diff options
| -rw-r--r-- | synapse/http/request_metrics.py | 160 | ||||
| -rw-r--r-- | synapse/http/site.py | 4 | ||||
| -rw-r--r-- | synapse/util/logcontext.py | 38 |
3 files changed, 189 insertions, 13 deletions
diff --git a/synapse/http/request_metrics.py b/synapse/http/request_metrics.py index 8c850bf23f..5917c83958 100644 --- a/synapse/http/request_metrics.py +++ b/synapse/http/request_metrics.py @@ -98,14 +98,87 @@ response_size = metrics.register_counter( "response_size", labels=["method", "servlet", "tag"] ) +# In flight metrics are incremented while the requests are in flight, rather +# than when the response was written. + +in_flight_requests_ru_utime = metrics.register_counter( + "in_flight_requests_ru_utime_seconds", labels=["method", "servlet"], +) + +in_flight_requests_ru_stime = metrics.register_counter( + "in_flight_requests_ru_stime_seconds", labels=["method", "servlet"], +) + +in_flight_requests_db_txn_count = metrics.register_counter( + "in_flight_requests_db_txn_count", labels=["method", "servlet"], +) + +# seconds spent waiting for db txns, excluding scheduling time, when processing +# this request +in_flight_requests_db_txn_duration = metrics.register_counter( + "in_flight_requests_db_txn_duration_seconds", labels=["method", "servlet"], +) + +# seconds spent waiting for a db connection, when processing this request +in_flight_requests_db_sched_duration = metrics.register_counter( + "in_flight_requests_db_sched_duration_seconds", labels=["method", "servlet"] +) + + +# The set of all in flight requests, set[RequestMetrics] +_in_flight_requests = set() + + +def _collect_in_flight(): + """Called just before metrics are collected, so we use it to update all + the in flight request metrics + """ + + for rm in _in_flight_requests: + rm.update_metrics() + + +metrics.register_collector(_collect_in_flight) + + +def _get_in_flight_counts(): + """Returns a count of all in flight requests by (method, server_name) + + Returns: + dict[tuple[str, str], int] + """ + + # Map from (method, name) -> int, the number of in flight requests of that + # type + counts = {} + for rm in _in_flight_requests: + key = (rm.method, rm.name,) + counts[key] = counts.get(key, 0) + 1 + + return counts + + +metrics.register_callback( + "in_flight_requests_count", + _get_in_flight_counts, + labels=["method", "servlet"] +) + class RequestMetrics(object): - def start(self, time_msec, name): + def start(self, time_msec, name, method): self.start = time_msec self.start_context = LoggingContext.current_context() self.name = name + self.method = method + + self._request_stats = _RequestStats.from_context(self.start_context) + + _in_flight_requests.add(self) def stop(self, time_msec, request): + _in_flight_requests.discard(self) + context = LoggingContext.current_context() tag = "" @@ -147,3 +220,88 @@ class RequestMetrics(object): ) response_size.inc_by(request.sentLength, request.method, self.name, tag) + + # We always call this at the end to ensure that we update the metrics + # regardless of whether a call to /metrics while the request was in + # flight. + self.update_metrics() + + def update_metrics(self): + """Updates the in flight metrics with values from this request. + """ + + diff = self._request_stats.update(self.start_context) + + in_flight_requests_ru_utime.inc_by( + diff.ru_utime, self.method, self.name, + ) + + in_flight_requests_ru_stime.inc_by( + diff.ru_stime, self.method, self.name, + ) + + in_flight_requests_db_txn_count.inc_by( + diff.db_txn_count, self.method, self.name, + ) + + in_flight_requests_db_txn_duration.inc_by( + diff.db_txn_duration_ms / 1000., self.method, self.name, + ) + + in_flight_requests_db_sched_duration.inc_by( + diff.db_sched_duration_ms / 1000., self.method, self.name, + ) + + +class _RequestStats(object): + """Keeps tracks of various metrics for an in flight request. + """ + + __slots__ = [ + "ru_utime", "ru_stime", + "db_txn_count", "db_txn_duration_ms", "db_sched_duration_ms", + ] + + def __init__(self, ru_utime, ru_stime, db_txn_count, + db_txn_duration_ms, db_sched_duration_ms): + self.ru_utime = ru_utime + self.ru_stime = ru_stime + self.db_txn_count = db_txn_count + self.db_txn_duration_ms = db_txn_duration_ms + self.db_sched_duration_ms = db_sched_duration_ms + + @staticmethod + def from_context(context): + ru_utime, ru_stime = context.get_resource_usage() + + return _RequestStats( + ru_utime, ru_stime, + context.db_txn_count, + context.db_txn_duration_ms, + context.db_sched_duration_ms, + ) + + def update(self, context): + """Updates the current values and returns the difference between the + old and new values. + + Returns: + _RequestStats: The difference between the old and new values + """ + new = _RequestStats.from_context(context) + + diff = _RequestStats( + new.ru_utime - self.ru_utime, + new.ru_stime - self.ru_stime, + new.db_txn_count - self.db_txn_count, + new.db_txn_duration_ms - self.db_txn_duration_ms, + new.db_sched_duration_ms - self.db_sched_duration_ms, + ) + + self.ru_utime = new.ru_utime + self.ru_stime = new.ru_stime + self.db_txn_count = new.db_txn_count + self.db_txn_duration_ms = new.db_txn_duration_ms + self.db_sched_duration_ms = new.db_sched_duration_ms + + return diff diff --git a/synapse/http/site.py b/synapse/http/site.py index 202a990508..b608504225 100644 --- a/synapse/http/site.py +++ b/synapse/http/site.py @@ -85,7 +85,9 @@ class SynapseRequest(Request): def _started_processing(self, servlet_name): self.start_time = int(time.time() * 1000) self.request_metrics = RequestMetrics() - self.request_metrics.start(self.start_time, name=servlet_name) + self.request_metrics.start( + self.start_time, name=servlet_name, method=self.method, + ) self.site.access_logger.info( "%s - %s - Received request: %s %s", diff --git a/synapse/util/logcontext.py b/synapse/util/logcontext.py index eab9d57650..914f616312 100644 --- a/synapse/util/logcontext.py +++ b/synapse/util/logcontext.py @@ -60,7 +60,7 @@ class LoggingContext(object): __slots__ = [ "previous_context", "name", "ru_stime", "ru_utime", "db_txn_count", "db_txn_duration_ms", "db_sched_duration_ms", - "usage_start", "usage_end", + "usage_start", "main_thread", "alive", "request", "tag", ] @@ -109,8 +109,10 @@ class LoggingContext(object): # ms spent waiting for db txns to be scheduled self.db_sched_duration_ms = 0 + # If alive has the thread resource usage when the logcontext last + # became active. self.usage_start = None - self.usage_end = None + self.main_thread = threading.current_thread() self.request = None self.tag = "" @@ -159,7 +161,7 @@ class LoggingContext(object): """Restore the logging context in thread local storage to the state it was before this context was entered. Returns: - None to avoid suppressing any exeptions that were thrown. + None to avoid suppressing any exceptions that were thrown. """ current = self.set_current_context(self.previous_context) if current is not self: @@ -185,29 +187,43 @@ class LoggingContext(object): def start(self): if threading.current_thread() is not self.main_thread: + logger.warning("Started logcontext %s on different thread", self) return - if self.usage_start and self.usage_end: - self.ru_utime += self.usage_end.ru_utime - self.usage_start.ru_utime - self.ru_stime += self.usage_end.ru_stime - self.usage_start.ru_stime - self.usage_start = None - self.usage_end = None - + # If we haven't already started record the thread resource usage so + # far if not self.usage_start: self.usage_start = get_thread_resource_usage() def stop(self): if threading.current_thread() is not self.main_thread: + logger.warning("Stopped logcontext %s on different thread", self) return + # When we stop, let's record the resource used since we started if self.usage_start: - self.usage_end = get_thread_resource_usage() + usage_end = get_thread_resource_usage() + + self.ru_utime += usage_end.ru_utime - self.usage_start.ru_utime + self.ru_stime += usage_end.ru_stime - self.usage_start.ru_stime + + self.usage_start = None + else: + logger.warning("Called stop on logcontext %s without calling start", self) def get_resource_usage(self): + """Get CPU time used by this logcontext so far. + + Returns: + tuple[float, float]: The user and system CPU usage in seconds + """ ru_utime = self.ru_utime ru_stime = self.ru_stime - if self.usage_start and threading.current_thread() is self.main_thread: + # If we are on the correct thread and we're currently running then we + # can include resource usage so far. + is_main_thread = threading.current_thread() is self.main_thread + if self.alive and self.usage_start and is_main_thread: current = get_thread_resource_usage() ru_utime += current.ru_utime - self.usage_start.ru_utime ru_stime += current.ru_stime - self.usage_start.ru_stime |