Merge pull request #3252 from matrix-org/erikj/in_flight_requests

Add in flight request metrics
pull/3258/head
Erik Johnston 2018-05-22 09:54:30 +01:00 committed by GitHub
commit 13a8dfba0d
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
2 changed files with 162 additions and 2 deletions

View File

@ -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

View File

@ -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",