Girish Gowdra | e09a620 | 2021-01-12 18:10:59 -0800 | [diff] [blame] | 1 | /* |
| 2 | * Copyright 2021-present Open Networking Foundation |
| 3 | |
| 4 | * Licensed under the Apache License, Version 2.0 (the "License"); |
| 5 | * you may not use this file except in compliance with the License. |
| 6 | * You may obtain a copy of the License at |
| 7 | |
| 8 | * http://www.apache.org/licenses/LICENSE-2.0 |
| 9 | |
| 10 | * Unless required by applicable law or agreed to in writing, software |
| 11 | * distributed under the License is distributed on an "AS IS" BASIS, |
| 12 | * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. |
| 13 | * See the License for the specific language governing permissions and |
| 14 | * limitations under the License. |
| 15 | */ |
| 16 | |
| 17 | //Package adaptercoreonu provides the utility for onu devices, flows and statistics |
| 18 | package adaptercoreonu |
| 19 | |
| 20 | import ( |
| 21 | "context" |
Girish Gowdra | 0e53364 | 2021-03-02 22:02:51 -0800 | [diff] [blame] | 22 | "encoding/json" |
Girish Gowdra | e09a620 | 2021-01-12 18:10:59 -0800 | [diff] [blame] | 23 | "fmt" |
Girish Gowdra | e0140f0 | 2021-02-02 16:55:09 -0800 | [diff] [blame] | 24 | "github.com/looplab/fsm" |
Girish Gowdra | e09a620 | 2021-01-12 18:10:59 -0800 | [diff] [blame] | 25 | "github.com/opencord/omci-lib-go" |
| 26 | me "github.com/opencord/omci-lib-go/generated" |
Girish Gowdra | 0e53364 | 2021-03-02 22:02:51 -0800 | [diff] [blame] | 27 | "github.com/opencord/voltha-lib-go/v4/pkg/db" |
| 28 | "github.com/opencord/voltha-lib-go/v4/pkg/db/kvstore" |
Girish Gowdra | e09a620 | 2021-01-12 18:10:59 -0800 | [diff] [blame] | 29 | "github.com/opencord/voltha-lib-go/v4/pkg/log" |
| 30 | "github.com/opencord/voltha-protos/v4/go/voltha" |
Girish Gowdra | 5a7c492 | 2021-01-22 18:33:41 -0800 | [diff] [blame] | 31 | "sync" |
Girish Gowdra | e09a620 | 2021-01-12 18:10:59 -0800 | [diff] [blame] | 32 | "time" |
| 33 | ) |
| 34 | |
Girish Gowdra | e0140f0 | 2021-02-02 16:55:09 -0800 | [diff] [blame] | 35 | const ( |
| 36 | // events of L2 PM FSM |
| 37 | l2PmEventInit = "l2PmEventInit" |
| 38 | l2PmEventTick = "l2PmEventTick" |
| 39 | l2PmEventSuccess = "l2PmEventSuccess" |
| 40 | l2PmEventFailure = "l2PmEventFailure" |
| 41 | l2PmEventAddMe = "l2PmEventAddMe" |
| 42 | l2PmEventDeleteMe = "l2PmEventDeleteMe" |
| 43 | l2PmEventStop = "l2PmEventStop" |
| 44 | ) |
| 45 | const ( |
| 46 | // states of L2 PM FSM |
| 47 | l2PmStNull = "l2PmStNull" |
| 48 | l2PmStStarting = "l2PmStStarting" |
| 49 | l2PmStSyncTime = "l2PmStSyncTime" |
| 50 | l2PmStIdle = "l2PmStIdle" |
| 51 | l2PmStCreatePmMe = "l2PmStCreatePm" |
| 52 | l2PmStDeletePmMe = "l2PmStDeletePmMe" |
| 53 | l2PmStCollectData = "l2PmStCollectData" |
| 54 | ) |
| 55 | |
| 56 | const cL2PmFsmIdleState = l2PmStIdle |
| 57 | |
Girish Gowdra | 5a7c492 | 2021-01-22 18:33:41 -0800 | [diff] [blame] | 58 | // general constants used for overall Metric Collection management |
| 59 | const ( |
| 60 | DefaultMetricCollectionFrequency = 15 * 60 // unit in seconds. This setting can be changed from voltha NBI PmConfig configuration |
| 61 | GroupMetricEnabled = true // This is READONLY and cannot be changed from VOLTHA NBI |
| 62 | DefaultFrequencyOverrideEnabled = true // This is READONLY and cannot be changed from VOLTHA NBI |
| 63 | FrequencyGranularity = 5 // The frequency (in seconds) has to be multiple of 5. This setting cannot changed later. |
| 64 | ) |
| 65 | |
| 66 | // OpticalPowerGroupMetrics are supported optical pm names |
| 67 | var OpticalPowerGroupMetrics = map[string]voltha.PmConfig_PmType{ |
| 68 | "ani_g_instance_id": voltha.PmConfig_CONTEXT, |
| 69 | "transmit_power": voltha.PmConfig_GAUGE, |
| 70 | "receive_power": voltha.PmConfig_GAUGE, |
| 71 | } |
| 72 | |
| 73 | // OpticalPowerGroupMetrics specific constants |
| 74 | const ( |
Girish Gowdra | e0140f0 | 2021-02-02 16:55:09 -0800 | [diff] [blame] | 75 | OpticalPowerGroupMetricName = "PON_Optical" |
Girish Gowdra | 5a7c492 | 2021-01-22 18:33:41 -0800 | [diff] [blame] | 76 | OpticalPowerGroupMetricEnabled = true // This setting can be changed from voltha NBI PmConfig configuration |
| 77 | OpticalPowerMetricGroupCollectionFrequency = 5 * 60 // unit in seconds. This setting can be changed from voltha NBI PmConfig configuration |
| 78 | ) |
| 79 | |
| 80 | // UniStatusGroupMetrics are supported UNI status names |
| 81 | var UniStatusGroupMetrics = map[string]voltha.PmConfig_PmType{ |
| 82 | "uni_port_no": voltha.PmConfig_CONTEXT, |
Girish Gowdra | 0e53364 | 2021-03-02 22:02:51 -0800 | [diff] [blame] | 83 | "entity_id": voltha.PmConfig_CONTEXT, |
Girish Gowdra | 5a7c492 | 2021-01-22 18:33:41 -0800 | [diff] [blame] | 84 | "ethernet_type": voltha.PmConfig_GAUGE, |
| 85 | "oper_status": voltha.PmConfig_GAUGE, |
| 86 | "uni_admin_state": voltha.PmConfig_GAUGE, |
| 87 | } |
| 88 | |
| 89 | // UniStatusGroupMetrics specific constants |
| 90 | const ( |
Girish Gowdra | e0140f0 | 2021-02-02 16:55:09 -0800 | [diff] [blame] | 91 | UniStatusGroupMetricName = "UNI_Status" |
Girish Gowdra | 5a7c492 | 2021-01-22 18:33:41 -0800 | [diff] [blame] | 92 | UniStatusGroupMetricEnabled = true // This setting can be changed from voltha NBI PmConfig configuration |
| 93 | UniStatusMetricGroupCollectionFrequency = 5 * 60 // unit in seconds. This setting can be changed from voltha NBI PmConfig configuration |
| 94 | ) |
| 95 | |
Girish Gowdra | e0140f0 | 2021-02-02 16:55:09 -0800 | [diff] [blame] | 96 | // *** Classical L2 PM Counters begin *** |
| 97 | |
| 98 | // EthernetBridgeHistory are supported ethernet bridge history counters fetched from |
| 99 | // Ethernet Frame Performance Monitoring History Data Downstream and Ethernet Frame Performance Monitoring History Data Upstream MEs. |
| 100 | var EthernetBridgeHistory = map[string]voltha.PmConfig_PmType{ |
| 101 | "class_id": voltha.PmConfig_CONTEXT, |
| 102 | "entity_id": voltha.PmConfig_CONTEXT, |
| 103 | "interval_end_time": voltha.PmConfig_CONTEXT, |
| 104 | "parent_class_id": voltha.PmConfig_CONTEXT, |
| 105 | "parent_entity_id": voltha.PmConfig_CONTEXT, |
| 106 | "upstream": voltha.PmConfig_CONTEXT, |
| 107 | |
| 108 | "drop_events": voltha.PmConfig_COUNTER, |
| 109 | "octets": voltha.PmConfig_COUNTER, |
| 110 | "packets": voltha.PmConfig_COUNTER, |
| 111 | "broadcast_packets": voltha.PmConfig_COUNTER, |
| 112 | "multicast_packets": voltha.PmConfig_COUNTER, |
| 113 | "crc_errored_packets": voltha.PmConfig_COUNTER, |
| 114 | "undersize_packets": voltha.PmConfig_COUNTER, |
| 115 | "oversize_packets": voltha.PmConfig_COUNTER, |
| 116 | "64_octets": voltha.PmConfig_COUNTER, |
| 117 | "65_to_127_octets": voltha.PmConfig_COUNTER, |
| 118 | "128_to_255_octets": voltha.PmConfig_COUNTER, |
| 119 | "256_to_511_octets": voltha.PmConfig_COUNTER, |
| 120 | "512_to_1023_octets": voltha.PmConfig_COUNTER, |
| 121 | "1024_to_1518_octets": voltha.PmConfig_COUNTER, |
| 122 | } |
| 123 | |
| 124 | // EthernetUniHistory are supported ethernet uni history counters fetched from |
| 125 | // Ethernet Performance Monitoring History Data ME. |
| 126 | var EthernetUniHistory = map[string]voltha.PmConfig_PmType{ |
| 127 | "class_id": voltha.PmConfig_CONTEXT, |
| 128 | "entity_id": voltha.PmConfig_CONTEXT, |
| 129 | "interval_end_time": voltha.PmConfig_CONTEXT, |
| 130 | |
| 131 | "fcs_errors": voltha.PmConfig_COUNTER, |
| 132 | "excessive_collision_counter": voltha.PmConfig_COUNTER, |
| 133 | "late_collision_counter": voltha.PmConfig_COUNTER, |
| 134 | "frames_too_long": voltha.PmConfig_COUNTER, |
| 135 | "buffer_overflows_on_rx": voltha.PmConfig_COUNTER, |
| 136 | "buffer_overflows_on_tx": voltha.PmConfig_COUNTER, |
| 137 | "single_collision_frame_counter": voltha.PmConfig_COUNTER, |
| 138 | "multiple_collisions_frame_counter": voltha.PmConfig_COUNTER, |
| 139 | "sqe_counter": voltha.PmConfig_COUNTER, |
| 140 | "deferred_tx_counter": voltha.PmConfig_COUNTER, |
| 141 | "internal_mac_tx_error_counter": voltha.PmConfig_COUNTER, |
| 142 | "carrier_sense_error_counter": voltha.PmConfig_COUNTER, |
| 143 | "alignment_error_counter": voltha.PmConfig_COUNTER, |
| 144 | "internal_mac_rx_error_counter": voltha.PmConfig_COUNTER, |
| 145 | } |
| 146 | |
Girish Gowdra | 5c5aaf4 | 2021-02-17 19:40:50 -0800 | [diff] [blame] | 147 | // FecHistory is supported FEC Performance Monitoring History Data related metrics |
| 148 | var FecHistory = map[string]voltha.PmConfig_PmType{ |
| 149 | "class_id": voltha.PmConfig_CONTEXT, |
| 150 | "entity_id": voltha.PmConfig_CONTEXT, |
| 151 | "interval_end_time": voltha.PmConfig_CONTEXT, |
| 152 | |
| 153 | "corrected_bytes": voltha.PmConfig_COUNTER, |
| 154 | "corrected_code_words": voltha.PmConfig_COUNTER, |
| 155 | "uncorrectable_code_words": voltha.PmConfig_COUNTER, |
| 156 | "total_code_words": voltha.PmConfig_COUNTER, |
| 157 | "fec_seconds": voltha.PmConfig_COUNTER, |
| 158 | } |
| 159 | |
| 160 | // GemPortHistory is supported GEM Port Network Ctp Performance Monitoring History Data |
| 161 | // related metrics |
| 162 | var GemPortHistory = map[string]voltha.PmConfig_PmType{ |
| 163 | "class_id": voltha.PmConfig_CONTEXT, |
| 164 | "entity_id": voltha.PmConfig_CONTEXT, |
| 165 | "interval_end_time": voltha.PmConfig_CONTEXT, |
| 166 | |
| 167 | "transmitted_gem_frames": voltha.PmConfig_COUNTER, |
| 168 | "received_gem_frames": voltha.PmConfig_COUNTER, |
| 169 | "received_payload_bytes": voltha.PmConfig_COUNTER, |
| 170 | "transmitted_payload_bytes": voltha.PmConfig_COUNTER, |
| 171 | "encryption_key_errors": voltha.PmConfig_COUNTER, |
| 172 | } |
| 173 | |
Girish Gowdra | e0140f0 | 2021-02-02 16:55:09 -0800 | [diff] [blame] | 174 | // Constants specific for L2 PM collection |
| 175 | const ( |
| 176 | L2PmCollectionInterval = 15 * 60 // Unit in seconds. Do not change this as this fixed by OMCI specification for L2 PM counters |
| 177 | SyncTimeRetryInterval = 15 // Unit seconds |
| 178 | L2PmCreateAttempts = 3 |
Girish Gowdra | 36ccf7d | 2021-02-25 20:42:51 -0800 | [diff] [blame] | 179 | L2PmDeleteAttempts = 3 |
Girish Gowdra | e0140f0 | 2021-02-02 16:55:09 -0800 | [diff] [blame] | 180 | L2PmCollectAttempts = 3 |
Girish Gowdra | 453750f | 2021-02-16 16:36:46 -0800 | [diff] [blame] | 181 | // Per Table 11.2.9-1 – OMCI baseline message limitations in G.988 spec, the max GET Response |
| 182 | // payload size is 25. We define 24 (one less) to allow for dynamic insertion of IntervalEndTime |
| 183 | // attribute (1 byte) in L2 PM GET Requests. |
| 184 | MaxL2PMGetPayLoadSize = 24 |
Girish Gowdra | e0140f0 | 2021-02-02 16:55:09 -0800 | [diff] [blame] | 185 | ) |
| 186 | |
| 187 | // EthernetUniHistoryName specific constants |
| 188 | const ( |
| 189 | EthernetBridgeHistoryName = "Ethernet_Bridge_Port_History" |
| 190 | EthernetBridgeHistoryEnabled = true // This setting can be changed from voltha NBI PmConfig configuration |
| 191 | EthernetBridgeHistoryFrequency = L2PmCollectionInterval |
| 192 | ) |
| 193 | |
| 194 | // EthernetBridgeHistory specific constants |
| 195 | const ( |
| 196 | EthernetUniHistoryName = "Ethernet_UNI_History" |
| 197 | EthernetUniHistoryEnabled = true // This setting can be changed from voltha NBI PmConfig configuration |
| 198 | EthernetUniHistoryFrequency = L2PmCollectionInterval |
| 199 | ) |
| 200 | |
Girish Gowdra | 5c5aaf4 | 2021-02-17 19:40:50 -0800 | [diff] [blame] | 201 | // FecHistory specific constants |
| 202 | const ( |
| 203 | FecHistoryName = "FEC_History" |
| 204 | FecHistoryEnabled = true // This setting can be changed from voltha NBI PmConfig configuration |
| 205 | FecHistoryFrequency = L2PmCollectionInterval |
| 206 | ) |
| 207 | |
| 208 | // GemPortHistory specific constants |
| 209 | const ( |
| 210 | GemPortHistoryName = "GEM_Port_History" |
| 211 | GemPortHistoryEnabled = true // This setting can be changed from voltha NBI PmConfig configuration |
| 212 | GemPortHistoryFrequency = L2PmCollectionInterval |
| 213 | ) |
| 214 | |
Girish Gowdra | 0e53364 | 2021-03-02 22:02:51 -0800 | [diff] [blame] | 215 | // KV Store related constants |
| 216 | const ( |
| 217 | cPmKvStorePrefix = "%s/openonu/pm-data/%s" // <some-base-path>/openonu/pm-data/<onu-device-id> |
| 218 | cPmAdd = "add" |
| 219 | cPmAdded = "added" |
| 220 | cPmRemove = "remove" |
| 221 | cPmRemoved = "removed" |
| 222 | ) |
| 223 | |
Girish Gowdra | 5c5aaf4 | 2021-02-17 19:40:50 -0800 | [diff] [blame] | 224 | // Defines the type for generic metric population function |
| 225 | type groupMetricPopulateFunc func(context.Context, me.ClassID, uint16, me.AttributeValueMap, me.AttributeValueMap, map[string]float32, *int) error |
| 226 | |
Girish Gowdra | e0140f0 | 2021-02-02 16:55:09 -0800 | [diff] [blame] | 227 | // *** Classical L2 PM Counters end *** |
| 228 | |
Girish Gowdra | 0e53364 | 2021-03-02 22:02:51 -0800 | [diff] [blame] | 229 | type pmMEData struct { |
| 230 | InstancesActive []uint16 `json:"instances_active"` // list of active ME instance IDs for the group |
| 231 | InstancesToDelete []uint16 `json:"instances_to_delete"` // list of ME instance IDs marked for deletion for the group |
| 232 | InstancesToAdd []uint16 `json:"instances_to_add"` // list of ME instance IDs marked for addition for the group |
| 233 | } |
| 234 | |
Girish Gowdra | 5a7c492 | 2021-01-22 18:33:41 -0800 | [diff] [blame] | 235 | type groupMetric struct { |
| 236 | groupName string |
| 237 | enabled bool |
| 238 | frequency uint32 // valid only if FrequencyOverride is enabled. |
| 239 | metricMap map[string]voltha.PmConfig_PmType |
| 240 | nextCollectionInterval time.Time // valid only if FrequencyOverride is enabled. |
Girish Gowdra | e0140f0 | 2021-02-02 16:55:09 -0800 | [diff] [blame] | 241 | isL2PMCounter bool // true for only L2 PM counters |
| 242 | collectAttempts uint32 // number of attempts to collect L2 PM data |
Girish Gowdra | 0e53364 | 2021-03-02 22:02:51 -0800 | [diff] [blame] | 243 | pmMEData *pmMEData |
Girish Gowdra | 5a7c492 | 2021-01-22 18:33:41 -0800 | [diff] [blame] | 244 | } |
| 245 | |
| 246 | type standaloneMetric struct { |
| 247 | metricName string |
| 248 | enabled bool |
| 249 | frequency uint32 // valid only if FrequencyOverride is enabled. |
| 250 | nextCollectionInterval time.Time // valid only if FrequencyOverride is enabled. |
| 251 | } |
| 252 | |
Girish Gowdra | e09a620 | 2021-01-12 18:10:59 -0800 | [diff] [blame] | 253 | type onuMetricsManager struct { |
| 254 | pDeviceHandler *deviceHandler |
Girish Gowdra | e0140f0 | 2021-02-02 16:55:09 -0800 | [diff] [blame] | 255 | pAdaptFsm *AdapterFsm |
Girish Gowdra | e09a620 | 2021-01-12 18:10:59 -0800 | [diff] [blame] | 256 | |
Girish Gowdra | e0140f0 | 2021-02-02 16:55:09 -0800 | [diff] [blame] | 257 | opticalMetricsChan chan me.AttributeValueMap |
| 258 | uniStatusMetricsChan chan me.AttributeValueMap |
| 259 | l2PmChan chan me.AttributeValueMap |
| 260 | syncTimeResponseChan chan bool // true is success, false is fail |
| 261 | l2PmCreateOrDeleteResponseChan chan bool // true is success, false is fail |
| 262 | |
| 263 | activeL2Pms []string // list of active l2 pm MEs created on the ONU. |
| 264 | l2PmToDelete []string // list of L2 PMs to delete |
| 265 | l2PmToAdd []string // list of L2 PM to add |
Girish Gowdra | e09a620 | 2021-01-12 18:10:59 -0800 | [diff] [blame] | 266 | |
Girish Gowdra | 5a7c492 | 2021-01-22 18:33:41 -0800 | [diff] [blame] | 267 | groupMetricMap map[string]*groupMetric |
| 268 | standaloneMetricMap map[string]*standaloneMetric |
| 269 | |
Girish Gowdra | e09a620 | 2021-01-12 18:10:59 -0800 | [diff] [blame] | 270 | stopProcessingOmciResponses chan bool |
Girish Gowdra | 5a7c492 | 2021-01-22 18:33:41 -0800 | [diff] [blame] | 271 | |
Girish Gowdra | e0140f0 | 2021-02-02 16:55:09 -0800 | [diff] [blame] | 272 | stopTicks chan bool |
| 273 | |
Girish Gowdra | 5a7c492 | 2021-01-22 18:33:41 -0800 | [diff] [blame] | 274 | nextGlobalMetricCollectionTime time.Time // valid only if pmConfig.FreqOverride is set to false. |
| 275 | |
| 276 | onuMetricsManagerLock sync.RWMutex |
Girish Gowdra | 0e53364 | 2021-03-02 22:02:51 -0800 | [diff] [blame] | 277 | |
| 278 | pmKvStore *db.Backend |
Girish Gowdra | e09a620 | 2021-01-12 18:10:59 -0800 | [diff] [blame] | 279 | } |
| 280 | |
| 281 | // newonuMetricsManager returns a new instance of the newonuMetricsManager |
Girish Gowdra | 0e53364 | 2021-03-02 22:02:51 -0800 | [diff] [blame] | 282 | // The metrics manager module is responsible for configuration and management of individual and group metrics. |
| 283 | // Currently all the metrics are managed as a group which fall into two categories - L2 PM and "all others" |
| 284 | // The L2 PM counters have a fixed 15min interval for PM collection while all other group counters have |
| 285 | // the collection interval configurable. |
| 286 | // The global PM config is part of the voltha.Device struct and is backed up on KV store (by rw-core). |
| 287 | // This module also implements resiliency for L2 PM ME instances that are active/pending-delete/pending-add. |
Girish Gowdra | e09a620 | 2021-01-12 18:10:59 -0800 | [diff] [blame] | 288 | func newonuMetricsManager(ctx context.Context, dh *deviceHandler) *onuMetricsManager { |
| 289 | |
| 290 | var metricsManager onuMetricsManager |
| 291 | logger.Debugw(ctx, "init-onuMetricsManager", log.Fields{"device-id": dh.deviceID}) |
| 292 | metricsManager.pDeviceHandler = dh |
| 293 | |
Girish Gowdra | e0140f0 | 2021-02-02 16:55:09 -0800 | [diff] [blame] | 294 | commMetricsChan := make(chan Message) |
Girish Gowdra | e09a620 | 2021-01-12 18:10:59 -0800 | [diff] [blame] | 295 | metricsManager.opticalMetricsChan = make(chan me.AttributeValueMap) |
| 296 | metricsManager.uniStatusMetricsChan = make(chan me.AttributeValueMap) |
Girish Gowdra | e0140f0 | 2021-02-02 16:55:09 -0800 | [diff] [blame] | 297 | metricsManager.l2PmChan = make(chan me.AttributeValueMap) |
| 298 | |
| 299 | metricsManager.syncTimeResponseChan = make(chan bool) |
| 300 | metricsManager.l2PmCreateOrDeleteResponseChan = make(chan bool) |
| 301 | |
Girish Gowdra | e09a620 | 2021-01-12 18:10:59 -0800 | [diff] [blame] | 302 | metricsManager.stopProcessingOmciResponses = make(chan bool) |
Girish Gowdra | e0140f0 | 2021-02-02 16:55:09 -0800 | [diff] [blame] | 303 | metricsManager.stopTicks = make(chan bool) |
Girish Gowdra | e09a620 | 2021-01-12 18:10:59 -0800 | [diff] [blame] | 304 | |
Girish Gowdra | 5a7c492 | 2021-01-22 18:33:41 -0800 | [diff] [blame] | 305 | metricsManager.groupMetricMap = make(map[string]*groupMetric) |
| 306 | metricsManager.standaloneMetricMap = make(map[string]*standaloneMetric) |
| 307 | |
| 308 | if dh.pmConfigs == nil { // dh.pmConfigs is NOT nil if adapter comes back from a restart. We should NOT go back to defaults in this case |
Girish Gowdra | 5c5aaf4 | 2021-02-17 19:40:50 -0800 | [diff] [blame] | 309 | metricsManager.initializeAllGroupMetrics() |
Girish Gowdra | 5a7c492 | 2021-01-22 18:33:41 -0800 | [diff] [blame] | 310 | } |
| 311 | |
Girish Gowdra | 5c5aaf4 | 2021-02-17 19:40:50 -0800 | [diff] [blame] | 312 | metricsManager.populateLocalGroupMetricData(ctx) |
Girish Gowdra | 5a7c492 | 2021-01-22 18:33:41 -0800 | [diff] [blame] | 313 | |
Girish Gowdra | 5c5aaf4 | 2021-02-17 19:40:50 -0800 | [diff] [blame] | 314 | if err := metricsManager.initializeL2PmFsm(ctx, commMetricsChan); err != nil { |
Girish Gowdra | e0140f0 | 2021-02-02 16:55:09 -0800 | [diff] [blame] | 315 | return nil |
| 316 | } |
Girish Gowdra | e0140f0 | 2021-02-02 16:55:09 -0800 | [diff] [blame] | 317 | |
Girish Gowdra | 5a7c492 | 2021-01-22 18:33:41 -0800 | [diff] [blame] | 318 | // initialize the next metric collection intervals. |
| 319 | metricsManager.initializeMetricCollectionTime(ctx) |
Girish Gowdra | 0e53364 | 2021-03-02 22:02:51 -0800 | [diff] [blame] | 320 | |
| 321 | baseKvStorePath := fmt.Sprintf(cPmKvStorePrefix, dh.pOpenOnuAc.cm.Backend.PathPrefix, dh.deviceID) |
| 322 | metricsManager.pmKvStore = dh.setBackend(ctx, baseKvStorePath) |
| 323 | if metricsManager.pmKvStore == nil { |
| 324 | logger.Errorw(ctx, "Can't initialize pmKvStore - no backend connection to PM module", |
| 325 | log.Fields{"device-id": dh.deviceID, "service": baseKvStorePath}) |
| 326 | return nil |
| 327 | } |
| 328 | |
Girish Gowdra | 5a7c492 | 2021-01-22 18:33:41 -0800 | [diff] [blame] | 329 | logger.Info(ctx, "init-onuMetricsManager completed", log.Fields{"device-id": dh.deviceID}) |
Girish Gowdra | e09a620 | 2021-01-12 18:10:59 -0800 | [diff] [blame] | 330 | return &metricsManager |
| 331 | } |
| 332 | |
Girish Gowdra | 5a7c492 | 2021-01-22 18:33:41 -0800 | [diff] [blame] | 333 | func (mm *onuMetricsManager) initializeMetricCollectionTime(ctx context.Context) { |
| 334 | if mm.pDeviceHandler.pmConfigs.FreqOverride { |
| 335 | // If mm.pDeviceHandler.pmConfigs.FreqOverride is set to true, then group/standalone metric specific interval applies |
| 336 | mm.onuMetricsManagerLock.Lock() |
| 337 | defer mm.onuMetricsManagerLock.Unlock() |
| 338 | for _, v := range mm.groupMetricMap { |
Girish Gowdra | e0140f0 | 2021-02-02 16:55:09 -0800 | [diff] [blame] | 339 | if v.enabled && !v.isL2PMCounter { // L2 PM counter collection is managed in a L2PmFsm |
Girish Gowdra | 5a7c492 | 2021-01-22 18:33:41 -0800 | [diff] [blame] | 340 | v.nextCollectionInterval = time.Now().Add(time.Duration(v.frequency) * time.Second) |
| 341 | } |
| 342 | } |
| 343 | |
| 344 | for _, v := range mm.standaloneMetricMap { |
| 345 | if v.enabled { |
| 346 | v.nextCollectionInterval = time.Now().Add(time.Duration(v.frequency) * time.Second) |
| 347 | } |
| 348 | } |
| 349 | } else { |
| 350 | // If mm.pDeviceHandler.pmConfigs.FreqOverride is set to false, then overall metric specific interval applies |
| 351 | mm.nextGlobalMetricCollectionTime = time.Now().Add(time.Duration(mm.pDeviceHandler.pmConfigs.DefaultFreq) * time.Second) |
| 352 | } |
| 353 | logger.Infow(ctx, "initialized standalone group/metric collection time", log.Fields{"device-id": mm.pDeviceHandler.deviceID}) |
| 354 | } |
| 355 | |
| 356 | func (mm *onuMetricsManager) updateDefaultFrequency(ctx context.Context, pmConfigs *voltha.PmConfigs) error { |
| 357 | // Verify that the configured DefaultFrequency is > 0 and is a multiple of FrequencyGranularity |
Girish Gowdra | af0ad63 | 2021-01-27 13:00:01 -0800 | [diff] [blame] | 358 | if pmConfigs.DefaultFreq == 0 || (pmConfigs.DefaultFreq > 0 && pmConfigs.DefaultFreq%FrequencyGranularity != 0) { |
Girish Gowdra | 5a7c492 | 2021-01-22 18:33:41 -0800 | [diff] [blame] | 359 | logger.Errorf(ctx, "frequency-%u-should-be-a-multiple-of-%u", pmConfigs.DefaultFreq, FrequencyGranularity) |
| 360 | return fmt.Errorf("frequency-%d-should-be-a-multiple-of-%d", pmConfigs.DefaultFreq, FrequencyGranularity) |
| 361 | } |
| 362 | mm.pDeviceHandler.pmConfigs.DefaultFreq = pmConfigs.DefaultFreq |
| 363 | // re-set the nextGlobalMetricCollectionTime based on the new DefaultFreq |
| 364 | mm.nextGlobalMetricCollectionTime = time.Now().Add(time.Duration(mm.pDeviceHandler.pmConfigs.DefaultFreq) * time.Second) |
| 365 | logger.Debugw(ctx, "frequency-updated--new-frequency", log.Fields{"device-id": mm.pDeviceHandler.deviceID, "frequency": mm.pDeviceHandler.pmConfigs.DefaultFreq}) |
| 366 | return nil |
| 367 | } |
| 368 | |
| 369 | func (mm *onuMetricsManager) updateGroupFreq(ctx context.Context, aGroupName string, pmConfigs *voltha.PmConfigs) error { |
| 370 | var newGroupFreq uint32 |
| 371 | found := false |
| 372 | groupSliceIdx := 0 |
| 373 | var group *voltha.PmGroupConfig |
| 374 | for groupSliceIdx, group = range pmConfigs.Groups { |
| 375 | if group.GroupName == aGroupName { |
Girish Gowdra | af0ad63 | 2021-01-27 13:00:01 -0800 | [diff] [blame] | 376 | // freq 0 is not allowed and it should be multiple of FrequencyGranularity |
| 377 | if group.GroupFreq == 0 || (group.GroupFreq > 0 && group.GroupFreq%FrequencyGranularity != 0) { |
| 378 | logger.Errorf(ctx, "frequency-%u-should-be-a-multiple-of-%u", group.GroupFreq, FrequencyGranularity) |
| 379 | return fmt.Errorf("frequency-%d-should-be-a-multiple-of-%d", group.GroupFreq, FrequencyGranularity) |
Girish Gowdra | 5a7c492 | 2021-01-22 18:33:41 -0800 | [diff] [blame] | 380 | } |
Girish Gowdra | af0ad63 | 2021-01-27 13:00:01 -0800 | [diff] [blame] | 381 | newGroupFreq = group.GroupFreq |
| 382 | found = true |
| 383 | break |
Girish Gowdra | 5a7c492 | 2021-01-22 18:33:41 -0800 | [diff] [blame] | 384 | } |
| 385 | } |
| 386 | // if not found update group freq and next collection interval for the group |
| 387 | if !found { |
| 388 | logger.Errorw(ctx, "group name not found", log.Fields{"device-id": mm.pDeviceHandler.deviceID, "groupName": aGroupName}) |
| 389 | return fmt.Errorf("group-name-not-found-%v", aGroupName) |
| 390 | } |
| 391 | |
| 392 | updated := false |
| 393 | mm.onuMetricsManagerLock.Lock() |
| 394 | defer mm.onuMetricsManagerLock.Unlock() |
| 395 | for k, v := range mm.groupMetricMap { |
Girish Gowdra | e0140f0 | 2021-02-02 16:55:09 -0800 | [diff] [blame] | 396 | if k == aGroupName && !v.isL2PMCounter { // We cannot allow the L2 PM counter frequency to be updated. It is 15min fixed by OMCI spec |
Girish Gowdra | 5a7c492 | 2021-01-22 18:33:41 -0800 | [diff] [blame] | 397 | v.frequency = newGroupFreq |
| 398 | // update internal pm config |
| 399 | mm.pDeviceHandler.pmConfigs.Groups[groupSliceIdx].GroupFreq = newGroupFreq |
| 400 | // Also updated the next group metric collection time from now |
| 401 | v.nextCollectionInterval = time.Now().Add(time.Duration(newGroupFreq) * time.Second) |
| 402 | updated = true |
| 403 | logger.Infow(ctx, "group frequency updated", log.Fields{"device-id": mm.pDeviceHandler.deviceID, "newGroupFreq": newGroupFreq, "groupName": aGroupName}) |
Girish Gowdra | 36ccf7d | 2021-02-25 20:42:51 -0800 | [diff] [blame] | 404 | break |
Girish Gowdra | 5a7c492 | 2021-01-22 18:33:41 -0800 | [diff] [blame] | 405 | } |
| 406 | } |
| 407 | if !updated { |
| 408 | logger.Errorw(ctx, "group frequency not updated", log.Fields{"device-id": mm.pDeviceHandler.deviceID, "newGroupFreq": newGroupFreq, "groupName": aGroupName}) |
| 409 | return fmt.Errorf("internal-error-during-group-freq-update--groupname-%s-freq-%d", aGroupName, newGroupFreq) |
| 410 | } |
| 411 | return nil |
| 412 | } |
| 413 | |
| 414 | func (mm *onuMetricsManager) updateMetricFreq(ctx context.Context, aMetricName string, pmConfigs *voltha.PmConfigs) error { |
| 415 | var newMetricFreq uint32 |
| 416 | found := false |
| 417 | metricSliceIdx := 0 |
| 418 | var metric *voltha.PmConfig |
| 419 | for metricSliceIdx, metric = range pmConfigs.Metrics { |
| 420 | if metric.Name == aMetricName { |
Girish Gowdra | af0ad63 | 2021-01-27 13:00:01 -0800 | [diff] [blame] | 421 | // freq 0 is not allowed and it should be multiple of FrequencyGranularity |
| 422 | if metric.SampleFreq == 0 || (metric.SampleFreq > 0 && metric.SampleFreq%FrequencyGranularity != 0) { |
| 423 | logger.Errorf(ctx, "frequency-%u-should-be-a-multiple-of-%u", metric.SampleFreq, FrequencyGranularity) |
| 424 | return fmt.Errorf("frequency-%d-should-be-a-multiple-of-%d", metric.SampleFreq, FrequencyGranularity) |
Girish Gowdra | 5a7c492 | 2021-01-22 18:33:41 -0800 | [diff] [blame] | 425 | } |
Girish Gowdra | af0ad63 | 2021-01-27 13:00:01 -0800 | [diff] [blame] | 426 | newMetricFreq = metric.SampleFreq |
| 427 | found = true |
| 428 | break |
Girish Gowdra | 5a7c492 | 2021-01-22 18:33:41 -0800 | [diff] [blame] | 429 | } |
| 430 | } |
| 431 | if !found { |
| 432 | logger.Errorw(ctx, "metric name not found", log.Fields{"device-id": mm.pDeviceHandler.deviceID, "metricName": aMetricName}) |
| 433 | return fmt.Errorf("metric-name-not-found-%v", aMetricName) |
| 434 | } |
| 435 | |
| 436 | updated := false |
| 437 | mm.onuMetricsManagerLock.Lock() |
| 438 | defer mm.onuMetricsManagerLock.Unlock() |
| 439 | for k, v := range mm.groupMetricMap { |
Girish Gowdra | af0ad63 | 2021-01-27 13:00:01 -0800 | [diff] [blame] | 440 | if k == aMetricName { |
Girish Gowdra | 5a7c492 | 2021-01-22 18:33:41 -0800 | [diff] [blame] | 441 | v.frequency = newMetricFreq |
| 442 | // update internal pm config |
| 443 | mm.pDeviceHandler.pmConfigs.Metrics[metricSliceIdx].SampleFreq = newMetricFreq |
| 444 | // Also updated the next standalone metric collection time from now |
| 445 | v.nextCollectionInterval = time.Now().Add(time.Duration(newMetricFreq) * time.Second) |
| 446 | updated = true |
| 447 | logger.Infow(ctx, "metric frequency updated", log.Fields{"device-id": mm.pDeviceHandler.deviceID, "newMetricFreq": newMetricFreq, "aMetricName": aMetricName}) |
Girish Gowdra | 36ccf7d | 2021-02-25 20:42:51 -0800 | [diff] [blame] | 448 | break |
Girish Gowdra | 5a7c492 | 2021-01-22 18:33:41 -0800 | [diff] [blame] | 449 | } |
| 450 | } |
| 451 | if !updated { |
| 452 | logger.Errorw(ctx, "metric frequency not updated", log.Fields{"device-id": mm.pDeviceHandler.deviceID, "newMetricFreq": newMetricFreq, "aMetricName": aMetricName}) |
| 453 | return fmt.Errorf("internal-error-during-standalone-metric-update--matricnane-%s-freq-%d", aMetricName, newMetricFreq) |
| 454 | } |
| 455 | return nil |
| 456 | } |
| 457 | |
| 458 | func (mm *onuMetricsManager) updateGroupSupport(ctx context.Context, aGroupName string, pmConfigs *voltha.PmConfigs) error { |
| 459 | groupSliceIdx := 0 |
| 460 | var group *voltha.PmGroupConfig |
| 461 | |
| 462 | for groupSliceIdx, group = range pmConfigs.Groups { |
| 463 | if group.GroupName == aGroupName { |
| 464 | break |
| 465 | } |
| 466 | } |
| 467 | if group == nil { |
| 468 | logger.Errorw(ctx, "group metric not found", log.Fields{"device-id": mm.pDeviceHandler.deviceID, "groupName": aGroupName}) |
| 469 | return fmt.Errorf("group-not-found--groupName-%s", aGroupName) |
| 470 | } |
| 471 | |
| 472 | updated := false |
| 473 | mm.onuMetricsManagerLock.Lock() |
| 474 | defer mm.onuMetricsManagerLock.Unlock() |
| 475 | for k, v := range mm.groupMetricMap { |
| 476 | if k == aGroupName && v.enabled != group.Enabled { |
| 477 | mm.pDeviceHandler.pmConfigs.Groups[groupSliceIdx].Enabled = group.Enabled |
| 478 | v.enabled = group.Enabled |
Girish Gowdra | e0140f0 | 2021-02-02 16:55:09 -0800 | [diff] [blame] | 479 | if group.Enabled { |
| 480 | if v.isL2PMCounter { |
| 481 | // If it is a L2 PM counter we need to mark the PM to be added |
Girish Gowdra | 5c5aaf4 | 2021-02-17 19:40:50 -0800 | [diff] [blame] | 482 | mm.l2PmToAdd = mm.appendIfMissingString(mm.l2PmToAdd, v.groupName) |
Girish Gowdra | e0140f0 | 2021-02-02 16:55:09 -0800 | [diff] [blame] | 483 | // If the group support flag toggles too soon, we need to delete the group name from l2PmToDelete slice |
Girish Gowdra | 5c5aaf4 | 2021-02-17 19:40:50 -0800 | [diff] [blame] | 484 | mm.l2PmToDelete = mm.removeIfFoundString(mm.l2PmToDelete, v.groupName) |
| 485 | |
| 486 | // The GemPortHistory group requires some special handling as the instance IDs are not pre-defined |
| 487 | // unlike other L2 PM counters. We need to fetch the active gemport instance IDs in the system to |
| 488 | // take further action |
| 489 | if v.groupName == GemPortHistoryName { |
Girish Gowdra | 36ccf7d | 2021-02-25 20:42:51 -0800 | [diff] [blame] | 490 | mm.updateGemPortNTPInstanceToAddForPerfMonitoring(ctx) |
Girish Gowdra | 5c5aaf4 | 2021-02-17 19:40:50 -0800 | [diff] [blame] | 491 | } |
Girish Gowdra | e0140f0 | 2021-02-02 16:55:09 -0800 | [diff] [blame] | 492 | } else if mm.pDeviceHandler.pmConfigs.FreqOverride { // otherwise just update the next collection interval |
| 493 | v.nextCollectionInterval = time.Now().Add(time.Duration(v.frequency) * time.Second) |
| 494 | } |
| 495 | } else { // group counter is disabled |
| 496 | if v.isL2PMCounter { |
| 497 | // If it is a L2 PM counter we need to mark the PM to be deleted |
Girish Gowdra | 5c5aaf4 | 2021-02-17 19:40:50 -0800 | [diff] [blame] | 498 | mm.l2PmToDelete = mm.appendIfMissingString(mm.l2PmToDelete, v.groupName) |
Girish Gowdra | e0140f0 | 2021-02-02 16:55:09 -0800 | [diff] [blame] | 499 | // If the group support flag toggles too soon, we need to delete the group name from l2PmToAdd slice |
Girish Gowdra | 5c5aaf4 | 2021-02-17 19:40:50 -0800 | [diff] [blame] | 500 | mm.l2PmToAdd = mm.removeIfFoundString(mm.l2PmToAdd, v.groupName) |
| 501 | |
| 502 | // The GemPortHistory group requires some special handling as the instance IDs are not pre-defined |
| 503 | // unlike other L2 PM counters. We need to fetch the active gemport instance IDs in the system to |
| 504 | // take further action |
| 505 | if v.groupName == GemPortHistoryName { |
Girish Gowdra | 36ccf7d | 2021-02-25 20:42:51 -0800 | [diff] [blame] | 506 | mm.updateGemPortNTPInstanceToDeleteForPerfMonitoring(ctx) |
Girish Gowdra | 5c5aaf4 | 2021-02-17 19:40:50 -0800 | [diff] [blame] | 507 | } |
Girish Gowdra | e0140f0 | 2021-02-02 16:55:09 -0800 | [diff] [blame] | 508 | } |
Girish Gowdra | 5a7c492 | 2021-01-22 18:33:41 -0800 | [diff] [blame] | 509 | } |
| 510 | updated = true |
Girish Gowdra | e0140f0 | 2021-02-02 16:55:09 -0800 | [diff] [blame] | 511 | if v.isL2PMCounter { |
| 512 | logger.Infow(ctx, "l2 pm group metric support updated", |
| 513 | log.Fields{"device-id": mm.pDeviceHandler.deviceID, "groupName": aGroupName, "enabled": group.Enabled, "l2PmToAdd": mm.l2PmToAdd, "l2PmToDelete": mm.l2PmToDelete}) |
| 514 | } else { |
| 515 | logger.Infow(ctx, "group metric support updated", log.Fields{"device-id": mm.pDeviceHandler.deviceID, "groupName": aGroupName, "enabled": group.Enabled}) |
| 516 | } |
| 517 | break |
Girish Gowdra | 5a7c492 | 2021-01-22 18:33:41 -0800 | [diff] [blame] | 518 | } |
| 519 | } |
| 520 | |
| 521 | if !updated { |
| 522 | logger.Errorw(ctx, "group metric support not updated", log.Fields{"device-id": mm.pDeviceHandler.deviceID, "groupName": aGroupName}) |
| 523 | return fmt.Errorf("internal-error-during-group-support-update--groupName-%s", aGroupName) |
| 524 | } |
| 525 | return nil |
| 526 | } |
| 527 | |
| 528 | func (mm *onuMetricsManager) updateMetricSupport(ctx context.Context, aMetricName string, pmConfigs *voltha.PmConfigs) error { |
| 529 | metricSliceIdx := 0 |
| 530 | var metric *voltha.PmConfig |
| 531 | |
| 532 | for metricSliceIdx, metric = range pmConfigs.Metrics { |
| 533 | if metric.Name == aMetricName { |
| 534 | break |
| 535 | } |
| 536 | } |
| 537 | |
| 538 | if metric == nil { |
| 539 | logger.Errorw(ctx, "standalone metric not found", log.Fields{"device-id": mm.pDeviceHandler.deviceID, "metricName": aMetricName}) |
| 540 | return fmt.Errorf("metric-not-found--metricname-%s", aMetricName) |
| 541 | } |
| 542 | |
| 543 | updated := false |
| 544 | mm.onuMetricsManagerLock.Lock() |
| 545 | defer mm.onuMetricsManagerLock.Unlock() |
| 546 | for k, v := range mm.standaloneMetricMap { |
| 547 | if k == aMetricName && v.enabled != metric.Enabled { |
| 548 | mm.pDeviceHandler.pmConfigs.Metrics[metricSliceIdx].Enabled = metric.Enabled |
| 549 | v.enabled = metric.Enabled |
| 550 | // If the standalone metric is now enabled and frequency override is enabled, set the next metric collection time |
| 551 | if metric.Enabled && mm.pDeviceHandler.pmConfigs.FreqOverride { |
| 552 | v.nextCollectionInterval = time.Now().Add(time.Duration(v.frequency) * time.Second) |
| 553 | } |
| 554 | updated = true |
| 555 | logger.Infow(ctx, "standalone metric support updated", log.Fields{"device-id": mm.pDeviceHandler.deviceID, "metricName": aMetricName, "enabled": metric.Enabled}) |
Girish Gowdra | 36ccf7d | 2021-02-25 20:42:51 -0800 | [diff] [blame] | 556 | break |
Girish Gowdra | 5a7c492 | 2021-01-22 18:33:41 -0800 | [diff] [blame] | 557 | } |
| 558 | } |
| 559 | if !updated { |
| 560 | logger.Errorw(ctx, "standalone metric support not updated", log.Fields{"device-id": mm.pDeviceHandler.deviceID, "metricName": aMetricName}) |
| 561 | return fmt.Errorf("internal-error-during-standalone-support-update--metricname-%s", aMetricName) |
| 562 | } |
| 563 | return nil |
| 564 | } |
| 565 | |
| 566 | func (mm *onuMetricsManager) collectAllGroupAndStandaloneMetrics(ctx context.Context) { |
| 567 | if mm.pDeviceHandler.pmConfigs.Grouped { // metrics are managed as a group. |
| 568 | go mm.collectAllGroupMetrics(ctx) |
| 569 | } else { |
| 570 | go mm.collectAllStandaloneMetrics(ctx) |
| 571 | } |
| 572 | } |
| 573 | |
| 574 | func (mm *onuMetricsManager) collectAllGroupMetrics(ctx context.Context) { |
| 575 | go func() { |
| 576 | logger.Debug(ctx, "startCollector before collecting optical metrics") |
| 577 | metricInfo := mm.collectOpticalMetrics(ctx) |
| 578 | if metricInfo != nil { |
| 579 | mm.publishMetrics(ctx, metricInfo) |
| 580 | } |
| 581 | }() |
| 582 | |
| 583 | go func() { |
| 584 | logger.Debug(ctx, "startCollector before collecting uni metrics") |
| 585 | metricInfo := mm.collectUniStatusMetrics(ctx) |
| 586 | if metricInfo != nil { |
| 587 | mm.publishMetrics(ctx, metricInfo) |
| 588 | } |
| 589 | }() |
| 590 | |
| 591 | // Add more here |
| 592 | } |
| 593 | |
| 594 | func (mm *onuMetricsManager) collectAllStandaloneMetrics(ctx context.Context) { |
| 595 | // None exists as of now, add when available here |
| 596 | } |
| 597 | |
| 598 | func (mm *onuMetricsManager) collectGroupMetric(ctx context.Context, groupName string) { |
| 599 | switch groupName { |
| 600 | case OpticalPowerGroupMetricName: |
| 601 | go func() { |
| 602 | if mi := mm.collectOpticalMetrics(ctx); mm != nil { |
| 603 | mm.publishMetrics(ctx, mi) |
| 604 | } |
| 605 | }() |
| 606 | case UniStatusGroupMetricName: |
| 607 | go func() { |
| 608 | if mi := mm.collectUniStatusMetrics(ctx); mm != nil { |
| 609 | mm.publishMetrics(ctx, mi) |
| 610 | } |
| 611 | }() |
| 612 | default: |
| 613 | logger.Errorw(ctx, "unhandled group metric name", log.Fields{"device-id": mm.pDeviceHandler.deviceID, "groupName": groupName}) |
| 614 | } |
| 615 | } |
| 616 | |
| 617 | func (mm *onuMetricsManager) collectStandaloneMetric(ctx context.Context, metricName string) { |
| 618 | switch metricName { |
| 619 | // None exist as of now, add when available |
| 620 | default: |
| 621 | logger.Errorw(ctx, "unhandled standalone metric name", log.Fields{"device-id": mm.pDeviceHandler.deviceID, "metricName": metricName}) |
| 622 | } |
| 623 | } |
| 624 | |
| 625 | // collectOpticalMetrics collects groups metrics related to optical power from ani-g ME. |
Girish Gowdra | e09a620 | 2021-01-12 18:10:59 -0800 | [diff] [blame] | 626 | func (mm *onuMetricsManager) collectOpticalMetrics(ctx context.Context) []*voltha.MetricInformation { |
| 627 | logger.Debugw(ctx, "collectOpticalMetrics", log.Fields{"device-id": mm.pDeviceHandler.deviceID}) |
Girish Gowdra | 5a7c492 | 2021-01-22 18:33:41 -0800 | [diff] [blame] | 628 | |
| 629 | mm.onuMetricsManagerLock.RLock() |
| 630 | if !mm.groupMetricMap[OpticalPowerGroupMetricName].enabled { |
| 631 | mm.onuMetricsManagerLock.RUnlock() |
| 632 | logger.Debugw(ctx, "optical power group metric is not enabled", log.Fields{"device-id": mm.pDeviceHandler.deviceID}) |
| 633 | return nil |
| 634 | } |
| 635 | mm.onuMetricsManagerLock.RUnlock() |
| 636 | |
Girish Gowdra | e09a620 | 2021-01-12 18:10:59 -0800 | [diff] [blame] | 637 | var metricInfoSlice []*voltha.MetricInformation |
| 638 | metricsContext := make(map[string]string) |
| 639 | metricsContext["onuID"] = fmt.Sprintf("%d", mm.pDeviceHandler.device.ProxyAddress.OnuId) |
| 640 | metricsContext["intfID"] = fmt.Sprintf("%d", mm.pDeviceHandler.device.ProxyAddress.ChannelId) |
| 641 | metricsContext["devicetype"] = mm.pDeviceHandler.DeviceType |
| 642 | |
Girish Gowdra | 5a7c492 | 2021-01-22 18:33:41 -0800 | [diff] [blame] | 643 | raisedTs := time.Now().Unix() |
Girish Gowdra | e09a620 | 2021-01-12 18:10:59 -0800 | [diff] [blame] | 644 | mmd := voltha.MetricMetaData{ |
Girish Gowdra | 5a7c492 | 2021-01-22 18:33:41 -0800 | [diff] [blame] | 645 | Title: OpticalPowerGroupMetricName, |
Girish Gowdra | e09a620 | 2021-01-12 18:10:59 -0800 | [diff] [blame] | 646 | Ts: float64(raisedTs), |
| 647 | Context: metricsContext, |
| 648 | DeviceId: mm.pDeviceHandler.deviceID, |
| 649 | LogicalDeviceId: mm.pDeviceHandler.logicalDeviceID, |
| 650 | SerialNo: mm.pDeviceHandler.device.SerialNumber, |
| 651 | } |
| 652 | |
Girish Gowdra | e09a620 | 2021-01-12 18:10:59 -0800 | [diff] [blame] | 653 | // get the ANI-G instance IDs |
| 654 | anigInstKeys := mm.pDeviceHandler.pOnuOmciDevice.pOnuDB.getSortedInstKeys(ctx, me.AniGClassID) |
| 655 | loop: |
| 656 | for _, anigInstID := range anigInstKeys { |
| 657 | var meAttributes me.AttributeValueMap |
| 658 | opticalMetrics := make(map[string]float32) |
| 659 | // Get the ANI-G instance optical power attributes |
| 660 | requestedAttributes := me.AttributeValueMap{"OpticalSignalLevel": 0, "TransmitOpticalLevel": 0} |
Girish Gowdra | e0140f0 | 2021-02-02 16:55:09 -0800 | [diff] [blame] | 661 | if meInstance := mm.pDeviceHandler.pOnuOmciDevice.PDevOmciCC.sendGetMe(ctx, me.AniGClassID, anigInstID, requestedAttributes, ConstDefaultOmciTimeout, true, mm.pAdaptFsm.commChan); meInstance != nil { |
Girish Gowdra | e09a620 | 2021-01-12 18:10:59 -0800 | [diff] [blame] | 662 | select { |
| 663 | case meAttributes = <-mm.opticalMetricsChan: |
| 664 | logger.Debugw(ctx, "received optical metrics", log.Fields{"device-id": mm.pDeviceHandler.deviceID}) |
| 665 | case <-time.After(time.Duration(ConstDefaultOmciTimeout) * time.Second): |
Girish Gowdra | 36ccf7d | 2021-02-25 20:42:51 -0800 | [diff] [blame] | 666 | logger.Errorw(ctx, "timeout waiting for omci-get response for optical metrics", log.Fields{"device-id": mm.pDeviceHandler.deviceID}) |
Girish Gowdra | e09a620 | 2021-01-12 18:10:59 -0800 | [diff] [blame] | 667 | // The metrics will be empty in this case |
| 668 | break loop |
| 669 | } |
| 670 | // Populate metric only if it was enabled. |
Girish Gowdra | 5a7c492 | 2021-01-22 18:33:41 -0800 | [diff] [blame] | 671 | for k := range OpticalPowerGroupMetrics { |
| 672 | switch k { |
| 673 | case "ani_g_instance_id": |
| 674 | if val, ok := meAttributes["ManagedEntityId"]; ok && val != nil { |
| 675 | opticalMetrics[k] = float32(val.(uint16)) |
| 676 | } |
Girish Gowdra | e09a620 | 2021-01-12 18:10:59 -0800 | [diff] [blame] | 677 | case "transmit_power": |
Girish Gowdra | 5a7c492 | 2021-01-22 18:33:41 -0800 | [diff] [blame] | 678 | if val, ok := meAttributes["TransmitOpticalLevel"]; ok && val != nil { |
| 679 | opticalMetrics[k] = float32(val.(uint16)) |
| 680 | } |
Girish Gowdra | e09a620 | 2021-01-12 18:10:59 -0800 | [diff] [blame] | 681 | case "receive_power": |
Girish Gowdra | 5a7c492 | 2021-01-22 18:33:41 -0800 | [diff] [blame] | 682 | if val, ok := meAttributes["OpticalSignalLevel"]; ok && val != nil { |
| 683 | opticalMetrics[k] = float32(val.(uint16)) |
| 684 | } |
Girish Gowdra | e09a620 | 2021-01-12 18:10:59 -0800 | [diff] [blame] | 685 | default: |
| 686 | // do nothing |
| 687 | } |
| 688 | } |
| 689 | } |
| 690 | // create slice of metrics given that there could be more than one ANI-G instance and |
| 691 | // optical metrics are collected per ANI-G instance |
| 692 | metricInfo := voltha.MetricInformation{Metadata: &mmd, Metrics: opticalMetrics} |
| 693 | metricInfoSlice = append(metricInfoSlice, &metricInfo) |
| 694 | } |
| 695 | |
| 696 | return metricInfoSlice |
| 697 | } |
| 698 | |
Girish Gowdra | 5a7c492 | 2021-01-22 18:33:41 -0800 | [diff] [blame] | 699 | // collectUniStatusMetrics collects UNI status group metric from various MEs (uni-g, pptp and veip). |
Girish Gowdra | e09a620 | 2021-01-12 18:10:59 -0800 | [diff] [blame] | 700 | // nolint: gocyclo |
| 701 | func (mm *onuMetricsManager) collectUniStatusMetrics(ctx context.Context) []*voltha.MetricInformation { |
| 702 | logger.Debugw(ctx, "collectUniStatusMetrics", log.Fields{"device-id": mm.pDeviceHandler.deviceID}) |
Girish Gowdra | 5a7c492 | 2021-01-22 18:33:41 -0800 | [diff] [blame] | 703 | mm.onuMetricsManagerLock.RLock() |
| 704 | if !mm.groupMetricMap[UniStatusGroupMetricName].enabled { |
| 705 | mm.onuMetricsManagerLock.RUnlock() |
| 706 | logger.Debugw(ctx, "uni status group metric is not enabled", log.Fields{"device-id": mm.pDeviceHandler.deviceID}) |
| 707 | return nil |
| 708 | } |
| 709 | mm.onuMetricsManagerLock.RUnlock() |
| 710 | |
Girish Gowdra | e09a620 | 2021-01-12 18:10:59 -0800 | [diff] [blame] | 711 | var metricInfoSlice []*voltha.MetricInformation |
| 712 | metricsContext := make(map[string]string) |
| 713 | metricsContext["onuID"] = fmt.Sprintf("%d", mm.pDeviceHandler.device.ProxyAddress.OnuId) |
| 714 | metricsContext["intfID"] = fmt.Sprintf("%d", mm.pDeviceHandler.device.ProxyAddress.ChannelId) |
| 715 | metricsContext["devicetype"] = mm.pDeviceHandler.DeviceType |
| 716 | |
Girish Gowdra | 5a7c492 | 2021-01-22 18:33:41 -0800 | [diff] [blame] | 717 | raisedTs := time.Now().Unix() |
Girish Gowdra | e09a620 | 2021-01-12 18:10:59 -0800 | [diff] [blame] | 718 | mmd := voltha.MetricMetaData{ |
| 719 | Title: "UniStatus", // Is this ok to hard code? |
| 720 | Ts: float64(raisedTs), |
| 721 | Context: metricsContext, |
| 722 | DeviceId: mm.pDeviceHandler.deviceID, |
| 723 | LogicalDeviceId: mm.pDeviceHandler.logicalDeviceID, |
| 724 | SerialNo: mm.pDeviceHandler.device.SerialNumber, |
| 725 | } |
| 726 | |
Girish Gowdra | e09a620 | 2021-01-12 18:10:59 -0800 | [diff] [blame] | 727 | // get the UNI-G instance IDs |
| 728 | unigInstKeys := mm.pDeviceHandler.pOnuOmciDevice.pOnuDB.getSortedInstKeys(ctx, me.UniGClassID) |
| 729 | loop1: |
| 730 | for _, unigInstID := range unigInstKeys { |
| 731 | // TODO: Include additional information in the voltha.MetricMetaData - like portno, uni-id, instance-id |
| 732 | // to uniquely identify this ME instance and also to correlate the ME instance to physical instance |
| 733 | unigMetrics := make(map[string]float32) |
| 734 | var meAttributes me.AttributeValueMap |
| 735 | // Get the UNI-G instance optical power attributes |
| 736 | requestedAttributes := me.AttributeValueMap{"AdministrativeState": 0} |
Girish Gowdra | e0140f0 | 2021-02-02 16:55:09 -0800 | [diff] [blame] | 737 | if meInstance := mm.pDeviceHandler.pOnuOmciDevice.PDevOmciCC.sendGetMe(ctx, me.UniGClassID, unigInstID, requestedAttributes, ConstDefaultOmciTimeout, true, mm.pAdaptFsm.commChan); meInstance != nil { |
Girish Gowdra | e09a620 | 2021-01-12 18:10:59 -0800 | [diff] [blame] | 738 | // Wait for metrics or timeout |
| 739 | select { |
| 740 | case meAttributes = <-mm.uniStatusMetricsChan: |
| 741 | logger.Debugw(ctx, "received uni-g metrics", log.Fields{"device-id": mm.pDeviceHandler.deviceID}) |
| 742 | case <-time.After(time.Duration(ConstDefaultOmciTimeout) * time.Second): |
| 743 | logger.Errorw(ctx, "timeout waiting for omci-get response for uni status", log.Fields{"device-id": mm.pDeviceHandler.deviceID}) |
| 744 | // The metrics could be empty in this case |
| 745 | break loop1 |
| 746 | } |
| 747 | // Populate metric only if it was enabled. |
Girish Gowdra | 5a7c492 | 2021-01-22 18:33:41 -0800 | [diff] [blame] | 748 | for k := range UniStatusGroupMetrics { |
| 749 | switch k { |
Girish Gowdra | e09a620 | 2021-01-12 18:10:59 -0800 | [diff] [blame] | 750 | case "uni_admin_state": |
Girish Gowdra | 5a7c492 | 2021-01-22 18:33:41 -0800 | [diff] [blame] | 751 | if val, ok := meAttributes["AdministrativeState"]; ok && val != nil { |
| 752 | unigMetrics[k] = float32(val.(byte)) |
| 753 | } |
Girish Gowdra | e09a620 | 2021-01-12 18:10:59 -0800 | [diff] [blame] | 754 | default: |
| 755 | // do nothing |
| 756 | } |
| 757 | } |
Girish Gowdra | 5a7c492 | 2021-01-22 18:33:41 -0800 | [diff] [blame] | 758 | if val, ok := meAttributes["ManagedEntityId"]; ok && val != nil { |
Girish Gowdra | 0e53364 | 2021-03-02 22:02:51 -0800 | [diff] [blame] | 759 | entityID := val.(uint16) |
| 760 | unigMetrics["entity_id"] = float32(entityID) |
| 761 | // TODO: Rlock needed for reading uniEntityMap? May not be needed given uniEntityMap is populated setup at initial ONU bring up |
| 762 | for _, uni := range mm.pDeviceHandler.uniEntityMap { |
| 763 | if uni.entityID == entityID { |
| 764 | unigMetrics["uni_port_no"] = float32(uni.portNo) |
| 765 | } |
| 766 | } |
Girish Gowdra | 5a7c492 | 2021-01-22 18:33:41 -0800 | [diff] [blame] | 767 | } |
Girish Gowdra | 0e53364 | 2021-03-02 22:02:51 -0800 | [diff] [blame] | 768 | |
Girish Gowdra | e09a620 | 2021-01-12 18:10:59 -0800 | [diff] [blame] | 769 | // create slice of metrics given that there could be more than one UNI-G instance |
| 770 | metricInfo := voltha.MetricInformation{Metadata: &mmd, Metrics: unigMetrics} |
| 771 | metricInfoSlice = append(metricInfoSlice, &metricInfo) |
| 772 | } |
| 773 | } |
| 774 | |
| 775 | // get the PPTP instance IDs |
| 776 | pptpInstKeys := mm.pDeviceHandler.pOnuOmciDevice.pOnuDB.getSortedInstKeys(ctx, me.PhysicalPathTerminationPointEthernetUniClassID) |
| 777 | loop2: |
| 778 | for _, pptpInstID := range pptpInstKeys { |
| 779 | // TODO: Include additional information in the voltha.MetricMetaData - like portno, uni-id, instance-id |
| 780 | // to uniquely identify this ME instance and also to correlate the ME instance to physical instance |
| 781 | var meAttributes me.AttributeValueMap |
| 782 | pptpMetrics := make(map[string]float32) |
| 783 | |
| 784 | requestedAttributes := me.AttributeValueMap{"SensedType": 0, "OperationalState": 0, "AdministrativeState": 0} |
Girish Gowdra | e0140f0 | 2021-02-02 16:55:09 -0800 | [diff] [blame] | 785 | if meInstance := mm.pDeviceHandler.pOnuOmciDevice.PDevOmciCC.sendGetMe(ctx, me.PhysicalPathTerminationPointEthernetUniClassID, pptpInstID, requestedAttributes, ConstDefaultOmciTimeout, true, mm.pAdaptFsm.commChan); meInstance != nil { |
Girish Gowdra | e09a620 | 2021-01-12 18:10:59 -0800 | [diff] [blame] | 786 | // Wait for metrics or timeout |
| 787 | select { |
| 788 | case meAttributes = <-mm.uniStatusMetricsChan: |
| 789 | logger.Debugw(ctx, "received pptp metrics", log.Fields{"device-id": mm.pDeviceHandler.deviceID}) |
| 790 | case <-time.After(time.Duration(ConstDefaultOmciTimeout) * time.Second): |
| 791 | logger.Errorw(ctx, "timeout waiting for omci-get response for uni status", log.Fields{"device-id": mm.pDeviceHandler.deviceID}) |
| 792 | // The metrics could be empty in this case |
| 793 | break loop2 |
| 794 | } |
| 795 | |
| 796 | // Populate metric only if it was enabled. |
Girish Gowdra | 5a7c492 | 2021-01-22 18:33:41 -0800 | [diff] [blame] | 797 | for k := range UniStatusGroupMetrics { |
| 798 | switch k { |
Girish Gowdra | e09a620 | 2021-01-12 18:10:59 -0800 | [diff] [blame] | 799 | case "ethernet_type": |
Girish Gowdra | 5a7c492 | 2021-01-22 18:33:41 -0800 | [diff] [blame] | 800 | if val, ok := meAttributes["SensedType"]; ok && val != nil { |
| 801 | pptpMetrics[k] = float32(val.(byte)) |
| 802 | } |
Girish Gowdra | e09a620 | 2021-01-12 18:10:59 -0800 | [diff] [blame] | 803 | case "oper_status": |
Girish Gowdra | 5a7c492 | 2021-01-22 18:33:41 -0800 | [diff] [blame] | 804 | if val, ok := meAttributes["OperationalState"]; ok && val != nil { |
| 805 | pptpMetrics[k] = float32(val.(byte)) |
| 806 | } |
Girish Gowdra | e09a620 | 2021-01-12 18:10:59 -0800 | [diff] [blame] | 807 | case "uni_admin_state": |
Girish Gowdra | 5a7c492 | 2021-01-22 18:33:41 -0800 | [diff] [blame] | 808 | if val, ok := meAttributes["AdministrativeState"]; ok && val != nil { |
| 809 | pptpMetrics[k] = float32(val.(byte)) |
| 810 | } |
Girish Gowdra | e09a620 | 2021-01-12 18:10:59 -0800 | [diff] [blame] | 811 | default: |
| 812 | // do nothing |
| 813 | } |
| 814 | } |
| 815 | } |
Girish Gowdra | 5a7c492 | 2021-01-22 18:33:41 -0800 | [diff] [blame] | 816 | if val, ok := meAttributes["ManagedEntityId"]; ok && val != nil { |
Girish Gowdra | 0e53364 | 2021-03-02 22:02:51 -0800 | [diff] [blame] | 817 | entityID := val.(uint16) |
| 818 | pptpMetrics["entity_id"] = float32(entityID) |
| 819 | // TODO: Rlock needed for reading uniEntityMap? May not be needed given uniEntityMap is populated setup at initial ONU bring up |
| 820 | for _, uni := range mm.pDeviceHandler.uniEntityMap { |
| 821 | if uni.entityID == entityID { |
| 822 | pptpMetrics["uni_port_no"] = float32(uni.portNo) |
| 823 | } |
| 824 | } |
Girish Gowdra | 5a7c492 | 2021-01-22 18:33:41 -0800 | [diff] [blame] | 825 | } |
| 826 | |
Girish Gowdra | e09a620 | 2021-01-12 18:10:59 -0800 | [diff] [blame] | 827 | // create slice of metrics given that there could be more than one PPTP instance and |
| 828 | metricInfo := voltha.MetricInformation{Metadata: &mmd, Metrics: pptpMetrics} |
| 829 | metricInfoSlice = append(metricInfoSlice, &metricInfo) |
| 830 | } |
| 831 | |
| 832 | // get the VEIP instance IDs |
| 833 | veipInstKeys := mm.pDeviceHandler.pOnuOmciDevice.pOnuDB.getSortedInstKeys(ctx, me.VirtualEthernetInterfacePointClassID) |
| 834 | loop3: |
| 835 | for _, veipInstID := range veipInstKeys { |
| 836 | // TODO: Include additional information in the voltha.MetricMetaData - like portno, uni-id, instance-id |
| 837 | // to uniquely identify this ME instance and also to correlate the ME instance to physical instance |
| 838 | var meAttributes me.AttributeValueMap |
Girish Gowdra | 5a7c492 | 2021-01-22 18:33:41 -0800 | [diff] [blame] | 839 | veipMetrics := make(map[string]float32) |
Girish Gowdra | e09a620 | 2021-01-12 18:10:59 -0800 | [diff] [blame] | 840 | |
| 841 | requestedAttributes := me.AttributeValueMap{"OperationalState": 0, "AdministrativeState": 0} |
Girish Gowdra | e0140f0 | 2021-02-02 16:55:09 -0800 | [diff] [blame] | 842 | if meInstance := mm.pDeviceHandler.pOnuOmciDevice.PDevOmciCC.sendGetMe(ctx, me.VirtualEthernetInterfacePointClassID, veipInstID, requestedAttributes, ConstDefaultOmciTimeout, true, mm.pAdaptFsm.commChan); meInstance != nil { |
Girish Gowdra | e09a620 | 2021-01-12 18:10:59 -0800 | [diff] [blame] | 843 | // Wait for metrics or timeout |
| 844 | select { |
| 845 | case meAttributes = <-mm.uniStatusMetricsChan: |
| 846 | logger.Debugw(ctx, "received veip metrics", log.Fields{"device-id": mm.pDeviceHandler.deviceID}) |
| 847 | case <-time.After(time.Duration(ConstDefaultOmciTimeout) * time.Second): |
| 848 | logger.Errorw(ctx, "timeout waiting for omci-get response for uni status", log.Fields{"device-id": mm.pDeviceHandler.deviceID}) |
| 849 | // The metrics could be empty in this case |
| 850 | break loop3 |
| 851 | } |
| 852 | |
| 853 | // Populate metric only if it was enabled. |
Girish Gowdra | 5a7c492 | 2021-01-22 18:33:41 -0800 | [diff] [blame] | 854 | for k := range UniStatusGroupMetrics { |
| 855 | switch k { |
Girish Gowdra | e09a620 | 2021-01-12 18:10:59 -0800 | [diff] [blame] | 856 | case "oper_status": |
Girish Gowdra | 5a7c492 | 2021-01-22 18:33:41 -0800 | [diff] [blame] | 857 | if val, ok := meAttributes["OperationalState"]; ok && val != nil { |
| 858 | veipMetrics[k] = float32(val.(byte)) |
| 859 | } |
Girish Gowdra | e09a620 | 2021-01-12 18:10:59 -0800 | [diff] [blame] | 860 | case "uni_admin_state": |
Girish Gowdra | 5a7c492 | 2021-01-22 18:33:41 -0800 | [diff] [blame] | 861 | if val, ok := meAttributes["AdministrativeState"]; ok && val != nil { |
| 862 | veipMetrics[k] = float32(val.(byte)) |
| 863 | } |
Girish Gowdra | e09a620 | 2021-01-12 18:10:59 -0800 | [diff] [blame] | 864 | default: |
| 865 | // do nothing |
| 866 | } |
| 867 | } |
| 868 | } |
Girish Gowdra | 5a7c492 | 2021-01-22 18:33:41 -0800 | [diff] [blame] | 869 | |
Girish Gowdra | 5a7c492 | 2021-01-22 18:33:41 -0800 | [diff] [blame] | 870 | if val, ok := meAttributes["ManagedEntityId"]; ok && val != nil { |
Girish Gowdra | 0e53364 | 2021-03-02 22:02:51 -0800 | [diff] [blame] | 871 | entityID := val.(uint16) |
| 872 | veipMetrics["entity_id"] = float32(entityID) |
| 873 | // TODO: Rlock needed for reading uniEntityMap? May not be needed given uniEntityMap is populated setup at initial ONU bring up |
| 874 | for _, uni := range mm.pDeviceHandler.uniEntityMap { |
| 875 | if uni.entityID == entityID { |
| 876 | veipMetrics["uni_port_no"] = float32(uni.portNo) |
| 877 | } |
| 878 | } |
Girish Gowdra | 5a7c492 | 2021-01-22 18:33:41 -0800 | [diff] [blame] | 879 | } |
| 880 | |
Girish Gowdra | e09a620 | 2021-01-12 18:10:59 -0800 | [diff] [blame] | 881 | // create slice of metrics given that there could be more than one VEIP instance |
Girish Gowdra | 5a7c492 | 2021-01-22 18:33:41 -0800 | [diff] [blame] | 882 | metricInfo := voltha.MetricInformation{Metadata: &mmd, Metrics: veipMetrics} |
Girish Gowdra | e09a620 | 2021-01-12 18:10:59 -0800 | [diff] [blame] | 883 | metricInfoSlice = append(metricInfoSlice, &metricInfo) |
| 884 | } |
| 885 | |
| 886 | return metricInfoSlice |
| 887 | } |
| 888 | |
| 889 | // publishMetrics publishes the metrics on kafka |
| 890 | func (mm *onuMetricsManager) publishMetrics(ctx context.Context, metricInfo []*voltha.MetricInformation) { |
| 891 | var ke voltha.KpiEvent2 |
Girish Gowdra | 5a7c492 | 2021-01-22 18:33:41 -0800 | [diff] [blame] | 892 | ts := time.Now().Unix() |
Girish Gowdra | e09a620 | 2021-01-12 18:10:59 -0800 | [diff] [blame] | 893 | ke.SliceData = metricInfo |
| 894 | ke.Type = voltha.KpiEventType_slice |
| 895 | ke.Ts = float64(ts) |
| 896 | |
| 897 | if err := mm.pDeviceHandler.EventProxy.SendKpiEvent(ctx, "STATS_EVENT", &ke, voltha.EventCategory_EQUIPMENT, voltha.EventSubCategory_ONU, ts); err != nil { |
| 898 | logger.Errorw(ctx, "failed-to-send-pon-stats", log.Fields{"err": err}) |
| 899 | } |
| 900 | } |
| 901 | |
| 902 | func (mm *onuMetricsManager) processOmciMessages(ctx context.Context) { |
| 903 | logger.Infow(ctx, "Start routine to process OMCI-GET messages for device-id", log.Fields{"device-id": mm.pDeviceHandler.deviceID}) |
| 904 | // Flush metric collection channels to be safe. |
| 905 | // It is possible that there is stale data on this channel if the processOmciMessages routine |
| 906 | // is stopped right after issuing a OMCI-GET request and started again. |
| 907 | // The processOmciMessages routine will get stopped if startCollector routine (in device_handler.go) |
| 908 | // is stopped - as a result of ONU going down. |
| 909 | mm.flushMetricCollectionChannels(ctx) |
| 910 | |
| 911 | for { |
| 912 | select { |
| 913 | case <-mm.stopProcessingOmciResponses: // stop this routine |
| 914 | logger.Infow(ctx, "Stop routine to process OMCI-GET messages for device-id", log.Fields{"device-id": mm.pDeviceHandler.deviceID}) |
| 915 | return |
Girish Gowdra | e0140f0 | 2021-02-02 16:55:09 -0800 | [diff] [blame] | 916 | case message, ok := <-mm.pAdaptFsm.commChan: |
Girish Gowdra | e09a620 | 2021-01-12 18:10:59 -0800 | [diff] [blame] | 917 | if !ok { |
| 918 | logger.Errorw(ctx, "Message couldn't be read from channel", log.Fields{"device-id": mm.pDeviceHandler.deviceID}) |
| 919 | continue |
| 920 | } |
| 921 | logger.Debugw(ctx, "Received message on ONU metrics channel", log.Fields{"device-id": mm.pDeviceHandler.deviceID}) |
| 922 | |
| 923 | switch message.Type { |
| 924 | case OMCI: |
| 925 | msg, _ := message.Data.(OmciMessage) |
| 926 | mm.handleOmciMessage(ctx, msg) |
| 927 | default: |
| 928 | logger.Warn(ctx, "Unknown message type received", log.Fields{"device-id": mm.pDeviceHandler.deviceID, "message.Type": message.Type}) |
| 929 | } |
| 930 | } |
| 931 | } |
| 932 | } |
| 933 | |
| 934 | func (mm *onuMetricsManager) handleOmciMessage(ctx context.Context, msg OmciMessage) { |
| 935 | logger.Debugw(ctx, "omci Msg", log.Fields{"device-id": mm.pDeviceHandler.deviceID, |
| 936 | "msgType": msg.OmciMsg.MessageType, "msg": msg}) |
| 937 | switch msg.OmciMsg.MessageType { |
| 938 | case omci.GetResponseType: |
| 939 | //TODO: error handling |
| 940 | _ = mm.handleOmciGetResponseMessage(ctx, msg) |
Girish Gowdra | e0140f0 | 2021-02-02 16:55:09 -0800 | [diff] [blame] | 941 | case omci.SynchronizeTimeResponseType: |
| 942 | _ = mm.handleOmciSynchronizeTimeResponseMessage(ctx, msg) |
| 943 | case omci.CreateResponseType: |
| 944 | _ = mm.handleOmciCreateResponseMessage(ctx, msg) |
| 945 | case omci.DeleteResponseType: |
| 946 | _ = mm.handleOmciDeleteResponseMessage(ctx, msg) |
Girish Gowdra | e09a620 | 2021-01-12 18:10:59 -0800 | [diff] [blame] | 947 | default: |
| 948 | logger.Warnw(ctx, "Unknown Message Type", log.Fields{"msgType": msg.OmciMsg.MessageType}) |
| 949 | |
| 950 | } |
| 951 | } |
| 952 | |
| 953 | func (mm *onuMetricsManager) handleOmciGetResponseMessage(ctx context.Context, msg OmciMessage) error { |
| 954 | msgLayer := (*msg.OmciPacket).Layer(omci.LayerTypeGetResponse) |
| 955 | if msgLayer == nil { |
| 956 | logger.Errorw(ctx, "omci Msg layer could not be detected for GetResponse - handling stopped", log.Fields{"device-id": mm.pDeviceHandler.deviceID}) |
| 957 | return fmt.Errorf("omci Msg layer could not be detected for GetResponse - handling stopped: %s", mm.pDeviceHandler.deviceID) |
| 958 | } |
| 959 | msgObj, msgOk := msgLayer.(*omci.GetResponse) |
| 960 | if !msgOk { |
| 961 | logger.Errorw(ctx, "omci Msg layer could not be assigned for GetResponse - handling stopped", log.Fields{"device-id": mm.pDeviceHandler.deviceID}) |
| 962 | return fmt.Errorf("omci Msg layer could not be assigned for GetResponse - handling stopped: %s", mm.pDeviceHandler.deviceID) |
| 963 | } |
| 964 | logger.Debugw(ctx, "OMCI GetResponse Data", log.Fields{"device-id": mm.pDeviceHandler.deviceID, "data-fields": msgObj}) |
| 965 | if msgObj.Result == me.Success { |
| 966 | meAttributes := msgObj.Attributes |
| 967 | switch msgObj.EntityClass { |
| 968 | case me.AniGClassID: |
| 969 | mm.opticalMetricsChan <- meAttributes |
| 970 | return nil |
| 971 | case me.UniGClassID: |
| 972 | mm.uniStatusMetricsChan <- meAttributes |
| 973 | return nil |
| 974 | case me.PhysicalPathTerminationPointEthernetUniClassID: |
| 975 | mm.uniStatusMetricsChan <- meAttributes |
| 976 | return nil |
| 977 | case me.VirtualEthernetInterfacePointClassID: |
| 978 | mm.uniStatusMetricsChan <- meAttributes |
| 979 | return nil |
Girish Gowdra | e0140f0 | 2021-02-02 16:55:09 -0800 | [diff] [blame] | 980 | case me.EthernetFramePerformanceMonitoringHistoryDataUpstreamClassID, |
| 981 | me.EthernetFramePerformanceMonitoringHistoryDataDownstreamClassID, |
Girish Gowdra | 5c5aaf4 | 2021-02-17 19:40:50 -0800 | [diff] [blame] | 982 | me.EthernetPerformanceMonitoringHistoryDataClassID, |
| 983 | me.FecPerformanceMonitoringHistoryDataClassID, |
| 984 | me.GemPortNetworkCtpPerformanceMonitoringHistoryDataClassID: |
Girish Gowdra | e0140f0 | 2021-02-02 16:55:09 -0800 | [diff] [blame] | 985 | mm.l2PmChan <- meAttributes |
Girish Gowdra | e09a620 | 2021-01-12 18:10:59 -0800 | [diff] [blame] | 986 | default: |
| 987 | logger.Errorw(ctx, "unhandled omci get response message", |
| 988 | log.Fields{"device-id": mm.pDeviceHandler.deviceID, "class-id": msgObj.EntityClass}) |
| 989 | } |
| 990 | } |
| 991 | |
Girish Gowdra | e0140f0 | 2021-02-02 16:55:09 -0800 | [diff] [blame] | 992 | return fmt.Errorf("unhandled-omci-get-response-message") |
| 993 | } |
| 994 | |
| 995 | func (mm *onuMetricsManager) handleOmciSynchronizeTimeResponseMessage(ctx context.Context, msg OmciMessage) error { |
| 996 | msgLayer := (*msg.OmciPacket).Layer(omci.LayerTypeSynchronizeTimeResponse) |
| 997 | if msgLayer == nil { |
| 998 | logger.Errorw(ctx, "omci Msg layer could not be detected for synchronize time response - handling stopped", log.Fields{"device-id": mm.pDeviceHandler.deviceID}) |
| 999 | return fmt.Errorf("omci Msg layer could not be detected for synchronize time response - handling stopped: %s", mm.pDeviceHandler.deviceID) |
| 1000 | } |
| 1001 | msgObj, msgOk := msgLayer.(*omci.SynchronizeTimeResponse) |
| 1002 | if !msgOk { |
| 1003 | logger.Errorw(ctx, "omci Msg layer could not be assigned for synchronize time response - handling stopped", log.Fields{"device-id": mm.pDeviceHandler.deviceID}) |
| 1004 | return fmt.Errorf("omci Msg layer could not be assigned for synchronize time response - handling stopped: %s", mm.pDeviceHandler.deviceID) |
| 1005 | } |
| 1006 | logger.Debugw(ctx, "OMCI synchronize time response Data", log.Fields{"device-id": mm.pDeviceHandler.deviceID, "data-fields": msgObj}) |
| 1007 | if msgObj.Result == me.Success { |
| 1008 | switch msgObj.EntityClass { |
| 1009 | case me.OnuGClassID: |
| 1010 | logger.Infow(ctx, "omci synchronize time success", log.Fields{"device-id": mm.pDeviceHandler.deviceID}) |
| 1011 | mm.syncTimeResponseChan <- true |
| 1012 | return nil |
| 1013 | default: |
| 1014 | logger.Errorw(ctx, "unhandled omci message", |
| 1015 | log.Fields{"device-id": mm.pDeviceHandler.deviceID, "class-id": msgObj.EntityClass}) |
| 1016 | } |
| 1017 | } |
| 1018 | mm.syncTimeResponseChan <- false |
| 1019 | logger.Errorf(ctx, "unhandled-omci-synchronize-time-response-message--error-code-%v", msgObj.Result) |
| 1020 | return fmt.Errorf("unhandled-omci-synchronize-time-response-message--error-code-%v", msgObj.Result) |
Girish Gowdra | e09a620 | 2021-01-12 18:10:59 -0800 | [diff] [blame] | 1021 | } |
| 1022 | |
| 1023 | // flushMetricCollectionChannels flushes all metric collection channels for any stale OMCI responses |
| 1024 | func (mm *onuMetricsManager) flushMetricCollectionChannels(ctx context.Context) { |
| 1025 | // flush commMetricsChan |
| 1026 | select { |
Girish Gowdra | e0140f0 | 2021-02-02 16:55:09 -0800 | [diff] [blame] | 1027 | case <-mm.pAdaptFsm.commChan: |
Girish Gowdra | e09a620 | 2021-01-12 18:10:59 -0800 | [diff] [blame] | 1028 | logger.Debug(ctx, "flushed common metrics channel") |
| 1029 | default: |
| 1030 | } |
| 1031 | |
| 1032 | // flush opticalMetricsChan |
| 1033 | select { |
| 1034 | case <-mm.opticalMetricsChan: |
| 1035 | logger.Debug(ctx, "flushed optical metrics channel") |
| 1036 | default: |
| 1037 | } |
| 1038 | |
| 1039 | // flush uniStatusMetricsChan |
| 1040 | select { |
| 1041 | case <-mm.uniStatusMetricsChan: |
| 1042 | logger.Debug(ctx, "flushed uni status metrics channel") |
| 1043 | default: |
| 1044 | } |
Girish Gowdra | e0140f0 | 2021-02-02 16:55:09 -0800 | [diff] [blame] | 1045 | |
| 1046 | // flush syncTimeResponseChan |
| 1047 | select { |
| 1048 | case <-mm.syncTimeResponseChan: |
| 1049 | logger.Debug(ctx, "flushed sync time response channel") |
| 1050 | default: |
| 1051 | } |
| 1052 | |
| 1053 | // flush l2PmChan |
| 1054 | select { |
| 1055 | case <-mm.l2PmChan: |
| 1056 | logger.Debug(ctx, "flushed L2 PM collection channel") |
| 1057 | default: |
| 1058 | } |
| 1059 | |
| 1060 | // flush stopTicks |
| 1061 | select { |
| 1062 | case <-mm.stopTicks: |
| 1063 | logger.Debug(ctx, "flushed stopTicks channel") |
| 1064 | default: |
| 1065 | } |
| 1066 | |
| 1067 | } |
| 1068 | |
| 1069 | // ** L2 PM FSM Handlers start ** |
| 1070 | |
| 1071 | func (mm *onuMetricsManager) l2PMFsmStarting(ctx context.Context, e *fsm.Event) { |
Girish Gowdra | 0e53364 | 2021-03-02 22:02:51 -0800 | [diff] [blame] | 1072 | // restore data from KV store |
| 1073 | if err := mm.restorePmData(ctx); err != nil { |
| 1074 | logger.Errorw(ctx, "error restoring pm data", log.Fields{"err": err}) |
| 1075 | // we continue given that it does not effect the actual services for the ONU, |
| 1076 | // but there may be some negative effect on PM collection (there may be some mismatch in |
| 1077 | // the actual PM config and what is present on the device). |
| 1078 | } |
| 1079 | |
Girish Gowdra | e0140f0 | 2021-02-02 16:55:09 -0800 | [diff] [blame] | 1080 | // Loop through all the group metrics |
| 1081 | // If it is a L2 PM Interval metric and it is enabled, then if it is not in the |
| 1082 | // list of active L2 PM list then mark it for creation |
| 1083 | // It it is a L2 PM Interval metric and it is disabled, then if it is in the |
| 1084 | // list of active L2 PM list then mark it for deletion |
| 1085 | mm.onuMetricsManagerLock.Lock() |
| 1086 | for n, g := range mm.groupMetricMap { |
| 1087 | if g.isL2PMCounter { // it is a l2 pm counter |
| 1088 | if g.enabled { // metric enabled. |
| 1089 | found := false |
| 1090 | inner1: |
| 1091 | for _, v := range mm.activeL2Pms { |
| 1092 | if v == n { |
| 1093 | found = true // metric already present in active l2 pm list |
| 1094 | break inner1 |
| 1095 | } |
| 1096 | } |
| 1097 | if !found { // metric not in active l2 pm list. Mark this to be added later |
Girish Gowdra | 5c5aaf4 | 2021-02-17 19:40:50 -0800 | [diff] [blame] | 1098 | mm.l2PmToAdd = mm.appendIfMissingString(mm.l2PmToAdd, n) |
Girish Gowdra | e0140f0 | 2021-02-02 16:55:09 -0800 | [diff] [blame] | 1099 | } |
| 1100 | } else { // metric not enabled. |
| 1101 | found := false |
| 1102 | inner2: |
| 1103 | for _, v := range mm.activeL2Pms { |
| 1104 | if v == n { |
| 1105 | found = true // metric is found in active l2 pm list |
| 1106 | break inner2 |
| 1107 | } |
| 1108 | } |
| 1109 | if found { // metric is found in active l2 pm list. Mark this to be deleted later |
Girish Gowdra | 5c5aaf4 | 2021-02-17 19:40:50 -0800 | [diff] [blame] | 1110 | mm.l2PmToDelete = mm.appendIfMissingString(mm.l2PmToDelete, n) |
Girish Gowdra | e0140f0 | 2021-02-02 16:55:09 -0800 | [diff] [blame] | 1111 | } |
| 1112 | } |
| 1113 | } |
| 1114 | } |
| 1115 | mm.onuMetricsManagerLock.Unlock() |
| 1116 | logger.Debugw(ctx, "pms to add and delete", |
| 1117 | log.Fields{"device-id": mm.pDeviceHandler.deviceID, "pms-to-add": mm.l2PmToAdd, "pms-to-delete": mm.l2PmToDelete}) |
| 1118 | go func() { |
| 1119 | // push a tick event to move to next state |
| 1120 | if err := mm.pAdaptFsm.pFsm.Event(l2PmEventTick); err != nil { |
| 1121 | logger.Errorw(ctx, "error calling event", log.Fields{"device-id": mm.pDeviceHandler.deviceID, "err": err}) |
| 1122 | } |
| 1123 | }() |
| 1124 | } |
| 1125 | |
| 1126 | func (mm *onuMetricsManager) l2PMFsmSyncTime(ctx context.Context, e *fsm.Event) { |
| 1127 | // Sync time with the ONU to establish 15min boundary for PM collection. |
| 1128 | if err := mm.syncTime(ctx); err != nil { |
| 1129 | go func() { |
| 1130 | time.Sleep(SyncTimeRetryInterval * time.Second) // retry to sync time after this timeout |
| 1131 | // This will result in FSM attempting to sync time again |
| 1132 | if err := mm.pAdaptFsm.pFsm.Event(l2PmEventFailure); err != nil { |
| 1133 | logger.Errorw(ctx, "error calling event", log.Fields{"device-id": mm.pDeviceHandler.deviceID, "err": err}) |
| 1134 | } |
| 1135 | }() |
| 1136 | } |
| 1137 | // Initiate a tick generation routine every L2PmCollectionInterval |
| 1138 | go mm.generateTicks(ctx) |
| 1139 | |
| 1140 | go func() { |
| 1141 | if err := mm.pAdaptFsm.pFsm.Event(l2PmEventSuccess); err != nil { |
| 1142 | logger.Errorw(ctx, "error calling event", log.Fields{"device-id": mm.pDeviceHandler.deviceID, "err": err}) |
| 1143 | } |
| 1144 | }() |
| 1145 | } |
| 1146 | |
| 1147 | func (mm *onuMetricsManager) l2PMFsmNull(ctx context.Context, e *fsm.Event) { |
| 1148 | // We need to reset the local data so that the L2 PM MEs are re-provisioned once the ONU is back up based on the latest PM CONFIG |
| 1149 | mm.onuMetricsManagerLock.Lock() |
| 1150 | mm.activeL2Pms = nil |
| 1151 | mm.l2PmToAdd = nil |
| 1152 | mm.l2PmToDelete = nil |
| 1153 | mm.onuMetricsManagerLock.Unlock() |
Girish Gowdra | 0e53364 | 2021-03-02 22:02:51 -0800 | [diff] [blame] | 1154 | // If the FSM was stopped, then clear PM data from KV store |
| 1155 | // The FSM is stopped when ONU goes down. It is time to clear its data from store |
| 1156 | if e.Event == l2PmEventStop { |
| 1157 | _ = mm.clearPmGroupData(ctx) // ignore error |
| 1158 | } |
| 1159 | |
Girish Gowdra | e0140f0 | 2021-02-02 16:55:09 -0800 | [diff] [blame] | 1160 | } |
| 1161 | func (mm *onuMetricsManager) l2PMFsmIdle(ctx context.Context, e *fsm.Event) { |
| 1162 | logger.Debugw(ctx, "Enter state idle", log.Fields{"device-id": mm.pDeviceHandler.deviceID}) |
| 1163 | |
| 1164 | mm.onuMetricsManagerLock.RLock() |
| 1165 | numOfPmToDelete := len(mm.l2PmToDelete) |
| 1166 | numOfPmToAdd := len(mm.l2PmToAdd) |
| 1167 | mm.onuMetricsManagerLock.RUnlock() |
| 1168 | |
| 1169 | if numOfPmToDelete > 0 { |
| 1170 | logger.Debugw(ctx, "state idle - pms to delete", log.Fields{"device-id": mm.pDeviceHandler.deviceID, "pms-to-delete": numOfPmToDelete}) |
| 1171 | go func() { |
| 1172 | if err := mm.pAdaptFsm.pFsm.Event(l2PmEventDeleteMe); err != nil { |
| 1173 | logger.Errorw(ctx, "error calling event", log.Fields{"device-id": mm.pDeviceHandler.deviceID, "err": err}) |
| 1174 | } |
| 1175 | }() |
| 1176 | } else if numOfPmToAdd > 0 { |
| 1177 | logger.Debugw(ctx, "state idle - pms to add", log.Fields{"device-id": mm.pDeviceHandler.deviceID, "pms-to-add": numOfPmToAdd}) |
| 1178 | go func() { |
| 1179 | if err := mm.pAdaptFsm.pFsm.Event(l2PmEventAddMe); err != nil { |
| 1180 | logger.Errorw(ctx, "error calling event", log.Fields{"device-id": mm.pDeviceHandler.deviceID, "err": err}) |
| 1181 | } |
| 1182 | }() |
| 1183 | } |
| 1184 | } |
| 1185 | |
| 1186 | func (mm *onuMetricsManager) l2PmFsmCollectData(ctx context.Context, e *fsm.Event) { |
| 1187 | logger.Debugw(ctx, "state collect data", log.Fields{"device-id": mm.pDeviceHandler.deviceID}) |
| 1188 | // Copy the activeL2Pms for which we want to collect the metrics since activeL2Pms can change dynamically |
| 1189 | mm.onuMetricsManagerLock.RLock() |
| 1190 | copyOfActiveL2Pms := make([]string, len(mm.activeL2Pms)) |
| 1191 | _ = copy(copyOfActiveL2Pms, mm.activeL2Pms) |
| 1192 | mm.onuMetricsManagerLock.RUnlock() |
| 1193 | |
| 1194 | for _, n := range copyOfActiveL2Pms { |
Girish Gowdra | 5c5aaf4 | 2021-02-17 19:40:50 -0800 | [diff] [blame] | 1195 | var metricInfoSlice []*voltha.MetricInformation |
Girish Gowdra | 0e53364 | 2021-03-02 22:02:51 -0800 | [diff] [blame] | 1196 | |
| 1197 | // mm.groupMetricMap[n].pmMEData.InstancesActive could dynamically change, so make a copy |
| 1198 | mm.onuMetricsManagerLock.RLock() |
| 1199 | copyOfEntityIDs := make([]uint16, len(mm.groupMetricMap[n].pmMEData.InstancesActive)) |
| 1200 | _ = copy(copyOfEntityIDs, mm.groupMetricMap[n].pmMEData.InstancesActive) |
| 1201 | mm.onuMetricsManagerLock.RUnlock() |
| 1202 | |
Girish Gowdra | e0140f0 | 2021-02-02 16:55:09 -0800 | [diff] [blame] | 1203 | switch n { |
| 1204 | case EthernetBridgeHistoryName: |
| 1205 | logger.Debugw(ctx, "state collect data - collecting data for EthernetFramePerformanceMonitoringHistoryData ME", log.Fields{"device-id": mm.pDeviceHandler.deviceID}) |
Girish Gowdra | 0e53364 | 2021-03-02 22:02:51 -0800 | [diff] [blame] | 1206 | for _, entityID := range copyOfEntityIDs { |
Girish Gowdra | e0140f0 | 2021-02-02 16:55:09 -0800 | [diff] [blame] | 1207 | if metricInfo := mm.collectEthernetFramePerformanceMonitoringHistoryData(ctx, true, entityID); metricInfo != nil { // upstream |
| 1208 | metricInfoSlice = append(metricInfoSlice, metricInfo) |
| 1209 | } |
| 1210 | if metricInfo := mm.collectEthernetFramePerformanceMonitoringHistoryData(ctx, false, entityID); metricInfo != nil { // downstream |
| 1211 | metricInfoSlice = append(metricInfoSlice, metricInfo) |
| 1212 | } |
| 1213 | } |
Girish Gowdra | e0140f0 | 2021-02-02 16:55:09 -0800 | [diff] [blame] | 1214 | case EthernetUniHistoryName: |
| 1215 | logger.Debugw(ctx, "state collect data - collecting data for EthernetPerformanceMonitoringHistoryData ME", log.Fields{"device-id": mm.pDeviceHandler.deviceID}) |
Girish Gowdra | 0e53364 | 2021-03-02 22:02:51 -0800 | [diff] [blame] | 1216 | for _, entityID := range copyOfEntityIDs { |
| 1217 | if metricInfo := mm.collectEthernetUniHistoryData(ctx, entityID); metricInfo != nil { // upstream |
| 1218 | metricInfoSlice = append(metricInfoSlice, metricInfo) |
Girish Gowdra | e0140f0 | 2021-02-02 16:55:09 -0800 | [diff] [blame] | 1219 | } |
| 1220 | } |
Girish Gowdra | 0e53364 | 2021-03-02 22:02:51 -0800 | [diff] [blame] | 1221 | |
Girish Gowdra | 5c5aaf4 | 2021-02-17 19:40:50 -0800 | [diff] [blame] | 1222 | case FecHistoryName: |
Girish Gowdra | 0e53364 | 2021-03-02 22:02:51 -0800 | [diff] [blame] | 1223 | for _, entityID := range copyOfEntityIDs { |
| 1224 | if metricInfo := mm.collectFecHistoryData(ctx, entityID); metricInfo != nil { // upstream |
Girish Gowdra | 5c5aaf4 | 2021-02-17 19:40:50 -0800 | [diff] [blame] | 1225 | metricInfoSlice = append(metricInfoSlice, metricInfo) |
Girish Gowdra | e0140f0 | 2021-02-02 16:55:09 -0800 | [diff] [blame] | 1226 | } |
Girish Gowdra | e0140f0 | 2021-02-02 16:55:09 -0800 | [diff] [blame] | 1227 | } |
Girish Gowdra | 5c5aaf4 | 2021-02-17 19:40:50 -0800 | [diff] [blame] | 1228 | case GemPortHistoryName: |
Girish Gowdra | 0e53364 | 2021-03-02 22:02:51 -0800 | [diff] [blame] | 1229 | for _, entityID := range copyOfEntityIDs { |
| 1230 | if metricInfo := mm.collectGemHistoryData(ctx, entityID); metricInfo != nil { // upstream |
Girish Gowdra | 5c5aaf4 | 2021-02-17 19:40:50 -0800 | [diff] [blame] | 1231 | metricInfoSlice = append(metricInfoSlice, metricInfo) |
| 1232 | } |
| 1233 | } |
| 1234 | |
Girish Gowdra | e0140f0 | 2021-02-02 16:55:09 -0800 | [diff] [blame] | 1235 | default: |
| 1236 | logger.Errorw(ctx, "unsupported l2 pm", log.Fields{"device-id": mm.pDeviceHandler.deviceID, "name": n}) |
| 1237 | } |
Girish Gowdra | 5c5aaf4 | 2021-02-17 19:40:50 -0800 | [diff] [blame] | 1238 | mm.handleMetricsPublish(ctx, n, metricInfoSlice) |
Girish Gowdra | e0140f0 | 2021-02-02 16:55:09 -0800 | [diff] [blame] | 1239 | } |
| 1240 | // Does not matter we send success or failure here. |
| 1241 | // Those PMs that we failed to collect data will be attempted to collect again in the next PM collection cycle (assuming |
| 1242 | // we have not exceed max attempts to collect the PM data) |
| 1243 | go func() { |
| 1244 | if err := mm.pAdaptFsm.pFsm.Event(l2PmEventSuccess); err != nil { |
| 1245 | logger.Errorw(ctx, "error calling event", log.Fields{"device-id": mm.pDeviceHandler.deviceID, "err": err}) |
| 1246 | } |
| 1247 | }() |
| 1248 | } |
| 1249 | |
Girish Gowdra | 0e53364 | 2021-03-02 22:02:51 -0800 | [diff] [blame] | 1250 | // nolint: gocyclo |
Girish Gowdra | e0140f0 | 2021-02-02 16:55:09 -0800 | [diff] [blame] | 1251 | func (mm *onuMetricsManager) l2PmFsmCreatePM(ctx context.Context, e *fsm.Event) { |
| 1252 | // Copy the l2PmToAdd for which we want to collect the metrics since l2PmToAdd can change dynamically |
| 1253 | mm.onuMetricsManagerLock.RLock() |
| 1254 | copyOfL2PmToAdd := make([]string, len(mm.l2PmToAdd)) |
| 1255 | _ = copy(copyOfL2PmToAdd, mm.l2PmToAdd) |
| 1256 | mm.onuMetricsManagerLock.RUnlock() |
| 1257 | |
Girish Gowdra | 5c5aaf4 | 2021-02-17 19:40:50 -0800 | [diff] [blame] | 1258 | logger.Debugw(ctx, "state create pm - start", log.Fields{"device-id": mm.pDeviceHandler.deviceID, "pms-to-add": copyOfL2PmToAdd}) |
Girish Gowdra | e0140f0 | 2021-02-02 16:55:09 -0800 | [diff] [blame] | 1259 | for _, n := range copyOfL2PmToAdd { |
| 1260 | resp := false |
Girish Gowdra | 0e53364 | 2021-03-02 22:02:51 -0800 | [diff] [blame] | 1261 | atLeastOneSuccess := false // flag indicates if at least one ME instance of the PM was successfully created. |
| 1262 | cnt := 0 |
Girish Gowdra | e0140f0 | 2021-02-02 16:55:09 -0800 | [diff] [blame] | 1263 | switch n { |
| 1264 | case EthernetBridgeHistoryName: |
| 1265 | boolForDirection := make([]bool, 2) // stores true and false to indicate upstream and downstream directions. |
| 1266 | boolForDirection = append(boolForDirection, true, false) |
| 1267 | // Create ME twice, one for each direction. Boolean true is used to indicate upstream and false for downstream. |
| 1268 | for _, direction := range boolForDirection { |
Girish Gowdra | e0140f0 | 2021-02-02 16:55:09 -0800 | [diff] [blame] | 1269 | for _, uniPort := range mm.pDeviceHandler.uniEntityMap { |
| 1270 | // Attach the EthernetFramePerformanceMonitoringHistoryData ME to MacBridgePortConfigData on the UNI port |
| 1271 | entityID := macBridgePortAniEID + uniPort.entityID |
Girish Gowdra | 0e53364 | 2021-03-02 22:02:51 -0800 | [diff] [blame] | 1272 | _ = mm.updatePmData(ctx, n, entityID, cPmAdd) // TODO: ignore error for now |
| 1273 | inner1: |
| 1274 | // retry L2PmCreateAttempts times to create the instance of PM |
| 1275 | for cnt = 0; cnt < L2PmCreateAttempts; cnt++ { |
| 1276 | mm.pDeviceHandler.pOnuOmciDevice.PDevOmciCC.sendCreateOrDeleteEthernetPerformanceMonitoringHistoryME( |
| 1277 | ctx, ConstDefaultOmciTimeout, true, direction, true, mm.pAdaptFsm.commChan, entityID) |
| 1278 | if resp = mm.waitForResponseOrTimeout(ctx, true, entityID, "EthernetFramePerformanceMonitoringHistoryData"); resp { |
| 1279 | atLeastOneSuccess = true |
| 1280 | _ = mm.updatePmData(ctx, n, entityID, cPmAdded) // TODO: ignore error for now |
| 1281 | break inner1 |
| 1282 | } |
| 1283 | } |
| 1284 | if cnt == L2PmCreateAttempts { // if we reached max attempts just give up hope on this given instance of the PM ME |
| 1285 | _ = mm.updatePmData(ctx, n, entityID, cPmRemoved) // TODO: ignore error for now |
Girish Gowdra | e0140f0 | 2021-02-02 16:55:09 -0800 | [diff] [blame] | 1286 | } |
| 1287 | } |
| 1288 | } |
| 1289 | case EthernetUniHistoryName: |
Girish Gowdra | e0140f0 | 2021-02-02 16:55:09 -0800 | [diff] [blame] | 1290 | for _, uniPort := range mm.pDeviceHandler.uniEntityMap { |
| 1291 | if uniPort.portType == uniPPTP { // This metric is only applicable for PPTP Uni Type |
Girish Gowdra | 0e53364 | 2021-03-02 22:02:51 -0800 | [diff] [blame] | 1292 | // Attach the EthernetPerformanceMonitoringHistoryData ME to PPTP port instance |
Girish Gowdra | e0140f0 | 2021-02-02 16:55:09 -0800 | [diff] [blame] | 1293 | entityID := uniPort.entityID |
Girish Gowdra | 0e53364 | 2021-03-02 22:02:51 -0800 | [diff] [blame] | 1294 | _ = mm.updatePmData(ctx, n, entityID, cPmAdd) // TODO: ignore error for now |
| 1295 | inner2: |
| 1296 | // retry L2PmCreateAttempts times to create the instance of PM |
| 1297 | for cnt = 0; cnt < L2PmCreateAttempts; cnt++ { |
| 1298 | mm.pDeviceHandler.pOnuOmciDevice.PDevOmciCC.sendCreateOrDeleteEthernetUniHistoryME( |
| 1299 | ctx, ConstDefaultOmciTimeout, true, true, mm.pAdaptFsm.commChan, entityID) |
| 1300 | if resp = mm.waitForResponseOrTimeout(ctx, true, entityID, "EthernetPerformanceMonitoringHistoryData"); resp { |
| 1301 | atLeastOneSuccess = true |
| 1302 | _ = mm.updatePmData(ctx, n, entityID, cPmAdded) // TODO: ignore error for now |
| 1303 | break inner2 |
| 1304 | } |
| 1305 | } |
| 1306 | if cnt == L2PmCreateAttempts { // if we reached max attempts just give up hope on this given instance of the PM ME |
| 1307 | _ = mm.updatePmData(ctx, n, entityID, cPmRemoved) // TODO: ignore error for now |
Girish Gowdra | e0140f0 | 2021-02-02 16:55:09 -0800 | [diff] [blame] | 1308 | } |
| 1309 | } |
| 1310 | } |
Girish Gowdra | 5c5aaf4 | 2021-02-17 19:40:50 -0800 | [diff] [blame] | 1311 | case FecHistoryName: |
Girish Gowdra | 5c5aaf4 | 2021-02-17 19:40:50 -0800 | [diff] [blame] | 1312 | for _, anigInstID := range mm.pDeviceHandler.pOnuOmciDevice.pOnuDB.getSortedInstKeys(ctx, me.AniGClassID) { |
Girish Gowdra | 0e53364 | 2021-03-02 22:02:51 -0800 | [diff] [blame] | 1313 | // Attach the FecPerformanceMonitoringHistoryData ME to the ANI-G ME instance |
Girish Gowdra | 5c5aaf4 | 2021-02-17 19:40:50 -0800 | [diff] [blame] | 1314 | mm.pDeviceHandler.pOnuOmciDevice.PDevOmciCC.sendCreateOrDeleteFecHistoryME( |
| 1315 | ctx, ConstDefaultOmciTimeout, true, true, mm.pAdaptFsm.commChan, anigInstID) |
Girish Gowdra | 0e53364 | 2021-03-02 22:02:51 -0800 | [diff] [blame] | 1316 | _ = mm.updatePmData(ctx, n, anigInstID, cPmAdd) // TODO: ignore error for now |
| 1317 | inner3: |
| 1318 | // retry L2PmCreateAttempts times to create the instance of PM |
| 1319 | for cnt = 0; cnt < L2PmCreateAttempts; cnt++ { |
| 1320 | if resp = mm.waitForResponseOrTimeout(ctx, true, anigInstID, "FecPerformanceMonitoringHistoryData"); resp { |
| 1321 | atLeastOneSuccess = true |
| 1322 | _ = mm.updatePmData(ctx, n, anigInstID, cPmAdded) // TODO: ignore error for now |
| 1323 | break inner3 |
| 1324 | } |
| 1325 | } |
| 1326 | if cnt == L2PmCreateAttempts { // if we reached max attempts just give up hope on this given instance of the PM ME |
| 1327 | _ = mm.updatePmData(ctx, n, anigInstID, cPmRemoved) // TODO: ignore error for now |
Girish Gowdra | 5c5aaf4 | 2021-02-17 19:40:50 -0800 | [diff] [blame] | 1328 | } |
| 1329 | } |
| 1330 | case GemPortHistoryName: |
| 1331 | |
| 1332 | mm.onuMetricsManagerLock.RLock() |
Girish Gowdra | 0e53364 | 2021-03-02 22:02:51 -0800 | [diff] [blame] | 1333 | copyOfGemPortInstIDsToAdd := make([]uint16, len(mm.groupMetricMap[GemPortHistoryName].pmMEData.InstancesToAdd)) |
| 1334 | _ = copy(copyOfGemPortInstIDsToAdd, mm.groupMetricMap[GemPortHistoryName].pmMEData.InstancesToAdd) |
Girish Gowdra | 5c5aaf4 | 2021-02-17 19:40:50 -0800 | [diff] [blame] | 1335 | mm.onuMetricsManagerLock.RUnlock() |
Girish Gowdra | 0e53364 | 2021-03-02 22:02:51 -0800 | [diff] [blame] | 1336 | |
| 1337 | if len(copyOfGemPortInstIDsToAdd) == 0 { |
| 1338 | // If there are no gemport history MEs to be created, just skip further processing |
| 1339 | // Otherwise down below (after 'switch' case handling) we assume the ME creation failed because resp and atLeastOneSuccess flag are false. |
| 1340 | // Normally there are no GemPortHistory MEs to create at start up. They come in only after provisioning service on the ONU. |
| 1341 | mm.onuMetricsManagerLock.Lock() |
| 1342 | mm.l2PmToAdd = mm.removeIfFoundString(mm.l2PmToAdd, n) |
| 1343 | mm.onuMetricsManagerLock.Unlock() |
| 1344 | continue |
| 1345 | } |
| 1346 | |
Girish Gowdra | 5c5aaf4 | 2021-02-17 19:40:50 -0800 | [diff] [blame] | 1347 | for _, v := range copyOfGemPortInstIDsToAdd { |
| 1348 | mm.pDeviceHandler.pOnuOmciDevice.PDevOmciCC.sendCreateOrDeleteGemPortHistoryME( |
| 1349 | ctx, ConstDefaultOmciTimeout, true, true, mm.pAdaptFsm.commChan, v) |
Girish Gowdra | 0e53364 | 2021-03-02 22:02:51 -0800 | [diff] [blame] | 1350 | _ = mm.updatePmData(ctx, n, v, cPmAdd) // TODO: ignore error for now |
| 1351 | inner4: |
| 1352 | // retry L2PmCreateAttempts times to create the instance of PM |
| 1353 | for cnt = 0; cnt < L2PmCreateAttempts; cnt++ { |
| 1354 | if resp = mm.waitForResponseOrTimeout(ctx, true, v, "GemPortNetworkCtpPerformanceMonitoringHistoryData"); resp { |
| 1355 | atLeastOneSuccess = true |
| 1356 | _ = mm.updatePmData(ctx, n, v, cPmAdded) // TODO: ignore error for now |
| 1357 | break inner4 |
| 1358 | } |
Girish Gowdra | 5c5aaf4 | 2021-02-17 19:40:50 -0800 | [diff] [blame] | 1359 | } |
Girish Gowdra | 0e53364 | 2021-03-02 22:02:51 -0800 | [diff] [blame] | 1360 | if cnt == L2PmCreateAttempts { // if we reached max attempts just give up hope on this given instance of the PM ME |
| 1361 | _ = mm.updatePmData(ctx, n, v, cPmRemoved) // TODO: ignore error for now |
| 1362 | } |
Girish Gowdra | 5c5aaf4 | 2021-02-17 19:40:50 -0800 | [diff] [blame] | 1363 | } |
| 1364 | |
Girish Gowdra | e0140f0 | 2021-02-02 16:55:09 -0800 | [diff] [blame] | 1365 | default: |
| 1366 | logger.Errorw(ctx, "unsupported l2 pm", log.Fields{"device-id": mm.pDeviceHandler.deviceID, "name": n}) |
| 1367 | } |
Girish Gowdra | 0e53364 | 2021-03-02 22:02:51 -0800 | [diff] [blame] | 1368 | // On success of at least one instance of the PM for a given ME, update the local list maintained for active PMs and PMs to add |
| 1369 | if atLeastOneSuccess { |
Girish Gowdra | e0140f0 | 2021-02-02 16:55:09 -0800 | [diff] [blame] | 1370 | mm.onuMetricsManagerLock.Lock() |
Girish Gowdra | 5c5aaf4 | 2021-02-17 19:40:50 -0800 | [diff] [blame] | 1371 | mm.activeL2Pms = mm.appendIfMissingString(mm.activeL2Pms, n) |
| 1372 | mm.l2PmToAdd = mm.removeIfFoundString(mm.l2PmToAdd, n) |
| 1373 | logger.Debugw(ctx, "success-resp", log.Fields{"pm-name": n, "active-l2-pms": mm.activeL2Pms, "pms-to-add": mm.l2PmToAdd}) |
Girish Gowdra | e0140f0 | 2021-02-02 16:55:09 -0800 | [diff] [blame] | 1374 | mm.onuMetricsManagerLock.Unlock() |
| 1375 | } else { |
Girish Gowdra | 0e53364 | 2021-03-02 22:02:51 -0800 | [diff] [blame] | 1376 | // If we are here then no instance of the PM of the given ME were created successfully, so locally disable the PM |
Girish Gowdra | e0140f0 | 2021-02-02 16:55:09 -0800 | [diff] [blame] | 1377 | // and also remove it from l2PmToAdd slice so that we do not try to create the PM ME anymore |
| 1378 | mm.onuMetricsManagerLock.Lock() |
Girish Gowdra | 0e53364 | 2021-03-02 22:02:51 -0800 | [diff] [blame] | 1379 | logger.Debugw(ctx, "exceeded-max-add-retry-attempts--disabling-group", log.Fields{"groupName": n}) |
| 1380 | mm.groupMetricMap[n].enabled = false |
| 1381 | mm.l2PmToAdd = mm.removeIfFoundString(mm.l2PmToAdd, n) |
Girish Gowdra | 5c5aaf4 | 2021-02-17 19:40:50 -0800 | [diff] [blame] | 1382 | |
Girish Gowdra | e0140f0 | 2021-02-02 16:55:09 -0800 | [diff] [blame] | 1383 | logger.Warnw(ctx, "state create pm - failed to create pm", |
Girish Gowdra | 5c5aaf4 | 2021-02-17 19:40:50 -0800 | [diff] [blame] | 1384 | log.Fields{"device-id": mm.pDeviceHandler.deviceID, "metricName": n, |
Girish Gowdra | 0e53364 | 2021-03-02 22:02:51 -0800 | [diff] [blame] | 1385 | "active-l2-pms": mm.activeL2Pms, "pms-to-add": mm.l2PmToAdd}) |
Girish Gowdra | e0140f0 | 2021-02-02 16:55:09 -0800 | [diff] [blame] | 1386 | mm.onuMetricsManagerLock.Unlock() |
| 1387 | } |
| 1388 | } |
Girish Gowdra | 5c5aaf4 | 2021-02-17 19:40:50 -0800 | [diff] [blame] | 1389 | logger.Debugw(ctx, "state create pm - done", log.Fields{"device-id": mm.pDeviceHandler.deviceID, "active-l2-pms": mm.activeL2Pms, "pms-to-add": mm.l2PmToAdd}) |
Girish Gowdra | e0140f0 | 2021-02-02 16:55:09 -0800 | [diff] [blame] | 1390 | // Does not matter we send success or failure here. |
| 1391 | // Those PMs that we failed to create will be attempted to create again in the next PM creation cycle (assuming |
| 1392 | // we have not exceed max attempts to create the PM ME) |
| 1393 | go func() { |
| 1394 | if err := mm.pAdaptFsm.pFsm.Event(l2PmEventSuccess); err != nil { |
| 1395 | logger.Errorw(ctx, "error calling event", log.Fields{"device-id": mm.pDeviceHandler.deviceID, "err": err}) |
| 1396 | } |
| 1397 | }() |
| 1398 | } |
| 1399 | |
Girish Gowdra | 0e53364 | 2021-03-02 22:02:51 -0800 | [diff] [blame] | 1400 | // nolint: gocyclo |
Girish Gowdra | e0140f0 | 2021-02-02 16:55:09 -0800 | [diff] [blame] | 1401 | func (mm *onuMetricsManager) l2PmFsmDeletePM(ctx context.Context, e *fsm.Event) { |
| 1402 | // Copy the l2PmToDelete for which we want to collect the metrics since l2PmToDelete can change dynamically |
| 1403 | mm.onuMetricsManagerLock.RLock() |
| 1404 | copyOfL2PmToDelete := make([]string, len(mm.l2PmToDelete)) |
| 1405 | _ = copy(copyOfL2PmToDelete, mm.l2PmToDelete) |
| 1406 | mm.onuMetricsManagerLock.RUnlock() |
| 1407 | |
| 1408 | logger.Debugw(ctx, "state delete pm", log.Fields{"device-id": mm.pDeviceHandler.deviceID, "pms-to-delete": mm.l2PmToDelete}) |
| 1409 | for _, n := range copyOfL2PmToDelete { |
| 1410 | resp := false |
Girish Gowdra | 0e53364 | 2021-03-02 22:02:51 -0800 | [diff] [blame] | 1411 | cnt := 0 |
| 1412 | atLeastOneDeleteFailure := false |
| 1413 | |
| 1414 | // mm.groupMetricMap[n].pmMEData.InstancesActive could dynamically change, so make a copy |
| 1415 | mm.onuMetricsManagerLock.RLock() |
| 1416 | copyOfEntityIDs := make([]uint16, len(mm.groupMetricMap[n].pmMEData.InstancesActive)) |
| 1417 | _ = copy(copyOfEntityIDs, mm.groupMetricMap[n].pmMEData.InstancesActive) |
| 1418 | mm.onuMetricsManagerLock.RUnlock() |
| 1419 | |
| 1420 | if len(copyOfEntityIDs) == 0 { |
| 1421 | // if there are no enityIDs to remove for the PM ME just clear the PM name entry from cache and continue |
| 1422 | mm.onuMetricsManagerLock.Lock() |
| 1423 | mm.activeL2Pms = mm.removeIfFoundString(mm.activeL2Pms, n) |
| 1424 | mm.l2PmToDelete = mm.removeIfFoundString(mm.l2PmToDelete, n) |
| 1425 | logger.Debugw(ctx, "success-resp", log.Fields{"pm-name": n, "active-l2-pms": mm.activeL2Pms, "pms-to-delete": mm.l2PmToDelete}) |
| 1426 | mm.onuMetricsManagerLock.Unlock() |
| 1427 | continue |
| 1428 | } |
| 1429 | logger.Debugw(ctx, "entities to delete", log.Fields{"device-id": mm.pDeviceHandler.deviceID, "metricName": n, "entityIDs": copyOfEntityIDs}) |
Girish Gowdra | e0140f0 | 2021-02-02 16:55:09 -0800 | [diff] [blame] | 1430 | switch n { |
| 1431 | case EthernetBridgeHistoryName: |
| 1432 | boolForDirection := make([]bool, 2) // stores true and false to indicate upstream and downstream directions. |
| 1433 | boolForDirection = append(boolForDirection, true, false) |
| 1434 | // Create ME twice, one for each direction. Boolean true is used to indicate upstream and false for downstream. |
| 1435 | for _, direction := range boolForDirection { |
Girish Gowdra | 0e53364 | 2021-03-02 22:02:51 -0800 | [diff] [blame] | 1436 | for _, entityID := range copyOfEntityIDs { |
| 1437 | inner1: |
| 1438 | // retry L2PmDeleteAttempts times to delete the instance of PM |
| 1439 | for cnt = 0; cnt < L2PmDeleteAttempts; cnt++ { |
| 1440 | mm.pDeviceHandler.pOnuOmciDevice.PDevOmciCC.sendCreateOrDeleteEthernetPerformanceMonitoringHistoryME( |
| 1441 | ctx, ConstDefaultOmciTimeout, true, direction, false, mm.pAdaptFsm.commChan, entityID) |
| 1442 | _ = mm.updatePmData(ctx, n, entityID, cPmRemove) // TODO: ignore error for now |
| 1443 | if resp = mm.waitForResponseOrTimeout(ctx, false, entityID, "EthernetFramePerformanceMonitoringHistoryData"); !resp { |
| 1444 | atLeastOneDeleteFailure = true |
| 1445 | } else { |
| 1446 | _ = mm.updatePmData(ctx, n, entityID, cPmRemoved) // TODO: ignore error for now |
| 1447 | break inner1 |
| 1448 | } |
| 1449 | } |
| 1450 | if cnt == L2PmDeleteAttempts { // if we reached max attempts just give up hope on this given instance of the PM ME |
| 1451 | _ = mm.updatePmData(ctx, n, entityID, cPmRemoved) // TODO: ignore error for now |
Girish Gowdra | e0140f0 | 2021-02-02 16:55:09 -0800 | [diff] [blame] | 1452 | } |
| 1453 | } |
| 1454 | } |
| 1455 | case EthernetUniHistoryName: |
Girish Gowdra | 0e53364 | 2021-03-02 22:02:51 -0800 | [diff] [blame] | 1456 | for _, entityID := range copyOfEntityIDs { |
| 1457 | inner2: |
| 1458 | // retry L2PmDeleteAttempts times to delete the instance of PM |
| 1459 | for cnt = 0; cnt < L2PmDeleteAttempts; cnt++ { |
Girish Gowdra | e0140f0 | 2021-02-02 16:55:09 -0800 | [diff] [blame] | 1460 | mm.pDeviceHandler.pOnuOmciDevice.PDevOmciCC.sendCreateOrDeleteEthernetUniHistoryME( |
| 1461 | ctx, ConstDefaultOmciTimeout, true, false, mm.pAdaptFsm.commChan, entityID) |
Girish Gowdra | 5c5aaf4 | 2021-02-17 19:40:50 -0800 | [diff] [blame] | 1462 | if resp = mm.waitForResponseOrTimeout(ctx, false, entityID, "EthernetPerformanceMonitoringHistoryData"); !resp { |
Girish Gowdra | 0e53364 | 2021-03-02 22:02:51 -0800 | [diff] [blame] | 1463 | atLeastOneDeleteFailure = true |
| 1464 | } else { |
| 1465 | _ = mm.updatePmData(ctx, n, entityID, cPmRemoved) // TODO: ignore error for now |
Girish Gowdra | 5c5aaf4 | 2021-02-17 19:40:50 -0800 | [diff] [blame] | 1466 | break inner2 |
Girish Gowdra | e0140f0 | 2021-02-02 16:55:09 -0800 | [diff] [blame] | 1467 | } |
| 1468 | } |
Girish Gowdra | 0e53364 | 2021-03-02 22:02:51 -0800 | [diff] [blame] | 1469 | if cnt == L2PmDeleteAttempts { // if we reached max attempts just give up hope on this given instance of the PM ME |
| 1470 | _ = mm.updatePmData(ctx, n, entityID, cPmRemoved) // TODO: ignore error for now |
| 1471 | } |
Girish Gowdra | e0140f0 | 2021-02-02 16:55:09 -0800 | [diff] [blame] | 1472 | } |
Girish Gowdra | 5c5aaf4 | 2021-02-17 19:40:50 -0800 | [diff] [blame] | 1473 | case FecHistoryName: |
Girish Gowdra | 0e53364 | 2021-03-02 22:02:51 -0800 | [diff] [blame] | 1474 | for _, entityID := range copyOfEntityIDs { |
| 1475 | inner3: |
| 1476 | // retry L2PmDeleteAttempts times to delete the instance of PM |
| 1477 | for cnt = 0; cnt < L2PmDeleteAttempts; cnt++ { |
| 1478 | mm.pDeviceHandler.pOnuOmciDevice.PDevOmciCC.sendCreateOrDeleteFecHistoryME( |
| 1479 | ctx, ConstDefaultOmciTimeout, true, false, mm.pAdaptFsm.commChan, entityID) |
| 1480 | if resp := mm.waitForResponseOrTimeout(ctx, false, entityID, "FecPerformanceMonitoringHistoryData"); !resp { |
| 1481 | atLeastOneDeleteFailure = true |
| 1482 | } else { |
| 1483 | _ = mm.updatePmData(ctx, n, entityID, cPmRemoved) // TODO: ignore error for now |
| 1484 | break inner3 |
| 1485 | } |
| 1486 | } |
| 1487 | if cnt == L2PmDeleteAttempts { // if we reached max attempts just give up hope on this given instance of the PM ME |
| 1488 | _ = mm.updatePmData(ctx, n, entityID, cPmRemoved) // TODO: ignore error for now |
Girish Gowdra | 5c5aaf4 | 2021-02-17 19:40:50 -0800 | [diff] [blame] | 1489 | } |
| 1490 | } |
| 1491 | case GemPortHistoryName: |
Girish Gowdra | 0e53364 | 2021-03-02 22:02:51 -0800 | [diff] [blame] | 1492 | for _, entityID := range copyOfEntityIDs { |
| 1493 | inner4: |
| 1494 | // retry L2PmDeleteAttempts times to delete the instance of PM |
| 1495 | for cnt = 0; cnt < L2PmDeleteAttempts; cnt++ { |
| 1496 | mm.pDeviceHandler.pOnuOmciDevice.PDevOmciCC.sendCreateOrDeleteGemPortHistoryME( |
| 1497 | ctx, ConstDefaultOmciTimeout, true, false, mm.pAdaptFsm.commChan, entityID) |
| 1498 | if resp = mm.waitForResponseOrTimeout(ctx, false, entityID, "GemPortNetworkCtpPerformanceMonitoringHistoryData"); !resp { |
| 1499 | atLeastOneDeleteFailure = true |
| 1500 | } else { |
| 1501 | _ = mm.updatePmData(ctx, n, entityID, cPmRemoved) // TODO: ignore error for now |
| 1502 | break inner4 |
| 1503 | } |
Girish Gowdra | 5c5aaf4 | 2021-02-17 19:40:50 -0800 | [diff] [blame] | 1504 | } |
Girish Gowdra | 0e53364 | 2021-03-02 22:02:51 -0800 | [diff] [blame] | 1505 | if cnt == L2PmDeleteAttempts { // if we reached max attempts just give up hope on this given instance of the PM ME |
| 1506 | _ = mm.updatePmData(ctx, n, entityID, cPmRemoved) // TODO: ignore error for now |
| 1507 | } |
Girish Gowdra | 5c5aaf4 | 2021-02-17 19:40:50 -0800 | [diff] [blame] | 1508 | } |
Girish Gowdra | e0140f0 | 2021-02-02 16:55:09 -0800 | [diff] [blame] | 1509 | default: |
| 1510 | logger.Errorw(ctx, "unsupported l2 pm", log.Fields{"device-id": mm.pDeviceHandler.deviceID, "name": n}) |
| 1511 | } |
Girish Gowdra | 0e53364 | 2021-03-02 22:02:51 -0800 | [diff] [blame] | 1512 | // If we could not completely clean up the PM ME then just give up. |
| 1513 | if atLeastOneDeleteFailure { |
| 1514 | logger.Warnw(ctx, "state delete pm - failed to delete at least one instance of the PM ME", |
| 1515 | log.Fields{"device-id": mm.pDeviceHandler.deviceID, "metricName": n, |
| 1516 | "active-l2-pms": mm.activeL2Pms, "pms-to-delete": mm.l2PmToDelete}) |
| 1517 | mm.onuMetricsManagerLock.Lock() |
| 1518 | logger.Debugw(ctx, "exceeded-max-delete-retry-attempts--disabling-group", log.Fields{"groupName": n}) |
| 1519 | mm.activeL2Pms = mm.removeIfFoundString(mm.activeL2Pms, n) |
| 1520 | mm.l2PmToDelete = mm.removeIfFoundString(mm.l2PmToDelete, n) |
| 1521 | mm.groupMetricMap[n].enabled = false |
| 1522 | mm.onuMetricsManagerLock.Unlock() |
| 1523 | } else { // success case |
Girish Gowdra | e0140f0 | 2021-02-02 16:55:09 -0800 | [diff] [blame] | 1524 | mm.onuMetricsManagerLock.Lock() |
Girish Gowdra | 5c5aaf4 | 2021-02-17 19:40:50 -0800 | [diff] [blame] | 1525 | mm.activeL2Pms = mm.removeIfFoundString(mm.activeL2Pms, n) |
| 1526 | mm.l2PmToDelete = mm.removeIfFoundString(mm.l2PmToDelete, n) |
| 1527 | logger.Debugw(ctx, "success-resp", log.Fields{"pm-name": n, "active-l2-pms": mm.activeL2Pms, "pms-to-delete": mm.l2PmToDelete}) |
Girish Gowdra | e0140f0 | 2021-02-02 16:55:09 -0800 | [diff] [blame] | 1528 | mm.onuMetricsManagerLock.Unlock() |
| 1529 | } |
| 1530 | } |
Girish Gowdra | 5c5aaf4 | 2021-02-17 19:40:50 -0800 | [diff] [blame] | 1531 | logger.Debugw(ctx, "state delete pm - done", log.Fields{"device-id": mm.pDeviceHandler.deviceID, "active-l2-pms": mm.activeL2Pms, "pms-to-delete": mm.l2PmToDelete}) |
Girish Gowdra | e0140f0 | 2021-02-02 16:55:09 -0800 | [diff] [blame] | 1532 | // Does not matter we send success or failure here. |
| 1533 | // Those PMs that we failed to delete will be attempted to create again in the next PM collection cycle |
| 1534 | go func() { |
| 1535 | if err := mm.pAdaptFsm.pFsm.Event(l2PmEventSuccess); err != nil { |
| 1536 | logger.Errorw(ctx, "error calling event", log.Fields{"device-id": mm.pDeviceHandler.deviceID, "err": err}) |
| 1537 | } |
| 1538 | }() |
| 1539 | } |
| 1540 | |
| 1541 | // ** L2 PM FSM Handlers end ** |
| 1542 | |
| 1543 | // syncTime synchronizes time with the ONU to establish a 15 min boundary for PM collection and reporting. |
| 1544 | func (mm *onuMetricsManager) syncTime(ctx context.Context) error { |
| 1545 | if err := mm.pDeviceHandler.pOnuOmciDevice.PDevOmciCC.sendSyncTime(ctx, ConstDefaultOmciTimeout, true, mm.pAdaptFsm.commChan); err != nil { |
| 1546 | logger.Errorw(ctx, "cannot send sync time request", log.Fields{"device-id": mm.pDeviceHandler.deviceID}) |
| 1547 | return err |
| 1548 | } |
| 1549 | |
| 1550 | select { |
| 1551 | case <-time.After(time.Duration(ConstDefaultOmciTimeout) * time.Second): |
| 1552 | logger.Errorf(ctx, "timed out waiting for sync time response from onu", log.Fields{"device-id": mm.pDeviceHandler.deviceID}) |
| 1553 | return fmt.Errorf("timed-out-waiting-for-sync-time-response-%v", mm.pDeviceHandler.deviceID) |
| 1554 | case syncTimeRes := <-mm.syncTimeResponseChan: |
| 1555 | if !syncTimeRes { |
| 1556 | return fmt.Errorf("failed-to-sync-time-%v", mm.pDeviceHandler.deviceID) |
| 1557 | } |
| 1558 | logger.Infow(ctx, "sync time success", log.Fields{"device-id": mm.pDeviceHandler.deviceID}) |
| 1559 | return nil |
| 1560 | } |
| 1561 | } |
| 1562 | |
| 1563 | func (mm *onuMetricsManager) collectEthernetFramePerformanceMonitoringHistoryData(ctx context.Context, upstream bool, entityID uint16) *voltha.MetricInformation { |
| 1564 | var mEnt *me.ManagedEntity |
| 1565 | var omciErr me.OmciErrors |
| 1566 | var classID me.ClassID |
| 1567 | var meAttributes me.AttributeValueMap |
| 1568 | logger.Debugw(ctx, "collecting data for EthernetFramePerformanceMonitoringHistoryData", log.Fields{"device-id": mm.pDeviceHandler.deviceID, "entityID": entityID, "upstream": upstream}) |
| 1569 | meParam := me.ParamData{EntityID: entityID} |
| 1570 | if upstream { |
| 1571 | if mEnt, omciErr = me.NewEthernetFramePerformanceMonitoringHistoryDataUpstream(meParam); omciErr == nil || mEnt == nil || omciErr.GetError() != nil { |
| 1572 | logger.Errorw(ctx, "error creating me", log.Fields{"device-id": mm.pDeviceHandler.deviceID, "entityID": entityID, "upstream": upstream}) |
| 1573 | return nil |
| 1574 | } |
| 1575 | classID = me.EthernetFramePerformanceMonitoringHistoryDataUpstreamClassID |
| 1576 | } else { |
| 1577 | if mEnt, omciErr = me.NewEthernetFramePerformanceMonitoringHistoryDataDownstream(meParam); omciErr == nil || mEnt == nil || omciErr.GetError() != nil { |
| 1578 | logger.Errorw(ctx, "error creating me", log.Fields{"device-id": mm.pDeviceHandler.deviceID, "entityID": entityID, "upstream": upstream}) |
| 1579 | return nil |
| 1580 | } |
| 1581 | classID = me.EthernetFramePerformanceMonitoringHistoryDataDownstreamClassID |
| 1582 | } |
| 1583 | |
Girish Gowdra | e0140f0 | 2021-02-02 16:55:09 -0800 | [diff] [blame] | 1584 | intervalEndTime := -1 |
| 1585 | ethPMHistData := make(map[string]float32) |
Girish Gowdra | 5c5aaf4 | 2021-02-17 19:40:50 -0800 | [diff] [blame] | 1586 | if err := mm.populateGroupSpecificMetrics(ctx, mEnt, classID, entityID, meAttributes, ethPMHistData, &intervalEndTime); err != nil { |
| 1587 | return nil |
Girish Gowdra | e0140f0 | 2021-02-02 16:55:09 -0800 | [diff] [blame] | 1588 | } |
| 1589 | |
| 1590 | // Populate some relevant context for the EthernetFramePerformanceMonitoringHistoryData PM |
| 1591 | ethPMHistData["class_id"] = float32(classID) |
| 1592 | ethPMHistData["interval_end_time"] = float32(intervalEndTime) |
| 1593 | ethPMHistData["parent_class_id"] = float32(me.MacBridgeConfigurationDataClassID) // EthernetFramePerformanceMonitoringHistoryData is attached to MBPCD ME |
| 1594 | ethPMHistData["parent_entity_id"] = float32(entityID) |
| 1595 | if upstream { |
| 1596 | ethPMHistData["upstream"] = float32(1) |
| 1597 | } else { |
| 1598 | ethPMHistData["upstream"] = float32(0) |
| 1599 | } |
| 1600 | |
Girish Gowdra | 5c5aaf4 | 2021-02-17 19:40:50 -0800 | [diff] [blame] | 1601 | metricInfo := mm.populateOnuMetricInfo(EthernetBridgeHistoryName, ethPMHistData) |
Girish Gowdra | e0140f0 | 2021-02-02 16:55:09 -0800 | [diff] [blame] | 1602 | |
Girish Gowdra | e0140f0 | 2021-02-02 16:55:09 -0800 | [diff] [blame] | 1603 | logger.Debugw(ctx, "collecting data for EthernetFramePerformanceMonitoringHistoryData successful", |
| 1604 | log.Fields{"device-id": mm.pDeviceHandler.deviceID, "entityID": entityID, "upstream": upstream, "metricInfo": metricInfo}) |
| 1605 | return &metricInfo |
| 1606 | } |
| 1607 | |
| 1608 | func (mm *onuMetricsManager) collectEthernetUniHistoryData(ctx context.Context, entityID uint16) *voltha.MetricInformation { |
| 1609 | var mEnt *me.ManagedEntity |
| 1610 | var omciErr me.OmciErrors |
| 1611 | var classID me.ClassID |
| 1612 | var meAttributes me.AttributeValueMap |
| 1613 | logger.Debugw(ctx, "collecting data for EthernetFramePerformanceMonitoringHistoryData", log.Fields{"device-id": mm.pDeviceHandler.deviceID, "entityID": entityID}) |
| 1614 | meParam := me.ParamData{EntityID: entityID} |
| 1615 | if mEnt, omciErr = me.NewEthernetPerformanceMonitoringHistoryData(meParam); omciErr == nil || mEnt == nil || omciErr.GetError() != nil { |
| 1616 | logger.Errorw(ctx, "error creating me", log.Fields{"device-id": mm.pDeviceHandler.deviceID, "entityID": entityID}) |
| 1617 | return nil |
| 1618 | } |
| 1619 | classID = me.EthernetPerformanceMonitoringHistoryDataClassID |
| 1620 | |
Girish Gowdra | e0140f0 | 2021-02-02 16:55:09 -0800 | [diff] [blame] | 1621 | intervalEndTime := -1 |
| 1622 | ethUniHistData := make(map[string]float32) |
Girish Gowdra | 5c5aaf4 | 2021-02-17 19:40:50 -0800 | [diff] [blame] | 1623 | if err := mm.populateGroupSpecificMetrics(ctx, mEnt, classID, entityID, meAttributes, ethUniHistData, &intervalEndTime); err != nil { |
| 1624 | return nil |
Girish Gowdra | e0140f0 | 2021-02-02 16:55:09 -0800 | [diff] [blame] | 1625 | } |
| 1626 | |
| 1627 | // Populate some relevant context for the EthernetPerformanceMonitoringHistoryData PM |
| 1628 | ethUniHistData["class_id"] = float32(classID) |
| 1629 | ethUniHistData["interval_end_time"] = float32(intervalEndTime) |
| 1630 | |
Girish Gowdra | 5c5aaf4 | 2021-02-17 19:40:50 -0800 | [diff] [blame] | 1631 | metricInfo := mm.populateOnuMetricInfo(EthernetUniHistoryName, ethUniHistData) |
Girish Gowdra | e0140f0 | 2021-02-02 16:55:09 -0800 | [diff] [blame] | 1632 | |
Girish Gowdra | e0140f0 | 2021-02-02 16:55:09 -0800 | [diff] [blame] | 1633 | logger.Debugw(ctx, "collecting data for EthernetPerformanceMonitoringHistoryData successful", |
| 1634 | log.Fields{"device-id": mm.pDeviceHandler.deviceID, "entityID": entityID, "metricInfo": metricInfo}) |
| 1635 | return &metricInfo |
| 1636 | } |
| 1637 | |
Girish Gowdra | 5c5aaf4 | 2021-02-17 19:40:50 -0800 | [diff] [blame] | 1638 | func (mm *onuMetricsManager) collectFecHistoryData(ctx context.Context, entityID uint16) *voltha.MetricInformation { |
| 1639 | var mEnt *me.ManagedEntity |
| 1640 | var omciErr me.OmciErrors |
| 1641 | var classID me.ClassID |
| 1642 | var meAttributes me.AttributeValueMap |
| 1643 | logger.Debugw(ctx, "collecting data for FecPerformanceMonitoringHistoryData", log.Fields{"device-id": mm.pDeviceHandler.deviceID, "entityID": entityID}) |
| 1644 | meParam := me.ParamData{EntityID: entityID} |
| 1645 | if mEnt, omciErr = me.NewFecPerformanceMonitoringHistoryData(meParam); omciErr == nil || mEnt == nil || omciErr.GetError() != nil { |
| 1646 | logger.Errorw(ctx, "error creating me", log.Fields{"device-id": mm.pDeviceHandler.deviceID, "entityID": entityID}) |
| 1647 | return nil |
| 1648 | } |
| 1649 | classID = me.FecPerformanceMonitoringHistoryDataClassID |
| 1650 | |
| 1651 | intervalEndTime := -1 |
| 1652 | fecHistData := make(map[string]float32) |
| 1653 | if err := mm.populateGroupSpecificMetrics(ctx, mEnt, classID, entityID, meAttributes, fecHistData, &intervalEndTime); err != nil { |
| 1654 | return nil |
| 1655 | } |
| 1656 | |
| 1657 | // Populate some relevant context for the EthernetPerformanceMonitoringHistoryData PM |
| 1658 | fecHistData["class_id"] = float32(classID) |
| 1659 | fecHistData["interval_end_time"] = float32(intervalEndTime) |
| 1660 | |
| 1661 | metricInfo := mm.populateOnuMetricInfo(FecHistoryName, fecHistData) |
| 1662 | |
| 1663 | logger.Debugw(ctx, "collecting data for FecPerformanceMonitoringHistoryData successful", |
| 1664 | log.Fields{"device-id": mm.pDeviceHandler.deviceID, "entityID": entityID, "metricInfo": metricInfo}) |
| 1665 | return &metricInfo |
| 1666 | } |
| 1667 | |
| 1668 | func (mm *onuMetricsManager) collectGemHistoryData(ctx context.Context, entityID uint16) *voltha.MetricInformation { |
| 1669 | var mEnt *me.ManagedEntity |
| 1670 | var omciErr me.OmciErrors |
| 1671 | var classID me.ClassID |
| 1672 | var meAttributes me.AttributeValueMap |
| 1673 | logger.Debugw(ctx, "collecting data for GemPortNetworkCtpPerformanceMonitoringHistoryData", log.Fields{"device-id": mm.pDeviceHandler.deviceID, "entityID": entityID}) |
| 1674 | meParam := me.ParamData{EntityID: entityID} |
| 1675 | if mEnt, omciErr = me.NewGemPortNetworkCtpPerformanceMonitoringHistoryData(meParam); omciErr == nil || mEnt == nil || omciErr.GetError() != nil { |
| 1676 | logger.Errorw(ctx, "error creating me", log.Fields{"device-id": mm.pDeviceHandler.deviceID, "entityID": entityID}) |
| 1677 | return nil |
| 1678 | } |
| 1679 | classID = me.GemPortNetworkCtpPerformanceMonitoringHistoryDataClassID |
| 1680 | |
| 1681 | intervalEndTime := -1 |
| 1682 | gemHistData := make(map[string]float32) |
| 1683 | if err := mm.populateGroupSpecificMetrics(ctx, mEnt, classID, entityID, meAttributes, gemHistData, &intervalEndTime); err != nil { |
| 1684 | return nil |
| 1685 | } |
| 1686 | |
| 1687 | // Populate some relevant context for the GemPortNetworkCtpPerformanceMonitoringHistoryData PM |
| 1688 | gemHistData["class_id"] = float32(classID) |
| 1689 | gemHistData["interval_end_time"] = float32(intervalEndTime) |
| 1690 | |
| 1691 | metricInfo := mm.populateOnuMetricInfo(GemPortHistoryName, gemHistData) |
| 1692 | |
| 1693 | logger.Debugw(ctx, "collecting data for GemPortNetworkCtpPerformanceMonitoringHistoryData successful", |
| 1694 | log.Fields{"device-id": mm.pDeviceHandler.deviceID, "entityID": entityID, "metricInfo": metricInfo}) |
| 1695 | return &metricInfo |
| 1696 | } |
| 1697 | |
Girish Gowdra | e0140f0 | 2021-02-02 16:55:09 -0800 | [diff] [blame] | 1698 | // nolint: gocyclo |
Girish Gowdra | 5c5aaf4 | 2021-02-17 19:40:50 -0800 | [diff] [blame] | 1699 | func (mm *onuMetricsManager) populateEthernetBridgeHistoryMetrics(ctx context.Context, classID me.ClassID, entityID uint16, |
Girish Gowdra | e0140f0 | 2021-02-02 16:55:09 -0800 | [diff] [blame] | 1700 | meAttributes me.AttributeValueMap, requestedAttributes me.AttributeValueMap, ethPMHistData map[string]float32, intervalEndTime *int) error { |
Girish Gowdra | 5c5aaf4 | 2021-02-17 19:40:50 -0800 | [diff] [blame] | 1701 | upstream := false |
| 1702 | if classID == me.EthernetFramePerformanceMonitoringHistoryDataUpstreamClassID { |
| 1703 | upstream = true |
| 1704 | } |
Girish Gowdra | e0140f0 | 2021-02-02 16:55:09 -0800 | [diff] [blame] | 1705 | // Make sure "IntervalEndTime" is part of the requested attributes as we need this to compare the get responses when get request is multipart |
| 1706 | if _, ok := requestedAttributes["IntervalEndTime"]; !ok { |
| 1707 | requestedAttributes["IntervalEndTime"] = 0 |
| 1708 | } |
| 1709 | if meInstance := mm.pDeviceHandler.pOnuOmciDevice.PDevOmciCC.sendGetMe(ctx, classID, entityID, requestedAttributes, ConstDefaultOmciTimeout, true, mm.pAdaptFsm.commChan); meInstance != nil { |
| 1710 | select { |
| 1711 | case meAttributes = <-mm.l2PmChan: |
| 1712 | logger.Debugw(ctx, "received ethernet pm history data metrics", |
| 1713 | log.Fields{"device-id": mm.pDeviceHandler.deviceID, "upstream": upstream, "entityID": entityID}) |
| 1714 | case <-time.After(time.Duration(ConstDefaultOmciTimeout) * time.Second): |
| 1715 | logger.Errorw(ctx, "timeout waiting for omci-get response for ethernet pm history data", |
| 1716 | log.Fields{"device-id": mm.pDeviceHandler.deviceID, "upstream": upstream, "entityID": entityID}) |
| 1717 | // The metrics will be empty in this case |
| 1718 | return fmt.Errorf("timeout-during-l2-pm-collection-for-ethernet-bridge-history-%v", mm.pDeviceHandler.deviceID) |
| 1719 | } |
| 1720 | // verify that interval end time has not changed during metric collection. If it changed, we abort the procedure |
Girish Gowdra | 5c5aaf4 | 2021-02-17 19:40:50 -0800 | [diff] [blame] | 1721 | if valid := mm.updateAndValidateIntervalEndTime(ctx, entityID, meAttributes, intervalEndTime); !valid { |
| 1722 | return fmt.Errorf("interval-end-time-changed-during-metric-collection-for-ethernet-bridge-history-%v", mm.pDeviceHandler.deviceID) |
Girish Gowdra | e0140f0 | 2021-02-02 16:55:09 -0800 | [diff] [blame] | 1723 | } |
| 1724 | } |
| 1725 | for k := range EthernetBridgeHistory { |
| 1726 | // populate ethPMHistData only if metric key not already present (or populated), since it is possible that we populate |
| 1727 | // the attributes in multiple iterations for a given L2 PM ME as there is a limit on the max OMCI GET payload size. |
| 1728 | if _, ok := ethPMHistData[k]; !ok { |
| 1729 | switch k { |
Girish Gowdra | 0e53364 | 2021-03-02 22:02:51 -0800 | [diff] [blame] | 1730 | case "entity_id": |
| 1731 | if val, ok := meAttributes["ManagedEntityId"]; ok && val != nil { |
| 1732 | ethPMHistData[k] = float32(val.(uint16)) |
| 1733 | } |
Girish Gowdra | e0140f0 | 2021-02-02 16:55:09 -0800 | [diff] [blame] | 1734 | case "drop_events": |
| 1735 | if val, ok := meAttributes["DropEvents"]; ok && val != nil { |
| 1736 | ethPMHistData[k] = float32(val.(uint32)) |
| 1737 | } |
| 1738 | case "octets": |
| 1739 | if val, ok := meAttributes["Octets"]; ok && val != nil { |
| 1740 | ethPMHistData[k] = float32(val.(uint32)) |
| 1741 | } |
| 1742 | case "packets": |
| 1743 | if val, ok := meAttributes["Packets"]; ok && val != nil { |
| 1744 | ethPMHistData[k] = float32(val.(uint32)) |
| 1745 | } |
| 1746 | case "broadcast_packets": |
| 1747 | if val, ok := meAttributes["BroadcastPackets"]; ok && val != nil { |
| 1748 | ethPMHistData[k] = float32(val.(uint32)) |
| 1749 | } |
| 1750 | case "multicast_packets": |
| 1751 | if val, ok := meAttributes["MulticastPackets"]; ok && val != nil { |
| 1752 | ethPMHistData[k] = float32(val.(uint32)) |
| 1753 | } |
| 1754 | case "crc_errored_packets": |
| 1755 | if val, ok := meAttributes["CrcErroredPackets"]; ok && val != nil { |
| 1756 | ethPMHistData[k] = float32(val.(uint32)) |
| 1757 | } |
| 1758 | case "undersize_packets": |
| 1759 | if val, ok := meAttributes["UndersizePackets"]; ok && val != nil { |
| 1760 | ethPMHistData[k] = float32(val.(uint32)) |
| 1761 | } |
| 1762 | case "oversize_packets": |
| 1763 | if val, ok := meAttributes["OversizePackets"]; ok && val != nil { |
| 1764 | ethPMHistData[k] = float32(val.(uint32)) |
| 1765 | } |
| 1766 | case "64_octets": |
| 1767 | if val, ok := meAttributes["Packets64Octets"]; ok && val != nil { |
| 1768 | ethPMHistData[k] = float32(val.(uint32)) |
| 1769 | } |
| 1770 | case "65_to_127_octets": |
| 1771 | if val, ok := meAttributes["Packets65To127Octets"]; ok && val != nil { |
| 1772 | ethPMHistData[k] = float32(val.(uint32)) |
| 1773 | } |
| 1774 | case "128_to_255_octets": |
| 1775 | if val, ok := meAttributes["Packets128To255Octets"]; ok && val != nil { |
| 1776 | ethPMHistData[k] = float32(val.(uint32)) |
| 1777 | } |
| 1778 | case "256_to_511_octets": |
| 1779 | if val, ok := meAttributes["Packets256To511Octets"]; ok && val != nil { |
| 1780 | ethPMHistData[k] = float32(val.(uint32)) |
| 1781 | } |
| 1782 | case "512_to_1023_octets": |
| 1783 | if val, ok := meAttributes["Packets512To1023Octets"]; ok && val != nil { |
| 1784 | ethPMHistData[k] = float32(val.(uint32)) |
| 1785 | } |
| 1786 | case "1024_to_1518_octets": |
| 1787 | if val, ok := meAttributes["Packets1024To1518Octets"]; ok && val != nil { |
| 1788 | ethPMHistData[k] = float32(val.(uint32)) |
| 1789 | } |
| 1790 | default: |
| 1791 | // do nothing |
| 1792 | } |
| 1793 | } |
| 1794 | } |
| 1795 | return nil |
| 1796 | } |
| 1797 | |
| 1798 | // nolint: gocyclo |
| 1799 | func (mm *onuMetricsManager) populateEthernetUniHistoryMetrics(ctx context.Context, classID me.ClassID, entityID uint16, |
| 1800 | meAttributes me.AttributeValueMap, requestedAttributes me.AttributeValueMap, ethPMUniHistData map[string]float32, intervalEndTime *int) error { |
| 1801 | // Make sure "IntervalEndTime" is part of the requested attributes as we need this to compare the get responses when get request is multipart |
| 1802 | if _, ok := requestedAttributes["IntervalEndTime"]; !ok { |
| 1803 | requestedAttributes["IntervalEndTime"] = 0 |
| 1804 | } |
| 1805 | if meInstance := mm.pDeviceHandler.pOnuOmciDevice.PDevOmciCC.sendGetMe(ctx, classID, entityID, requestedAttributes, ConstDefaultOmciTimeout, true, mm.pAdaptFsm.commChan); meInstance != nil { |
| 1806 | select { |
| 1807 | case meAttributes = <-mm.l2PmChan: |
| 1808 | logger.Debugw(ctx, "received ethernet uni history data metrics", |
| 1809 | log.Fields{"device-id": mm.pDeviceHandler.deviceID, "entityID": entityID}) |
| 1810 | case <-time.After(time.Duration(ConstDefaultOmciTimeout) * time.Second): |
| 1811 | logger.Errorw(ctx, "timeout waiting for omci-get response for ethernet uni history data", |
| 1812 | log.Fields{"device-id": mm.pDeviceHandler.deviceID, "entityID": entityID}) |
| 1813 | // The metrics will be empty in this case |
| 1814 | return fmt.Errorf("timeout-during-l2-pm-collection-for-ethernet-uni-history-%v", mm.pDeviceHandler.deviceID) |
| 1815 | } |
| 1816 | // verify that interval end time has not changed during metric collection. If it changed, we abort the procedure |
Girish Gowdra | 5c5aaf4 | 2021-02-17 19:40:50 -0800 | [diff] [blame] | 1817 | if valid := mm.updateAndValidateIntervalEndTime(ctx, entityID, meAttributes, intervalEndTime); !valid { |
| 1818 | return fmt.Errorf("interval-end-time-changed-during-metric-collection-for-ethernet-uni-history-%v", mm.pDeviceHandler.deviceID) |
Girish Gowdra | e0140f0 | 2021-02-02 16:55:09 -0800 | [diff] [blame] | 1819 | } |
| 1820 | } |
| 1821 | for k := range EthernetUniHistory { |
| 1822 | // populate ethPMUniHistData only if metric key not already present (or populated), since it is possible that we populate |
| 1823 | // the attributes in multiple iterations for a given L2 PM ME as there is a limit on the max OMCI GET payload size. |
| 1824 | if _, ok := ethPMUniHistData[k]; !ok { |
| 1825 | switch k { |
Girish Gowdra | 0e53364 | 2021-03-02 22:02:51 -0800 | [diff] [blame] | 1826 | case "entity_id": |
| 1827 | if val, ok := meAttributes["ManagedEntityId"]; ok && val != nil { |
| 1828 | ethPMUniHistData[k] = float32(val.(uint16)) |
| 1829 | } |
Girish Gowdra | e0140f0 | 2021-02-02 16:55:09 -0800 | [diff] [blame] | 1830 | case "fcs_errors": |
| 1831 | if val, ok := meAttributes["FcsErrors"]; ok && val != nil { |
| 1832 | ethPMUniHistData[k] = float32(val.(uint32)) |
| 1833 | } |
| 1834 | case "excessive_collision_counter": |
| 1835 | if val, ok := meAttributes["ExcessiveCollisionCounter"]; ok && val != nil { |
| 1836 | ethPMUniHistData[k] = float32(val.(uint32)) |
| 1837 | } |
| 1838 | case "late_collision_counter": |
| 1839 | if val, ok := meAttributes["LateCollisionCounter"]; ok && val != nil { |
| 1840 | ethPMUniHistData[k] = float32(val.(uint32)) |
| 1841 | } |
| 1842 | case "frames_too_long": |
| 1843 | if val, ok := meAttributes["FramesTooLong"]; ok && val != nil { |
| 1844 | ethPMUniHistData[k] = float32(val.(uint32)) |
| 1845 | } |
| 1846 | case "buffer_overflows_on_rx": |
| 1847 | if val, ok := meAttributes["BufferOverflowsOnReceive"]; ok && val != nil { |
| 1848 | ethPMUniHistData[k] = float32(val.(uint32)) |
| 1849 | } |
| 1850 | case "buffer_overflows_on_tx": |
| 1851 | if val, ok := meAttributes["BufferOverflowsOnTransmit"]; ok && val != nil { |
| 1852 | ethPMUniHistData[k] = float32(val.(uint32)) |
| 1853 | } |
| 1854 | case "single_collision_frame_counter": |
| 1855 | if val, ok := meAttributes["SingleCollisionFrameCounter"]; ok && val != nil { |
| 1856 | ethPMUniHistData[k] = float32(val.(uint32)) |
| 1857 | } |
| 1858 | case "multiple_collisions_frame_counter": |
| 1859 | if val, ok := meAttributes["MultipleCollisionsFrameCounter"]; ok && val != nil { |
| 1860 | ethPMUniHistData[k] = float32(val.(uint32)) |
| 1861 | } |
| 1862 | case "sqe_counter": |
| 1863 | if val, ok := meAttributes["SqeCounter"]; ok && val != nil { |
| 1864 | ethPMUniHistData[k] = float32(val.(uint32)) |
| 1865 | } |
| 1866 | case "deferred_tx_counter": |
| 1867 | if val, ok := meAttributes["DeferredTransmissionCounter"]; ok && val != nil { |
| 1868 | ethPMUniHistData[k] = float32(val.(uint32)) |
| 1869 | } |
| 1870 | case "internal_mac_tx_error_counter": |
| 1871 | if val, ok := meAttributes["InternalMacTransmitErrorCounter"]; ok && val != nil { |
| 1872 | ethPMUniHistData[k] = float32(val.(uint32)) |
| 1873 | } |
| 1874 | case "carrier_sense_error_counter": |
| 1875 | if val, ok := meAttributes["CarrierSenseErrorCounter"]; ok && val != nil { |
| 1876 | ethPMUniHistData[k] = float32(val.(uint32)) |
| 1877 | } |
| 1878 | case "alignment_error_counter": |
| 1879 | if val, ok := meAttributes["AlignmentErrorCounter"]; ok && val != nil { |
| 1880 | ethPMUniHistData[k] = float32(val.(uint32)) |
| 1881 | } |
| 1882 | case "internal_mac_rx_error_counter": |
| 1883 | if val, ok := meAttributes["InternalMacReceiveErrorCounter"]; ok && val != nil { |
| 1884 | ethPMUniHistData[k] = float32(val.(uint32)) |
| 1885 | } |
| 1886 | default: |
| 1887 | // do nothing |
| 1888 | } |
| 1889 | } |
| 1890 | } |
| 1891 | return nil |
| 1892 | } |
| 1893 | |
Girish Gowdra | 5c5aaf4 | 2021-02-17 19:40:50 -0800 | [diff] [blame] | 1894 | // nolint: gocyclo |
| 1895 | func (mm *onuMetricsManager) populateFecHistoryMetrics(ctx context.Context, classID me.ClassID, entityID uint16, |
| 1896 | meAttributes me.AttributeValueMap, requestedAttributes me.AttributeValueMap, fecHistData map[string]float32, intervalEndTime *int) error { |
| 1897 | // Make sure "IntervalEndTime" is part of the requested attributes as we need this to compare the get responses when get request is multipart |
| 1898 | if _, ok := requestedAttributes["IntervalEndTime"]; !ok { |
| 1899 | requestedAttributes["IntervalEndTime"] = 0 |
| 1900 | } |
| 1901 | if meInstance := mm.pDeviceHandler.pOnuOmciDevice.PDevOmciCC.sendGetMe(ctx, classID, entityID, requestedAttributes, ConstDefaultOmciTimeout, true, mm.pAdaptFsm.commChan); meInstance != nil { |
| 1902 | select { |
| 1903 | case meAttributes = <-mm.l2PmChan: |
| 1904 | logger.Debugw(ctx, "received fec history data metrics", |
| 1905 | log.Fields{"device-id": mm.pDeviceHandler.deviceID, "entityID": entityID}) |
| 1906 | case <-time.After(time.Duration(ConstDefaultOmciTimeout) * time.Second): |
| 1907 | logger.Errorw(ctx, "timeout waiting for omci-get response for fec history data", |
| 1908 | log.Fields{"device-id": mm.pDeviceHandler.deviceID, "entityID": entityID}) |
| 1909 | // The metrics will be empty in this case |
| 1910 | return fmt.Errorf("timeout-during-l2-pm-collection-for-fec-history-%v", mm.pDeviceHandler.deviceID) |
| 1911 | } |
| 1912 | // verify that interval end time has not changed during metric collection. If it changed, we abort the procedure |
| 1913 | if valid := mm.updateAndValidateIntervalEndTime(ctx, entityID, meAttributes, intervalEndTime); !valid { |
| 1914 | return fmt.Errorf("interval-end-time-changed-during-metric-collection-for-fec-history-%v", mm.pDeviceHandler.deviceID) |
| 1915 | } |
| 1916 | } |
| 1917 | for k := range FecHistory { |
| 1918 | // populate fecHistData only if metric key not already present (or populated), since it is possible that we populate |
| 1919 | // the attributes in multiple iterations for a given L2 PM ME as there is a limit on the max OMCI GET payload size. |
| 1920 | if _, ok := fecHistData[k]; !ok { |
| 1921 | switch k { |
Girish Gowdra | 0e53364 | 2021-03-02 22:02:51 -0800 | [diff] [blame] | 1922 | case "entity_id": |
| 1923 | if val, ok := meAttributes["ManagedEntityId"]; ok && val != nil { |
| 1924 | fecHistData[k] = float32(val.(uint16)) |
| 1925 | } |
Girish Gowdra | 5c5aaf4 | 2021-02-17 19:40:50 -0800 | [diff] [blame] | 1926 | case "corrected_bytes": |
| 1927 | if val, ok := meAttributes["CorrectedBytes"]; ok && val != nil { |
| 1928 | fecHistData[k] = float32(val.(uint32)) |
| 1929 | } |
| 1930 | case "corrected_code_words": |
| 1931 | if val, ok := meAttributes["CorrectedCodeWords"]; ok && val != nil { |
| 1932 | fecHistData[k] = float32(val.(uint32)) |
| 1933 | } |
| 1934 | case "uncorrectable_code_words": |
| 1935 | if val, ok := meAttributes["UncorrectableCodeWords"]; ok && val != nil { |
| 1936 | fecHistData[k] = float32(val.(uint32)) |
| 1937 | } |
| 1938 | case "total_code_words": |
| 1939 | if val, ok := meAttributes["TotalCodeWords"]; ok && val != nil { |
| 1940 | fecHistData[k] = float32(val.(uint32)) |
| 1941 | } |
| 1942 | case "fec_seconds": |
| 1943 | if val, ok := meAttributes["FecSeconds"]; ok && val != nil { |
| 1944 | fecHistData[k] = float32(val.(uint16)) |
| 1945 | } |
| 1946 | default: |
| 1947 | // do nothing |
| 1948 | } |
| 1949 | } |
| 1950 | } |
| 1951 | return nil |
| 1952 | } |
| 1953 | |
| 1954 | // nolint: gocyclo |
| 1955 | func (mm *onuMetricsManager) populateGemPortMetrics(ctx context.Context, classID me.ClassID, entityID uint16, |
| 1956 | meAttributes me.AttributeValueMap, requestedAttributes me.AttributeValueMap, gemPortHistData map[string]float32, intervalEndTime *int) error { |
| 1957 | // Make sure "IntervalEndTime" is part of the requested attributes as we need this to compare the get responses when get request is multipart |
| 1958 | if _, ok := requestedAttributes["IntervalEndTime"]; !ok { |
| 1959 | requestedAttributes["IntervalEndTime"] = 0 |
| 1960 | } |
| 1961 | if meInstance := mm.pDeviceHandler.pOnuOmciDevice.PDevOmciCC.sendGetMe(ctx, classID, entityID, requestedAttributes, ConstDefaultOmciTimeout, true, mm.pAdaptFsm.commChan); meInstance != nil { |
| 1962 | select { |
| 1963 | case meAttributes = <-mm.l2PmChan: |
| 1964 | logger.Debugw(ctx, "received gem port history data metrics", |
| 1965 | log.Fields{"device-id": mm.pDeviceHandler.deviceID, "entityID": entityID}) |
| 1966 | case <-time.After(time.Duration(ConstDefaultOmciTimeout) * time.Second): |
| 1967 | logger.Errorw(ctx, "timeout waiting for omci-get response for gem port history data", |
| 1968 | log.Fields{"device-id": mm.pDeviceHandler.deviceID, "entityID": entityID}) |
| 1969 | // The metrics will be empty in this case |
| 1970 | return fmt.Errorf("timeout-during-l2-pm-collection-for-gemport-history-%v", mm.pDeviceHandler.deviceID) |
| 1971 | } |
| 1972 | // verify that interval end time has not changed during metric collection. If it changed, we abort the procedure |
| 1973 | if valid := mm.updateAndValidateIntervalEndTime(ctx, entityID, meAttributes, intervalEndTime); !valid { |
| 1974 | return fmt.Errorf("interval-end-time-changed-during-metric-collection-for-gemport-history-%v", mm.pDeviceHandler.deviceID) |
| 1975 | } |
| 1976 | } |
| 1977 | for k := range GemPortHistory { |
| 1978 | // populate gemPortHistData only if metric key not already present (or populated), since it is possible that we populate |
| 1979 | // the attributes in multiple iterations for a given L2 PM ME as there is a limit on the max OMCI GET payload size. |
| 1980 | if _, ok := gemPortHistData[k]; !ok { |
| 1981 | switch k { |
Girish Gowdra | 0e53364 | 2021-03-02 22:02:51 -0800 | [diff] [blame] | 1982 | case "entity_id": |
| 1983 | if val, ok := meAttributes["ManagedEntityId"]; ok && val != nil { |
| 1984 | gemPortHistData[k] = float32(val.(uint16)) |
| 1985 | } |
Girish Gowdra | 5c5aaf4 | 2021-02-17 19:40:50 -0800 | [diff] [blame] | 1986 | case "transmitted_gem_frames": |
| 1987 | if val, ok := meAttributes["TransmittedGemFrames"]; ok && val != nil { |
| 1988 | gemPortHistData[k] = float32(val.(uint32)) |
| 1989 | } |
| 1990 | case "received_gem_frames": |
| 1991 | if val, ok := meAttributes["ReceivedGemFrames"]; ok && val != nil { |
| 1992 | gemPortHistData[k] = float32(val.(uint32)) |
| 1993 | } |
| 1994 | case "received_payload_bytes": |
| 1995 | if val, ok := meAttributes["ReceivedPayloadBytes"]; ok && val != nil { |
| 1996 | gemPortHistData[k] = float32(val.(uint64)) |
| 1997 | } |
| 1998 | case "transmitted_payload_bytes": |
| 1999 | if val, ok := meAttributes["TransmittedPayloadBytes"]; ok && val != nil { |
| 2000 | gemPortHistData[k] = float32(val.(uint64)) |
| 2001 | } |
| 2002 | case "encryption_key_errors": |
| 2003 | if val, ok := meAttributes["EncryptionKeyErrors"]; ok && val != nil { |
| 2004 | gemPortHistData[k] = float32(val.(uint32)) |
| 2005 | } |
| 2006 | default: |
| 2007 | // do nothing |
| 2008 | } |
| 2009 | } |
| 2010 | } |
| 2011 | return nil |
| 2012 | } |
| 2013 | |
Girish Gowdra | e0140f0 | 2021-02-02 16:55:09 -0800 | [diff] [blame] | 2014 | func (mm *onuMetricsManager) handleOmciCreateResponseMessage(ctx context.Context, msg OmciMessage) error { |
| 2015 | msgLayer := (*msg.OmciPacket).Layer(omci.LayerTypeCreateResponse) |
| 2016 | if msgLayer == nil { |
| 2017 | logger.Errorw(ctx, "omci Msg layer could not be detected for create response - handling stopped", log.Fields{"device-id": mm.pDeviceHandler.deviceID}) |
| 2018 | return fmt.Errorf("omci Msg layer could not be detected for create response - handling stopped: %s", mm.pDeviceHandler.deviceID) |
| 2019 | } |
| 2020 | msgObj, msgOk := msgLayer.(*omci.CreateResponse) |
| 2021 | if !msgOk { |
| 2022 | logger.Errorw(ctx, "omci Msg layer could not be assigned for create response - handling stopped", log.Fields{"device-id": mm.pDeviceHandler.deviceID}) |
| 2023 | return fmt.Errorf("omci Msg layer could not be assigned for delete response - handling stopped: %s", mm.pDeviceHandler.deviceID) |
| 2024 | } |
| 2025 | logger.Debugw(ctx, "OMCI create response Data", log.Fields{"device-id": mm.pDeviceHandler.deviceID, "data-fields": msgObj}) |
| 2026 | switch msgObj.EntityClass { |
| 2027 | case me.EthernetFramePerformanceMonitoringHistoryDataUpstreamClassID, |
| 2028 | me.EthernetFramePerformanceMonitoringHistoryDataDownstreamClassID, |
Girish Gowdra | 5c5aaf4 | 2021-02-17 19:40:50 -0800 | [diff] [blame] | 2029 | me.EthernetPerformanceMonitoringHistoryDataClassID, |
| 2030 | me.FecPerformanceMonitoringHistoryDataClassID, |
| 2031 | me.GemPortNetworkCtpPerformanceMonitoringHistoryDataClassID: |
Girish Gowdra | e0140f0 | 2021-02-02 16:55:09 -0800 | [diff] [blame] | 2032 | // If the result is me.InstanceExists it means the entity was already created. It is ok handled that as success |
| 2033 | if msgObj.Result == me.Success || msgObj.Result == me.InstanceExists { |
| 2034 | mm.l2PmCreateOrDeleteResponseChan <- true |
| 2035 | } else { |
| 2036 | logger.Warnw(ctx, "failed to create me", log.Fields{"device-id": mm.pDeviceHandler.deviceID, "class-id": msgObj.EntityClass}) |
| 2037 | mm.l2PmCreateOrDeleteResponseChan <- false |
| 2038 | } |
| 2039 | return nil |
| 2040 | default: |
| 2041 | logger.Errorw(ctx, "unhandled omci create response message", |
| 2042 | log.Fields{"device-id": mm.pDeviceHandler.deviceID, "class-id": msgObj.EntityClass}) |
| 2043 | } |
| 2044 | return fmt.Errorf("unhandled-omci-create-response-message-%v", mm.pDeviceHandler.deviceID) |
| 2045 | } |
| 2046 | |
| 2047 | func (mm *onuMetricsManager) handleOmciDeleteResponseMessage(ctx context.Context, msg OmciMessage) error { |
| 2048 | msgLayer := (*msg.OmciPacket).Layer(omci.LayerTypeDeleteResponse) |
| 2049 | if msgLayer == nil { |
| 2050 | logger.Errorw(ctx, "omci Msg layer could not be detected for delete response - handling stopped", log.Fields{"device-id": mm.pDeviceHandler.deviceID}) |
| 2051 | return fmt.Errorf("omci Msg layer could not be detected for create response - handling stopped: %s", mm.pDeviceHandler.deviceID) |
| 2052 | } |
| 2053 | msgObj, msgOk := msgLayer.(*omci.DeleteResponse) |
| 2054 | if !msgOk { |
| 2055 | logger.Errorw(ctx, "omci Msg layer could not be assigned for delete response - handling stopped", log.Fields{"device-id": mm.pDeviceHandler.deviceID}) |
| 2056 | return fmt.Errorf("omci Msg layer could not be assigned for delete response - handling stopped: %s", mm.pDeviceHandler.deviceID) |
| 2057 | } |
| 2058 | logger.Debugw(ctx, "OMCI delete response Data", log.Fields{"device-id": mm.pDeviceHandler.deviceID, "data-fields": msgObj}) |
| 2059 | switch msgObj.EntityClass { |
| 2060 | case me.EthernetFramePerformanceMonitoringHistoryDataUpstreamClassID, |
| 2061 | me.EthernetFramePerformanceMonitoringHistoryDataDownstreamClassID, |
Girish Gowdra | 5c5aaf4 | 2021-02-17 19:40:50 -0800 | [diff] [blame] | 2062 | me.EthernetPerformanceMonitoringHistoryDataClassID, |
| 2063 | me.FecPerformanceMonitoringHistoryDataClassID, |
| 2064 | me.GemPortNetworkCtpPerformanceMonitoringHistoryDataClassID: |
Girish Gowdra | e0140f0 | 2021-02-02 16:55:09 -0800 | [diff] [blame] | 2065 | // If the result is me.UnknownInstance it means the entity was already deleted. It is ok handled that as success |
| 2066 | if msgObj.Result == me.Success || msgObj.Result == me.UnknownInstance { |
| 2067 | mm.l2PmCreateOrDeleteResponseChan <- true |
| 2068 | } else { |
| 2069 | logger.Warnw(ctx, "failed to delete me", log.Fields{"device-id": mm.pDeviceHandler.deviceID, "class-id": msgObj.EntityClass}) |
| 2070 | mm.l2PmCreateOrDeleteResponseChan <- false |
| 2071 | } |
| 2072 | return nil |
| 2073 | default: |
| 2074 | logger.Errorw(ctx, "unhandled omci delete response message", |
| 2075 | log.Fields{"device-id": mm.pDeviceHandler.deviceID, "class-id": msgObj.EntityClass}) |
| 2076 | } |
| 2077 | return fmt.Errorf("unhandled-omci-delete-response-message-%v", mm.pDeviceHandler.deviceID) |
| 2078 | } |
| 2079 | |
| 2080 | func (mm *onuMetricsManager) generateTicks(ctx context.Context) { |
| 2081 | for { |
| 2082 | select { |
| 2083 | case <-time.After(L2PmCollectionInterval * time.Second): |
| 2084 | go func() { |
| 2085 | if err := mm.pAdaptFsm.pFsm.Event(l2PmEventTick); err != nil { |
| 2086 | logger.Errorw(ctx, "error calling event", log.Fields{"device-id": mm.pDeviceHandler.deviceID, "err": err}) |
| 2087 | } |
| 2088 | }() |
| 2089 | case <-mm.stopTicks: |
| 2090 | logger.Infow(ctx, "stopping ticks", log.Fields{"device-id": mm.pDeviceHandler.deviceID}) |
| 2091 | return |
| 2092 | } |
| 2093 | } |
| 2094 | } |
| 2095 | |
Girish Gowdra | 5c5aaf4 | 2021-02-17 19:40:50 -0800 | [diff] [blame] | 2096 | func (mm *onuMetricsManager) handleMetricsPublish(ctx context.Context, metricName string, metricInfoSlice []*voltha.MetricInformation) { |
| 2097 | // Publish metrics if it is valid |
| 2098 | if metricInfoSlice != nil { |
| 2099 | mm.publishMetrics(ctx, metricInfoSlice) |
| 2100 | } else { |
| 2101 | // If collectAttempts exceeds L2PmCollectAttempts then remove it from activeL2Pms |
| 2102 | // slice so that we do not collect data from that PM ME anymore |
| 2103 | mm.onuMetricsManagerLock.Lock() |
| 2104 | mm.groupMetricMap[metricName].collectAttempts++ |
| 2105 | if mm.groupMetricMap[metricName].collectAttempts > L2PmCollectAttempts { |
| 2106 | mm.activeL2Pms = mm.removeIfFoundString(mm.activeL2Pms, metricName) |
| 2107 | } |
| 2108 | logger.Warnw(ctx, "state collect data - no metrics collected", |
| 2109 | log.Fields{"device-id": mm.pDeviceHandler.deviceID, "metricName": metricName, "collectAttempts": mm.groupMetricMap[metricName].collectAttempts}) |
| 2110 | mm.onuMetricsManagerLock.Unlock() |
| 2111 | } |
| 2112 | } |
| 2113 | |
| 2114 | func (mm *onuMetricsManager) populateGroupSpecificMetrics(ctx context.Context, mEnt *me.ManagedEntity, classID me.ClassID, entityID uint16, |
| 2115 | meAttributes me.AttributeValueMap, data map[string]float32, intervalEndTime *int) error { |
| 2116 | var grpFunc groupMetricPopulateFunc |
| 2117 | switch classID { |
| 2118 | case me.EthernetFramePerformanceMonitoringHistoryDataUpstreamClassID, me.EthernetFramePerformanceMonitoringHistoryDataDownstreamClassID: |
| 2119 | grpFunc = mm.populateEthernetBridgeHistoryMetrics |
| 2120 | case me.EthernetPerformanceMonitoringHistoryDataClassID: |
| 2121 | grpFunc = mm.populateEthernetUniHistoryMetrics |
| 2122 | case me.FecPerformanceMonitoringHistoryDataClassID: |
| 2123 | grpFunc = mm.populateFecHistoryMetrics |
| 2124 | case me.GemPortNetworkCtpPerformanceMonitoringHistoryDataClassID: |
| 2125 | grpFunc = mm.populateGemPortMetrics |
| 2126 | default: |
| 2127 | return fmt.Errorf("unknown-classid-%v", classID) |
| 2128 | } |
| 2129 | |
| 2130 | size := 0 |
| 2131 | requestedAttributes := make(me.AttributeValueMap) |
| 2132 | for _, v := range mEnt.GetAttributeDefinitions() { |
| 2133 | if (v.Size + size) <= MaxL2PMGetPayLoadSize { |
| 2134 | requestedAttributes[v.Name] = v.DefValue |
| 2135 | size = v.Size + size |
| 2136 | } else { // We exceeded the allow omci get size |
| 2137 | // Let's collect the attributes via get now and collect remaining in the next iteration |
| 2138 | if err := grpFunc(ctx, classID, entityID, meAttributes, requestedAttributes, data, intervalEndTime); err != nil { |
| 2139 | logger.Errorw(ctx, "error during metric collection", |
| 2140 | log.Fields{"device-id": mm.pDeviceHandler.deviceID, "entityID": entityID, "err": err}) |
| 2141 | return err |
| 2142 | } |
| 2143 | size = 0 // reset size |
| 2144 | requestedAttributes = make(me.AttributeValueMap) // reset map |
| 2145 | } |
| 2146 | } |
| 2147 | // Collect the omci get attributes for the last bunch of attributes. |
| 2148 | if len(requestedAttributes) > 0 { |
| 2149 | if err := grpFunc(ctx, classID, entityID, meAttributes, requestedAttributes, data, intervalEndTime); err != nil { |
| 2150 | logger.Errorw(ctx, "error during metric collection", |
| 2151 | log.Fields{"device-id": mm.pDeviceHandler.deviceID, "entityID": entityID, "err": err}) |
| 2152 | return err |
| 2153 | } |
| 2154 | } |
| 2155 | return nil |
| 2156 | } |
| 2157 | |
| 2158 | func (mm *onuMetricsManager) populateOnuMetricInfo(title string, data map[string]float32) voltha.MetricInformation { |
| 2159 | metricsContext := make(map[string]string) |
| 2160 | metricsContext["onuID"] = fmt.Sprintf("%d", mm.pDeviceHandler.device.ProxyAddress.OnuId) |
| 2161 | metricsContext["intfID"] = fmt.Sprintf("%d", mm.pDeviceHandler.device.ProxyAddress.ChannelId) |
| 2162 | metricsContext["devicetype"] = mm.pDeviceHandler.DeviceType |
| 2163 | |
| 2164 | raisedTs := time.Now().Unix() |
| 2165 | mmd := voltha.MetricMetaData{ |
| 2166 | Title: title, |
| 2167 | Ts: float64(raisedTs), |
| 2168 | Context: metricsContext, |
| 2169 | DeviceId: mm.pDeviceHandler.deviceID, |
| 2170 | LogicalDeviceId: mm.pDeviceHandler.logicalDeviceID, |
| 2171 | SerialNo: mm.pDeviceHandler.device.SerialNumber, |
| 2172 | } |
| 2173 | |
| 2174 | // create slice of metrics given that there could be more than one VEIP instance |
| 2175 | metricInfo := voltha.MetricInformation{Metadata: &mmd, Metrics: data} |
| 2176 | return metricInfo |
| 2177 | } |
| 2178 | |
| 2179 | func (mm *onuMetricsManager) updateAndValidateIntervalEndTime(ctx context.Context, entityID uint16, meAttributes me.AttributeValueMap, intervalEndTime *int) bool { |
| 2180 | valid := false |
| 2181 | if *intervalEndTime == -1 { // first time |
| 2182 | // Update the interval end time |
| 2183 | if val, ok := meAttributes["IntervalEndTime"]; ok && val != nil { |
| 2184 | *intervalEndTime = int(meAttributes["IntervalEndTime"].(uint8)) |
| 2185 | valid = true |
| 2186 | } |
| 2187 | } else { |
| 2188 | var currIntervalEndTime int |
| 2189 | if val, ok := meAttributes["IntervalEndTime"]; ok && val != nil { |
| 2190 | currIntervalEndTime = int(meAttributes["IntervalEndTime"].(uint8)) |
| 2191 | } |
| 2192 | if currIntervalEndTime != *intervalEndTime { // interval end time changed during metric collection |
| 2193 | logger.Errorw(ctx, "interval end time changed during metrics collection for ethernet pm history data", |
| 2194 | log.Fields{"device-id": mm.pDeviceHandler.deviceID, "entityID": entityID, |
| 2195 | "currIntervalEndTime": *intervalEndTime, "newIntervalEndTime": currIntervalEndTime}) |
| 2196 | } else { |
| 2197 | valid = true |
| 2198 | } |
| 2199 | } |
| 2200 | return valid |
| 2201 | } |
| 2202 | |
| 2203 | func (mm *onuMetricsManager) waitForResponseOrTimeout(ctx context.Context, create bool, instID uint16, meClassName string) bool { |
| 2204 | logger.Debugw(ctx, "waitForResponseOrTimeout", log.Fields{"create": create, "instID": instID, "meClassName": meClassName}) |
| 2205 | select { |
| 2206 | case resp := <-mm.l2PmCreateOrDeleteResponseChan: |
Girish Gowdra | 36ccf7d | 2021-02-25 20:42:51 -0800 | [diff] [blame] | 2207 | logger.Debugw(ctx, "received l2 pm me response", |
Girish Gowdra | 5c5aaf4 | 2021-02-17 19:40:50 -0800 | [diff] [blame] | 2208 | log.Fields{"device-id": mm.pDeviceHandler.deviceID, "resp": resp, "create": create, "meClassName": meClassName, "instID": instID}) |
| 2209 | return resp |
| 2210 | case <-time.After(time.Duration(ConstDefaultOmciTimeout) * time.Second): |
Girish Gowdra | 36ccf7d | 2021-02-25 20:42:51 -0800 | [diff] [blame] | 2211 | logger.Errorw(ctx, "timeout waiting for l2 pm me response", |
Girish Gowdra | 5c5aaf4 | 2021-02-17 19:40:50 -0800 | [diff] [blame] | 2212 | log.Fields{"device-id": mm.pDeviceHandler.deviceID, "resp": false, "create": create, "meClassName": meClassName, "instID": instID}) |
| 2213 | } |
| 2214 | return false |
| 2215 | } |
| 2216 | |
| 2217 | func (mm *onuMetricsManager) initializeGroupMetric(grpMtrcs map[string]voltha.PmConfig_PmType, grpName string, grpEnabled bool, grpFreq uint32) { |
| 2218 | var pmConfigSlice []*voltha.PmConfig |
| 2219 | for k, v := range grpMtrcs { |
Girish Gowdra | 0e53364 | 2021-03-02 22:02:51 -0800 | [diff] [blame] | 2220 | pmConfigSlice = append(pmConfigSlice, |
| 2221 | &voltha.PmConfig{ |
| 2222 | Name: k, |
| 2223 | Type: v, |
| 2224 | Enabled: grpEnabled && mm.pDeviceHandler.pOpenOnuAc.metricsEnabled, |
| 2225 | SampleFreq: grpFreq}) |
Girish Gowdra | 5c5aaf4 | 2021-02-17 19:40:50 -0800 | [diff] [blame] | 2226 | } |
| 2227 | groupMetric := voltha.PmGroupConfig{ |
| 2228 | GroupName: grpName, |
| 2229 | Enabled: grpEnabled && mm.pDeviceHandler.pOpenOnuAc.metricsEnabled, |
| 2230 | GroupFreq: grpFreq, |
| 2231 | Metrics: pmConfigSlice, |
| 2232 | } |
| 2233 | mm.pDeviceHandler.pmConfigs.Groups = append(mm.pDeviceHandler.pmConfigs.Groups, &groupMetric) |
| 2234 | |
| 2235 | } |
| 2236 | |
| 2237 | func (mm *onuMetricsManager) initializeL2PmFsm(ctx context.Context, aCommChannel chan Message) error { |
| 2238 | mm.pAdaptFsm = NewAdapterFsm("L2PmFSM", mm.pDeviceHandler.deviceID, aCommChannel) |
| 2239 | if mm.pAdaptFsm == nil { |
| 2240 | logger.Errorw(ctx, "L2PMFsm AdapterFsm could not be instantiated!!", log.Fields{ |
| 2241 | "device-id": mm.pDeviceHandler.deviceID}) |
| 2242 | return fmt.Errorf("nil-adapter-fsm") |
| 2243 | } |
| 2244 | // L2 PM FSM related state machine |
| 2245 | mm.pAdaptFsm.pFsm = fsm.NewFSM( |
| 2246 | l2PmStNull, |
| 2247 | fsm.Events{ |
| 2248 | {Name: l2PmEventInit, Src: []string{l2PmStNull}, Dst: l2PmStStarting}, |
| 2249 | {Name: l2PmEventTick, Src: []string{l2PmStStarting}, Dst: l2PmStSyncTime}, |
| 2250 | {Name: l2PmEventTick, Src: []string{l2PmStIdle, l2PmEventDeleteMe, l2PmEventAddMe}, Dst: l2PmStCollectData}, |
| 2251 | {Name: l2PmEventSuccess, Src: []string{l2PmStSyncTime, l2PmStCreatePmMe, l2PmStDeletePmMe, l2PmStCollectData}, Dst: l2PmStIdle}, |
| 2252 | {Name: l2PmEventFailure, Src: []string{l2PmStCreatePmMe, l2PmStDeletePmMe, l2PmStCollectData}, Dst: l2PmStIdle}, |
| 2253 | {Name: l2PmEventFailure, Src: []string{l2PmStSyncTime}, Dst: l2PmStSyncTime}, |
| 2254 | {Name: l2PmEventAddMe, Src: []string{l2PmStIdle}, Dst: l2PmStCreatePmMe}, |
| 2255 | {Name: l2PmEventDeleteMe, Src: []string{l2PmStIdle}, Dst: l2PmStDeletePmMe}, |
| 2256 | {Name: l2PmEventStop, Src: []string{l2PmStNull, l2PmStStarting, l2PmStSyncTime, l2PmStIdle, l2PmStCreatePmMe, l2PmStDeletePmMe, l2PmStCollectData}, Dst: l2PmStNull}, |
| 2257 | }, |
| 2258 | fsm.Callbacks{ |
| 2259 | "enter_state": func(e *fsm.Event) { mm.pAdaptFsm.logFsmStateChange(ctx, e) }, |
| 2260 | "enter_" + l2PmStNull: func(e *fsm.Event) { mm.l2PMFsmNull(ctx, e) }, |
| 2261 | "enter_" + l2PmStIdle: func(e *fsm.Event) { mm.l2PMFsmIdle(ctx, e) }, |
| 2262 | "enter_" + l2PmStStarting: func(e *fsm.Event) { mm.l2PMFsmStarting(ctx, e) }, |
| 2263 | "enter_" + l2PmStSyncTime: func(e *fsm.Event) { mm.l2PMFsmSyncTime(ctx, e) }, |
| 2264 | "enter_" + l2PmStCollectData: func(e *fsm.Event) { mm.l2PmFsmCollectData(ctx, e) }, |
| 2265 | "enter_" + l2PmStCreatePmMe: func(e *fsm.Event) { mm.l2PmFsmCreatePM(ctx, e) }, |
| 2266 | "enter_" + l2PmStDeletePmMe: func(e *fsm.Event) { mm.l2PmFsmDeletePM(ctx, e) }, |
| 2267 | }, |
| 2268 | ) |
| 2269 | return nil |
| 2270 | } |
| 2271 | |
| 2272 | func (mm *onuMetricsManager) initializeAllGroupMetrics() { |
| 2273 | mm.pDeviceHandler.pmConfigs = &voltha.PmConfigs{} |
| 2274 | mm.pDeviceHandler.pmConfigs.Id = mm.pDeviceHandler.deviceID |
| 2275 | mm.pDeviceHandler.pmConfigs.DefaultFreq = DefaultMetricCollectionFrequency |
| 2276 | mm.pDeviceHandler.pmConfigs.Grouped = GroupMetricEnabled |
| 2277 | mm.pDeviceHandler.pmConfigs.FreqOverride = DefaultFrequencyOverrideEnabled |
| 2278 | |
| 2279 | // Populate group metrics. |
| 2280 | // Lets populate irrespective of GroupMetricEnabled is true or not. |
| 2281 | // The group metrics collection will decided on this flag later |
| 2282 | |
| 2283 | mm.initializeGroupMetric(OpticalPowerGroupMetrics, OpticalPowerGroupMetricName, |
| 2284 | OpticalPowerGroupMetricEnabled, OpticalPowerMetricGroupCollectionFrequency) |
| 2285 | |
| 2286 | mm.initializeGroupMetric(UniStatusGroupMetrics, UniStatusGroupMetricName, |
| 2287 | UniStatusGroupMetricEnabled, UniStatusMetricGroupCollectionFrequency) |
| 2288 | |
| 2289 | // classical l2 pm counter start |
| 2290 | |
| 2291 | mm.initializeGroupMetric(EthernetBridgeHistory, EthernetBridgeHistoryName, |
| 2292 | EthernetBridgeHistoryEnabled, EthernetBridgeHistoryFrequency) |
| 2293 | |
| 2294 | mm.initializeGroupMetric(EthernetUniHistory, EthernetUniHistoryName, |
| 2295 | EthernetUniHistoryEnabled, EthernetUniHistoryFrequency) |
| 2296 | |
| 2297 | mm.initializeGroupMetric(FecHistory, FecHistoryName, |
| 2298 | FecHistoryEnabled, FecHistoryFrequency) |
| 2299 | |
| 2300 | mm.initializeGroupMetric(GemPortHistory, GemPortHistoryName, |
| 2301 | GemPortHistoryEnabled, GemPortHistoryFrequency) |
| 2302 | |
| 2303 | // classical l2 pm counter end |
| 2304 | |
| 2305 | // Add standalone metric (if present) after this (will be added to dh.pmConfigs.Metrics) |
| 2306 | } |
| 2307 | |
| 2308 | func (mm *onuMetricsManager) populateLocalGroupMetricData(ctx context.Context) { |
| 2309 | // Populate local group metric structures |
| 2310 | for _, g := range mm.pDeviceHandler.pmConfigs.Groups { |
| 2311 | mm.groupMetricMap[g.GroupName] = &groupMetric{ |
| 2312 | groupName: g.GroupName, |
| 2313 | enabled: g.Enabled, |
| 2314 | frequency: g.GroupFreq, |
| 2315 | } |
| 2316 | switch g.GroupName { |
| 2317 | case OpticalPowerGroupMetricName: |
| 2318 | mm.groupMetricMap[g.GroupName].metricMap = OpticalPowerGroupMetrics |
| 2319 | case UniStatusGroupMetricName: |
| 2320 | mm.groupMetricMap[g.GroupName].metricMap = UniStatusGroupMetrics |
| 2321 | case EthernetBridgeHistoryName: |
| 2322 | mm.groupMetricMap[g.GroupName].metricMap = EthernetBridgeHistory |
| 2323 | mm.groupMetricMap[g.GroupName].isL2PMCounter = true |
| 2324 | case EthernetUniHistoryName: |
| 2325 | mm.groupMetricMap[g.GroupName].metricMap = EthernetUniHistory |
| 2326 | mm.groupMetricMap[g.GroupName].isL2PMCounter = true |
| 2327 | case FecHistoryName: |
| 2328 | mm.groupMetricMap[g.GroupName].metricMap = FecHistory |
| 2329 | mm.groupMetricMap[g.GroupName].isL2PMCounter = true |
| 2330 | case GemPortHistoryName: |
| 2331 | mm.groupMetricMap[g.GroupName].metricMap = GemPortHistory |
| 2332 | mm.groupMetricMap[g.GroupName].isL2PMCounter = true |
| 2333 | default: |
| 2334 | logger.Errorw(ctx, "unhandled-group-name", log.Fields{"groupName": g.GroupName}) |
| 2335 | } |
| 2336 | } |
| 2337 | |
| 2338 | // Populate local standalone metric structures |
| 2339 | for _, m := range mm.pDeviceHandler.pmConfigs.Metrics { |
| 2340 | mm.standaloneMetricMap[m.Name] = &standaloneMetric{ |
| 2341 | metricName: m.Name, |
| 2342 | enabled: m.Enabled, |
| 2343 | frequency: m.SampleFreq, |
| 2344 | } |
| 2345 | switch m.Name { |
| 2346 | // None exist as of now. Add when available. |
| 2347 | default: |
| 2348 | logger.Errorw(ctx, "unhandled-metric-name", log.Fields{"metricName": m.Name}) |
| 2349 | } |
| 2350 | } |
| 2351 | } |
| 2352 | |
| 2353 | func (mm *onuMetricsManager) AddGemPortForPerfMonitoring(gemPortNTPInstID uint16) { |
| 2354 | mm.onuMetricsManagerLock.Lock() |
| 2355 | defer mm.onuMetricsManagerLock.Unlock() |
| 2356 | // mark the instance for addition |
Girish Gowdra | 0e53364 | 2021-03-02 22:02:51 -0800 | [diff] [blame] | 2357 | mm.groupMetricMap[GemPortHistoryName].pmMEData.InstancesToAdd = mm.appendIfMissingUnt16(mm.groupMetricMap[GemPortHistoryName].pmMEData.InstancesToAdd, gemPortNTPInstID) |
Girish Gowdra | 5c5aaf4 | 2021-02-17 19:40:50 -0800 | [diff] [blame] | 2358 | // If the instance presence toggles too soon, we need to remove it from gemPortNCTPPerfHistInstToDelete slice |
Girish Gowdra | 0e53364 | 2021-03-02 22:02:51 -0800 | [diff] [blame] | 2359 | mm.groupMetricMap[GemPortHistoryName].pmMEData.InstancesToDelete = mm.removeIfFoundUint16(mm.groupMetricMap[GemPortHistoryName].pmMEData.InstancesToDelete, gemPortNTPInstID) |
Girish Gowdra | 5c5aaf4 | 2021-02-17 19:40:50 -0800 | [diff] [blame] | 2360 | |
| 2361 | mm.l2PmToAdd = mm.appendIfMissingString(mm.l2PmToAdd, GemPortHistoryName) |
| 2362 | // We do not need to remove from l2PmToDelete slice as we could have Add and Delete of |
| 2363 | // GemPortPerfHistory ME simultaneously for different instances of the ME. |
| 2364 | // The creation or deletion of an instance is decided based on its presence in gemPortNCTPPerfHistInstToDelete or |
| 2365 | // gemPortNCTPPerfHistInstToAdd slice |
| 2366 | } |
| 2367 | |
| 2368 | func (mm *onuMetricsManager) RemoveGemPortForPerfMonitoring(gemPortNTPInstID uint16) { |
| 2369 | mm.onuMetricsManagerLock.Lock() |
| 2370 | defer mm.onuMetricsManagerLock.Unlock() |
Girish Gowdra | 0e53364 | 2021-03-02 22:02:51 -0800 | [diff] [blame] | 2371 | mm.groupMetricMap[GemPortHistoryName].pmMEData.InstancesToDelete = mm.appendIfMissingUnt16(mm.groupMetricMap[GemPortHistoryName].pmMEData.InstancesToDelete, gemPortNTPInstID) |
Girish Gowdra | 5c5aaf4 | 2021-02-17 19:40:50 -0800 | [diff] [blame] | 2372 | // If the instance presence toggles too soon, we need to remove it from gemPortNCTPPerfHistInstToAdd slice |
Girish Gowdra | 0e53364 | 2021-03-02 22:02:51 -0800 | [diff] [blame] | 2373 | mm.groupMetricMap[GemPortHistoryName].pmMEData.InstancesToAdd = mm.removeIfFoundUint16(mm.groupMetricMap[GemPortHistoryName].pmMEData.InstancesToAdd, gemPortNTPInstID) |
Girish Gowdra | 5c5aaf4 | 2021-02-17 19:40:50 -0800 | [diff] [blame] | 2374 | |
| 2375 | mm.l2PmToDelete = mm.appendIfMissingString(mm.l2PmToDelete, GemPortHistoryName) |
| 2376 | // We do not need to remove from l2PmToAdd slice as we could have Add and Delete of |
| 2377 | // GemPortPerfHistory ME simultaneously for different instances of the ME. |
| 2378 | // The creation or deletion of an instance is decided based on its presence in gemPortNCTPPerfHistInstToDelete or |
| 2379 | // gemPortNCTPPerfHistInstToAdd slice |
| 2380 | } |
| 2381 | |
Girish Gowdra | 36ccf7d | 2021-02-25 20:42:51 -0800 | [diff] [blame] | 2382 | func (mm *onuMetricsManager) updateGemPortNTPInstanceToAddForPerfMonitoring(ctx context.Context) { |
Girish Gowdra | 5c5aaf4 | 2021-02-17 19:40:50 -0800 | [diff] [blame] | 2383 | if mm.pDeviceHandler.pOnuTP != nil { |
| 2384 | gemPortInstIDs := mm.pDeviceHandler.pOnuTP.GetAllBidirectionalGemPortIDsForOnu() |
Girish Gowdra | 36ccf7d | 2021-02-25 20:42:51 -0800 | [diff] [blame] | 2385 | // NOTE: It is expected that caller of this function has acquired the required mutex for synchronization purposes |
Girish Gowdra | 5c5aaf4 | 2021-02-17 19:40:50 -0800 | [diff] [blame] | 2386 | for _, v := range gemPortInstIDs { |
| 2387 | // mark the instance for addition |
Girish Gowdra | 0e53364 | 2021-03-02 22:02:51 -0800 | [diff] [blame] | 2388 | mm.groupMetricMap[GemPortHistoryName].pmMEData.InstancesToAdd = mm.appendIfMissingUnt16(mm.groupMetricMap[GemPortHistoryName].pmMEData.InstancesToAdd, v) |
Girish Gowdra | 5c5aaf4 | 2021-02-17 19:40:50 -0800 | [diff] [blame] | 2389 | // If the instance presence toggles too soon, we need to remove it from gemPortNCTPPerfHistInstToDelete slice |
Girish Gowdra | 0e53364 | 2021-03-02 22:02:51 -0800 | [diff] [blame] | 2390 | mm.groupMetricMap[GemPortHistoryName].pmMEData.InstancesToDelete = mm.removeIfFoundUint16(mm.groupMetricMap[GemPortHistoryName].pmMEData.InstancesToDelete, v) |
Girish Gowdra | 5c5aaf4 | 2021-02-17 19:40:50 -0800 | [diff] [blame] | 2391 | } |
Girish Gowdra | 36ccf7d | 2021-02-25 20:42:51 -0800 | [diff] [blame] | 2392 | logger.Debugw(ctx, "updateGemPortNTPInstanceToAddForPerfMonitoring", |
Girish Gowdra | 0e53364 | 2021-03-02 22:02:51 -0800 | [diff] [blame] | 2393 | log.Fields{"deviceID": mm.pDeviceHandler.deviceID, "gemToAdd": mm.groupMetricMap[GemPortHistoryName].pmMEData.InstancesToAdd, "gemToDel": mm.groupMetricMap[GemPortHistoryName].pmMEData.InstancesToDelete}) |
Girish Gowdra | 5c5aaf4 | 2021-02-17 19:40:50 -0800 | [diff] [blame] | 2394 | } |
| 2395 | } |
| 2396 | |
Girish Gowdra | 36ccf7d | 2021-02-25 20:42:51 -0800 | [diff] [blame] | 2397 | func (mm *onuMetricsManager) updateGemPortNTPInstanceToDeleteForPerfMonitoring(ctx context.Context) { |
Girish Gowdra | 5c5aaf4 | 2021-02-17 19:40:50 -0800 | [diff] [blame] | 2398 | if mm.pDeviceHandler.pOnuTP != nil { |
| 2399 | gemPortInstIDs := mm.pDeviceHandler.pOnuTP.GetAllBidirectionalGemPortIDsForOnu() |
Girish Gowdra | 36ccf7d | 2021-02-25 20:42:51 -0800 | [diff] [blame] | 2400 | // NOTE: It is expected that caller of this function has acquired the required mutex for synchronization purposes |
Girish Gowdra | 5c5aaf4 | 2021-02-17 19:40:50 -0800 | [diff] [blame] | 2401 | for _, v := range gemPortInstIDs { |
Girish Gowdra | 0e53364 | 2021-03-02 22:02:51 -0800 | [diff] [blame] | 2402 | mm.groupMetricMap[GemPortHistoryName].pmMEData.InstancesToDelete = mm.appendIfMissingUnt16(mm.groupMetricMap[GemPortHistoryName].pmMEData.InstancesToDelete, v) |
Girish Gowdra | 5c5aaf4 | 2021-02-17 19:40:50 -0800 | [diff] [blame] | 2403 | // If the instance presence toggles too soon, we need to remove it from gemPortNCTPPerfHistInstToAdd slice |
Girish Gowdra | 0e53364 | 2021-03-02 22:02:51 -0800 | [diff] [blame] | 2404 | mm.groupMetricMap[GemPortHistoryName].pmMEData.InstancesToAdd = mm.removeIfFoundUint16(mm.groupMetricMap[GemPortHistoryName].pmMEData.InstancesToAdd, v) |
Girish Gowdra | 5c5aaf4 | 2021-02-17 19:40:50 -0800 | [diff] [blame] | 2405 | } |
| 2406 | } |
Girish Gowdra | 36ccf7d | 2021-02-25 20:42:51 -0800 | [diff] [blame] | 2407 | logger.Debugw(ctx, "updateGemPortNTPInstanceToDeleteForPerfMonitoring", |
Girish Gowdra | 0e53364 | 2021-03-02 22:02:51 -0800 | [diff] [blame] | 2408 | log.Fields{"deviceID": mm.pDeviceHandler.deviceID, "gemToAdd": mm.groupMetricMap[GemPortHistoryName].pmMEData.InstancesToAdd, "gemToDel": mm.groupMetricMap[GemPortHistoryName].pmMEData.InstancesToDelete}) |
| 2409 | } |
| 2410 | |
| 2411 | // restorePmData restores any PM data available on the KV store to local cache |
| 2412 | func (mm *onuMetricsManager) restorePmData(ctx context.Context) error { |
| 2413 | logger.Debugw(ctx, "restorePmData - start", log.Fields{"device-id": mm.pDeviceHandler.deviceID}) |
| 2414 | if mm.pmKvStore == nil { |
| 2415 | logger.Errorw(ctx, "pmKvStore not set - abort", log.Fields{"device-id": mm.pDeviceHandler.deviceID}) |
| 2416 | return fmt.Errorf(fmt.Sprintf("pmKvStore-not-set-abort-%s", mm.pDeviceHandler.deviceID)) |
| 2417 | } |
| 2418 | var errorsList []error |
| 2419 | for groupName, group := range mm.groupMetricMap { |
| 2420 | group.pmMEData = &pmMEData{} |
| 2421 | Value, err := mm.pmKvStore.Get(ctx, groupName) |
| 2422 | if err == nil { |
| 2423 | if Value != nil { |
| 2424 | logger.Debugw(ctx, "PM data read", |
| 2425 | log.Fields{"Key": Value.Key, "device-id": mm.pDeviceHandler.deviceID}) |
| 2426 | tmpBytes, _ := kvstore.ToByte(Value.Value) |
| 2427 | |
| 2428 | if err = json.Unmarshal(tmpBytes, &group.pmMEData); err != nil { |
| 2429 | logger.Errorw(ctx, "unable to unmarshal PM data", log.Fields{"error": err, "device-id": mm.pDeviceHandler.deviceID}) |
| 2430 | errorsList = append(errorsList, fmt.Errorf(fmt.Sprintf("unable-to-unmarshal-PM-data-%s-for-group-%s", mm.pDeviceHandler.deviceID, groupName))) |
| 2431 | continue |
| 2432 | } |
| 2433 | logger.Debugw(ctx, "restorePmData - success", log.Fields{"pmData": group.pmMEData, "groupName": groupName, "device-id": mm.pDeviceHandler.deviceID}) |
| 2434 | } else { |
| 2435 | logger.Debugw(ctx, "no PM data found", log.Fields{"groupName": groupName, "device-id": mm.pDeviceHandler.deviceID}) |
| 2436 | continue |
| 2437 | } |
| 2438 | } else { |
| 2439 | logger.Errorw(ctx, "restorePmData - fail", log.Fields{"device-id": mm.pDeviceHandler.deviceID, "groupName": groupName, "err": err}) |
| 2440 | errorsList = append(errorsList, fmt.Errorf(fmt.Sprintf("unable-to-read-from-KVstore-%s-for-group-%s", mm.pDeviceHandler.deviceID, groupName))) |
| 2441 | continue |
| 2442 | } |
| 2443 | } |
| 2444 | if len(errorsList) > 0 { |
| 2445 | return fmt.Errorf("errors-restoring-pm-data-for-one-or-more-groups--errors:%v", errorsList) |
| 2446 | } |
| 2447 | logger.Debugw(ctx, "restorePmData - complete success", log.Fields{"device-id": mm.pDeviceHandler.deviceID}) |
| 2448 | return nil |
| 2449 | } |
| 2450 | |
| 2451 | // getPmData gets pmMEData from cache. Since we have write through cache implementation for pmMEData, |
| 2452 | // the data must be available in cache. |
| 2453 | // Note, it is expected that caller of this function manages the required synchronization (like using locks etc.). |
| 2454 | func (mm *onuMetricsManager) getPmData(ctx context.Context, groupName string) (*pmMEData, error) { |
| 2455 | if grp, ok := mm.groupMetricMap[groupName]; ok { |
| 2456 | return grp.pmMEData, nil |
| 2457 | } |
| 2458 | // Data not in cache, try to fetch from kv store. |
| 2459 | data := &pmMEData{} |
| 2460 | if mm.pmKvStore == nil { |
| 2461 | logger.Errorw(ctx, "pmKvStore not set - abort", log.Fields{"device-id": mm.pDeviceHandler.deviceID}) |
| 2462 | return data, fmt.Errorf("pmKvStore not set. device-id - %s", mm.pDeviceHandler.deviceID) |
| 2463 | } |
| 2464 | Value, err := mm.pmKvStore.Get(ctx, groupName) |
| 2465 | if err == nil { |
| 2466 | if Value != nil { |
| 2467 | logger.Debugw(ctx, "PM data read", |
| 2468 | log.Fields{"Key": Value.Key, "device-id": mm.pDeviceHandler.deviceID}) |
| 2469 | tmpBytes, _ := kvstore.ToByte(Value.Value) |
| 2470 | |
| 2471 | if err = json.Unmarshal(tmpBytes, data); err != nil { |
| 2472 | logger.Errorw(ctx, "unable to unmarshal PM data", log.Fields{"error": err, "device-id": mm.pDeviceHandler.deviceID}) |
| 2473 | return data, err |
| 2474 | } |
| 2475 | logger.Debugw(ctx, "PM data", log.Fields{"pmData": data, "groupName": groupName, "device-id": mm.pDeviceHandler.deviceID}) |
| 2476 | } else { |
| 2477 | logger.Debugw(ctx, "no PM data found", log.Fields{"groupName": groupName, "device-id": mm.pDeviceHandler.deviceID}) |
| 2478 | return data, err |
| 2479 | } |
| 2480 | } else { |
| 2481 | logger.Errorw(ctx, "unable to read from KVstore", log.Fields{"device-id": mm.pDeviceHandler.deviceID}) |
| 2482 | return data, err |
| 2483 | } |
| 2484 | |
| 2485 | return data, nil |
| 2486 | } |
| 2487 | |
| 2488 | // updatePmData update pmMEData to store. It is write through cache, i.e., write to cache first and then update store |
| 2489 | func (mm *onuMetricsManager) updatePmData(ctx context.Context, groupName string, meInstanceID uint16, pmAction string) error { |
| 2490 | logger.Debugw(ctx, "updatePmData - start", log.Fields{"device-id": mm.pDeviceHandler.deviceID, "groupName": groupName, "entityID": meInstanceID, "pmAction": pmAction}) |
| 2491 | mm.onuMetricsManagerLock.Lock() |
| 2492 | defer mm.onuMetricsManagerLock.Unlock() |
| 2493 | |
| 2494 | if mm.pmKvStore == nil { |
| 2495 | logger.Errorw(ctx, "pmKvStore not set - abort", log.Fields{"device-id": mm.pDeviceHandler.deviceID}) |
| 2496 | return fmt.Errorf(fmt.Sprintf("pmKvStore-not-set-abort-%s", mm.pDeviceHandler.deviceID)) |
| 2497 | } |
| 2498 | |
| 2499 | pmMEData, err := mm.getPmData(ctx, groupName) |
| 2500 | if err != nil || pmMEData == nil { |
| 2501 | // error already logged in called function. |
| 2502 | return err |
| 2503 | } |
| 2504 | switch pmAction { |
| 2505 | case cPmAdd: |
| 2506 | pmMEData.InstancesToAdd = mm.appendIfMissingUnt16(pmMEData.InstancesToAdd, meInstanceID) |
| 2507 | pmMEData.InstancesToDelete = mm.removeIfFoundUint16(pmMEData.InstancesToDelete, meInstanceID) |
| 2508 | pmMEData.InstancesActive = mm.removeIfFoundUint16(pmMEData.InstancesActive, meInstanceID) |
| 2509 | case cPmAdded: |
| 2510 | pmMEData.InstancesActive = mm.appendIfMissingUnt16(pmMEData.InstancesActive, meInstanceID) |
| 2511 | pmMEData.InstancesToAdd = mm.removeIfFoundUint16(pmMEData.InstancesToAdd, meInstanceID) |
| 2512 | pmMEData.InstancesToDelete = mm.removeIfFoundUint16(pmMEData.InstancesToDelete, meInstanceID) |
| 2513 | case cPmRemove: |
| 2514 | pmMEData.InstancesToDelete = mm.appendIfMissingUnt16(pmMEData.InstancesToDelete, meInstanceID) |
| 2515 | pmMEData.InstancesToAdd = mm.removeIfFoundUint16(pmMEData.InstancesToAdd, meInstanceID) |
| 2516 | pmMEData.InstancesActive = mm.removeIfFoundUint16(pmMEData.InstancesActive, meInstanceID) |
| 2517 | case cPmRemoved: |
| 2518 | pmMEData.InstancesToDelete = mm.removeIfFoundUint16(pmMEData.InstancesToDelete, meInstanceID) |
| 2519 | pmMEData.InstancesToAdd = mm.removeIfFoundUint16(pmMEData.InstancesToAdd, meInstanceID) |
| 2520 | pmMEData.InstancesActive = mm.removeIfFoundUint16(pmMEData.InstancesActive, meInstanceID) |
| 2521 | default: |
| 2522 | logger.Errorw(ctx, "unknown pm action", log.Fields{"device-id": mm.pDeviceHandler.deviceID, "pmAction": pmAction, "groupName": groupName}) |
| 2523 | return fmt.Errorf(fmt.Sprintf("unknown-pm-action-deviceid-%s-groupName-%s-pmaction-%s", mm.pDeviceHandler.deviceID, groupName, pmAction)) |
| 2524 | } |
| 2525 | // write through cache |
| 2526 | mm.groupMetricMap[groupName].pmMEData = pmMEData |
| 2527 | |
| 2528 | Value, err := json.Marshal(*pmMEData) |
| 2529 | if err != nil { |
| 2530 | logger.Errorw(ctx, "unable to marshal PM data", log.Fields{"groupName": groupName, "pmAction": pmAction, "pmData": *pmMEData, "err": err}) |
| 2531 | return err |
| 2532 | } |
| 2533 | // Update back to kv store |
| 2534 | if err = mm.pmKvStore.Put(ctx, groupName, Value); err != nil { |
| 2535 | logger.Errorw(ctx, "unable to put PM data to kv store", log.Fields{"groupName": groupName, "pmData": *pmMEData, "pmAction": pmAction, "err": err}) |
| 2536 | return err |
| 2537 | } |
| 2538 | logger.Debugw(ctx, "updatePmData - success", log.Fields{"device-id": mm.pDeviceHandler.deviceID, "groupName": groupName, "pmData": *pmMEData, "pmAction": pmAction}) |
| 2539 | |
| 2540 | return nil |
| 2541 | } |
| 2542 | |
| 2543 | // clearPmGroupData cleans PM Group data from store |
| 2544 | func (mm *onuMetricsManager) clearPmGroupData(ctx context.Context) error { |
| 2545 | mm.onuMetricsManagerLock.Lock() |
| 2546 | defer mm.onuMetricsManagerLock.Unlock() |
| 2547 | logger.Debugw(ctx, "clearPmGroupData - start", log.Fields{"device-id": mm.pDeviceHandler.deviceID}) |
| 2548 | if mm.pmKvStore == nil { |
| 2549 | logger.Errorw(ctx, "pmKvStore not set - abort", log.Fields{"device-id": mm.pDeviceHandler.deviceID}) |
| 2550 | return fmt.Errorf(fmt.Sprintf("pmKvStore-not-set-abort-%s", mm.pDeviceHandler.deviceID)) |
| 2551 | } |
| 2552 | |
| 2553 | for n := range mm.groupMetricMap { |
| 2554 | if err := mm.pmKvStore.Delete(ctx, n); err != nil { |
| 2555 | logger.Errorw(ctx, "clearPmGroupData - fail", log.Fields{"deviceID": mm.pDeviceHandler.deviceID, "groupName": n, "err": err}) |
| 2556 | // do not abort this procedure. continue to delete next group. |
| 2557 | } else { |
| 2558 | logger.Debugw(ctx, "clearPmGroupData - success", log.Fields{"device-id": mm.pDeviceHandler.deviceID, "groupName": n}) |
| 2559 | } |
| 2560 | } |
| 2561 | |
| 2562 | return nil |
| 2563 | } |
| 2564 | |
| 2565 | // clearAllPmData clears all PM data associated with the device from KV store |
| 2566 | func (mm *onuMetricsManager) clearAllPmData(ctx context.Context) error { |
| 2567 | mm.onuMetricsManagerLock.Lock() |
| 2568 | defer mm.onuMetricsManagerLock.Unlock() |
| 2569 | logger.Debugw(ctx, "clearAllPmData - start", log.Fields{"device-id": mm.pDeviceHandler.deviceID}) |
| 2570 | if mm.pmKvStore == nil { |
| 2571 | logger.Errorw(ctx, "pmKvStore not set - abort", log.Fields{"device-id": mm.pDeviceHandler.deviceID}) |
| 2572 | return fmt.Errorf(fmt.Sprintf("pmKvStore-not-set-abort-%s", mm.pDeviceHandler.deviceID)) |
| 2573 | } |
| 2574 | |
| 2575 | if err := mm.pmKvStore.Delete(ctx, ""); err != nil { |
| 2576 | logger.Errorw(ctx, "unable to delete PM data from kv store", log.Fields{"deviceID": mm.pDeviceHandler.deviceID, "err": err}) |
| 2577 | return err |
| 2578 | } |
| 2579 | logger.Debugw(ctx, "clearAllPmData - success", log.Fields{"device-id": mm.pDeviceHandler.deviceID}) |
| 2580 | return nil |
Girish Gowdra | 5c5aaf4 | 2021-02-17 19:40:50 -0800 | [diff] [blame] | 2581 | } |
| 2582 | |
| 2583 | func (mm *onuMetricsManager) appendIfMissingString(slice []string, n string) []string { |
Girish Gowdra | e0140f0 | 2021-02-02 16:55:09 -0800 | [diff] [blame] | 2584 | for _, ele := range slice { |
| 2585 | if ele == n { |
| 2586 | return slice |
| 2587 | } |
| 2588 | } |
| 2589 | return append(slice, n) |
| 2590 | } |
| 2591 | |
Girish Gowdra | 5c5aaf4 | 2021-02-17 19:40:50 -0800 | [diff] [blame] | 2592 | func (mm *onuMetricsManager) removeIfFoundString(slice []string, n string) []string { |
| 2593 | for i, ele := range slice { |
| 2594 | if ele == n { |
| 2595 | return append(slice[:i], slice[i+1:]...) |
| 2596 | } |
| 2597 | } |
| 2598 | return slice |
| 2599 | } |
| 2600 | |
| 2601 | func (mm *onuMetricsManager) appendIfMissingUnt16(slice []uint16, n uint16) []uint16 { |
| 2602 | for _, ele := range slice { |
| 2603 | if ele == n { |
| 2604 | return slice |
| 2605 | } |
| 2606 | } |
| 2607 | return append(slice, n) |
| 2608 | } |
| 2609 | |
| 2610 | func (mm *onuMetricsManager) removeIfFoundUint16(slice []uint16, n uint16) []uint16 { |
Girish Gowdra | e0140f0 | 2021-02-02 16:55:09 -0800 | [diff] [blame] | 2611 | for i, ele := range slice { |
| 2612 | if ele == n { |
| 2613 | return append(slice[:i], slice[i+1:]...) |
| 2614 | } |
| 2615 | } |
| 2616 | return slice |
Girish Gowdra | e09a620 | 2021-01-12 18:10:59 -0800 | [diff] [blame] | 2617 | } |