| 1 | // SPDX-License-Identifier: GPL-3.0-or-later |
| 2 | |
| 3 | package nats |
| 4 | |
| 5 | import ( |
| 6 | "fmt" |
| 7 | "maps" |
| 8 | "strconv" |
| 9 | "strings" |
| 10 | |
| 11 | "golang.org/x/text/cases" |
| 12 | "golang.org/x/text/language" |
| 13 | |
| 14 | "github.com/netdata/netdata/go/plugins/plugin/framework/collectorapi" |
| 15 | ) |
| 16 | |
| 17 | const ( |
| 18 | prioServerTraffic = collectorapi.Priority + iota |
| 19 | prioServerMessages |
| 20 | prioServerConnections |
| 21 | prioServerConnectionsRate |
| 22 | prioHttpEndpointRequests |
| 23 | prioServerHealthProbeStatus |
| 24 | prioServerCpuUsage |
| 25 | prioServerMemoryUsage |
| 26 | prioServerUptime |
| 27 | |
| 28 | prioJetStreamStatus |
| 29 | prioJetStreamStreams |
| 30 | prioJetStreamConsumers |
| 31 | prioJetStreamBytes |
| 32 | prioJetStreamMessages |
| 33 | prioJetStreamApiRequests |
| 34 | prioJetStreamApiErrors |
| 35 | prioJetStreamApiInflight |
| 36 | prioJetStreamMemoryUsed |
| 37 | prioJetStreamStorageUsed |
| 38 | |
| 39 | prioAccountTraffic |
| 40 | prioAccountMessages |
| 41 | prioAccountConnections |
| 42 | prioAccountConnectionsRate |
| 43 | prioAccountSubscriptions |
| 44 | prioAccountSlowConsumers |
| 45 | prioAccountLeafNodes |
| 46 | |
| 47 | prioRouteTraffic |
| 48 | prioRouteMessages |
| 49 | prioRouteSubscriptions |
| 50 | |
| 51 | prioGatewayConnTraffic |
| 52 | prioGatewayConnMessages |
| 53 | prioGatewayConnSubscriptions |
| 54 | prioGatewayConnUptime |
| 55 | |
| 56 | prioLeafConnTraffic |
| 57 | prioLeafConnMessages |
| 58 | prioLeafConnSubscriptions |
| 59 | prioLeafRTT |
| 60 | ) |
| 61 | |
| 62 | func serverCharts() *collectorapi.Charts { |
| 63 | charts := collectorapi.Charts{ |
| 64 | chartServerConnectionsCurrent.Copy(), |
| 65 | chartServerConnectionsRate.Copy(), |
| 66 | chartServerTraffic.Copy(), |
| 67 | chartServerMessages.Copy(), |
| 68 | chartServerHealthProbeStatus.Copy(), |
| 69 | chartServerCpuUsage.Copy(), |
| 70 | chartServerMemUsage.Copy(), |
| 71 | chartServerUptime.Copy(), |
| 72 | } |
| 73 | charts = append(charts, httpEndpointsCharts()...) |
| 74 | charts = append(charts, *jetStreamCharts.Copy()...) |
| 75 | return charts.Copy() |
| 76 | } |
| 77 | |
| 78 | var ( |
| 79 | chartServerTraffic = collectorapi.Chart{ |
| 80 | ID: "server_traffic", |
| 81 | Title: "Server Traffic", |
| 82 | Units: "bytes/s", |
| 83 | Fam: "traffic", |
| 84 | Ctx: "nats.server_traffic", |
| 85 | Priority: prioServerTraffic, |
| 86 | Type: collectorapi.Area, |
| 87 | Dims: collectorapi.Dims{ |
| 88 | {ID: "varz_srv_in_bytes", Name: "received", Algo: collectorapi.Incremental}, |
| 89 | {ID: "varz_srv_out_bytes", Name: "sent", Mul: -1, Algo: collectorapi.Incremental}, |
| 90 | }, |
| 91 | } |
| 92 | chartServerMessages = collectorapi.Chart{ |
| 93 | ID: "server_messages", |
| 94 | Title: "Server Messages", |
| 95 | Units: "messages/s", |
| 96 | Fam: "traffic", |
| 97 | Ctx: "nats.server_messages", |
| 98 | Priority: prioServerMessages, |
| 99 | Dims: collectorapi.Dims{ |
| 100 | {ID: "varz_srv_in_msgs", Name: "received", Algo: collectorapi.Incremental}, |
| 101 | {ID: "varz_srv_out_msgs", Name: "sent", Mul: -1, Algo: collectorapi.Incremental}, |
| 102 | }, |
| 103 | } |
| 104 | chartServerConnectionsCurrent = collectorapi.Chart{ |
| 105 | ID: "server_connections", |
| 106 | Title: "Server Active Connections", |
| 107 | Units: "connections", |
| 108 | Fam: "connections", |
| 109 | Ctx: "nats.server_connections", |
| 110 | Priority: prioServerConnections, |
| 111 | Dims: collectorapi.Dims{ |
| 112 | {ID: "varz_srv_connections", Name: "active"}, |
| 113 | }, |
| 114 | } |
| 115 | chartServerConnectionsRate = collectorapi.Chart{ |
| 116 | ID: "server_connections_rate", |
| 117 | Title: "Server Connections", |
| 118 | Units: "connections/s", |
| 119 | Fam: "connections", |
| 120 | Ctx: "nats.server_connections_rate", |
| 121 | Priority: prioServerConnectionsRate, |
| 122 | Dims: collectorapi.Dims{ |
| 123 | {ID: "varz_srv_total_connections", Name: "connections", Algo: collectorapi.Incremental}, |
| 124 | }, |
| 125 | } |
| 126 | chartServerHealthProbeStatus = collectorapi.Chart{ |
| 127 | ID: "server_health_probe_status", |
| 128 | Title: "Server Health Probe Status", |
| 129 | Units: "status", |
| 130 | Fam: "health", |
| 131 | Ctx: "nats.server_health_probe_status", |
| 132 | Priority: prioServerHealthProbeStatus, |
| 133 | Dims: collectorapi.Dims{ |
| 134 | {ID: "varz_srv_healthz_status_ok", Name: "ok"}, |
| 135 | {ID: "varz_srv_healthz_status_error", Name: "error"}, |
| 136 | }, |
| 137 | } |
| 138 | chartServerCpuUsage = collectorapi.Chart{ |
| 139 | ID: "server_cpu_usage", |
| 140 | Title: "Server CPU Usage", |
| 141 | Units: "percent", |
| 142 | Fam: "rusage", |
| 143 | Ctx: "nats.server_cpu_usage", |
| 144 | Priority: prioServerCpuUsage, |
| 145 | Type: collectorapi.Area, |
| 146 | Dims: collectorapi.Dims{ |
| 147 | {ID: "varz_srv_cpu", Name: "used"}, |
| 148 | }, |
| 149 | } |
| 150 | chartServerMemUsage = collectorapi.Chart{ |
| 151 | ID: "server_mem_usage", |
| 152 | Title: "Server Memory Usage", |
| 153 | Units: "bytes", |
| 154 | Fam: "rusage", |
| 155 | Ctx: "nats.server_mem_usage", |
| 156 | Priority: prioServerMemoryUsage, |
| 157 | Type: collectorapi.Area, |
| 158 | Dims: collectorapi.Dims{ |
| 159 | {ID: "varz_srv_mem", Name: "used"}, |
| 160 | }, |
| 161 | } |
| 162 | chartServerUptime = collectorapi.Chart{ |
| 163 | ID: "server_uptime", |
| 164 | Title: "Server Uptime", |
| 165 | Units: "seconds", |
| 166 | Fam: "uptime", |
| 167 | Ctx: "nats.server_uptime", |
| 168 | Priority: prioServerUptime, |
| 169 | Dims: collectorapi.Dims{ |
| 170 | {ID: "varz_srv_uptime", Name: "uptime"}, |
| 171 | }, |
| 172 | } |
| 173 | ) |
| 174 | |
| 175 | func httpEndpointsCharts() collectorapi.Charts { |
| 176 | var charts collectorapi.Charts |
| 177 | |
| 178 | for _, path := range httpEndpoints { |
| 179 | chart := httpEndpointRequestsChartTmpl.Copy() |
| 180 | |
| 181 | chart.ID = fmt.Sprintf(chart.ID, path) |
| 182 | chart.Labels = []collectorapi.Label{ |
| 183 | {Key: "http_endpoint", Value: path}, |
| 184 | } |
| 185 | for _, dim := range chart.Dims { |
| 186 | dim.ID = fmt.Sprintf(dim.ID, path) |
| 187 | } |
| 188 | charts = append(charts, chart) |
| 189 | } |
| 190 | |
| 191 | return charts |
| 192 | } |
| 193 | |
| 194 | var httpEndpointRequestsChartTmpl = collectorapi.Chart{ |
| 195 | ID: "http_endpoint_%s_requests", |
| 196 | Title: "HTTP Endpoint Requests", |
| 197 | Units: "requests/s", |
| 198 | Fam: "http requests", |
| 199 | Ctx: "nats.http_endpoint_requests", |
| 200 | Priority: prioHttpEndpointRequests, |
| 201 | Dims: collectorapi.Dims{ |
| 202 | {ID: "varz_http_endpoint_%s_req", Name: "requests", Algo: collectorapi.Incremental}, |
| 203 | }, |
| 204 | } |
| 205 | |
| 206 | var jetStreamCharts = collectorapi.Charts{ |
| 207 | jetStreamStatus.Copy(), |
| 208 | jetStreamStreams.Copy(), |
| 209 | jetStreamStreamsStorageBytes.Copy(), |
| 210 | jetStreamStreamsStorageMessages.Copy(), |
| 211 | jetStreamConsumers.Copy(), |
| 212 | jetStreamApiRequests.Copy(), |
| 213 | jetStreamApiInflightRequests.Copy(), |
| 214 | jetStreamApiErrors.Copy(), |
| 215 | jetStreamMemoryUsed.Copy(), |
| 216 | jetStreamStorageUsed.Copy(), |
| 217 | } |
| 218 | |
| 219 | var ( |
| 220 | jetStreamStatus = collectorapi.Chart{ |
| 221 | ID: "jetstream_status", |
| 222 | Title: "JetStream Status", |
| 223 | Units: "status", |
| 224 | Fam: "jstream streams", |
| 225 | Ctx: "nats.jetstream_status", |
| 226 | Priority: prioJetStreamStatus, |
| 227 | Dims: collectorapi.Dims{ |
| 228 | {ID: "jsz_enabled", Name: "enabled"}, |
| 229 | {ID: "jsz_disabled", Name: "disabled"}, |
| 230 | }, |
| 231 | } |
| 232 | jetStreamStreams = collectorapi.Chart{ |
| 233 | ID: "jetstream_streams", |
| 234 | Title: "JetStream Streams", |
| 235 | Units: "streams", |
| 236 | Fam: "jstream streams", |
| 237 | Ctx: "nats.jetstream_streams", |
| 238 | Priority: prioJetStreamStreams, |
| 239 | Dims: collectorapi.Dims{ |
| 240 | {ID: "jsz_streams", Name: "active"}, |
| 241 | }, |
| 242 | } |
| 243 | jetStreamStreamsStorageBytes = collectorapi.Chart{ |
| 244 | ID: "jetstream_streams_storage_bytes", |
| 245 | Title: "JetStream Bytes", |
| 246 | Units: "bytes", |
| 247 | Fam: "jstream streams", |
| 248 | Ctx: "nats.jetstream_streams_storage_bytes", |
| 249 | Priority: prioJetStreamBytes, |
| 250 | Type: collectorapi.Area, |
| 251 | Dims: collectorapi.Dims{ |
| 252 | {ID: "jsz_bytes", Name: "used"}, |
| 253 | }, |
| 254 | } |
| 255 | jetStreamStreamsStorageMessages = collectorapi.Chart{ |
| 256 | ID: "jetstream_streams_storage_messages", |
| 257 | Title: "JetStream Messages", |
| 258 | Units: "messages", |
| 259 | Fam: "jstream streams", |
| 260 | Ctx: "nats.jetstream_streams_storage_messages", |
| 261 | Priority: prioJetStreamMessages, |
| 262 | Dims: collectorapi.Dims{ |
| 263 | {ID: "jsz_messages", Name: "stored"}, |
| 264 | }, |
| 265 | } |
| 266 | jetStreamConsumers = collectorapi.Chart{ |
| 267 | ID: "jetstream_consumers", |
| 268 | Title: "JetStream Consumers", |
| 269 | Units: "consumers", |
| 270 | Fam: "jstream consumers", |
| 271 | Ctx: "nats.jetstream_consumers", |
| 272 | Priority: prioJetStreamConsumers, |
| 273 | Dims: collectorapi.Dims{ |
| 274 | {ID: "jsz_consumers", Name: "active"}, |
| 275 | }, |
| 276 | } |
| 277 | jetStreamApiRequests = collectorapi.Chart{ |
| 278 | ID: "jetstream_api_requests", |
| 279 | Title: "JetStream API Requests", |
| 280 | Units: "requests/s", |
| 281 | Fam: "jstream api", |
| 282 | Ctx: "nats.jetstream_api_requests", |
| 283 | Priority: prioJetStreamApiRequests, |
| 284 | Dims: collectorapi.Dims{ |
| 285 | {ID: "jsz_api_total", Name: "requests", Algo: collectorapi.Incremental}, |
| 286 | }, |
| 287 | } |
| 288 | jetStreamApiErrors = collectorapi.Chart{ |
| 289 | ID: "jetstream_api_errors", |
| 290 | Title: "JetStream API Errors", |
| 291 | Units: "errors/s", |
| 292 | Fam: "jstream api", |
| 293 | Ctx: "nats.jetstream_api_errors", |
| 294 | Priority: prioJetStreamApiErrors, |
| 295 | Dims: collectorapi.Dims{ |
| 296 | {ID: "jsz_api_errors", Name: "errors", Algo: collectorapi.Incremental}, |
| 297 | }, |
| 298 | } |
| 299 | jetStreamApiInflightRequests = collectorapi.Chart{ |
| 300 | ID: "jetstream_api_inflight", |
| 301 | Title: "JetStream API Inflight", |
| 302 | Units: "requests", |
| 303 | Fam: "jstream api", |
| 304 | Ctx: "nats.jetstream_api_inflight", |
| 305 | Priority: prioJetStreamApiInflight, |
| 306 | Dims: collectorapi.Dims{ |
| 307 | {ID: "jsz_api_inflight", Name: "inflight"}, |
| 308 | }, |
| 309 | } |
| 310 | jetStreamMemoryUsed = collectorapi.Chart{ |
| 311 | ID: "jetstream_memory_used", |
| 312 | Title: "JetStream Used Memory", |
| 313 | Units: "bytes", |
| 314 | Fam: "jstream rusage", |
| 315 | Ctx: "nats.jetstream_memory_used", |
| 316 | Priority: prioJetStreamMemoryUsed, |
| 317 | Type: collectorapi.Area, |
| 318 | Dims: collectorapi.Dims{ |
| 319 | {ID: "jsz_memory_used", Name: "used"}, |
| 320 | }, |
| 321 | } |
| 322 | jetStreamStorageUsed = collectorapi.Chart{ |
| 323 | ID: "jetstream_storage_used", |
| 324 | Title: "JetStream Used Storage", |
| 325 | Units: "bytes", |
| 326 | Fam: "jstream rusage", |
| 327 | Ctx: "nats.jetstream_storage_used", |
| 328 | Priority: prioJetStreamStorageUsed, |
| 329 | Dims: collectorapi.Dims{ |
| 330 | {ID: "jsz_store_used", Name: "used"}, |
| 331 | }, |
| 332 | } |
| 333 | ) |
| 334 | |
| 335 | var accountChartsTmpl = collectorapi.Charts{ |
| 336 | accountTrafficTmpl.Copy(), |
| 337 | accountMessagesTmpl.Copy(), |
| 338 | accountConnectionsCurrentTmpl.Copy(), |
| 339 | accountConnectionsRateTmpl.Copy(), |
| 340 | accountSubscriptionsTmpl.Copy(), |
| 341 | accountSlowConsumersTmpl.Copy(), |
| 342 | accountLeadNodesTmpl.Copy(), |
| 343 | } |
| 344 | |
| 345 | var ( |
| 346 | accountTrafficTmpl = collectorapi.Chart{ |
| 347 | ID: "account_%s_traffic", |
| 348 | Title: "Account Traffic", |
| 349 | Units: "bytes/s", |
| 350 | Fam: "acc traffic", |
| 351 | Ctx: "nats.account_traffic", |
| 352 | Priority: prioAccountTraffic, |
| 353 | Type: collectorapi.Area, |
| 354 | Dims: collectorapi.Dims{ |
| 355 | {ID: "accstatz_acc_%s_received_bytes", Name: "received", Algo: collectorapi.Incremental}, |
| 356 | {ID: "accstatz_acc_%s_sent_bytes", Name: "sent", Mul: -1, Algo: collectorapi.Incremental}, |
| 357 | }, |
| 358 | } |
| 359 | accountMessagesTmpl = collectorapi.Chart{ |
| 360 | ID: "account_%s_messages", |
| 361 | Title: "Account Messages", |
| 362 | Units: "messages/s", |
| 363 | Fam: "acc traffic", |
| 364 | Ctx: "nats.account_messages", |
| 365 | Priority: prioAccountMessages, |
| 366 | Type: collectorapi.Line, |
| 367 | Dims: collectorapi.Dims{ |
| 368 | {ID: "accstatz_acc_%s_received_msgs", Name: "received", Algo: collectorapi.Incremental}, |
| 369 | {ID: "accstatz_acc_%s_sent_msgs", Name: "sent", Mul: -1, Algo: collectorapi.Incremental}, |
| 370 | }, |
| 371 | } |
| 372 | accountConnectionsCurrentTmpl = collectorapi.Chart{ |
| 373 | ID: "account_%s_connections", |
| 374 | Title: "Account Active Connections", |
| 375 | Units: "connections", |
| 376 | Fam: "acc connections", |
| 377 | Ctx: "nats.account_connections", |
| 378 | Priority: prioAccountConnections, |
| 379 | Type: collectorapi.Line, |
| 380 | Dims: collectorapi.Dims{ |
| 381 | {ID: "accstatz_acc_%s_conns", Name: "active"}, |
| 382 | }, |
| 383 | } |
| 384 | accountConnectionsRateTmpl = collectorapi.Chart{ |
| 385 | ID: "account_%s_connections_rate", |
| 386 | Title: "Account Connections", |
| 387 | Units: "connections/s", |
| 388 | Fam: "acc connections", |
| 389 | Ctx: "nats.account_connections_rate", |
| 390 | Priority: prioAccountConnectionsRate, |
| 391 | Type: collectorapi.Line, |
| 392 | Dims: collectorapi.Dims{ |
| 393 | {ID: "accstatz_acc_%s_total_conns", Name: "connections", Algo: collectorapi.Incremental}, |
| 394 | }, |
| 395 | } |
| 396 | accountSubscriptionsTmpl = collectorapi.Chart{ |
| 397 | ID: "account_%s_subscriptions", |
| 398 | Title: "Account Active Subscriptions", |
| 399 | Units: "subscriptions", |
| 400 | Fam: "acc subscriptions", |
| 401 | Ctx: "nats.account_subscriptions", |
| 402 | Priority: prioAccountSubscriptions, |
| 403 | Type: collectorapi.Line, |
| 404 | Dims: collectorapi.Dims{ |
| 405 | {ID: "accstatz_acc_%s_num_subs", Name: "active"}, |
| 406 | }, |
| 407 | } |
| 408 | accountSlowConsumersTmpl = collectorapi.Chart{ |
| 409 | ID: "account_%s_slow_consumers", |
| 410 | Title: "Account Slow Consumers", |
| 411 | Units: "consumers/s", |
| 412 | Fam: "acc consumers", |
| 413 | Ctx: "nats.account_slow_consumers", |
| 414 | Priority: prioAccountSlowConsumers, |
| 415 | Type: collectorapi.Line, |
| 416 | Dims: collectorapi.Dims{ |
| 417 | {ID: "accstatz_acc_%s_slow_consumers", Name: "slow", Algo: collectorapi.Incremental}, |
| 418 | }, |
| 419 | } |
| 420 | accountLeadNodesTmpl = collectorapi.Chart{ |
| 421 | ID: "account_%s_leaf_nodes", |
| 422 | Title: "Account Leaf Nodes", |
| 423 | Units: "servers", |
| 424 | Fam: "acc leaf nodes", |
| 425 | Ctx: "nats.account_leaf_nodes", |
| 426 | Priority: prioAccountLeafNodes, |
| 427 | Type: collectorapi.Line, |
| 428 | Dims: collectorapi.Dims{ |
| 429 | {ID: "accstatz_acc_%s_leaf_nodes", Name: "leafnode"}, |
| 430 | }, |
| 431 | } |
| 432 | ) |
| 433 | |
| 434 | var routeChartsTmpl = collectorapi.Charts{ |
| 435 | routeTrafficTmpl.Copy(), |
| 436 | routeMessagesTmpl.Copy(), |
| 437 | routeSubscriptionsTmpl.Copy(), |
| 438 | } |
| 439 | |
| 440 | var ( |
| 441 | routeTrafficTmpl = collectorapi.Chart{ |
| 442 | ID: "route_%d_traffic", |
| 443 | Title: "Route Traffic", |
| 444 | Units: "bytes/s", |
| 445 | Fam: "route traffic", |
| 446 | Ctx: "nats.route_traffic", |
| 447 | Priority: prioRouteTraffic, |
| 448 | Type: collectorapi.Area, |
| 449 | Dims: collectorapi.Dims{ |
| 450 | {ID: "routez_route_id_%d_in_bytes", Name: "in", Algo: collectorapi.Incremental}, |
| 451 | {ID: "routez_route_id_%d_out_bytes", Name: "out", Mul: -1, Algo: collectorapi.Incremental}, |
| 452 | }, |
| 453 | } |
| 454 | routeMessagesTmpl = collectorapi.Chart{ |
| 455 | ID: "route_%d_messages", |
| 456 | Title: "Route Messages", |
| 457 | Units: "messages/s", |
| 458 | Fam: "route traffic", |
| 459 | Ctx: "nats.route_messages", |
| 460 | Priority: prioRouteMessages, |
| 461 | Type: collectorapi.Line, |
| 462 | Dims: collectorapi.Dims{ |
| 463 | {ID: "routez_route_id_%d_in_msgs", Name: "in", Algo: collectorapi.Incremental}, |
| 464 | {ID: "routez_route_id_%d_out_msgs", Name: "out", Mul: -1, Algo: collectorapi.Incremental}, |
| 465 | }, |
| 466 | } |
| 467 | routeSubscriptionsTmpl = collectorapi.Chart{ |
| 468 | ID: "route_%d_subscriptions", |
| 469 | Title: "Route Active Subscriptions", |
| 470 | Units: "subscriptions", |
| 471 | Fam: "route subscriptions", |
| 472 | Ctx: "nats.route_subscriptions", |
| 473 | Priority: prioRouteSubscriptions, |
| 474 | Type: collectorapi.Line, |
| 475 | Dims: collectorapi.Dims{ |
| 476 | {ID: "routez_route_id_%d_num_subs", Name: "active"}, |
| 477 | }, |
| 478 | } |
| 479 | ) |
| 480 | |
| 481 | var gatewayConnChartsTmpl = collectorapi.Charts{ |
| 482 | gatewayConnTrafficTmpl.Copy(), |
| 483 | gatewayConnMessagesTmpl.Copy(), |
| 484 | gatewayConnSubscriptionsTmpl.Copy(), |
| 485 | gatewayConnUptime.Copy(), |
| 486 | } |
| 487 | |
| 488 | var ( |
| 489 | gatewayConnTrafficTmpl = collectorapi.Chart{ |
| 490 | ID: "%s_gw_%s_cid_%d_traffic", |
| 491 | Title: "%s Gateway Traffic", |
| 492 | Units: "bytes/s", |
| 493 | Fam: "gw traffic", |
| 494 | Ctx: "nats.%s_gateway_conn_traffic", |
| 495 | Priority: prioGatewayConnTraffic, |
| 496 | Type: collectorapi.Area, |
| 497 | Dims: collectorapi.Dims{ |
| 498 | {ID: "gatewayz_%s_gw_%s_cid_%d_in_bytes", Name: "in", Algo: collectorapi.Incremental}, |
| 499 | {ID: "gatewayz_%s_gw_%s_cid_%d_out_bytes", Name: "out", Mul: -1, Algo: collectorapi.Incremental}, |
| 500 | }, |
| 501 | } |
| 502 | gatewayConnMessagesTmpl = collectorapi.Chart{ |
| 503 | ID: "%s_gw_%s_cid_%d_messages", |
| 504 | Title: "%s Gateway Messages", |
| 505 | Units: "messages/s", |
| 506 | Fam: "gw traffic", |
| 507 | Ctx: "nats.%s_gateway_conn_messages", |
| 508 | Priority: prioGatewayConnMessages, |
| 509 | Type: collectorapi.Line, |
| 510 | Dims: collectorapi.Dims{ |
| 511 | {ID: "gatewayz_%s_gw_%s_cid_%d_in_msgs", Name: "in", Algo: collectorapi.Incremental}, |
| 512 | {ID: "gatewayz_%s_gw_%s_cid_%d_out_msgs", Name: "out", Mul: -1, Algo: collectorapi.Incremental}, |
| 513 | }, |
| 514 | } |
| 515 | gatewayConnSubscriptionsTmpl = collectorapi.Chart{ |
| 516 | ID: "%s_gw_%s_cid_%d_subscriptions", |
| 517 | Title: "%s Gateway Active Subscriptions", |
| 518 | Units: "subscriptions", |
| 519 | Fam: "gw subscriptions", |
| 520 | Ctx: "nats.%s_gateway_conn_subscriptions", |
| 521 | Priority: prioGatewayConnSubscriptions, |
| 522 | Type: collectorapi.Line, |
| 523 | Dims: collectorapi.Dims{ |
| 524 | {ID: "gatewayz_%s_gw_%s_cid_%d_num_subs", Name: "active"}, |
| 525 | }, |
| 526 | } |
| 527 | gatewayConnUptime = collectorapi.Chart{ |
| 528 | ID: "%s_gw_%s_cid_%d_uptime", |
| 529 | Title: "%s Gateway Connection Uptime", |
| 530 | Units: "seconds", |
| 531 | Fam: "gw uptime", |
| 532 | Ctx: "nats.%s_gateway_conn_uptime", |
| 533 | Priority: prioGatewayConnUptime, |
| 534 | Dims: collectorapi.Dims{ |
| 535 | {ID: "gatewayz_%s_gw_%s_cid_%d_uptime", Name: "uptime"}, |
| 536 | }, |
| 537 | } |
| 538 | ) |
| 539 | |
| 540 | var leafConnChartsTmpl = collectorapi.Charts{ |
| 541 | leafConnTrafficTmpl.Copy(), |
| 542 | leafConnMessagesTmpl.Copy(), |
| 543 | leafConnSubscriptionsTmpl.Copy(), |
| 544 | leafConnRTT.Copy(), |
| 545 | } |
| 546 | |
| 547 | var ( |
| 548 | leafConnTrafficTmpl = collectorapi.Chart{ |
| 549 | ID: "leaf_node_conn_%s_%s_%s_%d_traffic", |
| 550 | Title: "Leaf Node Connection Traffic", |
| 551 | Units: "bytes/s", |
| 552 | Fam: "leaf traffic", |
| 553 | Ctx: "nats.leaf_node_conn_traffic", |
| 554 | Priority: prioLeafConnTraffic, |
| 555 | Type: collectorapi.Area, |
| 556 | Dims: collectorapi.Dims{ |
| 557 | {ID: "leafz_leaf_%s_%s_%s_%d_in_bytes", Name: "in", Algo: collectorapi.Incremental}, |
| 558 | {ID: "leafz_leaf_%s_%s_%s_%d_out_bytes", Name: "out", Mul: -1, Algo: collectorapi.Incremental}, |
| 559 | }, |
| 560 | } |
| 561 | leafConnMessagesTmpl = collectorapi.Chart{ |
| 562 | ID: "leaf_node_conn_%s_%s_%s_%d_messages", |
| 563 | Title: "Leaf Node Connection Messages", |
| 564 | Units: "messages/s", |
| 565 | Fam: "leaf traffic", |
| 566 | Ctx: "nats.leaf_node_conn_messages", |
| 567 | Priority: prioLeafConnMessages, |
| 568 | Type: collectorapi.Line, |
| 569 | Dims: collectorapi.Dims{ |
| 570 | {ID: "leafz_leaf_%s_%s_%s_%d_in_msgs", Name: "in", Algo: collectorapi.Incremental}, |
| 571 | {ID: "leafz_leaf_%s_%s_%s_%d_out_msgs", Name: "out", Mul: -1, Algo: collectorapi.Incremental}, |
| 572 | }, |
| 573 | } |
| 574 | leafConnSubscriptionsTmpl = collectorapi.Chart{ |
| 575 | ID: "leaf_node_conn_%s_%s_%s_%d_subscriptions", |
| 576 | Title: "Leaf Node Connection Active Subscriptions", |
| 577 | Units: "subscriptions", |
| 578 | Fam: "leaf subscriptions", |
| 579 | Ctx: "nats.leaf_node_conn_subscriptions", |
| 580 | Priority: prioLeafConnSubscriptions, |
| 581 | Type: collectorapi.Line, |
| 582 | Dims: collectorapi.Dims{ |
| 583 | {ID: "leafz_leaf_%s_%s_%s_%d_num_subs", Name: "active"}, |
| 584 | }, |
| 585 | } |
| 586 | leafConnRTT = collectorapi.Chart{ |
| 587 | ID: "leaf_node_conn_%s_%s_%s_%d_rtt", |
| 588 | Title: "Leaf Node Connection RTT", |
| 589 | Units: "microseconds", |
| 590 | Fam: "leaf rtt", |
| 591 | Ctx: "nats.leaf_node_conn_rtt", |
| 592 | Priority: prioLeafRTT, |
| 593 | Dims: collectorapi.Dims{ |
| 594 | {ID: "leafz_leaf_%s_%s_%s_%d_rtt", Name: "rtt"}, |
| 595 | }, |
| 596 | } |
| 597 | ) |
| 598 | |
| 599 | func (c *Collector) updateCharts() { |
| 600 | c.onceAddSrvCharts.Do(c.addServerCharts) |
| 601 | |
| 602 | maps.DeleteFunc(c.cache.accounts, func(_ string, acc *accCacheEntry) bool { |
| 603 | if !acc.updated { |
| 604 | c.removeAccountCharts(acc) |
| 605 | return true |
| 606 | } |
| 607 | if !acc.hasCharts { |
| 608 | acc.hasCharts = true |
| 609 | c.addAccountCharts(acc) |
| 610 | } |
| 611 | return false |
| 612 | }) |
| 613 | maps.DeleteFunc(c.cache.routes, func(_ uint64, route *routeCacheEntry) bool { |
| 614 | if !route.updated { |
| 615 | c.removeRouteCharts(route) |
| 616 | return true |
| 617 | } |
| 618 | if !route.hasCharts { |
| 619 | route.hasCharts = true |
| 620 | c.addRouteCharts(route) |
| 621 | } |
| 622 | return false |
| 623 | }) |
| 624 | maps.DeleteFunc(c.cache.inGateways, func(_ string, igw *gwCacheEntry) bool { |
| 625 | maps.DeleteFunc(igw.conns, func(_ uint64, inConn *gwConnCacheEntry) bool { |
| 626 | if !inConn.updated { |
| 627 | c.removeGatewayConnCharts(inConn, true) |
| 628 | return true |
| 629 | } |
| 630 | if !inConn.hasCharts { |
| 631 | inConn.hasCharts = true |
| 632 | c.addGatewayConnCharts(inConn, true) |
| 633 | } |
| 634 | return false |
| 635 | }) |
| 636 | return false |
| 637 | }) |
| 638 | maps.DeleteFunc(c.cache.outGateways, func(_ string, ogw *gwCacheEntry) bool { |
| 639 | maps.DeleteFunc(ogw.conns, func(_ uint64, outConn *gwConnCacheEntry) bool { |
| 640 | if !outConn.updated { |
| 641 | c.removeGatewayConnCharts(outConn, false) |
| 642 | return true |
| 643 | } |
| 644 | if !outConn.hasCharts { |
| 645 | outConn.hasCharts = true |
| 646 | c.addGatewayConnCharts(outConn, false) |
| 647 | } |
| 648 | return false |
| 649 | }) |
| 650 | return false |
| 651 | }) |
| 652 | maps.DeleteFunc(c.cache.leafs, func(_ string, leaf *leafCacheEntry) bool { |
| 653 | if !leaf.updated { |
| 654 | c.removeLeafCharts(leaf) |
| 655 | return true |
| 656 | } |
| 657 | if !leaf.hasCharts { |
| 658 | leaf.hasCharts = true |
| 659 | c.addLeafCharts(leaf) |
| 660 | } |
| 661 | return false |
| 662 | }) |
| 663 | } |
| 664 | |
| 665 | func (c *Collector) addServerCharts() { |
| 666 | charts := serverCharts() |
| 667 | |
| 668 | for _, chart := range *charts { |
| 669 | chart.Labels = []collectorapi.Label{ |
| 670 | {Key: "cluster_name", Value: c.srvMeta.clusterName}, |
| 671 | {Key: "server_id", Value: c.srvMeta.id}, |
| 672 | {Key: "server_name", Value: c.srvMeta.name}, |
| 673 | } |
| 674 | } |
| 675 | |
| 676 | if err := c.Charts().Add(*charts...); err != nil { |
| 677 | c.Warningf("failed to add server charts: %v", err) |
| 678 | } |
| 679 | } |
| 680 | |
| 681 | func (c *Collector) addAccountCharts(acc *accCacheEntry) { |
| 682 | charts := accountChartsTmpl.Copy() |
| 683 | |
| 684 | for _, chart := range *charts { |
| 685 | chart.ID = fmt.Sprintf(chart.ID, acc.accName) |
| 686 | chart.Labels = []collectorapi.Label{ |
| 687 | {Key: "cluster_name", Value: c.srvMeta.clusterName}, |
| 688 | {Key: "server_id", Value: c.srvMeta.id}, |
| 689 | {Key: "server_name", Value: c.srvMeta.name}, |
| 690 | {Key: "account", Value: acc.accName}, |
| 691 | } |
| 692 | for _, dim := range chart.Dims { |
| 693 | dim.ID = fmt.Sprintf(dim.ID, acc.accName) |
| 694 | } |
| 695 | } |
| 696 | |
| 697 | if err := c.Charts().Add(*charts...); err != nil { |
| 698 | c.Warningf("failed to add charts for account %s: %s", acc.accName, err) |
| 699 | } |
| 700 | } |
| 701 | |
| 702 | func (c *Collector) removeAccountCharts(acc *accCacheEntry) { |
| 703 | px := fmt.Sprintf("account_%s_", acc.accName) |
| 704 | c.removeCharts(px) |
| 705 | } |
| 706 | |
| 707 | func (c *Collector) addRouteCharts(route *routeCacheEntry) { |
| 708 | charts := routeChartsTmpl.Copy() |
| 709 | |
| 710 | for _, chart := range *charts { |
| 711 | chart.ID = fmt.Sprintf(chart.ID, route.rid) |
| 712 | chart.Labels = []collectorapi.Label{ |
| 713 | {Key: "cluster_name", Value: c.srvMeta.clusterName}, |
| 714 | {Key: "server_id", Value: c.srvMeta.id}, |
| 715 | {Key: "server_name", Value: c.srvMeta.name}, |
| 716 | {Key: "route_id", Value: strconv.FormatUint(route.rid, 10)}, |
| 717 | {Key: "remote_id", Value: route.remoteId}, |
| 718 | } |
| 719 | for _, dim := range chart.Dims { |
| 720 | dim.ID = fmt.Sprintf(dim.ID, route.rid) |
| 721 | } |
| 722 | } |
| 723 | |
| 724 | if err := c.Charts().Add(*charts...); err != nil { |
| 725 | c.Warningf("failed to add charts for route id %d: %s", route.rid, err) |
| 726 | } |
| 727 | } |
| 728 | |
| 729 | func (c *Collector) removeRouteCharts(route *routeCacheEntry) { |
| 730 | px := fmt.Sprintf("route_%d_", route.rid) |
| 731 | c.removeCharts(px) |
| 732 | } |
| 733 | |
| 734 | func (c *Collector) addGatewayConnCharts(gwConn *gwConnCacheEntry, isInbound bool) { |
| 735 | direction := "outbound" |
| 736 | if isInbound { |
| 737 | direction = "inbound" |
| 738 | } |
| 739 | |
| 740 | charts := gatewayConnChartsTmpl.Copy() |
| 741 | |
| 742 | for _, chart := range *charts { |
| 743 | chart.ID = fmt.Sprintf(chart.ID, direction, gwConn.rgwName, gwConn.cid) |
| 744 | chart.Title = fmt.Sprintf(chart.Title, cases.Title(language.English, cases.Compact).String(direction)) |
| 745 | chart.Ctx = fmt.Sprintf(chart.Ctx, direction) |
| 746 | chart.Labels = []collectorapi.Label{ |
| 747 | {Key: "cluster_name", Value: c.srvMeta.clusterName}, |
| 748 | {Key: "server_id", Value: c.srvMeta.id}, |
| 749 | {Key: "server_name", Value: c.srvMeta.name}, |
| 750 | {Key: "gateway", Value: gwConn.gwName}, |
| 751 | {Key: "remote_gateway", Value: gwConn.rgwName}, |
| 752 | {Key: "cid", Value: strconv.FormatUint(gwConn.cid, 10)}, |
| 753 | } |
| 754 | for _, dim := range chart.Dims { |
| 755 | dim.ID = fmt.Sprintf(dim.ID, direction, gwConn.rgwName, gwConn.cid) |
| 756 | } |
| 757 | } |
| 758 | |
| 759 | if err := c.Charts().Add(*charts...); err != nil { |
| 760 | c.Warningf("failed to add charts for gateway %s %s %d: %s", direction, gwConn.rgwName, gwConn.cid, err) |
| 761 | } |
| 762 | } |
| 763 | |
| 764 | func (c *Collector) removeGatewayConnCharts(gwConn *gwConnCacheEntry, isInbound bool) { |
| 765 | direction := "outbound" |
| 766 | if isInbound { |
| 767 | direction = "inbound" |
| 768 | } |
| 769 | px := fmt.Sprintf("%s_gw_%s_cid_%d_", direction, gwConn.rgwName, gwConn.cid) |
| 770 | c.removeCharts(px) |
| 771 | } |
| 772 | |
| 773 | func (c *Collector) addLeafCharts(leaf *leafCacheEntry) { |
| 774 | charts := leafConnChartsTmpl.Copy() |
| 775 | |
| 776 | for _, chart := range *charts { |
| 777 | chart.ID = fmt.Sprintf(chart.ID, leaf.leafName, leaf.account, leaf.ip, leaf.port) |
| 778 | chart.ID = cleanChartID(chart.ID) |
| 779 | chart.Labels = []collectorapi.Label{ |
| 780 | {Key: "cluster_name", Value: c.srvMeta.clusterName}, |
| 781 | {Key: "server_id", Value: c.srvMeta.id}, |
| 782 | {Key: "server_name", Value: c.srvMeta.name}, |
| 783 | {Key: "remote_name", Value: leaf.leafName}, |
| 784 | {Key: "account", Value: leaf.account}, |
| 785 | {Key: "ip", Value: leaf.ip}, |
| 786 | {Key: "port", Value: strconv.Itoa(leaf.port)}, |
| 787 | } |
| 788 | for _, dim := range chart.Dims { |
| 789 | dim.ID = fmt.Sprintf(dim.ID, leaf.leafName, leaf.account, leaf.ip, leaf.port) |
| 790 | } |
| 791 | } |
| 792 | |
| 793 | if err := c.Charts().Add(*charts...); err != nil { |
| 794 | c.Warningf("failed to add charts for leaf %s: %s", leaf.leafName, err) |
| 795 | } |
| 796 | } |
| 797 | |
| 798 | func (c *Collector) removeLeafCharts(leaf *leafCacheEntry) { |
| 799 | px := fmt.Sprintf("leaf_node_conn_%s_%s_%s_%d_", leaf.leafName, leaf.account, leaf.ip, leaf.port) |
| 800 | cleanChartID(px) |
| 801 | c.removeCharts(px) |
| 802 | } |
| 803 | |
| 804 | func (c *Collector) removeCharts(prefix string) { |
| 805 | for _, chart := range *c.Charts() { |
| 806 | if strings.HasPrefix(chart.ID, prefix) { |
| 807 | chart.MarkRemove() |
| 808 | chart.MarkNotCreated() |
| 809 | } |
| 810 | } |
| 811 | } |
| 812 | |
| 813 | func cleanChartID(id string) string { |
| 814 | r := strings.NewReplacer(".", "_", " ", "_") |
| 815 | return strings.ToLower(r.Replace(id)) |
| 816 | } |