summary refs log tree commit diff
path: root/synapse/metrics/__init__.py
blob: 9d18b0fbfe65657a5dc6d9db22d0978bffaa4f8f (plain) (blame)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
# -*- coding: utf-8 -*-
# Copyright 2015, 2016 OpenMarket Ltd
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
#     http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# 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.

# Because otherwise 'resource' collides with synapse.metrics.resource
from __future__ import absolute_import

import logging
from resource import getrusage, RUSAGE_SELF
import functools
import os
import stat
import time
import gc

from twisted.internet import reactor

from .metric import (
    CounterMetric, CallbackMetric, DistributionMetric, CacheMetric,
    MemoryUsageMetric,
)


logger = logging.getLogger(__name__)


all_metrics = []
all_collectors = []


class Metrics(object):
    """ A single Metrics object gives a (mutable) slice view of the all_metrics
    dict, allowing callers to easily register new metrics that are namespaced
    nicely."""

    def __init__(self, name):
        self.name_prefix = name

    def register_collector(self, func):
        all_collectors.append(func)

    def _register(self, metric_class, name, *args, **kwargs):
        full_name = "%s_%s" % (self.name_prefix, name)

        metric = metric_class(full_name, *args, **kwargs)

        all_metrics.append(metric)
        return metric

    def register_counter(self, *args, **kwargs):
        return self._register(CounterMetric, *args, **kwargs)

    def register_callback(self, *args, **kwargs):
        return self._register(CallbackMetric, *args, **kwargs)

    def register_distribution(self, *args, **kwargs):
        return self._register(DistributionMetric, *args, **kwargs)

    def register_cache(self, *args, **kwargs):
        return self._register(CacheMetric, *args, **kwargs)


def register_memory_metrics(hs):
    try:
        import psutil
        process = psutil.Process()
        process.memory_info().rss
    except (ImportError, AttributeError):
        logger.warn(
            "psutil is not installed or incorrect version."
            " Disabling memory metrics."
        )
        return
    metric = MemoryUsageMetric(hs, psutil)
    all_metrics.append(metric)


def get_metrics_for(pkg_name):
    """ Returns a Metrics instance for conveniently creating metrics
    namespaced with the given name prefix. """

    # Convert a "package.name" to "package_name" because Prometheus doesn't
    # let us use . in metric names
    return Metrics(pkg_name.replace(".", "_"))


def render_all():
    strs = []

    for collector in all_collectors:
        collector()

    for metric in all_metrics:
        try:
            strs += metric.render()
        except Exception:
            strs += ["# FAILED to render"]
            logger.exception("Failed to render metric")

    strs.append("")  # to generate a final CRLF

    return "\n".join(strs)


# Now register some standard process-wide state metrics, to give indications of
# process resource usage

TICKS_PER_SEC = 100
BYTES_PER_PAGE = 4096

HAVE_PROC_STAT = os.path.exists("/proc/stat")
HAVE_PROC_SELF_STAT = os.path.exists("/proc/self/stat")
HAVE_PROC_SELF_LIMITS = os.path.exists("/proc/self/limits")
HAVE_PROC_SELF_FDS = os.path.exists("/proc/self/fds")

rusage = None
stats = None
fd_counts = None

# In order to report process_start_time_seconds we need to know the machine's
# boot time, because the value in /proc/self/stat is relative to this
boot_time = None
if HAVE_PROC_STAT:
    with open("/proc/stat") as _procstat:
        for line in _procstat:
            if line.startswith("btime "):
                boot_time = int(line.split()[1])

TYPES = {
    stat.S_IFSOCK: "SOCK",
    stat.S_IFLNK: "LNK",
    stat.S_IFREG: "REG",
    stat.S_IFBLK: "BLK",
    stat.S_IFDIR: "DIR",
    stat.S_IFCHR: "CHR",
    stat.S_IFIFO: "FIFO",
}


def update_resource_metrics():
    global rusage
    rusage = getrusage(RUSAGE_SELF)

    if HAVE_PROC_SELF_STAT:
        global stats
        with open("/proc/self/stat") as s:
            line = s.read()
            # line is PID (command) more stats go here ...
            stats = line.split(") ", 1)[1].split(" ")

    global fd_counts
    fd_counts = _process_fds()


def _process_fds():
    counts = {(k,): 0 for k in TYPES.values()}
    counts[("other",)] = 0

    # Not every OS will have a /proc/self/fd directory
    if not HAVE_PROC_SELF_FDS:
        return counts

    for fd in os.listdir("/proc/self/fd"):
        try:
            s = os.stat("/proc/self/fd/%s" % (fd))
            fmt = stat.S_IFMT(s.st_mode)
            if fmt in TYPES:
                t = TYPES[fmt]
            else:
                t = "other"

            counts[(t,)] += 1
        except OSError:
            # the dirh itself used by listdir() is usually missing by now
            pass

    return counts


# Legacy synapse-invented metric names

resource_metrics = get_metrics_for("process.resource")

resource_metrics.register_collector(update_resource_metrics)

# msecs
resource_metrics.register_callback("utime", lambda: rusage.ru_utime * 1000)
resource_metrics.register_callback("stime", lambda: rusage.ru_stime * 1000)

# kilobytes
resource_metrics.register_callback("maxrss", lambda: rusage.ru_maxrss * 1024)

get_metrics_for("process").register_callback("fds", _process_fds, labels=["type"])

# New prometheus-standard metric names

process_metrics = get_metrics_for("process")

if HAVE_PROC_SELF_STAT:
    process_metrics.register_callback(
        "cpu_user_seconds_total", lambda: float(stats[11]) / TICKS_PER_SEC
    )
    process_metrics.register_callback(
        "cpu_system_seconds_total", lambda: float(stats[12]) / TICKS_PER_SEC
    )
    process_metrics.register_callback(
        "cpu_seconds_total", lambda: (float(stats[11]) + float(stats[12])) / TICKS_PER_SEC
    )

    process_metrics.register_callback(
        "virtual_memory_bytes", lambda: int(stats[20])
    )
    process_metrics.register_callback(
        "resident_memory_bytes", lambda: int(stats[21]) * BYTES_PER_PAGE
    )

    process_metrics.register_callback(
        "start_time_seconds", lambda: boot_time + int(stats[19]) / TICKS_PER_SEC
    )

if HAVE_PROC_SELF_FDS:
    process_metrics.register_callback(
        "open_fds", lambda: sum(fd_counts.values())
    )

if HAVE_PROC_SELF_LIMITS:
    def _get_max_fds():
        with open("/proc/self/limits") as limits:
            for line in limits:
                if not line.startswith("Max open files "):
                    continue
                # Line is  Max open files  $SOFT  $HARD
                return int(line.split()[3])
        return None

    process_metrics.register_callback(
        "max_fds", lambda: _get_max_fds()
    )

reactor_metrics = get_metrics_for("reactor")
tick_time = reactor_metrics.register_distribution("tick_time")
pending_calls_metric = reactor_metrics.register_distribution("pending_calls")

gc_time = reactor_metrics.register_distribution("gc_time", labels=["gen"])
gc_unreachable = reactor_metrics.register_counter("gc_unreachable", labels=["gen"])

reactor_metrics.register_callback(
    "gc_counts", lambda: {(i,): v for i, v in enumerate(gc.get_count())}, labels=["gen"]
)


def runUntilCurrentTimer(func):

    @functools.wraps(func)
    def f(*args, **kwargs):
        now = reactor.seconds()
        num_pending = 0

        # _newTimedCalls is one long list of *all* pending calls. Below loop
        # is based off of impl of reactor.runUntilCurrent
        for delayed_call in reactor._newTimedCalls:
            if delayed_call.time > now:
                break

            if delayed_call.delayed_time > 0:
                continue

            num_pending += 1

        num_pending += len(reactor.threadCallQueue)

        start = time.time() * 1000
        ret = func(*args, **kwargs)
        end = time.time() * 1000
        tick_time.inc_by(end - start)
        pending_calls_metric.inc_by(num_pending)

        # Check if we need to do a manual GC (since its been disabled), and do
        # one if necessary.
        threshold = gc.get_threshold()
        counts = gc.get_count()
        for i in (2, 1, 0):
            if threshold[i] < counts[i]:
                logger.info("Collecting gc %d", i)

                start = time.time() * 1000
                unreachable = gc.collect(i)
                end = time.time() * 1000

                gc_time.inc_by(end - start, i)
                gc_unreachable.inc_by(unreachable, i)

        return ret

    return f


try:
    # Ensure the reactor has all the attributes we expect
    reactor.runUntilCurrent
    reactor._newTimedCalls
    reactor.threadCallQueue

    # runUntilCurrent is called when we have pending calls. It is called once
    # per iteratation after fd polling.
    reactor.runUntilCurrent = runUntilCurrentTimer(reactor.runUntilCurrent)

    # We manually run the GC each reactor tick so that we can get some metrics
    # about time spent doing GC,
    gc.disable()
except AttributeError:
    pass