.1'] # type: manhole +# Forward extremities can build up in a room due to networking delays between +# homeservers. Once this happens in a large room, calculation of the state of +# that room can become quite expensive. To mitigate this, once the number of +# forward extremities reaches a given threshold, Synapse will send an +# org.matrix.dummy_event event, which will reduce the forward extremities +# in the room. +# +# This setting defines the threshold (i.e. number of forward extremities in the +# room) at which dummy events are sent. The default value is 10. +# +#dummy_events_threshold: 5 + ## Homeserver blocking ## diff --git a/synapse/config/server.py b/synapse/config/server.py index c6d58effd4..6d88231843 100644 --- a/synapse/config/server.py +++ b/synapse/config/server.py @@ -505,6 +505,9 @@ class ServerConfig(Config): "cleanup_extremities_with_dummy_events", True ) + # The number of forward extremities in a room needed to send a dummy event. + self.dummy_events_threshold = config.get("dummy_events_threshold", 10) + self.enable_ephemeral_messages = config.get("enable_ephemeral_messages", False) # Inhibits the /requestToken endpoints from returning an error that might leak @@ -823,6 +826,18 @@ class ServerConfig(Config): # bind_addresses: ['::1', '127.0.0.1'] # type: manhole + # Forward extremities can build up in a room due to networking delays between + # homeservers. Once this happens in a large room, calculation of the state of + # that room can become quite expensive. To mitigate this, once the number of + # forward extremities reaches a given threshold, Synapse will send an + # org.matrix.dummy_event event, which will reduce the forward extremities + # in the room. + # + # This setting defines the threshold (i.e. number of forward extremities in the + # room) at which dummy events are sent. The default value is 10. + # + #dummy_events_threshold: 5 + ## Homeserver blocking ## diff --git a/synapse/handlers/message.py b/synapse/handlers/message.py index a324f09340..a622a600b4 100644 --- a/synapse/handlers/message.py +++ b/synapse/handlers/message.py @@ -419,6 +419,8 @@ class EventCreationHandler(object): self._ephemeral_events_enabled = hs.config.enable_ephemeral_messages + self._dummy_events_threshold = hs.config.dummy_events_threshold + @defer.inlineCallbacks def create_event( self, @@ -1085,7 +1087,7 @@ class EventCreationHandler(object): """ self._expire_rooms_to_exclude_from_dummy_event_insertion() room_ids = await self.store.get_rooms_with_many_extremities( - min_count=10, + min_count=self._dummy_events_threshold, limit=5, room_id_filter=self._rooms_to_exclude_from_dummy_event_insertion.keys(), ) -- cgit 1.4.1