Jagpal Singh Gill | 23f091e | 2023-12-10 15:23:19 -0800 | [diff] [blame] | 1 | #include "health_metric.hpp" |
| 2 | |
| 3 | #include <phosphor-logging/lg2.hpp> |
| 4 | |
| 5 | #include <numeric> |
| 6 | #include <unordered_map> |
| 7 | |
| 8 | PHOSPHOR_LOG2_USING; |
| 9 | |
| 10 | namespace phosphor::health::metric |
| 11 | { |
| 12 | |
| 13 | using association_t = std::tuple<std::string, std::string, std::string>; |
| 14 | |
| 15 | auto HealthMetric::getPath(SubType subType) -> std::string |
| 16 | { |
| 17 | std::string path; |
| 18 | switch (subType) |
| 19 | { |
| 20 | case SubType::cpuTotal: |
| 21 | { |
| 22 | return std::string(BmcPath) + "/" + PathIntf::total_cpu; |
| 23 | } |
| 24 | case SubType::cpuKernel: |
| 25 | { |
| 26 | return std::string(BmcPath) + "/" + PathIntf::kernel_cpu; |
| 27 | } |
| 28 | case SubType::cpuUser: |
| 29 | { |
| 30 | return std::string(BmcPath) + "/" + PathIntf::user_cpu; |
| 31 | } |
| 32 | case SubType::memoryAvailable: |
| 33 | { |
| 34 | return std::string(BmcPath) + "/" + PathIntf::available_memory; |
| 35 | } |
| 36 | case SubType::memoryBufferedAndCached: |
| 37 | { |
| 38 | return std::string(BmcPath) + "/" + |
| 39 | PathIntf::buffered_and_cached_memory; |
| 40 | } |
| 41 | case SubType::memoryFree: |
| 42 | { |
| 43 | return std::string(BmcPath) + "/" + PathIntf::free_memory; |
| 44 | } |
| 45 | case SubType::memoryShared: |
| 46 | { |
| 47 | return std::string(BmcPath) + "/" + PathIntf::shared_memory; |
| 48 | } |
| 49 | case SubType::memoryTotal: |
| 50 | { |
| 51 | return std::string(BmcPath) + "/" + PathIntf::total_memory; |
| 52 | } |
| 53 | case SubType::storageReadWrite: |
| 54 | { |
| 55 | return std::string(BmcPath) + "/" + PathIntf::read_write_storage; |
| 56 | } |
Jagpal Singh Gill | dfe839f | 2024-02-16 09:54:02 -0800 | [diff] [blame] | 57 | case SubType::storageTmp: |
| 58 | { |
| 59 | return std::string(BmcPath) + "/" + PathIntf::tmp_storage; |
| 60 | } |
Jagpal Singh Gill | 23f091e | 2023-12-10 15:23:19 -0800 | [diff] [blame] | 61 | default: |
| 62 | { |
| 63 | error("Invalid Memory metric {TYPE}", "TYPE", |
| 64 | std::to_underlying(subType)); |
| 65 | return ""; |
| 66 | } |
| 67 | } |
| 68 | } |
| 69 | |
| 70 | void HealthMetric::initProperties() |
| 71 | { |
| 72 | switch (config.subType) |
| 73 | { |
| 74 | case SubType::cpuTotal: |
| 75 | case SubType::cpuKernel: |
| 76 | case SubType::cpuUser: |
| 77 | { |
| 78 | ValueIntf::unit(ValueIntf::Unit::Percent, true); |
| 79 | ValueIntf::minValue(0.0, true); |
| 80 | ValueIntf::maxValue(100.0, true); |
| 81 | break; |
| 82 | } |
| 83 | case SubType::memoryAvailable: |
| 84 | case SubType::memoryBufferedAndCached: |
| 85 | case SubType::memoryFree: |
| 86 | case SubType::memoryShared: |
| 87 | case SubType::memoryTotal: |
| 88 | case SubType::storageReadWrite: |
| 89 | default: |
| 90 | { |
| 91 | ValueIntf::unit(ValueIntf::Unit::Bytes, true); |
| 92 | ValueIntf::minValue(0.0, true); |
| 93 | } |
| 94 | } |
Jagpal Singh Gill | c5b18bc | 2024-02-09 15:58:12 -0800 | [diff] [blame] | 95 | ValueIntf::value(std::numeric_limits<double>::quiet_NaN(), true); |
Jagpal Singh Gill | 23f091e | 2023-12-10 15:23:19 -0800 | [diff] [blame] | 96 | |
| 97 | using bound_map_t = std::map<ThresholdIntf::Bound, double>; |
| 98 | std::map<ThresholdIntf::Type, bound_map_t> thresholds; |
| 99 | for (const auto& [key, value] : config.thresholds) |
| 100 | { |
| 101 | auto type = std::get<ThresholdIntf::Type>(key); |
| 102 | auto bound = std::get<ThresholdIntf::Bound>(key); |
| 103 | auto threshold = thresholds.find(type); |
| 104 | if (threshold == thresholds.end()) |
| 105 | { |
| 106 | bound_map_t bounds; |
Jagpal Singh Gill | 6a3884a | 2024-02-24 18:08:23 -0800 | [diff] [blame^] | 107 | bounds.emplace(bound, std::numeric_limits<double>::quiet_NaN()); |
Jagpal Singh Gill | 23f091e | 2023-12-10 15:23:19 -0800 | [diff] [blame] | 108 | thresholds.emplace(type, bounds); |
| 109 | } |
| 110 | else |
| 111 | { |
| 112 | threshold->second.emplace(bound, value.value); |
| 113 | } |
| 114 | } |
Jagpal Singh Gill | c5b18bc | 2024-02-09 15:58:12 -0800 | [diff] [blame] | 115 | ThresholdIntf::value(thresholds, true); |
Jagpal Singh Gill | 23f091e | 2023-12-10 15:23:19 -0800 | [diff] [blame] | 116 | } |
| 117 | |
Jagpal Singh Gill | 55fb0c9 | 2024-02-22 18:07:13 -0800 | [diff] [blame] | 118 | bool didThresholdViolate(ThresholdIntf::Bound bound, double thresholdValue, |
| 119 | double value) |
| 120 | { |
| 121 | switch (bound) |
| 122 | { |
| 123 | case ThresholdIntf::Bound::Lower: |
| 124 | { |
| 125 | return (value < thresholdValue); |
| 126 | } |
| 127 | case ThresholdIntf::Bound::Upper: |
| 128 | { |
| 129 | return (value > thresholdValue); |
| 130 | } |
| 131 | default: |
| 132 | { |
| 133 | error("Invalid threshold bound {BOUND}", "BOUND", |
| 134 | std::to_underlying(bound)); |
| 135 | return false; |
| 136 | } |
| 137 | } |
| 138 | } |
| 139 | |
Jagpal Singh Gill | 23f091e | 2023-12-10 15:23:19 -0800 | [diff] [blame] | 140 | void HealthMetric::checkThreshold(ThresholdIntf::Type type, |
Jagpal Singh Gill | 6a3884a | 2024-02-24 18:08:23 -0800 | [diff] [blame^] | 141 | ThresholdIntf::Bound bound, MValue value) |
Jagpal Singh Gill | 23f091e | 2023-12-10 15:23:19 -0800 | [diff] [blame] | 142 | { |
| 143 | auto threshold = std::make_tuple(type, bound); |
| 144 | auto thresholds = ThresholdIntf::value(); |
| 145 | |
| 146 | if (thresholds.contains(type) && thresholds[type].contains(bound)) |
| 147 | { |
Jagpal Singh Gill | 6a3884a | 2024-02-24 18:08:23 -0800 | [diff] [blame^] | 148 | auto tConfig = config.thresholds.at(threshold); |
| 149 | auto thresholdValue = tConfig.value / 100 * value.total; |
| 150 | thresholds[type][bound] = thresholdValue; |
| 151 | ThresholdIntf::value(thresholds); |
Jagpal Singh Gill | 23f091e | 2023-12-10 15:23:19 -0800 | [diff] [blame] | 152 | auto assertions = ThresholdIntf::asserted(); |
Jagpal Singh Gill | 6a3884a | 2024-02-24 18:08:23 -0800 | [diff] [blame^] | 153 | if (didThresholdViolate(bound, thresholdValue, value.current)) |
Jagpal Singh Gill | 23f091e | 2023-12-10 15:23:19 -0800 | [diff] [blame] | 154 | { |
| 155 | if (!assertions.contains(threshold)) |
| 156 | { |
| 157 | assertions.insert(threshold); |
| 158 | ThresholdIntf::asserted(assertions); |
Jagpal Singh Gill | 6a3884a | 2024-02-24 18:08:23 -0800 | [diff] [blame^] | 159 | ThresholdIntf::assertionChanged(type, bound, true, |
| 160 | value.current); |
Jagpal Singh Gill | 23f091e | 2023-12-10 15:23:19 -0800 | [diff] [blame] | 161 | if (tConfig.log) |
| 162 | { |
| 163 | error( |
| 164 | "ASSERT: Health Metric {METRIC} crossed {TYPE} upper threshold", |
| 165 | "METRIC", config.name, "TYPE", |
| 166 | sdbusplus::message::convert_to_string(type)); |
| 167 | startUnit(bus, tConfig.target); |
| 168 | } |
| 169 | } |
| 170 | return; |
| 171 | } |
| 172 | else if (assertions.contains(threshold)) |
| 173 | { |
| 174 | assertions.erase(threshold); |
| 175 | ThresholdIntf::asserted(assertions); |
Jagpal Singh Gill | 6a3884a | 2024-02-24 18:08:23 -0800 | [diff] [blame^] | 176 | ThresholdIntf::assertionChanged(type, bound, false, value.current); |
Jagpal Singh Gill | 23f091e | 2023-12-10 15:23:19 -0800 | [diff] [blame] | 177 | if (config.thresholds.find(threshold)->second.log) |
| 178 | { |
| 179 | info( |
| 180 | "DEASSERT: Health Metric {METRIC} is below {TYPE} upper threshold", |
| 181 | "METRIC", config.name, "TYPE", |
| 182 | sdbusplus::message::convert_to_string(type)); |
| 183 | } |
| 184 | } |
| 185 | } |
| 186 | } |
| 187 | |
Jagpal Singh Gill | 6a3884a | 2024-02-24 18:08:23 -0800 | [diff] [blame^] | 188 | void HealthMetric::checkThresholds(MValue value) |
Jagpal Singh Gill | 23f091e | 2023-12-10 15:23:19 -0800 | [diff] [blame] | 189 | { |
| 190 | if (!ThresholdIntf::value().empty()) |
| 191 | { |
| 192 | for (auto type : |
| 193 | {ThresholdIntf::Type::HardShutdown, |
| 194 | ThresholdIntf::Type::SoftShutdown, |
| 195 | ThresholdIntf::Type::PerformanceLoss, |
| 196 | ThresholdIntf::Type::Critical, ThresholdIntf::Type::Warning}) |
| 197 | { |
Jagpal Singh Gill | 55fb0c9 | 2024-02-22 18:07:13 -0800 | [diff] [blame] | 198 | checkThreshold(type, ThresholdIntf::Bound::Lower, value); |
Jagpal Singh Gill | 23f091e | 2023-12-10 15:23:19 -0800 | [diff] [blame] | 199 | checkThreshold(type, ThresholdIntf::Bound::Upper, value); |
| 200 | } |
| 201 | } |
| 202 | } |
| 203 | |
| 204 | void HealthMetric::update(MValue value) |
| 205 | { |
| 206 | // Maintain window size for metric |
| 207 | if (history.size() >= config.windowSize) |
| 208 | { |
| 209 | history.pop_front(); |
| 210 | } |
Jagpal Singh Gill | 6a3884a | 2024-02-24 18:08:23 -0800 | [diff] [blame^] | 211 | history.push_back(value.current); |
Jagpal Singh Gill | 23f091e | 2023-12-10 15:23:19 -0800 | [diff] [blame] | 212 | |
| 213 | if (history.size() < config.windowSize) |
| 214 | { |
| 215 | // Wait for the metric to have enough samples to calculate average |
Patrick Williams | 0f54d7a | 2024-02-22 12:39:46 -0600 | [diff] [blame] | 216 | debug("Not enough samples to calculate average"); |
Jagpal Singh Gill | 23f091e | 2023-12-10 15:23:19 -0800 | [diff] [blame] | 217 | return; |
| 218 | } |
| 219 | |
| 220 | double average = (std::accumulate(history.begin(), history.end(), 0.0)) / |
| 221 | history.size(); |
| 222 | ValueIntf::value(average); |
Jagpal Singh Gill | 6a3884a | 2024-02-24 18:08:23 -0800 | [diff] [blame^] | 223 | checkThresholds(value); |
Jagpal Singh Gill | 23f091e | 2023-12-10 15:23:19 -0800 | [diff] [blame] | 224 | } |
| 225 | |
| 226 | void HealthMetric::create(const paths_t& bmcPaths) |
| 227 | { |
| 228 | info("Create Health Metric: {METRIC}", "METRIC", config.name); |
| 229 | initProperties(); |
| 230 | |
| 231 | std::vector<association_t> associations; |
| 232 | static constexpr auto forwardAssociation = "measuring"; |
| 233 | static constexpr auto reverseAssociation = "measured_by"; |
| 234 | for (const auto& bmcPath : bmcPaths) |
| 235 | { |
| 236 | /* |
| 237 | * This metric is "measuring" the health for the BMC at bmcPath |
| 238 | * The BMC at bmcPath is "measured_by" this metric. |
| 239 | */ |
| 240 | associations.push_back( |
| 241 | {forwardAssociation, reverseAssociation, bmcPath}); |
| 242 | } |
| 243 | AssociationIntf::associations(associations); |
| 244 | } |
| 245 | |
| 246 | } // namespace phosphor::health::metric |