| 1 | // SPDX-License-Identifier: GPL-3.0-or-later |
| 2 | |
| 3 | package mssql |
| 4 | |
| 5 | import ( |
| 6 | "fmt" |
| 7 | "strings" |
| 8 | |
| 9 | "github.com/netdata/netdata/go/plugins/plugin/framework/collectorapi" |
| 10 | ) |
| 11 | |
| 12 | const ( |
| 13 | prioUserConnections = collectorapi.Priority + iota |
| 14 | prioSessionConnections |
| 15 | prioBlockedProcesses |
| 16 | |
| 17 | prioBatchRequests |
| 18 | prioCompilations |
| 19 | prioRecompilations |
| 20 | prioAutoParamAttempts |
| 21 | prioSQLErrors |
| 22 | |
| 23 | prioBufferCacheHitRatio |
| 24 | prioBufferPageIOPS |
| 25 | prioBufferCheckpointPages |
| 26 | prioBufferPageLifeExpectancy |
| 27 | prioBufferLazyWrites |
| 28 | prioBufferPageLookups |
| 29 | |
| 30 | prioPageSplits |
| 31 | |
| 32 | prioMemoryTotal |
| 33 | prioMemoryConnection |
| 34 | prioMemoryPendingGrants |
| 35 | prioMemoryExternalBenefit |
| 36 | |
| 37 | prioProcessMemoryResident |
| 38 | prioProcessMemoryVirtual |
| 39 | prioProcessMemoryUtilization |
| 40 | prioProcessPageFaults |
| 41 | |
| 42 | prioOSMemory |
| 43 | prioOSPageFile |
| 44 | |
| 45 | prioDatabaseActiveTransactions |
| 46 | prioDatabaseTransactions |
| 47 | prioDatabaseWriteTransactions |
| 48 | prioDatabaseLogFlushes |
| 49 | prioDatabaseLogFlushed |
| 50 | prioDatabaseLogGrowths |
| 51 | prioDatabaseLogFileSize |
| 52 | prioDatabaseLogPercentUsed |
| 53 | prioDatabaseLogTruncationsShrinks |
| 54 | prioDatabaseIOStall |
| 55 | prioDatabaseDeadlocks |
| 56 | prioDatabaseLockWaits |
| 57 | prioDatabaseLockTimeouts |
| 58 | prioDatabaseLockRequests |
| 59 | prioDatabaseDataFileSize |
| 60 | prioDatabaseState |
| 61 | prioDatabaseReadOnly |
| 62 | |
| 63 | prioLocksByResource |
| 64 | |
| 65 | prioWaitTotalTime |
| 66 | prioWaitResourceTime |
| 67 | prioWaitSignalTime |
| 68 | prioWaitMaxTime |
| 69 | prioWaitCount |
| 70 | |
| 71 | prioJobStatus |
| 72 | |
| 73 | prioReplicationStatus |
| 74 | prioReplicationLatency |
| 75 | |
| 76 | prioAGSyncHealth |
| 77 | prioAGRecoveryHealth |
| 78 | |
| 79 | prioAGReplicaRole |
| 80 | prioAGReplicaConnectedState |
| 81 | prioAGReplicaSyncHealth |
| 82 | |
| 83 | prioAGDBSyncState |
| 84 | prioAGDBLogSendQueue |
| 85 | prioAGDBLogSendRate |
| 86 | prioAGDBRedoQueue |
| 87 | prioAGDBRedoRate |
| 88 | prioAGDBFilestreamSendRate |
| 89 | prioAGDBSecondaryLag |
| 90 | prioAGDBSuspended |
| 91 | prioAGDBFailoverReadiness |
| 92 | prioAGDBJoined |
| 93 | |
| 94 | prioAGClusterQuorumState |
| 95 | prioAGClusterMemberState |
| 96 | prioAGClusterMemberQuorumVotes |
| 97 | |
| 98 | prioAGThreads |
| 99 | |
| 100 | prioAGPageRepair |
| 101 | ) |
| 102 | |
| 103 | // instanceCharts are charts for the SQL Server instance level metrics |
| 104 | var instanceCharts = collectorapi.Charts{ |
| 105 | userConnectionsChart.Copy(), |
| 106 | sessionConnectionsChart.Copy(), |
| 107 | blockedProcessesChart.Copy(), |
| 108 | |
| 109 | batchRequestsChart.Copy(), |
| 110 | compilationsChart.Copy(), |
| 111 | recompilationsChart.Copy(), |
| 112 | autoParamAttemptsChart.Copy(), |
| 113 | sqlErrorsChart.Copy(), |
| 114 | |
| 115 | bufferCacheHitRatioChart.Copy(), |
| 116 | bufferPageIOPSChart.Copy(), |
| 117 | bufferCheckpointPagesChart.Copy(), |
| 118 | bufferPageLifeExpectancyChart.Copy(), |
| 119 | bufferLazyWritesChart.Copy(), |
| 120 | bufferPageLookupsChart.Copy(), |
| 121 | |
| 122 | pageSplitsChart.Copy(), |
| 123 | |
| 124 | memoryTotalChart.Copy(), |
| 125 | memoryConnectionChart.Copy(), |
| 126 | memoryPendingGrantsChart.Copy(), |
| 127 | memoryExternalBenefitChart.Copy(), |
| 128 | |
| 129 | processMemoryResidentChart.Copy(), |
| 130 | processMemoryVirtualChart.Copy(), |
| 131 | processMemoryUtilizationChart.Copy(), |
| 132 | processPageFaultsChart.Copy(), |
| 133 | |
| 134 | osMemoryChart.Copy(), |
| 135 | osPageFileChart.Copy(), |
| 136 | } |
| 137 | |
| 138 | var ( |
| 139 | userConnectionsChart = collectorapi.Chart{ |
| 140 | ID: "user_connections", |
| 141 | Title: "User connections", |
| 142 | Units: "connections", |
| 143 | Fam: "connections", |
| 144 | Ctx: "mssql.user_connections", |
| 145 | Priority: prioUserConnections, |
| 146 | Dims: collectorapi.Dims{ |
| 147 | {ID: "user_connections", Name: "user"}, |
| 148 | }, |
| 149 | } |
| 150 | blockedProcessesChart = collectorapi.Chart{ |
| 151 | ID: "blocked_processes", |
| 152 | Title: "Blocked processes", |
| 153 | Units: "processes", |
| 154 | Fam: "processes", |
| 155 | Ctx: "mssql.blocked_processes", |
| 156 | Priority: prioBlockedProcesses, |
| 157 | Dims: collectorapi.Dims{ |
| 158 | {ID: "blocked_processes", Name: "blocked"}, |
| 159 | }, |
| 160 | } |
| 161 | sessionConnectionsChart = collectorapi.Chart{ |
| 162 | ID: "session_connections", |
| 163 | Title: "Session connections", |
| 164 | Units: "connections", |
| 165 | Fam: "connections", |
| 166 | Ctx: "mssql.session_connections", |
| 167 | Priority: prioSessionConnections, |
| 168 | Dims: collectorapi.Dims{ |
| 169 | {ID: "session_connections_user", Name: "user"}, |
| 170 | {ID: "session_connections_internal", Name: "internal"}, |
| 171 | }, |
| 172 | } |
| 173 | |
| 174 | batchRequestsChart = collectorapi.Chart{ |
| 175 | ID: "batch_requests", |
| 176 | Title: "Batch requests", |
| 177 | Units: "requests/s", |
| 178 | Fam: "queries", |
| 179 | Ctx: "mssql.batch_requests", |
| 180 | Priority: prioBatchRequests, |
| 181 | Dims: collectorapi.Dims{ |
| 182 | {ID: "batch_requests", Name: "batch", Algo: collectorapi.Incremental}, |
| 183 | }, |
| 184 | } |
| 185 | compilationsChart = collectorapi.Chart{ |
| 186 | ID: "compilations", |
| 187 | Title: "SQL compilations", |
| 188 | Units: "compilations/s", |
| 189 | Fam: "queries", |
| 190 | Ctx: "mssql.compilations", |
| 191 | Priority: prioCompilations, |
| 192 | Dims: collectorapi.Dims{ |
| 193 | {ID: "sql_compilations", Name: "compilations", Algo: collectorapi.Incremental}, |
| 194 | }, |
| 195 | } |
| 196 | recompilationsChart = collectorapi.Chart{ |
| 197 | ID: "recompilations", |
| 198 | Title: "SQL re-compilations", |
| 199 | Units: "recompilations/s", |
| 200 | Fam: "queries", |
| 201 | Ctx: "mssql.recompilations", |
| 202 | Priority: prioRecompilations, |
| 203 | Dims: collectorapi.Dims{ |
| 204 | {ID: "sql_recompilations", Name: "recompilations", Algo: collectorapi.Incremental}, |
| 205 | }, |
| 206 | } |
| 207 | autoParamAttemptsChart = collectorapi.Chart{ |
| 208 | ID: "auto_param_attempts", |
| 209 | Title: "Auto-parameterization attempts", |
| 210 | Units: "attempts/s", |
| 211 | Fam: "queries", |
| 212 | Ctx: "mssql.auto_param_attempts", |
| 213 | Priority: prioAutoParamAttempts, |
| 214 | Dims: collectorapi.Dims{ |
| 215 | {ID: "auto_param_attempts", Name: "total", Algo: collectorapi.Incremental}, |
| 216 | {ID: "auto_param_safe", Name: "safe", Algo: collectorapi.Incremental}, |
| 217 | {ID: "auto_param_failed", Name: "failed", Algo: collectorapi.Incremental, Mul: -1}, |
| 218 | }, |
| 219 | } |
| 220 | sqlErrorsChart = collectorapi.Chart{ |
| 221 | ID: "sql_errors", |
| 222 | Title: "SQL errors", |
| 223 | Units: "errors/s", |
| 224 | Fam: "errors", |
| 225 | Ctx: "mssql.sql_errors", |
| 226 | Priority: prioSQLErrors, |
| 227 | Dims: collectorapi.Dims{ |
| 228 | {ID: "sql_errors_total", Name: "errors", Algo: collectorapi.Incremental}, |
| 229 | }, |
| 230 | } |
| 231 | |
| 232 | bufferCacheHitRatioChart = collectorapi.Chart{ |
| 233 | ID: "buffer_cache_hit_ratio", |
| 234 | Title: "Buffer cache hit ratio", |
| 235 | Units: "percentage", |
| 236 | Fam: "buffer", |
| 237 | Ctx: "mssql.buffer_cache_hit_ratio", |
| 238 | Priority: prioBufferCacheHitRatio, |
| 239 | Dims: collectorapi.Dims{ |
| 240 | {ID: "buffer_cache_hit_ratio", Name: "hit_ratio"}, |
| 241 | }, |
| 242 | } |
| 243 | bufferPageIOPSChart = collectorapi.Chart{ |
| 244 | ID: "buffer_page_iops", |
| 245 | Title: "Buffer page I/O", |
| 246 | Units: "pages/s", |
| 247 | Fam: "buffer", |
| 248 | Ctx: "mssql.buffer_page_iops", |
| 249 | Priority: prioBufferPageIOPS, |
| 250 | Dims: collectorapi.Dims{ |
| 251 | {ID: "buffer_page_reads", Name: "read", Algo: collectorapi.Incremental}, |
| 252 | {ID: "buffer_page_writes", Name: "written", Algo: collectorapi.Incremental, Mul: -1}, |
| 253 | }, |
| 254 | } |
| 255 | bufferCheckpointPagesChart = collectorapi.Chart{ |
| 256 | ID: "buffer_checkpoint_pages", |
| 257 | Title: "Buffer checkpoint pages flushed", |
| 258 | Units: "pages/s", |
| 259 | Fam: "buffer", |
| 260 | Ctx: "mssql.buffer_checkpoint_pages", |
| 261 | Priority: prioBufferCheckpointPages, |
| 262 | Dims: collectorapi.Dims{ |
| 263 | {ID: "buffer_checkpoint_pages", Name: "flushed", Algo: collectorapi.Incremental}, |
| 264 | }, |
| 265 | } |
| 266 | bufferPageLifeExpectancyChart = collectorapi.Chart{ |
| 267 | ID: "buffer_page_life_expectancy", |
| 268 | Title: "Buffer page life expectancy", |
| 269 | Units: "seconds", |
| 270 | Fam: "buffer", |
| 271 | Ctx: "mssql.buffer_page_life_expectancy", |
| 272 | Priority: prioBufferPageLifeExpectancy, |
| 273 | Dims: collectorapi.Dims{ |
| 274 | {ID: "buffer_page_life_expectancy", Name: "life_expectancy"}, |
| 275 | }, |
| 276 | } |
| 277 | bufferLazyWritesChart = collectorapi.Chart{ |
| 278 | ID: "buffer_lazy_writes", |
| 279 | Title: "Buffer lazy writes", |
| 280 | Units: "writes/s", |
| 281 | Fam: "buffer", |
| 282 | Ctx: "mssql.buffer_lazy_writes", |
| 283 | Priority: prioBufferLazyWrites, |
| 284 | Dims: collectorapi.Dims{ |
| 285 | {ID: "buffer_lazy_writes", Name: "lazy_writes", Algo: collectorapi.Incremental}, |
| 286 | }, |
| 287 | } |
| 288 | bufferPageLookupsChart = collectorapi.Chart{ |
| 289 | ID: "buffer_page_lookups", |
| 290 | Title: "Buffer page lookups", |
| 291 | Units: "lookups/s", |
| 292 | Fam: "buffer", |
| 293 | Ctx: "mssql.buffer_page_lookups", |
| 294 | Priority: prioBufferPageLookups, |
| 295 | Dims: collectorapi.Dims{ |
| 296 | {ID: "buffer_page_lookups", Name: "lookups", Algo: collectorapi.Incremental}, |
| 297 | }, |
| 298 | } |
| 299 | |
| 300 | pageSplitsChart = collectorapi.Chart{ |
| 301 | ID: "page_splits", |
| 302 | Title: "Page splits", |
| 303 | Units: "splits/s", |
| 304 | Fam: "access", |
| 305 | Ctx: "mssql.page_splits", |
| 306 | Priority: prioPageSplits, |
| 307 | Dims: collectorapi.Dims{ |
| 308 | {ID: "page_splits", Name: "page", Algo: collectorapi.Incremental}, |
| 309 | }, |
| 310 | } |
| 311 | |
| 312 | memoryTotalChart = collectorapi.Chart{ |
| 313 | ID: "memory_total", |
| 314 | Title: "Total server memory", |
| 315 | Units: "bytes", |
| 316 | Fam: "memory", |
| 317 | Ctx: "mssql.memory_total", |
| 318 | Priority: prioMemoryTotal, |
| 319 | Dims: collectorapi.Dims{ |
| 320 | {ID: "memory_total", Name: "memory"}, |
| 321 | }, |
| 322 | } |
| 323 | memoryConnectionChart = collectorapi.Chart{ |
| 324 | ID: "memory_connection", |
| 325 | Title: "Connection memory", |
| 326 | Units: "bytes", |
| 327 | Fam: "memory", |
| 328 | Ctx: "mssql.memory_connection", |
| 329 | Priority: prioMemoryConnection, |
| 330 | Dims: collectorapi.Dims{ |
| 331 | {ID: "memory_connection", Name: "memory"}, |
| 332 | }, |
| 333 | } |
| 334 | memoryPendingGrantsChart = collectorapi.Chart{ |
| 335 | ID: "memory_pending_grants", |
| 336 | Title: "Pending memory grants", |
| 337 | Units: "processes", |
| 338 | Fam: "memory", |
| 339 | Ctx: "mssql.memory_pending_grants", |
| 340 | Priority: prioMemoryPendingGrants, |
| 341 | Dims: collectorapi.Dims{ |
| 342 | {ID: "memory_pending_grants", Name: "pending"}, |
| 343 | }, |
| 344 | } |
| 345 | memoryExternalBenefitChart = collectorapi.Chart{ |
| 346 | ID: "memory_external_benefit", |
| 347 | Title: "External benefit of memory", |
| 348 | Units: "benefit", |
| 349 | Fam: "memory", |
| 350 | Ctx: "mssql.memory_external_benefit", |
| 351 | Priority: prioMemoryExternalBenefit, |
| 352 | Dims: collectorapi.Dims{ |
| 353 | {ID: "memory_external_benefit", Name: "benefit"}, |
| 354 | }, |
| 355 | } |
| 356 | |
| 357 | processMemoryResidentChart = collectorapi.Chart{ |
| 358 | ID: "process_memory_resident", |
| 359 | Title: "Process resident memory (working set)", |
| 360 | Units: "bytes", |
| 361 | Fam: "process memory", |
| 362 | Ctx: "mssql.process_memory_resident", |
| 363 | Priority: prioProcessMemoryResident, |
| 364 | Dims: collectorapi.Dims{ |
| 365 | {ID: "process_memory_resident", Name: "resident"}, |
| 366 | }, |
| 367 | } |
| 368 | processMemoryVirtualChart = collectorapi.Chart{ |
| 369 | ID: "process_memory_virtual", |
| 370 | Title: "Process virtual memory committed", |
| 371 | Units: "bytes", |
| 372 | Fam: "process memory", |
| 373 | Ctx: "mssql.process_memory_virtual", |
| 374 | Priority: prioProcessMemoryVirtual, |
| 375 | Dims: collectorapi.Dims{ |
| 376 | {ID: "process_memory_virtual", Name: "virtual"}, |
| 377 | }, |
| 378 | } |
| 379 | processMemoryUtilizationChart = collectorapi.Chart{ |
| 380 | ID: "process_memory_utilization", |
| 381 | Title: "Process memory utilization", |
| 382 | Units: "percentage", |
| 383 | Fam: "process memory", |
| 384 | Ctx: "mssql.process_memory_utilization", |
| 385 | Priority: prioProcessMemoryUtilization, |
| 386 | Dims: collectorapi.Dims{ |
| 387 | {ID: "process_memory_utilization", Name: "utilization"}, |
| 388 | }, |
| 389 | } |
| 390 | processPageFaultsChart = collectorapi.Chart{ |
| 391 | ID: "process_page_faults", |
| 392 | Title: "Process page faults", |
| 393 | Units: "faults", |
| 394 | Fam: "process memory", |
| 395 | Ctx: "mssql.process_page_faults", |
| 396 | Priority: prioProcessPageFaults, |
| 397 | Dims: collectorapi.Dims{ |
| 398 | {ID: "process_page_faults", Name: "page_faults", Algo: collectorapi.Incremental}, |
| 399 | }, |
| 400 | } |
| 401 | |
| 402 | osMemoryChart = collectorapi.Chart{ |
| 403 | ID: "os_memory", |
| 404 | Title: "OS physical memory", |
| 405 | Units: "bytes", |
| 406 | Fam: "os memory", |
| 407 | Ctx: "mssql.os_memory", |
| 408 | Priority: prioOSMemory, |
| 409 | Dims: collectorapi.Dims{ |
| 410 | {ID: "os_memory_used", Name: "used"}, |
| 411 | {ID: "os_memory_available", Name: "available"}, |
| 412 | }, |
| 413 | } |
| 414 | osPageFileChart = collectorapi.Chart{ |
| 415 | ID: "os_pagefile", |
| 416 | Title: "OS page file", |
| 417 | Units: "bytes", |
| 418 | Fam: "os memory", |
| 419 | Ctx: "mssql.os_pagefile", |
| 420 | Priority: prioOSPageFile, |
| 421 | Dims: collectorapi.Dims{ |
| 422 | {ID: "os_pagefile_used", Name: "used"}, |
| 423 | {ID: "os_pagefile_available", Name: "available"}, |
| 424 | }, |
| 425 | } |
| 426 | ) |
| 427 | |
| 428 | // Database chart templates |
| 429 | var ( |
| 430 | databaseActiveTransactionsChartTmpl = collectorapi.Chart{ |
| 431 | ID: "database_%s_active_transactions", |
| 432 | Title: "Active transactions", |
| 433 | Units: "transactions", |
| 434 | Fam: "db transactions", |
| 435 | Ctx: "mssql.database_active_transactions", |
| 436 | Priority: prioDatabaseActiveTransactions, |
| 437 | Dims: collectorapi.Dims{ |
| 438 | {ID: "database_%s_active_transactions", Name: "active"}, |
| 439 | }, |
| 440 | } |
| 441 | databaseTransactionsChartTmpl = collectorapi.Chart{ |
| 442 | ID: "database_%s_transactions", |
| 443 | Title: "Transactions", |
| 444 | Units: "transactions/s", |
| 445 | Fam: "db transactions", |
| 446 | Ctx: "mssql.database_transactions", |
| 447 | Priority: prioDatabaseTransactions, |
| 448 | Dims: collectorapi.Dims{ |
| 449 | {ID: "database_%s_transactions", Name: "transactions", Algo: collectorapi.Incremental}, |
| 450 | }, |
| 451 | } |
| 452 | databaseWriteTransactionsChartTmpl = collectorapi.Chart{ |
| 453 | ID: "database_%s_write_transactions", |
| 454 | Title: "Write transactions", |
| 455 | Units: "transactions/s", |
| 456 | Fam: "db transactions", |
| 457 | Ctx: "mssql.database_write_transactions", |
| 458 | Priority: prioDatabaseWriteTransactions, |
| 459 | Dims: collectorapi.Dims{ |
| 460 | {ID: "database_%s_write_transactions", Name: "write", Algo: collectorapi.Incremental}, |
| 461 | }, |
| 462 | } |
| 463 | databaseLogFlushesChartTmpl = collectorapi.Chart{ |
| 464 | ID: "database_%s_log_flushes", |
| 465 | Title: "Log flushes", |
| 466 | Units: "flushes/s", |
| 467 | Fam: "db log", |
| 468 | Ctx: "mssql.database_log_flushes", |
| 469 | Priority: prioDatabaseLogFlushes, |
| 470 | Dims: collectorapi.Dims{ |
| 471 | {ID: "database_%s_log_flushes", Name: "flushes", Algo: collectorapi.Incremental}, |
| 472 | }, |
| 473 | } |
| 474 | databaseLogFlushedChartTmpl = collectorapi.Chart{ |
| 475 | ID: "database_%s_log_flushed", |
| 476 | Title: "Log bytes flushed", |
| 477 | Units: "bytes/s", |
| 478 | Fam: "db log", |
| 479 | Ctx: "mssql.database_log_flushed", |
| 480 | Priority: prioDatabaseLogFlushed, |
| 481 | Dims: collectorapi.Dims{ |
| 482 | {ID: "database_%s_log_flushed", Name: "flushed", Algo: collectorapi.Incremental}, |
| 483 | }, |
| 484 | } |
| 485 | databaseDataFileSizeChartTmpl = collectorapi.Chart{ |
| 486 | ID: "database_%s_data_file_size", |
| 487 | Title: "Data file size", |
| 488 | Units: "bytes", |
| 489 | Fam: "db size", |
| 490 | Ctx: "mssql.database_data_file_size", |
| 491 | Priority: prioDatabaseDataFileSize, |
| 492 | Dims: collectorapi.Dims{ |
| 493 | {ID: "database_%s_data_file_size", Name: "size"}, |
| 494 | }, |
| 495 | } |
| 496 | databaseBackupRestoreThroughputChartTmpl = collectorapi.Chart{ |
| 497 | ID: "database_%s_backup_restore_throughput", |
| 498 | Title: "Backup/Restore throughput", |
| 499 | Units: "bytes/s", |
| 500 | Fam: "db backup", |
| 501 | Ctx: "mssql.database_backup_restore_throughput", |
| 502 | Priority: prioDatabaseDataFileSize + 1, |
| 503 | Dims: collectorapi.Dims{ |
| 504 | {ID: "database_%s_backup_restore_throughput", Name: "throughput", Algo: collectorapi.Incremental}, |
| 505 | }, |
| 506 | } |
| 507 | databaseStateChartTmpl = collectorapi.Chart{ |
| 508 | ID: "database_%s_state", |
| 509 | Title: "Database state", |
| 510 | Units: "state", |
| 511 | Fam: "db status", |
| 512 | Ctx: "mssql.database_state", |
| 513 | Priority: prioDatabaseState, |
| 514 | Dims: collectorapi.Dims{ |
| 515 | {ID: "database_%s_state_online", Name: "online"}, |
| 516 | {ID: "database_%s_state_restoring", Name: "restoring"}, |
| 517 | {ID: "database_%s_state_recovering", Name: "recovering"}, |
| 518 | {ID: "database_%s_state_pending", Name: "pending"}, |
| 519 | {ID: "database_%s_state_suspect", Name: "suspect"}, |
| 520 | {ID: "database_%s_state_emergency", Name: "emergency"}, |
| 521 | {ID: "database_%s_state_offline", Name: "offline"}, |
| 522 | }, |
| 523 | } |
| 524 | databaseReadOnlyChartTmpl = collectorapi.Chart{ |
| 525 | ID: "database_%s_read_only", |
| 526 | Title: "Database read-only status", |
| 527 | Units: "status", |
| 528 | Fam: "db status", |
| 529 | Ctx: "mssql.database_read_only", |
| 530 | Priority: prioDatabaseReadOnly, |
| 531 | Dims: collectorapi.Dims{ |
| 532 | {ID: "database_%s_read_only", Name: "read_only"}, |
| 533 | {ID: "database_%s_read_write", Name: "read_write"}, |
| 534 | }, |
| 535 | } |
| 536 | databaseLogGrowthsChartTmpl = collectorapi.Chart{ |
| 537 | ID: "database_%s_log_growths", |
| 538 | Title: "Database log growths", |
| 539 | Units: "growths", |
| 540 | Fam: "db log", |
| 541 | Ctx: "mssql.database_log_growths", |
| 542 | Priority: prioDatabaseLogGrowths, |
| 543 | Dims: collectorapi.Dims{ |
| 544 | {ID: "database_%s_log_growths", Name: "growths", Algo: collectorapi.Incremental}, |
| 545 | }, |
| 546 | } |
| 547 | databaseLogFileSizeChartTmpl = collectorapi.Chart{ |
| 548 | ID: "database_%s_log_file_size", |
| 549 | Title: "Transaction log file size", |
| 550 | Units: "bytes", |
| 551 | Fam: "db log", |
| 552 | Ctx: "mssql.database_log_file_size", |
| 553 | Type: collectorapi.Stacked, |
| 554 | Priority: prioDatabaseLogFileSize, |
| 555 | Dims: collectorapi.Dims{ |
| 556 | {ID: "database_%s_log_size_used", Name: "used"}, |
| 557 | {ID: "database_%s_log_size_free", Name: "free"}, |
| 558 | }, |
| 559 | } |
| 560 | databaseLogPercentUsedChartTmpl = collectorapi.Chart{ |
| 561 | ID: "database_%s_log_percent_used", |
| 562 | Title: "Transaction log space utilization", |
| 563 | Units: "percentage", |
| 564 | Fam: "db log", |
| 565 | Ctx: "mssql.database_log_percent_used", |
| 566 | Type: collectorapi.Line, |
| 567 | Priority: prioDatabaseLogPercentUsed, |
| 568 | Dims: collectorapi.Dims{ |
| 569 | {ID: "database_%s_log_percent_used", Name: "used", Div: 100}, |
| 570 | }, |
| 571 | } |
| 572 | databaseLogTruncationsShrinksChartTmpl = collectorapi.Chart{ |
| 573 | ID: "database_%s_log_truncations_shrinks", |
| 574 | Title: "Transaction log truncations and shrinks", |
| 575 | Units: "events/s", |
| 576 | Fam: "db log", |
| 577 | Ctx: "mssql.database_log_truncations_shrinks", |
| 578 | Type: collectorapi.Line, |
| 579 | Priority: prioDatabaseLogTruncationsShrinks, |
| 580 | Dims: collectorapi.Dims{ |
| 581 | {ID: "database_%s_log_truncations", Name: "truncations", Algo: collectorapi.Incremental}, |
| 582 | {ID: "database_%s_log_shrinks", Name: "shrinks", Algo: collectorapi.Incremental}, |
| 583 | }, |
| 584 | } |
| 585 | databaseIOStallChartTmpl = collectorapi.Chart{ |
| 586 | ID: "database_%s_io_stall", |
| 587 | Title: "Database I/O stall time", |
| 588 | Units: "ms", |
| 589 | Fam: "db io", |
| 590 | Ctx: "mssql.database_io_stall", |
| 591 | Priority: prioDatabaseIOStall, |
| 592 | Dims: collectorapi.Dims{ |
| 593 | {ID: "database_%s_io_stall_read", Name: "read", Algo: collectorapi.Incremental}, |
| 594 | {ID: "database_%s_io_stall_write", Name: "write", Algo: collectorapi.Incremental}, |
| 595 | }, |
| 596 | } |
| 597 | ) |
| 598 | |
| 599 | // Lock stats chart templates (per lock resource type, from performance counters) |
| 600 | var ( |
| 601 | lockStatsDeadlocksChartTmpl = collectorapi.Chart{ |
| 602 | ID: "lock_stats_%s_deadlocks", |
| 603 | Title: "Deadlocks by lock resource type", |
| 604 | Units: "deadlocks/s", |
| 605 | Fam: "lock stats", |
| 606 | Ctx: "mssql.lock_stats_deadlocks", |
| 607 | Priority: prioDatabaseDeadlocks, |
| 608 | Dims: collectorapi.Dims{ |
| 609 | {ID: "lock_stats_%s_deadlocks", Name: "deadlocks", Algo: collectorapi.Incremental}, |
| 610 | }, |
| 611 | } |
| 612 | lockStatsWaitsChartTmpl = collectorapi.Chart{ |
| 613 | ID: "lock_stats_%s_waits", |
| 614 | Title: "Lock waits by lock resource type", |
| 615 | Units: "waits/s", |
| 616 | Fam: "lock stats", |
| 617 | Ctx: "mssql.lock_stats_waits", |
| 618 | Priority: prioDatabaseLockWaits, |
| 619 | Dims: collectorapi.Dims{ |
| 620 | {ID: "lock_stats_%s_waits", Name: "waits", Algo: collectorapi.Incremental}, |
| 621 | }, |
| 622 | } |
| 623 | lockStatsTimeoutsChartTmpl = collectorapi.Chart{ |
| 624 | ID: "lock_stats_%s_timeouts", |
| 625 | Title: "Lock timeouts by lock resource type", |
| 626 | Units: "timeouts/s", |
| 627 | Fam: "lock stats", |
| 628 | Ctx: "mssql.lock_stats_timeouts", |
| 629 | Priority: prioDatabaseLockTimeouts, |
| 630 | Dims: collectorapi.Dims{ |
| 631 | {ID: "lock_stats_%s_timeouts", Name: "timeouts", Algo: collectorapi.Incremental}, |
| 632 | }, |
| 633 | } |
| 634 | lockStatsRequestsChartTmpl = collectorapi.Chart{ |
| 635 | ID: "lock_stats_%s_requests", |
| 636 | Title: "Lock requests by lock resource type", |
| 637 | Units: "requests/s", |
| 638 | Fam: "lock stats", |
| 639 | Ctx: "mssql.lock_stats_requests", |
| 640 | Priority: prioDatabaseLockRequests, |
| 641 | Dims: collectorapi.Dims{ |
| 642 | {ID: "lock_stats_%s_requests", Name: "requests", Algo: collectorapi.Incremental}, |
| 643 | }, |
| 644 | } |
| 645 | ) |
| 646 | |
| 647 | // Lock resource chart template |
| 648 | var locksByResourceChartTmpl = collectorapi.Chart{ |
| 649 | ID: "locks_by_resource_%s", |
| 650 | Title: "Locks by resource type", |
| 651 | Units: "locks", |
| 652 | Fam: "locks", |
| 653 | Ctx: "mssql.locks_by_resource", |
| 654 | Priority: prioLocksByResource, |
| 655 | Dims: collectorapi.Dims{ |
| 656 | {ID: "locks_%s_count", Name: "locks"}, |
| 657 | }, |
| 658 | } |
| 659 | |
| 660 | // Wait type chart templates |
| 661 | var ( |
| 662 | waitTotalTimeChartTmpl = collectorapi.Chart{ |
| 663 | ID: "wait_%s_total_time", |
| 664 | Title: "Total wait time", |
| 665 | Units: "ms", |
| 666 | Fam: "waits", |
| 667 | Ctx: "mssql.wait_total_time", |
| 668 | Priority: prioWaitTotalTime, |
| 669 | Dims: collectorapi.Dims{ |
| 670 | {ID: "wait_%s_total_ms", Name: "duration", Algo: collectorapi.Incremental}, |
| 671 | }, |
| 672 | } |
| 673 | waitResourceTimeChartTmpl = collectorapi.Chart{ |
| 674 | ID: "wait_%s_resource_time", |
| 675 | Title: "Resource wait time", |
| 676 | Units: "ms", |
| 677 | Fam: "waits", |
| 678 | Ctx: "mssql.wait_resource_time", |
| 679 | Priority: prioWaitResourceTime, |
| 680 | Dims: collectorapi.Dims{ |
| 681 | {ID: "wait_%s_resource_ms", Name: "duration", Algo: collectorapi.Incremental}, |
| 682 | }, |
| 683 | } |
| 684 | waitSignalTimeChartTmpl = collectorapi.Chart{ |
| 685 | ID: "wait_%s_signal_time", |
| 686 | Title: "Signal wait time", |
| 687 | Units: "ms", |
| 688 | Fam: "waits", |
| 689 | Ctx: "mssql.wait_signal_time", |
| 690 | Priority: prioWaitSignalTime, |
| 691 | Dims: collectorapi.Dims{ |
| 692 | {ID: "wait_%s_signal_ms", Name: "duration", Algo: collectorapi.Incremental}, |
| 693 | }, |
| 694 | } |
| 695 | waitCountChartTmpl = collectorapi.Chart{ |
| 696 | ID: "wait_%s_count", |
| 697 | Title: "Wait count", |
| 698 | Units: "waits/s", |
| 699 | Fam: "waits", |
| 700 | Ctx: "mssql.wait_count", |
| 701 | Priority: prioWaitCount, |
| 702 | Dims: collectorapi.Dims{ |
| 703 | {ID: "wait_%s_tasks", Name: "waits", Algo: collectorapi.Incremental}, |
| 704 | }, |
| 705 | } |
| 706 | waitMaxTimeChartTmpl = collectorapi.Chart{ |
| 707 | ID: "wait_%s_max_time", |
| 708 | Title: "Maximum wait time", |
| 709 | Units: "ms", |
| 710 | Fam: "waits", |
| 711 | Ctx: "mssql.wait_max_time", |
| 712 | Priority: prioWaitMaxTime, |
| 713 | Dims: collectorapi.Dims{ |
| 714 | {ID: "wait_%s_max_ms", Name: "max_time"}, |
| 715 | }, |
| 716 | } |
| 717 | ) |
| 718 | |
| 719 | // Job status chart template |
| 720 | var jobStatusChartTmpl = collectorapi.Chart{ |
| 721 | ID: "job_%s_status", |
| 722 | Title: "Job status", |
| 723 | Units: "status", |
| 724 | Fam: "jobs", |
| 725 | Ctx: "mssql.job_status", |
| 726 | Priority: prioJobStatus, |
| 727 | Dims: collectorapi.Dims{ |
| 728 | {ID: "job_%s_enabled", Name: "enabled"}, |
| 729 | {ID: "job_%s_disabled", Name: "disabled"}, |
| 730 | }, |
| 731 | } |
| 732 | |
| 733 | // Replication chart templates |
| 734 | var ( |
| 735 | replicationStatusChartTmpl = collectorapi.Chart{ |
| 736 | ID: "replication_%s_status", |
| 737 | Title: "Replication status", |
| 738 | Units: "status", |
| 739 | Fam: "replication", |
| 740 | Ctx: "mssql.replication_status", |
| 741 | Priority: prioReplicationStatus, |
| 742 | Dims: collectorapi.Dims{ |
| 743 | {ID: "replication_%s_status_started", Name: "started"}, |
| 744 | {ID: "replication_%s_status_succeeded", Name: "succeeded"}, |
| 745 | {ID: "replication_%s_status_in_progress", Name: "in_progress"}, |
| 746 | {ID: "replication_%s_status_idle", Name: "idle"}, |
| 747 | {ID: "replication_%s_status_retrying", Name: "retrying"}, |
| 748 | {ID: "replication_%s_status_failed", Name: "failed"}, |
| 749 | }, |
| 750 | } |
| 751 | replicationWarningChartTmpl = collectorapi.Chart{ |
| 752 | ID: "replication_%s_warning", |
| 753 | Title: "Replication warnings", |
| 754 | Units: "flags", |
| 755 | Fam: "replication", |
| 756 | Ctx: "mssql.replication_warning", |
| 757 | Priority: prioReplicationStatus + 1, |
| 758 | Dims: collectorapi.Dims{ |
| 759 | {ID: "replication_%s_warning_expiration", Name: "expiration"}, |
| 760 | {ID: "replication_%s_warning_latency", Name: "latency"}, |
| 761 | {ID: "replication_%s_warning_mergeexpiration", Name: "merge_expiration"}, |
| 762 | {ID: "replication_%s_warning_mergeslowrunduration", Name: "merge_slow_duration"}, |
| 763 | {ID: "replication_%s_warning_mergefastrunduration", Name: "merge_fast_duration"}, |
| 764 | {ID: "replication_%s_warning_mergefastrunspeed", Name: "merge_fast_speed"}, |
| 765 | {ID: "replication_%s_warning_mergeslowrunspeed", Name: "merge_slow_speed"}, |
| 766 | }, |
| 767 | } |
| 768 | replicationLatencyChartTmpl = collectorapi.Chart{ |
| 769 | ID: "replication_%s_latency", |
| 770 | Title: "Replication latency", |
| 771 | Units: "seconds", |
| 772 | Fam: "replication", |
| 773 | Ctx: "mssql.replication_latency", |
| 774 | Priority: prioReplicationLatency, |
| 775 | Dims: collectorapi.Dims{ |
| 776 | {ID: "replication_%s_latency_avg", Name: "average"}, |
| 777 | {ID: "replication_%s_latency_best", Name: "best"}, |
| 778 | {ID: "replication_%s_latency_worst", Name: "worst"}, |
| 779 | }, |
| 780 | } |
| 781 | replicationSubscriptionsChartTmpl = collectorapi.Chart{ |
| 782 | ID: "replication_%s_subscriptions", |
| 783 | Title: "Replication subscriptions", |
| 784 | Units: "subscriptions", |
| 785 | Fam: "replication", |
| 786 | Ctx: "mssql.replication_subscriptions", |
| 787 | Priority: prioReplicationStatus + 1, |
| 788 | Dims: collectorapi.Dims{ |
| 789 | {ID: "replication_%s_subscriptions", Name: "total"}, |
| 790 | {ID: "replication_%s_agents_running", Name: "agents_running"}, |
| 791 | }, |
| 792 | } |
| 793 | ) |
| 794 | |
| 795 | func (c *Collector) addDatabaseCharts(dbName string) { |
| 796 | // internal databases used by SQL Server's engine |
| 797 | // Their metrics don't represent actual workload or user activity |
| 798 | if strings.HasPrefix(dbName, "model_") { |
| 799 | return |
| 800 | } |
| 801 | |
| 802 | charts := &collectorapi.Charts{ |
| 803 | databaseActiveTransactionsChartTmpl.Copy(), |
| 804 | databaseTransactionsChartTmpl.Copy(), |
| 805 | databaseWriteTransactionsChartTmpl.Copy(), |
| 806 | databaseLogFlushesChartTmpl.Copy(), |
| 807 | databaseLogFlushedChartTmpl.Copy(), |
| 808 | databaseLogGrowthsChartTmpl.Copy(), |
| 809 | databaseIOStallChartTmpl.Copy(), |
| 810 | databaseDataFileSizeChartTmpl.Copy(), |
| 811 | databaseBackupRestoreThroughputChartTmpl.Copy(), |
| 812 | databaseStateChartTmpl.Copy(), |
| 813 | databaseReadOnlyChartTmpl.Copy(), |
| 814 | } |
| 815 | |
| 816 | dbID := cleanDatabaseName(dbName) |
| 817 | |
| 818 | for _, chart := range *charts { |
| 819 | chart.ID = fmt.Sprintf(chart.ID, dbID) |
| 820 | chart.Labels = []collectorapi.Label{ |
| 821 | {Key: "database", Value: dbName}, |
| 822 | } |
| 823 | for _, dim := range chart.Dims { |
| 824 | dim.ID = fmt.Sprintf(dim.ID, dbID) |
| 825 | } |
| 826 | } |
| 827 | |
| 828 | if err := c.Charts().Add(*charts...); err != nil { |
| 829 | c.Warning(err) |
| 830 | } |
| 831 | } |
| 832 | |
| 833 | func (c *Collector) addDatabaseLogCharts(dbName string) { |
| 834 | charts := &collectorapi.Charts{ |
| 835 | databaseLogFileSizeChartTmpl.Copy(), |
| 836 | databaseLogPercentUsedChartTmpl.Copy(), |
| 837 | databaseLogTruncationsShrinksChartTmpl.Copy(), |
| 838 | } |
| 839 | |
| 840 | dbID := cleanDatabaseName(dbName) |
| 841 | |
| 842 | for _, chart := range *charts { |
| 843 | chart.ID = fmt.Sprintf(chart.ID, dbID) |
| 844 | chart.Labels = []collectorapi.Label{ |
| 845 | {Key: "database", Value: dbName}, |
| 846 | } |
| 847 | for _, dim := range chart.Dims { |
| 848 | dim.ID = fmt.Sprintf(dim.ID, dbID) |
| 849 | } |
| 850 | } |
| 851 | |
| 852 | if err := c.Charts().Add(*charts...); err != nil { |
| 853 | c.Warning(err) |
| 854 | } |
| 855 | } |
| 856 | |
| 857 | func (c *Collector) addWaitTypeCharts(waitType string, waitCategory string) { |
| 858 | charts := &collectorapi.Charts{ |
| 859 | waitTotalTimeChartTmpl.Copy(), |
| 860 | waitResourceTimeChartTmpl.Copy(), |
| 861 | waitSignalTimeChartTmpl.Copy(), |
| 862 | waitMaxTimeChartTmpl.Copy(), |
| 863 | waitCountChartTmpl.Copy(), |
| 864 | } |
| 865 | |
| 866 | waitID := cleanWaitTypeName(waitType) |
| 867 | |
| 868 | for _, chart := range *charts { |
| 869 | chart.ID = fmt.Sprintf(chart.ID, waitID) |
| 870 | chart.Labels = []collectorapi.Label{ |
| 871 | {Key: "wait_type", Value: waitType}, |
| 872 | {Key: "wait_category", Value: waitCategory}, |
| 873 | } |
| 874 | for _, dim := range chart.Dims { |
| 875 | dim.ID = fmt.Sprintf(dim.ID, waitID) |
| 876 | } |
| 877 | } |
| 878 | |
| 879 | if err := c.Charts().Add(*charts...); err != nil { |
| 880 | c.Warning(err) |
| 881 | } |
| 882 | } |
| 883 | |
| 884 | func (c *Collector) addLockResourceCharts(resourceType string) { |
| 885 | chart := locksByResourceChartTmpl.Copy() |
| 886 | |
| 887 | resID := cleanResourceTypeName(resourceType) |
| 888 | |
| 889 | chart.ID = fmt.Sprintf(chart.ID, resID) |
| 890 | chart.Labels = []collectorapi.Label{ |
| 891 | {Key: "resource", Value: resourceType}, |
| 892 | } |
| 893 | for _, dim := range chart.Dims { |
| 894 | dim.ID = fmt.Sprintf(dim.ID, resID) |
| 895 | } |
| 896 | |
| 897 | if err := c.Charts().Add(chart); err != nil { |
| 898 | c.Warning(err) |
| 899 | } |
| 900 | } |
| 901 | |
| 902 | func (c *Collector) addLockStatsCharts(resourceType string) { |
| 903 | charts := &collectorapi.Charts{ |
| 904 | lockStatsDeadlocksChartTmpl.Copy(), |
| 905 | lockStatsWaitsChartTmpl.Copy(), |
| 906 | lockStatsTimeoutsChartTmpl.Copy(), |
| 907 | lockStatsRequestsChartTmpl.Copy(), |
| 908 | } |
| 909 | |
| 910 | resID := cleanResourceTypeName(resourceType) |
| 911 | |
| 912 | for _, chart := range *charts { |
| 913 | chart.ID = fmt.Sprintf(chart.ID, resID) |
| 914 | chart.Labels = []collectorapi.Label{ |
| 915 | {Key: "resource", Value: resourceType}, |
| 916 | } |
| 917 | for _, dim := range chart.Dims { |
| 918 | dim.ID = fmt.Sprintf(dim.ID, resID) |
| 919 | } |
| 920 | } |
| 921 | |
| 922 | if err := c.Charts().Add(*charts...); err != nil { |
| 923 | c.Warning(err) |
| 924 | } |
| 925 | } |
| 926 | |
| 927 | func (c *Collector) addJobCharts(jobName string) { |
| 928 | chart := jobStatusChartTmpl.Copy() |
| 929 | |
| 930 | jobID := cleanJobName(jobName) |
| 931 | |
| 932 | chart.ID = fmt.Sprintf(chart.ID, jobID) |
| 933 | chart.Labels = []collectorapi.Label{ |
| 934 | {Key: "job_name", Value: jobName}, |
| 935 | } |
| 936 | for _, dim := range chart.Dims { |
| 937 | dim.ID = fmt.Sprintf(dim.ID, jobID) |
| 938 | } |
| 939 | |
| 940 | if err := c.Charts().Add(chart); err != nil { |
| 941 | c.Warning(err) |
| 942 | } |
| 943 | } |
| 944 | |
| 945 | func (c *Collector) addReplicationCharts(pubDB, publication string) { |
| 946 | charts := &collectorapi.Charts{ |
| 947 | replicationStatusChartTmpl.Copy(), |
| 948 | replicationWarningChartTmpl.Copy(), |
| 949 | replicationLatencyChartTmpl.Copy(), |
| 950 | replicationSubscriptionsChartTmpl.Copy(), |
| 951 | } |
| 952 | |
| 953 | pubID := cleanPublicationName(pubDB, publication) |
| 954 | |
| 955 | for _, chart := range *charts { |
| 956 | chart.ID = fmt.Sprintf(chart.ID, pubID) |
| 957 | chart.Labels = []collectorapi.Label{ |
| 958 | {Key: "publisher_db", Value: pubDB}, |
| 959 | {Key: "publication", Value: publication}, |
| 960 | } |
| 961 | for _, dim := range chart.Dims { |
| 962 | dim.ID = fmt.Sprintf(dim.ID, pubID) |
| 963 | } |
| 964 | } |
| 965 | |
| 966 | if err := c.Charts().Add(*charts...); err != nil { |
| 967 | c.Warning(err) |
| 968 | } |
| 969 | } |
| 970 | |
| 971 | func cleanDatabaseName(name string) string { |
| 972 | return strings.ReplaceAll(strings.ToLower(name), " ", "_") |
| 973 | } |
| 974 | |
| 975 | func cleanWaitTypeName(name string) string { |
| 976 | return strings.ToLower(name) |
| 977 | } |
| 978 | |
| 979 | func cleanResourceTypeName(name string) string { |
| 980 | return strings.ToLower(name) |
| 981 | } |
| 982 | |
| 983 | func cleanJobName(name string) string { |
| 984 | r := strings.NewReplacer(" ", "_", ".", "_", "-", "_") |
| 985 | return strings.ToLower(r.Replace(name)) |
| 986 | } |
| 987 | |
| 988 | func cleanPublicationName(pubDB, publication string) string { |
| 989 | r := strings.NewReplacer(" ", "_", ".", "_", "-", "_") |
| 990 | return strings.ToLower(r.Replace(pubDB + "_" + publication)) |
| 991 | } |
| 992 | |
| 993 | func cleanAGName(name string) string { |
| 994 | r := strings.NewReplacer(" ", "_", ".", "_", "-", "_", "\\", "_") |
| 995 | return strings.ToLower(r.Replace(name)) |
| 996 | } |
| 997 | |
| 998 | func cleanAGReplicaName(agName, replicaServer string) string { |
| 999 | return cleanAGName(agName) + "_" + cleanAGName(replicaServer) |
| 1000 | } |
| 1001 | |
| 1002 | func cleanAGDatabaseReplicaName(agName, replicaServer, dbName string) string { |
| 1003 | return cleanAGName(agName) + "_" + cleanAGName(replicaServer) + "_" + cleanDatabaseName(dbName) |
| 1004 | } |
| 1005 | |
| 1006 | // AG-level chart templates |
| 1007 | var ( |
| 1008 | agSyncHealthChartTmpl = collectorapi.Chart{ |
| 1009 | ID: "ag_%s_sync_health", |
| 1010 | Title: "Availability group synchronization health", |
| 1011 | Units: "state", |
| 1012 | Fam: "ag health", |
| 1013 | Ctx: "mssql.ag_sync_health", |
| 1014 | Priority: prioAGSyncHealth, |
| 1015 | Dims: collectorapi.Dims{ |
| 1016 | {ID: "ag_%s_sync_health_not_healthy", Name: "not_healthy"}, |
| 1017 | {ID: "ag_%s_sync_health_partially_healthy", Name: "partially_healthy"}, |
| 1018 | {ID: "ag_%s_sync_health_healthy", Name: "healthy"}, |
| 1019 | }, |
| 1020 | } |
| 1021 | agRecoveryHealthChartTmpl = collectorapi.Chart{ |
| 1022 | ID: "ag_%s_recovery_health", |
| 1023 | Title: "Availability group recovery health", |
| 1024 | Units: "state", |
| 1025 | Fam: "ag health", |
| 1026 | Ctx: "mssql.ag_recovery_health", |
| 1027 | Priority: prioAGRecoveryHealth, |
| 1028 | Dims: collectorapi.Dims{ |
| 1029 | {ID: "ag_%s_primary_recovery_online", Name: "primary_online"}, |
| 1030 | {ID: "ag_%s_primary_recovery_in_progress", Name: "primary_in_progress"}, |
| 1031 | {ID: "ag_%s_secondary_recovery_online", Name: "secondary_online"}, |
| 1032 | {ID: "ag_%s_secondary_recovery_in_progress", Name: "secondary_in_progress"}, |
| 1033 | }, |
| 1034 | } |
| 1035 | ) |
| 1036 | |
| 1037 | // Replica-level chart templates |
| 1038 | var ( |
| 1039 | agReplicaRoleChartTmpl = collectorapi.Chart{ |
| 1040 | ID: "ag_replica_%s_role", |
| 1041 | Title: "Availability group replica role", |
| 1042 | Units: "state", |
| 1043 | Fam: "ag replicas", |
| 1044 | Ctx: "mssql.ag_replica_role", |
| 1045 | Priority: prioAGReplicaRole, |
| 1046 | Dims: collectorapi.Dims{ |
| 1047 | {ID: "ag_replica_%s_role_primary", Name: "primary"}, |
| 1048 | {ID: "ag_replica_%s_role_secondary", Name: "secondary"}, |
| 1049 | {ID: "ag_replica_%s_role_resolving", Name: "resolving"}, |
| 1050 | {ID: "ag_replica_%s_role_unknown", Name: "unknown"}, |
| 1051 | }, |
| 1052 | } |
| 1053 | agReplicaConnectedStateChartTmpl = collectorapi.Chart{ |
| 1054 | ID: "ag_replica_%s_connected_state", |
| 1055 | Title: "Availability group replica connected state", |
| 1056 | Units: "state", |
| 1057 | Fam: "ag replicas", |
| 1058 | Ctx: "mssql.ag_replica_connected_state", |
| 1059 | Priority: prioAGReplicaConnectedState, |
| 1060 | Dims: collectorapi.Dims{ |
| 1061 | {ID: "ag_replica_%s_connected", Name: "connected"}, |
| 1062 | {ID: "ag_replica_%s_disconnected", Name: "disconnected"}, |
| 1063 | {ID: "ag_replica_%s_conn_unknown", Name: "unknown"}, |
| 1064 | }, |
| 1065 | } |
| 1066 | agReplicaSyncHealthChartTmpl = collectorapi.Chart{ |
| 1067 | ID: "ag_replica_%s_sync_health", |
| 1068 | Title: "Availability group replica synchronization health", |
| 1069 | Units: "state", |
| 1070 | Fam: "ag replicas", |
| 1071 | Ctx: "mssql.ag_replica_sync_health", |
| 1072 | Priority: prioAGReplicaSyncHealth, |
| 1073 | Dims: collectorapi.Dims{ |
| 1074 | {ID: "ag_replica_%s_sync_health_not_healthy", Name: "not_healthy"}, |
| 1075 | {ID: "ag_replica_%s_sync_health_partially_healthy", Name: "partially_healthy"}, |
| 1076 | {ID: "ag_replica_%s_sync_health_healthy", Name: "healthy"}, |
| 1077 | }, |
| 1078 | } |
| 1079 | ) |
| 1080 | |
| 1081 | // Database-replica-level chart templates |
| 1082 | var ( |
| 1083 | agDBSyncStateChartTmpl = collectorapi.Chart{ |
| 1084 | ID: "ag_db_%s_sync_state", |
| 1085 | Title: "AG database synchronization state", |
| 1086 | Units: "state", |
| 1087 | Fam: "ag databases", |
| 1088 | Ctx: "mssql.ag_db_sync_state", |
| 1089 | Priority: prioAGDBSyncState, |
| 1090 | Dims: collectorapi.Dims{ |
| 1091 | {ID: "ag_db_%s_sync_state_not_synchronizing", Name: "not_synchronizing"}, |
| 1092 | {ID: "ag_db_%s_sync_state_synchronizing", Name: "synchronizing"}, |
| 1093 | {ID: "ag_db_%s_sync_state_synchronized", Name: "synchronized"}, |
| 1094 | {ID: "ag_db_%s_sync_state_reverting", Name: "reverting"}, |
| 1095 | {ID: "ag_db_%s_sync_state_initializing", Name: "initializing"}, |
| 1096 | }, |
| 1097 | } |
| 1098 | agDBLogSendQueueChartTmpl = collectorapi.Chart{ |
| 1099 | ID: "ag_db_%s_log_send_queue", |
| 1100 | Title: "AG database log send queue size", |
| 1101 | Units: "bytes", |
| 1102 | Fam: "ag databases", |
| 1103 | Ctx: "mssql.ag_db_log_send_queue", |
| 1104 | Priority: prioAGDBLogSendQueue, |
| 1105 | Dims: collectorapi.Dims{ |
| 1106 | {ID: "ag_db_%s_log_send_queue_size", Name: "queue_size"}, |
| 1107 | }, |
| 1108 | } |
| 1109 | agDBLogSendRateChartTmpl = collectorapi.Chart{ |
| 1110 | ID: "ag_db_%s_log_send_rate", |
| 1111 | Title: "AG database log send rate", |
| 1112 | Units: "bytes/s", |
| 1113 | Fam: "ag databases", |
| 1114 | Ctx: "mssql.ag_db_log_send_rate", |
| 1115 | Priority: prioAGDBLogSendRate, |
| 1116 | Dims: collectorapi.Dims{ |
| 1117 | {ID: "ag_db_%s_log_send_rate", Name: "send_rate"}, |
| 1118 | }, |
| 1119 | } |
| 1120 | agDBRedoQueueChartTmpl = collectorapi.Chart{ |
| 1121 | ID: "ag_db_%s_redo_queue", |
| 1122 | Title: "AG database redo queue size", |
| 1123 | Units: "bytes", |
| 1124 | Fam: "ag databases", |
| 1125 | Ctx: "mssql.ag_db_redo_queue", |
| 1126 | Priority: prioAGDBRedoQueue, |
| 1127 | Dims: collectorapi.Dims{ |
| 1128 | {ID: "ag_db_%s_redo_queue_size", Name: "queue_size"}, |
| 1129 | }, |
| 1130 | } |
| 1131 | agDBRedoRateChartTmpl = collectorapi.Chart{ |
| 1132 | ID: "ag_db_%s_redo_rate", |
| 1133 | Title: "AG database redo rate", |
| 1134 | Units: "bytes/s", |
| 1135 | Fam: "ag databases", |
| 1136 | Ctx: "mssql.ag_db_redo_rate", |
| 1137 | Priority: prioAGDBRedoRate, |
| 1138 | Dims: collectorapi.Dims{ |
| 1139 | {ID: "ag_db_%s_redo_rate", Name: "redo_rate"}, |
| 1140 | }, |
| 1141 | } |
| 1142 | agDBFilestreamSendRateChartTmpl = collectorapi.Chart{ |
| 1143 | ID: "ag_db_%s_filestream_send_rate", |
| 1144 | Title: "AG database filestream send rate", |
| 1145 | Units: "bytes/s", |
| 1146 | Fam: "ag databases", |
| 1147 | Ctx: "mssql.ag_db_filestream_send_rate", |
| 1148 | Priority: prioAGDBFilestreamSendRate, |
| 1149 | Dims: collectorapi.Dims{ |
| 1150 | {ID: "ag_db_%s_filestream_send_rate", Name: "send_rate"}, |
| 1151 | }, |
| 1152 | } |
| 1153 | agDBSecondaryLagChartTmpl = collectorapi.Chart{ |
| 1154 | ID: "ag_db_%s_secondary_lag", |
| 1155 | Title: "AG database secondary lag", |
| 1156 | Units: "seconds", |
| 1157 | Fam: "ag databases", |
| 1158 | Ctx: "mssql.ag_db_secondary_lag", |
| 1159 | Priority: prioAGDBSecondaryLag, |
| 1160 | Dims: collectorapi.Dims{ |
| 1161 | {ID: "ag_db_%s_secondary_lag_seconds", Name: "lag"}, |
| 1162 | }, |
| 1163 | } |
| 1164 | agDBSuspendedChartTmpl = collectorapi.Chart{ |
| 1165 | ID: "ag_db_%s_suspended", |
| 1166 | Title: "AG database data movement suspended state", |
| 1167 | Units: "state", |
| 1168 | Fam: "ag databases", |
| 1169 | Ctx: "mssql.ag_db_suspended", |
| 1170 | Priority: prioAGDBSuspended, |
| 1171 | Dims: collectorapi.Dims{ |
| 1172 | {ID: "ag_db_%s_not_suspended", Name: "active"}, |
| 1173 | {ID: "ag_db_%s_suspended", Name: "suspended"}, |
| 1174 | }, |
| 1175 | } |
| 1176 | agDBFailoverReadinessChartTmpl = collectorapi.Chart{ |
| 1177 | ID: "ag_db_%s_failover_readiness", |
| 1178 | Title: "AG database failover readiness", |
| 1179 | Units: "state", |
| 1180 | Fam: "ag databases", |
| 1181 | Ctx: "mssql.ag_db_failover_readiness", |
| 1182 | Priority: prioAGDBFailoverReadiness, |
| 1183 | Dims: collectorapi.Dims{ |
| 1184 | {ID: "ag_db_%s_failover_ready", Name: "ready"}, |
| 1185 | {ID: "ag_db_%s_failover_not_ready", Name: "not_ready"}, |
| 1186 | }, |
| 1187 | } |
| 1188 | agDBJoinedChartTmpl = collectorapi.Chart{ |
| 1189 | ID: "ag_db_%s_joined_state", |
| 1190 | Title: "AG database joined state", |
| 1191 | Units: "state", |
| 1192 | Fam: "ag databases", |
| 1193 | Ctx: "mssql.ag_db_joined_state", |
| 1194 | Priority: prioAGDBJoined, |
| 1195 | Dims: collectorapi.Dims{ |
| 1196 | {ID: "ag_db_%s_joined", Name: "joined"}, |
| 1197 | {ID: "ag_db_%s_not_joined", Name: "not_joined"}, |
| 1198 | }, |
| 1199 | } |
| 1200 | ) |
| 1201 | |
| 1202 | // Cluster-level chart |
| 1203 | var agClusterQuorumStateChart = collectorapi.Chart{ |
| 1204 | ID: "ag_cluster_quorum_state", |
| 1205 | Title: "WSFC cluster quorum state", |
| 1206 | Units: "state", |
| 1207 | Fam: "ag cluster", |
| 1208 | Ctx: "mssql.ag_cluster_quorum_state", |
| 1209 | Priority: prioAGClusterQuorumState, |
| 1210 | Dims: collectorapi.Dims{ |
| 1211 | {ID: "ag_cluster_quorum_state_normal", Name: "normal"}, |
| 1212 | {ID: "ag_cluster_quorum_state_forced", Name: "forced"}, |
| 1213 | {ID: "ag_cluster_quorum_state_unknown", Name: "unknown"}, |
| 1214 | }, |
| 1215 | } |
| 1216 | |
| 1217 | // Cluster member chart templates |
| 1218 | var ( |
| 1219 | agClusterMemberStateChartTmpl = collectorapi.Chart{ |
| 1220 | ID: "ag_cluster_member_%s_state", |
| 1221 | Title: "WSFC cluster member state", |
| 1222 | Units: "state", |
| 1223 | Fam: "ag cluster", |
| 1224 | Ctx: "mssql.ag_cluster_member_state", |
| 1225 | Priority: prioAGClusterMemberState, |
| 1226 | Dims: collectorapi.Dims{ |
| 1227 | {ID: "ag_cluster_member_%s_up", Name: "up"}, |
| 1228 | {ID: "ag_cluster_member_%s_down", Name: "down"}, |
| 1229 | }, |
| 1230 | } |
| 1231 | agClusterMemberQuorumVotesChartTmpl = collectorapi.Chart{ |
| 1232 | ID: "ag_cluster_member_%s_quorum_votes", |
| 1233 | Title: "WSFC cluster member quorum votes", |
| 1234 | Units: "votes", |
| 1235 | Fam: "ag cluster", |
| 1236 | Ctx: "mssql.ag_cluster_member_quorum_votes", |
| 1237 | Priority: prioAGClusterMemberQuorumVotes, |
| 1238 | Dims: collectorapi.Dims{ |
| 1239 | {ID: "ag_cluster_member_%s_quorum_votes", Name: "votes"}, |
| 1240 | }, |
| 1241 | } |
| 1242 | ) |
| 1243 | |
| 1244 | // Page repair chart template (keyed by database name) |
| 1245 | var agPageRepairChartTmpl = collectorapi.Chart{ |
| 1246 | ID: "ag_page_repair_%s", |
| 1247 | Title: "AG automatic page repair events", |
| 1248 | Units: "repairs", |
| 1249 | Fam: "ag page repair", |
| 1250 | Ctx: "mssql.ag_page_repair", |
| 1251 | Priority: prioAGPageRepair, |
| 1252 | Dims: collectorapi.Dims{ |
| 1253 | {ID: "ag_page_repair_%s_successful", Name: "successful"}, |
| 1254 | {ID: "ag_page_repair_%s_failed", Name: "failed"}, |
| 1255 | }, |
| 1256 | } |
| 1257 | |
| 1258 | // Thread chart template |
| 1259 | var agThreadsChartTmpl = collectorapi.Chart{ |
| 1260 | ID: "ag_%s_threads", |
| 1261 | Title: "Availability group threads", |
| 1262 | Units: "threads", |
| 1263 | Fam: "ag threads", |
| 1264 | Ctx: "mssql.ag_threads", |
| 1265 | Priority: prioAGThreads, |
| 1266 | Dims: collectorapi.Dims{ |
| 1267 | {ID: "ag_%s_capture_threads", Name: "capture"}, |
| 1268 | {ID: "ag_%s_redo_threads", Name: "redo"}, |
| 1269 | {ID: "ag_%s_parallel_redo_threads", Name: "parallel_redo"}, |
| 1270 | }, |
| 1271 | } |
| 1272 | |
| 1273 | func (c *Collector) addAGCharts(agName string) { |
| 1274 | charts := &collectorapi.Charts{ |
| 1275 | agSyncHealthChartTmpl.Copy(), |
| 1276 | agRecoveryHealthChartTmpl.Copy(), |
| 1277 | } |
| 1278 | |
| 1279 | if c.majorVersion >= 15 { // SQL Server 2019+ |
| 1280 | *charts = append(*charts, agThreadsChartTmpl.Copy()) |
| 1281 | } |
| 1282 | |
| 1283 | agID := cleanAGName(agName) |
| 1284 | |
| 1285 | for _, chart := range *charts { |
| 1286 | chart.ID = fmt.Sprintf(chart.ID, agID) |
| 1287 | chart.Labels = []collectorapi.Label{ |
| 1288 | {Key: "ag_name", Value: agName}, |
| 1289 | } |
| 1290 | for _, dim := range chart.Dims { |
| 1291 | dim.ID = fmt.Sprintf(dim.ID, agID) |
| 1292 | } |
| 1293 | } |
| 1294 | |
| 1295 | if err := c.Charts().Add(*charts...); err != nil { |
| 1296 | c.Warning(err) |
| 1297 | } |
| 1298 | } |
| 1299 | |
| 1300 | func (c *Collector) addAGReplicaCharts(agName, replicaServer, availMode, failoverMode string) { |
| 1301 | charts := &collectorapi.Charts{ |
| 1302 | agReplicaRoleChartTmpl.Copy(), |
| 1303 | agReplicaConnectedStateChartTmpl.Copy(), |
| 1304 | agReplicaSyncHealthChartTmpl.Copy(), |
| 1305 | } |
| 1306 | |
| 1307 | rID := cleanAGReplicaName(agName, replicaServer) |
| 1308 | |
| 1309 | for _, chart := range *charts { |
| 1310 | chart.ID = fmt.Sprintf(chart.ID, rID) |
| 1311 | chart.Labels = []collectorapi.Label{ |
| 1312 | {Key: "ag_name", Value: agName}, |
| 1313 | {Key: "replica_server", Value: replicaServer}, |
| 1314 | {Key: "availability_mode", Value: availMode}, |
| 1315 | {Key: "failover_mode", Value: failoverMode}, |
| 1316 | } |
| 1317 | for _, dim := range chart.Dims { |
| 1318 | dim.ID = fmt.Sprintf(dim.ID, rID) |
| 1319 | } |
| 1320 | } |
| 1321 | |
| 1322 | if err := c.Charts().Add(*charts...); err != nil { |
| 1323 | c.Warning(err) |
| 1324 | } |
| 1325 | } |
| 1326 | |
| 1327 | func (c *Collector) addAGDatabaseReplicaCharts(agName, replicaServer, dbName string) { |
| 1328 | charts := &collectorapi.Charts{ |
| 1329 | agDBSyncStateChartTmpl.Copy(), |
| 1330 | agDBLogSendQueueChartTmpl.Copy(), |
| 1331 | agDBLogSendRateChartTmpl.Copy(), |
| 1332 | agDBRedoQueueChartTmpl.Copy(), |
| 1333 | agDBRedoRateChartTmpl.Copy(), |
| 1334 | agDBFilestreamSendRateChartTmpl.Copy(), |
| 1335 | agDBSuspendedChartTmpl.Copy(), |
| 1336 | agDBFailoverReadinessChartTmpl.Copy(), |
| 1337 | agDBJoinedChartTmpl.Copy(), |
| 1338 | } |
| 1339 | |
| 1340 | if c.majorVersion >= 13 { // SQL Server 2016+ |
| 1341 | *charts = append(*charts, agDBSecondaryLagChartTmpl.Copy()) |
| 1342 | } |
| 1343 | |
| 1344 | drID := cleanAGDatabaseReplicaName(agName, replicaServer, dbName) |
| 1345 | |
| 1346 | for _, chart := range *charts { |
| 1347 | chart.ID = fmt.Sprintf(chart.ID, drID) |
| 1348 | chart.Labels = []collectorapi.Label{ |
| 1349 | {Key: "ag_name", Value: agName}, |
| 1350 | {Key: "replica_server", Value: replicaServer}, |
| 1351 | {Key: "database", Value: dbName}, |
| 1352 | } |
| 1353 | for _, dim := range chart.Dims { |
| 1354 | dim.ID = fmt.Sprintf(dim.ID, drID) |
| 1355 | } |
| 1356 | } |
| 1357 | |
| 1358 | if err := c.Charts().Add(*charts...); err != nil { |
| 1359 | c.Warning(err) |
| 1360 | } |
| 1361 | } |
| 1362 | |
| 1363 | func (c *Collector) addAGClusterMemberCharts(memberName string) { |
| 1364 | charts := &collectorapi.Charts{ |
| 1365 | agClusterMemberStateChartTmpl.Copy(), |
| 1366 | agClusterMemberQuorumVotesChartTmpl.Copy(), |
| 1367 | } |
| 1368 | |
| 1369 | mID := cleanAGName(memberName) |
| 1370 | |
| 1371 | for _, chart := range *charts { |
| 1372 | chart.ID = fmt.Sprintf(chart.ID, mID) |
| 1373 | chart.Labels = []collectorapi.Label{ |
| 1374 | {Key: "cluster_member", Value: memberName}, |
| 1375 | } |
| 1376 | for _, dim := range chart.Dims { |
| 1377 | dim.ID = fmt.Sprintf(dim.ID, mID) |
| 1378 | } |
| 1379 | } |
| 1380 | |
| 1381 | if err := c.Charts().Add(*charts...); err != nil { |
| 1382 | c.Warning(err) |
| 1383 | } |
| 1384 | } |
| 1385 | |
| 1386 | func (c *Collector) addAGPageRepairCharts(dbName string) { |
| 1387 | chart := agPageRepairChartTmpl.Copy() |
| 1388 | |
| 1389 | dbID := cleanDatabaseName(dbName) |
| 1390 | |
| 1391 | chart.ID = fmt.Sprintf(chart.ID, dbID) |
| 1392 | chart.Labels = []collectorapi.Label{ |
| 1393 | {Key: "database", Value: dbName}, |
| 1394 | } |
| 1395 | for _, dim := range chart.Dims { |
| 1396 | dim.ID = fmt.Sprintf(dim.ID, dbID) |
| 1397 | } |
| 1398 | |
| 1399 | if err := c.Charts().Add(chart); err != nil { |
| 1400 | c.Warning(err) |
| 1401 | } |
| 1402 | } |