summary refs log tree commit diff
path: root/synapse/metrics/metric.py
blob: c5f0bcbc155cebf2ebce39c935dc5a3fefacea19 (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
# -*- 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.


from itertools import chain


# TODO(paul): I can't believe Python doesn't have one of these
def map_concat(func, items):
    # flatten a list-of-lists
    return list(chain.from_iterable(map(func, items)))


class BaseMetric(object):
    """Base class for metrics which report a single value per label set
    """

    def __init__(self, name, labels=[]):
        """
        Args:
            name (str): principal name for this metric
            labels (list(str)): names of the labels which will be reported
                for this metric
        """
        self.name = name
        self.labels = labels  # OK not to clone as we never write it

    def dimension(self):
        return len(self.labels)

    def is_scalar(self):
        return not len(self.labels)

    def _render_labelvalue(self, value):
        # TODO: escape backslashes, quotes and newlines
        return '"%s"' % (value)

    def _render_key(self, values):
        if self.is_scalar():
            return ""
        return "{%s}" % (
            ",".join(["%s=%s" % (k, self._render_labelvalue(v))
                      for k, v in zip(self.labels, values)])
        )

    def render(self):
        """Render this metric

        Each metric is rendered as:

            name{label1="val1",label2="val2"} value

        https://prometheus.io/docs/instrumenting/exposition_formats/#text-format-details

        Returns:
            iterable[str]: rendered metrics
        """
        raise NotImplementedError()


class CounterMetric(BaseMetric):
    """The simplest kind of metric; one that stores a monotonically-increasing
    value that counts events or running totals.

    Example use cases for Counters:
    - Number of requests processed
    - Number of items that were inserted into a queue
    - Total amount of data that a system has processed
    Counters can only go up (and be reset when the process restarts).
    """

    def __init__(self, *args, **kwargs):
        super(CounterMetric, self).__init__(*args, **kwargs)

        # dict[list[str]]: value for each set of label values. the keys are the
        # label values, in the same order as the labels in self.labels.
        #
        # (if the metric is a scalar, the (single) key is the empty list).
        self.counts = {}

        # Scalar metrics are never empty
        if self.is_scalar():
            self.counts[()] = 0.

    def inc_by(self, incr, *values):
        if len(values) != self.dimension():
            raise ValueError(
                "Expected as many values to inc() as labels (%d)" % (self.dimension())
            )

        # TODO: should assert that the tag values are all strings

        if values not in self.counts:
            self.counts[values] = incr
        else:
            self.counts[values] += incr

    def inc(self, *values):
        self.inc_by(1, *values)

    def render_item(self, k):
        return ["%s%s %.12g" % (self.name, self._render_key(k), self.counts[k])]

    def render(self):
        return map_concat(self.render_item, sorted(self.counts.keys()))


class CallbackMetric(BaseMetric):
    """A metric that returns the numeric value returned by a callback whenever
    it is rendered. Typically this is used to implement gauges that yield the
    size or other state of some in-memory object by actively querying it."""

    def __init__(self, name, callback, labels=[]):
        super(CallbackMetric, self).__init__(name, labels=labels)

        self.callback = callback

    def render(self):
        value = self.callback()

        if self.is_scalar():
            return ["%s %.12g" % (self.name, value)]

        return ["%s%s %.12g" % (self.name, self._render_key(k), value[k])
                for k in sorted(value.keys())]


class DistributionMetric(object):
    """A combination of an event counter and an accumulator, which counts
    both the number of events and accumulates the total value. Typically this
    could be used to keep track of method-running times, or other distributions
    of values that occur in discrete occurances.

    TODO(paul): Try to export some heatmap-style stats?
    """

    def __init__(self, name, *args, **kwargs):
        self.counts = CounterMetric(name + ":count", **kwargs)
        self.totals = CounterMetric(name + ":total", **kwargs)

    def inc_by(self, inc, *values):
        self.counts.inc(*values)
        self.totals.inc_by(inc, *values)

    def render(self):
        return self.counts.render() + self.totals.render()


class CacheMetric(object):
    __slots__ = ("name", "cache_name", "hits", "misses", "size_callback")

    def __init__(self, name, size_callback, cache_name):
        self.name = name
        self.cache_name = cache_name

        self.hits = 0
        self.misses = 0

        self.size_callback = size_callback

    def inc_hits(self):
        self.hits += 1

    def inc_misses(self):
        self.misses += 1

    def render(self):
        size = self.size_callback()
        hits = self.hits
        total = self.misses + self.hits

        return [
            """%s:hits{name="%s"} %d""" % (self.name, self.cache_name, hits),
            """%s:total{name="%s"} %d""" % (self.name, self.cache_name, total),
            """%s:size{name="%s"} %d""" % (self.name, self.cache_name, size),
        ]


class MemoryUsageMetric(object):
    """Keeps track of the current memory usage, using psutil.

    The class will keep the current min/max/sum/counts of rss over the last
    WINDOW_SIZE_SEC, by polling UPDATE_HZ times per second
    """

    UPDATE_HZ = 2  # number of times to get memory per second
    WINDOW_SIZE_SEC = 30  # the size of the window in seconds

    def __init__(self, hs, psutil):
        clock = hs.get_clock()
        self.memory_snapshots = []

        self.process = psutil.Process()

        clock.looping_call(self._update_curr_values, 1000 / self.UPDATE_HZ)

    def _update_curr_values(self):
        max_size = self.UPDATE_HZ * self.WINDOW_SIZE_SEC
        self.memory_snapshots.append(self.process.memory_info().rss)
        self.memory_snapshots[:] = self.memory_snapshots[-max_size:]

    def render(self):
        if not self.memory_snapshots:
            return []

        max_rss = max(self.memory_snapshots)
        min_rss = min(self.memory_snapshots)
        sum_rss = sum(self.memory_snapshots)
        len_rss = len(self.memory_snapshots)

        return [
            "process_psutil_rss:max %d" % max_rss,
            "process_psutil_rss:min %d" % min_rss,
            "process_psutil_rss:total %d" % sum_rss,
            "process_psutil_rss:count %d" % len_rss,
        ]