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
|
// SPDX-License-Identifier: GPL-3.0-or-later
#include "Config.h"
#include "ml-private.h"
using namespace ml;
/*
* Global configuration instance to be shared between training and
* prediction threads.
*/
Config ml::Cfg;
template <typename T>
static T clamp(const T& Value, const T& Min, const T& Max) {
return std::max(Min, std::min(Value, Max));
}
/*
* Initialize global configuration variable.
*/
void Config::readMLConfig(void) {
const char *ConfigSectionML = CONFIG_SECTION_ML;
bool EnableAnomalyDetection = config_get_boolean(ConfigSectionML, "enabled", false);
/*
* Read values
*/
unsigned MaxTrainSamples = config_get_number(ConfigSectionML, "maximum num samples to train", 4 * 3600);
unsigned MinTrainSamples = config_get_number(ConfigSectionML, "minimum num samples to train", 1 * 3600);
unsigned TrainEvery = config_get_number(ConfigSectionML, "train every", 1 * 3600);
unsigned DBEngineAnomalyRateEvery = config_get_number(ConfigSectionML, "dbengine anomaly rate every", 30);
unsigned DiffN = config_get_number(ConfigSectionML, "num samples to diff", 1);
unsigned SmoothN = config_get_number(ConfigSectionML, "num samples to smooth", 3);
unsigned LagN = config_get_number(ConfigSectionML, "num samples to lag", 5);
double RandomSamplingRatio = config_get_float(ConfigSectionML, "random sampling ratio", 1.0 / LagN);
unsigned MaxKMeansIters = config_get_number(ConfigSectionML, "maximum number of k-means iterations", 1000);
double DimensionAnomalyScoreThreshold = config_get_float(ConfigSectionML, "dimension anomaly score threshold", 0.99);
double HostAnomalyRateThreshold = config_get_float(ConfigSectionML, "host anomaly rate threshold", 0.02);
double ADMinWindowSize = config_get_float(ConfigSectionML, "minimum window size", 30);
double ADMaxWindowSize = config_get_float(ConfigSectionML, "maximum window size", 600);
double ADIdleWindowSize = config_get_float(ConfigSectionML, "idle window size", 30);
double ADWindowRateThreshold = config_get_float(ConfigSectionML, "window minimum anomaly rate", 0.25);
double ADDimensionRateThreshold = config_get_float(ConfigSectionML, "anomaly event min dimension rate threshold", 0.05);
std::stringstream SS;
SS << netdata_configured_cache_dir << "/anomaly-detection.db";
Cfg.AnomalyDBPath = SS.str();
/*
* Clamp
*/
MaxTrainSamples = clamp(MaxTrainSamples, 1 * 3600u, 24 * 3600u);
MinTrainSamples = clamp(MinTrainSamples, 1 * 900u, 6 * 3600u);
TrainEvery = clamp(TrainEvery, 1 * 3600u, 6 * 3600u);
DBEngineAnomalyRateEvery = clamp(DBEngineAnomalyRateEvery, 1 * 30u, 15 * 60u);
DiffN = clamp(DiffN, 0u, 1u);
SmoothN = clamp(SmoothN, 0u, 5u);
LagN = clamp(LagN, 1u, 5u);
RandomSamplingRatio = clamp(RandomSamplingRatio, 0.2, 1.0);
MaxKMeansIters = clamp(MaxKMeansIters, 500u, 1000u);
DimensionAnomalyScoreThreshold = clamp(DimensionAnomalyScoreThreshold, 0.01, 5.00);
HostAnomalyRateThreshold = clamp(HostAnomalyRateThreshold, 0.01, 1.0);
ADMinWindowSize = clamp(ADMinWindowSize, 30.0, 300.0);
ADMaxWindowSize = clamp(ADMaxWindowSize, 60.0, 900.0);
ADIdleWindowSize = clamp(ADIdleWindowSize, 30.0, 900.0);
ADWindowRateThreshold = clamp(ADWindowRateThreshold, 0.01, 0.99);
ADDimensionRateThreshold = clamp(ADDimensionRateThreshold, 0.01, 0.99);
/*
* Validate
*/
if (MinTrainSamples >= MaxTrainSamples) {
error("invalid min/max train samples found (%u >= %u)", MinTrainSamples, MaxTrainSamples);
MinTrainSamples = 1 * 3600;
MaxTrainSamples = 4 * 3600;
}
if (ADMinWindowSize >= ADMaxWindowSize) {
error("invalid min/max anomaly window size found (%lf >= %lf)", ADMinWindowSize, ADMaxWindowSize);
ADMinWindowSize = 30.0;
ADMaxWindowSize = 600.0;
}
/*
* Assign to config instance
*/
Cfg.EnableAnomalyDetection = EnableAnomalyDetection;
Cfg.MaxTrainSamples = MaxTrainSamples;
Cfg.MinTrainSamples = MinTrainSamples;
Cfg.TrainEvery = TrainEvery;
Cfg.DBEngineAnomalyRateEvery = DBEngineAnomalyRateEvery;
Cfg.DiffN = DiffN;
Cfg.SmoothN = SmoothN;
Cfg.LagN = LagN;
Cfg.RandomSamplingRatio = RandomSamplingRatio;
Cfg.MaxKMeansIters = MaxKMeansIters;
Cfg.DimensionAnomalyScoreThreshold = DimensionAnomalyScoreThreshold;
Cfg.HostAnomalyRateThreshold = HostAnomalyRateThreshold;
Cfg.ADMinWindowSize = ADMinWindowSize;
Cfg.ADMaxWindowSize = ADMaxWindowSize;
Cfg.ADIdleWindowSize = ADIdleWindowSize;
Cfg.ADWindowRateThreshold = ADWindowRateThreshold;
Cfg.ADDimensionRateThreshold = ADDimensionRateThreshold;
Cfg.HostsToSkip = config_get(ConfigSectionML, "hosts to skip from training", "!*");
Cfg.SP_HostsToSkip = simple_pattern_create(Cfg.HostsToSkip.c_str(), NULL, SIMPLE_PATTERN_EXACT);
// Always exclude anomaly_detection charts from training.
Cfg.ChartsToSkip = "anomaly_detection.* ";
Cfg.ChartsToSkip += config_get(ConfigSectionML, "charts to skip from training", "netdata.*");
Cfg.SP_ChartsToSkip = simple_pattern_create(ChartsToSkip.c_str(), NULL, SIMPLE_PATTERN_EXACT);
Cfg.StreamADCharts = config_get_boolean(ConfigSectionML, "stream anomaly detection charts", true);
}
|
