forked-synapse/synapse/http/request_metrics.py

264 lines
8.3 KiB
Python
Raw Normal View History

#
2023-11-21 15:29:58 -05:00
# This file is licensed under the Affero General Public License (AGPL) version 3.
#
# Copyright 2014-2016 OpenMarket Ltd
2023-11-21 15:29:58 -05:00
# Copyright (C) 2023 New Vector, Ltd
#
# This program is free software: you can redistribute it and/or modify
# it under the terms of the GNU Affero General Public License as
# published by the Free Software Foundation, either version 3 of the
# License, or (at your option) any later version.
#
# See the GNU Affero General Public License for more details:
# <https://www.gnu.org/licenses/agpl-3.0.html>.
#
# Originally licensed under the Apache License, Version 2.0:
# <http://www.apache.org/licenses/LICENSE-2.0>.
#
# [This file includes modifications made by New Vector Limited]
#
#
import logging
import threading
import traceback
from typing import Dict, Mapping, Set, Tuple
2018-05-21 20:48:57 -04:00
from prometheus_client.core import Counter, Histogram
from synapse.logging.context import current_context
2018-07-09 02:09:20 -04:00
from synapse.metrics import LaterGauge
logger = logging.getLogger(__name__)
# total number of responses served, split by method/servlet/tag
2018-05-22 17:54:22 -04:00
response_count = Counter(
"synapse_http_server_response_count", "", ["method", "servlet", "tag"]
)
2018-05-22 17:54:22 -04:00
requests_counter = Counter(
"synapse_http_server_requests_received", "", ["method", "servlet"]
)
2018-05-22 17:54:22 -04:00
outgoing_responses_counter = Counter(
"synapse_http_server_responses", "", ["method", "code"]
)
2018-05-22 17:54:22 -04:00
response_timer = Histogram(
2018-10-19 06:45:45 -04:00
"synapse_http_server_response_time_seconds",
"sec",
2018-07-18 09:19:00 -04:00
["method", "servlet", "tag", "code"],
2018-05-22 17:54:22 -04:00
)
2018-05-22 17:54:22 -04:00
response_ru_utime = Counter(
2018-05-28 05:10:27 -04:00
"synapse_http_server_response_ru_utime_seconds", "sec", ["method", "servlet", "tag"]
2018-05-22 17:54:22 -04:00
)
2018-05-22 17:54:22 -04:00
response_ru_stime = Counter(
2018-05-28 05:10:27 -04:00
"synapse_http_server_response_ru_stime_seconds", "sec", ["method", "servlet", "tag"]
2018-05-22 17:54:22 -04:00
)
2018-05-22 17:54:22 -04:00
response_db_txn_count = Counter(
"synapse_http_server_response_db_txn_count", "", ["method", "servlet", "tag"]
)
# seconds spent waiting for db txns, excluding scheduling time, when processing
# this request
2018-05-22 17:54:22 -04:00
response_db_txn_duration = Counter(
"synapse_http_server_response_db_txn_duration_seconds",
"",
["method", "servlet", "tag"],
)
# seconds spent waiting for a db connection, when processing this request
2018-05-22 17:54:22 -04:00
response_db_sched_duration = Counter(
2018-05-23 14:03:51 -04:00
"synapse_http_server_response_db_sched_duration_seconds",
2018-05-22 17:54:22 -04:00
"",
["method", "servlet", "tag"],
)
# size in bytes of the response written
2018-05-22 17:54:22 -04:00
response_size = Counter(
2018-05-23 14:03:51 -04:00
"synapse_http_server_response_size", "", ["method", "servlet", "tag"]
)
# In flight metrics are incremented while the requests are in flight, rather
# than when the response was written.
2018-05-22 17:54:22 -04:00
in_flight_requests_ru_utime = Counter(
2018-10-19 06:45:45 -04:00
"synapse_http_server_in_flight_requests_ru_utime_seconds", "", ["method", "servlet"]
2018-05-22 17:54:22 -04:00
)
2018-05-22 17:54:22 -04:00
in_flight_requests_ru_stime = Counter(
2018-10-19 06:45:45 -04:00
"synapse_http_server_in_flight_requests_ru_stime_seconds", "", ["method", "servlet"]
2018-05-22 17:54:22 -04:00
)
2018-05-22 17:54:22 -04:00
in_flight_requests_db_txn_count = Counter(
2018-05-23 14:03:51 -04:00
"synapse_http_server_in_flight_requests_db_txn_count", "", ["method", "servlet"]
2018-05-22 17:54:22 -04:00
)
# seconds spent waiting for db txns, excluding scheduling time, when processing
# this request
2018-05-22 17:54:22 -04:00
in_flight_requests_db_txn_duration = Counter(
2018-05-23 14:03:51 -04:00
"synapse_http_server_in_flight_requests_db_txn_duration_seconds",
2018-05-22 17:54:22 -04:00
"",
["method", "servlet"],
)
# seconds spent waiting for a db connection, when processing this request
2018-05-22 17:54:22 -04:00
in_flight_requests_db_sched_duration = Counter(
2018-05-23 14:03:51 -04:00
"synapse_http_server_in_flight_requests_db_sched_duration_seconds",
2018-05-22 17:54:22 -04:00
"",
["method", "servlet"],
)
_in_flight_requests: Set["RequestMetrics"] = set()
2020-10-23 12:38:40 -04:00
# Protects the _in_flight_requests set from concurrent access
2018-08-21 11:28:10 -04:00
_in_flight_requests_lock = threading.Lock()
def _get_in_flight_counts() -> Mapping[Tuple[str, ...], int]:
"""Returns a count of all in flight requests by (method, server_name)"""
# Cast to a list to prevent it changing while the Prometheus
# thread is collecting metrics
2018-08-21 11:28:10 -04:00
with _in_flight_requests_lock:
reqs = list(_in_flight_requests)
for rm in reqs:
2018-05-22 11:56:03 -04:00
rm.update_metrics()
# Map from (method, name) -> int, the number of in flight requests of that
# type. The key type is Tuple[str, str], but we leave the length unspecified
# for compatability with LaterGauge's annotations.
counts: Dict[Tuple[str, ...], int] = {}
for rm in reqs:
2018-10-19 06:45:45 -04:00
key = (rm.method, rm.name)
counts[key] = counts.get(key, 0) + 1
return counts
2018-05-22 11:56:03 -04:00
LaterGauge(
"synapse_http_server_in_flight_requests_count",
2018-05-22 17:54:22 -04:00
"",
["method", "servlet"],
2018-05-22 17:54:22 -04:00
_get_in_flight_counts,
)
2020-09-04 06:54:56 -04:00
class RequestMetrics:
def start(self, time_sec: float, name: str, method: str) -> None:
self.start_ts = time_sec
self.start_context = current_context()
self.name = name
self.method = method
if self.start_context:
# _request_stats records resource usage that we have already added
# to the "in flight" metrics.
self._request_stats = self.start_context.get_resource_usage()
else:
logger.error(
"Tried to start a RequestMetric from the sentinel context.\n%s",
"".join(traceback.format_stack()),
)
2018-08-21 11:28:10 -04:00
with _in_flight_requests_lock:
_in_flight_requests.add(self)
def stop(self, time_sec: float, response_code: int, sent_bytes: int) -> None:
2018-08-21 11:28:10 -04:00
with _in_flight_requests_lock:
_in_flight_requests.discard(self)
context = current_context()
tag = ""
if context:
tag = context.tag
if context != self.start_context:
logger.error(
"Context have unexpectedly changed %r, %r",
2018-10-19 06:45:45 -04:00
context,
self.start_context,
)
return
else:
logger.error(
"Trying to stop RequestMetrics in the sentinel context.\n%s",
"".join(traceback.format_stack()),
)
return
response_code_str = str(response_code)
outgoing_responses_counter.labels(self.method, response_code_str).inc()
2018-09-17 12:16:50 -04:00
response_count.labels(self.method, self.name, tag).inc()
response_timer.labels(self.method, self.name, tag, response_code_str).observe(
time_sec - self.start_ts
2018-05-22 17:54:22 -04:00
)
resource_usage = context.get_resource_usage()
2018-09-17 12:16:50 -04:00
response_ru_utime.labels(self.method, self.name, tag).inc(
2018-10-19 06:45:45 -04:00
resource_usage.ru_utime
)
2018-09-17 12:16:50 -04:00
response_ru_stime.labels(self.method, self.name, tag).inc(
2018-10-19 06:45:45 -04:00
resource_usage.ru_stime
)
2018-09-17 12:16:50 -04:00
response_db_txn_count.labels(self.method, self.name, tag).inc(
resource_usage.db_txn_count
2018-05-22 17:54:22 -04:00
)
2018-09-17 12:16:50 -04:00
response_db_txn_duration.labels(self.method, self.name, tag).inc(
resource_usage.db_txn_duration_sec
2018-05-22 17:54:22 -04:00
)
2018-09-17 12:16:50 -04:00
response_db_sched_duration.labels(self.method, self.name, tag).inc(
resource_usage.db_sched_duration_sec
2018-05-22 17:54:22 -04:00
)
2018-05-21 20:48:57 -04:00
2018-09-17 12:16:50 -04:00
response_size.labels(self.method, self.name, tag).inc(sent_bytes)
# 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) -> None:
"""Updates the in flight metrics with values from this request."""
if not self.start_context:
logger.error(
"Tried to update a RequestMetric from the sentinel context.\n%s",
"".join(traceback.format_stack()),
)
return
new_stats = self.start_context.get_resource_usage()
diff = new_stats - self._request_stats
self._request_stats = new_stats
2018-10-19 06:45:45 -04:00
# max() is used since rapid use of ru_stime/ru_utime can end up with the
# count going backwards due to NTP, time smearing, fine-grained
# correction, or floating points. Who knows, really?
in_flight_requests_ru_utime.labels(self.method, self.name).inc(
max(diff.ru_utime, 0)
)
in_flight_requests_ru_stime.labels(self.method, self.name).inc(
max(diff.ru_stime, 0)
)
2018-05-22 17:54:22 -04:00
in_flight_requests_db_txn_count.labels(self.method, self.name).inc(
diff.db_txn_count
)
2018-05-22 17:54:22 -04:00
in_flight_requests_db_txn_duration.labels(self.method, self.name).inc(
diff.db_txn_duration_sec
2018-05-22 17:54:22 -04:00
)
2018-05-22 17:54:22 -04:00
in_flight_requests_db_sched_duration.labels(self.method, self.name).inc(
diff.db_sched_duration_sec
2018-05-22 17:54:22 -04:00
)