| 1 | // SPDX-License-Identifier: GPL-3.0-or-later |
| 2 | |
| 3 | package mssql |
| 4 | |
| 5 | import ( |
| 6 | "context" |
| 7 | "database/sql" |
| 8 | "fmt" |
| 9 | "strconv" |
| 10 | "strings" |
| 11 | "time" |
| 12 | |
| 13 | "github.com/netdata/netdata/go/plugins/plugin/go.d/pkg/cloudauth/sqladapter" |
| 14 | ) |
| 15 | |
| 16 | // noLatencySentinel is the value SQL Server returns when no latency data is available |
| 17 | const noLatencySentinel = 999999 |
| 18 | |
| 19 | const ( |
| 20 | maxKBToBytes = int64(1<<63-1) / 1024 |
| 21 | maxKBToCentiPercent = int64(1<<63-1) / 10000 |
| 22 | ) |
| 23 | |
| 24 | func (c *Collector) collect() (map[string]int64, error) { |
| 25 | if c.db == nil { |
| 26 | db, err := c.openConnection() |
| 27 | if err != nil { |
| 28 | return nil, err |
| 29 | } |
| 30 | c.db = db |
| 31 | } |
| 32 | |
| 33 | if c.version == "" { |
| 34 | ver, err := c.queryVersion() |
| 35 | if err != nil { |
| 36 | return nil, fmt.Errorf("failed to query version: %v", err) |
| 37 | } |
| 38 | c.version = ver |
| 39 | c.majorVersion = parseMajorVersion(c.version) |
| 40 | c.Debugf("connected to SQL Server version %s (major: %d)", c.version, c.majorVersion) |
| 41 | } |
| 42 | |
| 43 | if !c.hadrChecked { |
| 44 | if err := c.checkHadrEnabled(); err != nil { |
| 45 | c.Debugf("HADR check failed: %v", err) |
| 46 | } |
| 47 | c.hadrChecked = true |
| 48 | } |
| 49 | |
| 50 | mx := make(map[string]int64) |
| 51 | |
| 52 | if err := c.collectInstanceMetrics(mx); err != nil { |
| 53 | return nil, err |
| 54 | } |
| 55 | if err := c.collectDatabaseMetrics(mx); err != nil { |
| 56 | return nil, err |
| 57 | } |
| 58 | if err := c.collectLockMetrics(mx); err != nil { |
| 59 | return nil, err |
| 60 | } |
| 61 | if err := c.collectWaitStats(mx); err != nil { |
| 62 | return nil, err |
| 63 | } |
| 64 | if err := c.collectJobStatus(mx); err != nil { |
| 65 | return nil, err |
| 66 | } |
| 67 | if err := c.collectReplicationStatus(mx); err != nil { |
| 68 | return nil, err |
| 69 | } |
| 70 | if c.hadrEnabled { |
| 71 | if err := c.collectAvailabilityGroups(mx); err != nil { |
| 72 | c.Warningf("AG metrics collection failed: %v", err) |
| 73 | } |
| 74 | } |
| 75 | |
| 76 | return mx, nil |
| 77 | } |
| 78 | |
| 79 | func (c *Collector) openConnection() (*sql.DB, error) { |
| 80 | driverName, dsn, err := c.resolveConnectionParams() |
| 81 | if err != nil { |
| 82 | return nil, err |
| 83 | } |
| 84 | |
| 85 | db, err := sql.Open(driverName, dsn) |
| 86 | if err != nil { |
| 87 | return nil, fmt.Errorf("error opening connection: %v", err) |
| 88 | } |
| 89 | |
| 90 | db.SetMaxOpenConns(1) |
| 91 | db.SetMaxIdleConns(1) |
| 92 | db.SetConnMaxLifetime(10 * time.Minute) |
| 93 | |
| 94 | ctx, cancel := context.WithTimeout(context.Background(), c.Timeout.Duration()) |
| 95 | defer cancel() |
| 96 | |
| 97 | if err := db.PingContext(ctx); err != nil { |
| 98 | _ = db.Close() |
| 99 | return nil, fmt.Errorf("error pinging database: %v", err) |
| 100 | } |
| 101 | |
| 102 | return db, nil |
| 103 | } |
| 104 | |
| 105 | func (c *Collector) resolveConnectionParams() (string, string, error) { |
| 106 | driverName := sqladapter.MSSQLDriver(c.CloudAuth) |
| 107 | dsn := c.DSN |
| 108 | if c.CloudAuth.IsEnabled() { |
| 109 | var err error |
| 110 | dsn, err = sqladapter.BuildMSSQLAzureADDSN(c.DSN, c.CloudAuth) |
| 111 | if err != nil { |
| 112 | return "", "", fmt.Errorf("error preparing cloud auth SQL Server DSN: %v", err) |
| 113 | } |
| 114 | } |
| 115 | return driverName, dsn, nil |
| 116 | } |
| 117 | |
| 118 | func (c *Collector) queryVersion() (string, error) { |
| 119 | ctx, cancel := context.WithTimeout(context.Background(), c.Timeout.Duration()) |
| 120 | defer cancel() |
| 121 | |
| 122 | var version string |
| 123 | err := c.db.QueryRowContext(ctx, queryVersion).Scan(&version) |
| 124 | if err != nil { |
| 125 | return "", err |
| 126 | } |
| 127 | |
| 128 | return version, nil |
| 129 | } |
| 130 | |
| 131 | func (c *Collector) collectInstanceMetrics(mx map[string]int64) error { |
| 132 | if err := c.collectUserConnections(mx); err != nil { |
| 133 | return err |
| 134 | } |
| 135 | if err := c.collectBlockedProcesses(mx); err != nil { |
| 136 | return err |
| 137 | } |
| 138 | if err := c.collectBatchRequests(mx); err != nil { |
| 139 | return err |
| 140 | } |
| 141 | if err := c.collectCompilations(mx); err != nil { |
| 142 | return err |
| 143 | } |
| 144 | if err := c.collectSQLErrors(mx); err != nil { |
| 145 | return err |
| 146 | } |
| 147 | if err := c.collectBufferManager(mx); err != nil { |
| 148 | return err |
| 149 | } |
| 150 | if err := c.collectMemoryManager(mx); err != nil { |
| 151 | return err |
| 152 | } |
| 153 | if err := c.collectAccessMethods(mx); err != nil { |
| 154 | return err |
| 155 | } |
| 156 | // Process and OS memory metrics (always collected) |
| 157 | if err := c.collectProcessMemory(mx); err != nil { |
| 158 | return err |
| 159 | } |
| 160 | if err := c.collectOSMemory(mx); err != nil { |
| 161 | return err |
| 162 | } |
| 163 | |
| 164 | return nil |
| 165 | } |
| 166 | |
| 167 | func (c *Collector) collectUserConnections(mx map[string]int64) error { |
| 168 | ctx, cancel := context.WithTimeout(context.Background(), c.Timeout.Duration()) |
| 169 | defer cancel() |
| 170 | |
| 171 | var userConns, sysConns int64 |
| 172 | err := c.db.QueryRowContext(ctx, queryUserConnections).Scan(&userConns, &sysConns) |
| 173 | if err != nil { |
| 174 | return fmt.Errorf("user connections query failed: %v", err) |
| 175 | } |
| 176 | |
| 177 | mx["user_connections"] = userConns |
| 178 | // Session connections: user vs internal (system) |
| 179 | mx["session_connections_user"] = userConns |
| 180 | mx["session_connections_internal"] = sysConns |
| 181 | |
| 182 | return nil |
| 183 | } |
| 184 | |
| 185 | func (c *Collector) collectBlockedProcesses(mx map[string]int64) error { |
| 186 | ctx, cancel := context.WithTimeout(context.Background(), c.Timeout.Duration()) |
| 187 | defer cancel() |
| 188 | |
| 189 | var blocked int64 |
| 190 | err := c.db.QueryRowContext(ctx, queryBlockedProcesses).Scan(&blocked) |
| 191 | if err != nil { |
| 192 | return fmt.Errorf("blocked processes query failed: %v", err) |
| 193 | } |
| 194 | |
| 195 | mx["blocked_processes"] = blocked |
| 196 | |
| 197 | return nil |
| 198 | } |
| 199 | |
| 200 | func (c *Collector) collectBatchRequests(mx map[string]int64) error { |
| 201 | ctx, cancel := context.WithTimeout(context.Background(), c.Timeout.Duration()) |
| 202 | defer cancel() |
| 203 | |
| 204 | var value int64 |
| 205 | err := c.db.QueryRowContext(ctx, queryBatchRequests).Scan(&value) |
| 206 | if err != nil { |
| 207 | return fmt.Errorf("batch requests query failed: %v", err) |
| 208 | } |
| 209 | |
| 210 | mx["batch_requests"] = value |
| 211 | |
| 212 | return nil |
| 213 | } |
| 214 | |
| 215 | func (c *Collector) collectCompilations(mx map[string]int64) error { |
| 216 | ctx, cancel := context.WithTimeout(context.Background(), c.Timeout.Duration()) |
| 217 | defer cancel() |
| 218 | |
| 219 | rows, err := c.db.QueryContext(ctx, queryCompilations) |
| 220 | if err != nil { |
| 221 | return fmt.Errorf("compilations query failed: %v", err) |
| 222 | } |
| 223 | defer rows.Close() |
| 224 | |
| 225 | for rows.Next() { |
| 226 | var counterName string |
| 227 | var value int64 |
| 228 | if err := rows.Scan(&counterName, &value); err != nil { |
| 229 | continue |
| 230 | } |
| 231 | |
| 232 | counterName = strings.TrimSpace(counterName) |
| 233 | switch counterName { |
| 234 | case "SQL Compilations/sec": |
| 235 | mx["sql_compilations"] = value |
| 236 | case "SQL Re-Compilations/sec": |
| 237 | mx["sql_recompilations"] = value |
| 238 | case "Auto-Param Attempts/sec": |
| 239 | mx["auto_param_attempts"] = value |
| 240 | case "Safe Auto-Params/sec": |
| 241 | mx["auto_param_safe"] = value |
| 242 | case "Failed Auto-Params/sec": |
| 243 | mx["auto_param_failed"] = value |
| 244 | } |
| 245 | } |
| 246 | |
| 247 | return rows.Err() |
| 248 | } |
| 249 | |
| 250 | func (c *Collector) collectBufferManager(mx map[string]int64) error { |
| 251 | ctx, cancel := context.WithTimeout(context.Background(), c.Timeout.Duration()) |
| 252 | defer cancel() |
| 253 | |
| 254 | rows, err := c.db.QueryContext(ctx, queryBufferManager) |
| 255 | if err != nil { |
| 256 | return fmt.Errorf("buffer manager query failed: %v", err) |
| 257 | } |
| 258 | defer rows.Close() |
| 259 | |
| 260 | var cacheHitRatio, cacheHitRatioBase int64 |
| 261 | |
| 262 | for rows.Next() { |
| 263 | var counterName string |
| 264 | var value int64 |
| 265 | if err := rows.Scan(&counterName, &value); err != nil { |
| 266 | continue |
| 267 | } |
| 268 | |
| 269 | counterName = strings.TrimSpace(counterName) |
| 270 | switch counterName { |
| 271 | case "Page reads/sec": |
| 272 | mx["buffer_page_reads"] = value |
| 273 | case "Page writes/sec": |
| 274 | mx["buffer_page_writes"] = value |
| 275 | case "Buffer cache hit ratio": |
| 276 | cacheHitRatio = value |
| 277 | case "Buffer cache hit ratio base": |
| 278 | cacheHitRatioBase = value |
| 279 | case "Checkpoint pages/sec": |
| 280 | mx["buffer_checkpoint_pages"] = value |
| 281 | case "Page life expectancy": |
| 282 | mx["buffer_page_life_expectancy"] = value |
| 283 | case "Lazy writes/sec": |
| 284 | mx["buffer_lazy_writes"] = value |
| 285 | case "Page lookups/sec": |
| 286 | mx["buffer_page_lookups"] = value |
| 287 | } |
| 288 | } |
| 289 | |
| 290 | // Calculate hit ratio as percentage |
| 291 | if cacheHitRatioBase > 0 { |
| 292 | mx["buffer_cache_hit_ratio"] = (cacheHitRatio * 100) / cacheHitRatioBase |
| 293 | } |
| 294 | |
| 295 | return rows.Err() |
| 296 | } |
| 297 | |
| 298 | func (c *Collector) collectMemoryManager(mx map[string]int64) error { |
| 299 | ctx, cancel := context.WithTimeout(context.Background(), c.Timeout.Duration()) |
| 300 | defer cancel() |
| 301 | |
| 302 | rows, err := c.db.QueryContext(ctx, queryMemoryManager) |
| 303 | if err != nil { |
| 304 | return fmt.Errorf("memory manager query failed: %v", err) |
| 305 | } |
| 306 | defer rows.Close() |
| 307 | |
| 308 | for rows.Next() { |
| 309 | var counterName string |
| 310 | var value int64 |
| 311 | if err := rows.Scan(&counterName, &value); err != nil { |
| 312 | continue |
| 313 | } |
| 314 | |
| 315 | counterName = strings.TrimSpace(counterName) |
| 316 | switch counterName { |
| 317 | case "Total Server Memory (KB)": |
| 318 | mx["memory_total"] = value * 1024 // Convert to bytes |
| 319 | case "Connection Memory (KB)": |
| 320 | mx["memory_connection"] = value * 1024 |
| 321 | case "Memory Grants Pending": |
| 322 | mx["memory_pending_grants"] = value |
| 323 | case "External benefit of memory": |
| 324 | mx["memory_external_benefit"] = value |
| 325 | } |
| 326 | } |
| 327 | |
| 328 | return rows.Err() |
| 329 | } |
| 330 | |
| 331 | func (c *Collector) collectAccessMethods(mx map[string]int64) error { |
| 332 | ctx, cancel := context.WithTimeout(context.Background(), c.Timeout.Duration()) |
| 333 | defer cancel() |
| 334 | |
| 335 | var value int64 |
| 336 | err := c.db.QueryRowContext(ctx, queryAccessMethods).Scan(&value) |
| 337 | if err != nil { |
| 338 | return fmt.Errorf("access methods query failed: %v", err) |
| 339 | } |
| 340 | |
| 341 | mx["page_splits"] = value |
| 342 | |
| 343 | return nil |
| 344 | } |
| 345 | |
| 346 | func (c *Collector) collectDatabaseMetrics(mx map[string]int64) error { |
| 347 | if err := c.collectDatabaseCounters(mx); err != nil { |
| 348 | return err |
| 349 | } |
| 350 | if err := c.collectLockStatsByResourceType(mx); err != nil { |
| 351 | return err |
| 352 | } |
| 353 | if err := c.collectDatabaseSize(mx); err != nil { |
| 354 | return err |
| 355 | } |
| 356 | if err := c.collectDatabaseStatus(mx); err != nil { |
| 357 | return err |
| 358 | } |
| 359 | // Collect I/O stall and log growth metrics per database |
| 360 | if err := c.collectIOStall(mx); err != nil { |
| 361 | return err |
| 362 | } |
| 363 | if err := c.collectLogGrowths(mx); err != nil { |
| 364 | return err |
| 365 | } |
| 366 | if err := c.collectDatabaseLogCounters(mx); err != nil { |
| 367 | return err |
| 368 | } |
| 369 | |
| 370 | return nil |
| 371 | } |
| 372 | |
| 373 | func (c *Collector) collectDatabaseCounters(mx map[string]int64) error { |
| 374 | ctx, cancel := context.WithTimeout(context.Background(), c.Timeout.Duration()) |
| 375 | defer cancel() |
| 376 | |
| 377 | rows, err := c.db.QueryContext(ctx, queryDatabaseCounters) |
| 378 | if err != nil { |
| 379 | return fmt.Errorf("database counters query failed: %v", err) |
| 380 | } |
| 381 | defer rows.Close() |
| 382 | |
| 383 | for rows.Next() { |
| 384 | var dbName, counterName string |
| 385 | var value int64 |
| 386 | if err := rows.Scan(&dbName, &counterName, &value); err != nil { |
| 387 | continue |
| 388 | } |
| 389 | |
| 390 | dbName = strings.TrimSpace(dbName) |
| 391 | counterName = strings.TrimSpace(counterName) |
| 392 | |
| 393 | if !c.seenDatabases[dbName] { |
| 394 | c.seenDatabases[dbName] = true |
| 395 | c.addDatabaseCharts(dbName) |
| 396 | } |
| 397 | |
| 398 | dbID := cleanDatabaseName(dbName) |
| 399 | switch counterName { |
| 400 | case "Active Transactions": |
| 401 | mx[fmt.Sprintf("database_%s_active_transactions", dbID)] = value |
| 402 | case "Transactions/sec": |
| 403 | mx[fmt.Sprintf("database_%s_transactions", dbID)] = value |
| 404 | case "Write Transactions/sec": |
| 405 | mx[fmt.Sprintf("database_%s_write_transactions", dbID)] = value |
| 406 | case "Backup/Restore Throughput/sec": |
| 407 | mx[fmt.Sprintf("database_%s_backup_restore_throughput", dbID)] = value |
| 408 | case "Log Bytes Flushed/sec": |
| 409 | mx[fmt.Sprintf("database_%s_log_flushed", dbID)] = value |
| 410 | case "Log Flushes/sec": |
| 411 | mx[fmt.Sprintf("database_%s_log_flushes", dbID)] = value |
| 412 | } |
| 413 | } |
| 414 | |
| 415 | return rows.Err() |
| 416 | } |
| 417 | |
| 418 | type databaseLogCounters struct { |
| 419 | sizeKB int64 |
| 420 | usedKB int64 |
| 421 | truncations int64 |
| 422 | shrinks int64 |
| 423 | hasSize bool |
| 424 | hasUsed bool |
| 425 | hasTruncations bool |
| 426 | hasShrinks bool |
| 427 | hasKnownCounter bool |
| 428 | } |
| 429 | |
| 430 | func (c *Collector) collectDatabaseLogCounters(mx map[string]int64) error { |
| 431 | ctx, cancel := context.WithTimeout(context.Background(), c.Timeout.Duration()) |
| 432 | defer cancel() |
| 433 | |
| 434 | rows, err := c.db.QueryContext(ctx, queryDatabaseLogCounters) |
| 435 | if err != nil { |
| 436 | return fmt.Errorf("database log counters query failed: %v", err) |
| 437 | } |
| 438 | defer rows.Close() |
| 439 | |
| 440 | counters := make(map[string]*databaseLogCounters) |
| 441 | |
| 442 | for rows.Next() { |
| 443 | var dbName, counterName string |
| 444 | var value int64 |
| 445 | if err := rows.Scan(&dbName, &counterName, &value); err != nil { |
| 446 | continue |
| 447 | } |
| 448 | if value < 0 { |
| 449 | continue |
| 450 | } |
| 451 | |
| 452 | dbName = strings.TrimSpace(dbName) |
| 453 | counterName = strings.TrimSpace(counterName) |
| 454 | if dbName == "" { |
| 455 | continue |
| 456 | } |
| 457 | |
| 458 | if !c.seenDatabasesWithLog[dbName] { |
| 459 | c.seenDatabasesWithLog[dbName] = true |
| 460 | c.addDatabaseLogCharts(dbName) |
| 461 | } |
| 462 | |
| 463 | if counters[dbName] == nil { |
| 464 | counters[dbName] = &databaseLogCounters{} |
| 465 | } |
| 466 | |
| 467 | switch counterName { |
| 468 | case "Log File(s) Size (KB)": |
| 469 | counters[dbName].sizeKB = value |
| 470 | counters[dbName].hasSize = true |
| 471 | counters[dbName].hasKnownCounter = true |
| 472 | case "Log File(s) Used Size (KB)": |
| 473 | counters[dbName].usedKB = value |
| 474 | counters[dbName].hasUsed = true |
| 475 | counters[dbName].hasKnownCounter = true |
| 476 | case "Log Truncations": |
| 477 | counters[dbName].truncations = value |
| 478 | counters[dbName].hasTruncations = true |
| 479 | counters[dbName].hasKnownCounter = true |
| 480 | case "Log Shrinks": |
| 481 | counters[dbName].shrinks = value |
| 482 | counters[dbName].hasShrinks = true |
| 483 | counters[dbName].hasKnownCounter = true |
| 484 | } |
| 485 | } |
| 486 | |
| 487 | if err := rows.Err(); err != nil { |
| 488 | return err |
| 489 | } |
| 490 | |
| 491 | for dbName, values := range counters { |
| 492 | if values == nil || !values.hasKnownCounter { |
| 493 | continue |
| 494 | } |
| 495 | |
| 496 | dbID := cleanDatabaseName(dbName) |
| 497 | |
| 498 | if values.hasSize && values.hasUsed && values.sizeKB > 0 && |
| 499 | values.sizeKB <= maxKBToBytes && values.usedKB <= maxKBToBytes { |
| 500 | usedBytes := values.usedKB * 1024 |
| 501 | freeKB := max(values.sizeKB-values.usedKB, 0) |
| 502 | |
| 503 | mx[fmt.Sprintf("database_%s_log_size_used", dbID)] = usedBytes |
| 504 | mx[fmt.Sprintf("database_%s_log_size_free", dbID)] = freeKB * 1024 |
| 505 | |
| 506 | if values.usedKB <= maxKBToCentiPercent { |
| 507 | mx[fmt.Sprintf("database_%s_log_percent_used", dbID)] = values.usedKB * 10000 / values.sizeKB |
| 508 | } |
| 509 | } |
| 510 | |
| 511 | if values.hasTruncations { |
| 512 | mx[fmt.Sprintf("database_%s_log_truncations", dbID)] = values.truncations |
| 513 | } |
| 514 | if values.hasShrinks { |
| 515 | mx[fmt.Sprintf("database_%s_log_shrinks", dbID)] = values.shrinks |
| 516 | } |
| 517 | } |
| 518 | |
| 519 | return nil |
| 520 | } |
| 521 | |
| 522 | func (c *Collector) collectLockStatsByResourceType(mx map[string]int64) error { |
| 523 | ctx, cancel := context.WithTimeout(context.Background(), c.Timeout.Duration()) |
| 524 | defer cancel() |
| 525 | |
| 526 | rows, err := c.db.QueryContext(ctx, queryDatabaseLocks) |
| 527 | if err != nil { |
| 528 | return fmt.Errorf("lock stats query failed: %v", err) |
| 529 | } |
| 530 | defer rows.Close() |
| 531 | |
| 532 | for rows.Next() { |
| 533 | var resourceType, counterName string |
| 534 | var value int64 |
| 535 | if err := rows.Scan(&resourceType, &counterName, &value); err != nil { |
| 536 | continue |
| 537 | } |
| 538 | |
| 539 | resourceType = strings.TrimSpace(resourceType) |
| 540 | counterName = strings.TrimSpace(counterName) |
| 541 | |
| 542 | if !c.seenLockStatsTypes[resourceType] { |
| 543 | c.seenLockStatsTypes[resourceType] = true |
| 544 | c.addLockStatsCharts(resourceType) |
| 545 | } |
| 546 | |
| 547 | // Note: instance_name from Locks counter is the lock resource type, not database name |
| 548 | resID := cleanResourceTypeName(resourceType) |
| 549 | switch counterName { |
| 550 | case "Number of Deadlocks/sec": |
| 551 | mx[fmt.Sprintf("lock_stats_%s_deadlocks", resID)] = value |
| 552 | case "Lock Waits/sec": |
| 553 | mx[fmt.Sprintf("lock_stats_%s_waits", resID)] = value |
| 554 | case "Lock Timeouts/sec": |
| 555 | mx[fmt.Sprintf("lock_stats_%s_timeouts", resID)] = value |
| 556 | case "Lock Requests/sec": |
| 557 | mx[fmt.Sprintf("lock_stats_%s_requests", resID)] = value |
| 558 | } |
| 559 | } |
| 560 | |
| 561 | return rows.Err() |
| 562 | } |
| 563 | |
| 564 | func (c *Collector) collectDatabaseSize(mx map[string]int64) error { |
| 565 | ctx, cancel := context.WithTimeout(context.Background(), c.Timeout.Duration()) |
| 566 | defer cancel() |
| 567 | |
| 568 | rows, err := c.db.QueryContext(ctx, queryDatabaseSize) |
| 569 | if err != nil { |
| 570 | return fmt.Errorf("database size query failed: %v", err) |
| 571 | } |
| 572 | defer rows.Close() |
| 573 | |
| 574 | for rows.Next() { |
| 575 | var dbName string |
| 576 | var size int64 |
| 577 | if err := rows.Scan(&dbName, &size); err != nil { |
| 578 | continue |
| 579 | } |
| 580 | |
| 581 | dbName = strings.TrimSpace(dbName) |
| 582 | |
| 583 | if !c.seenDatabases[dbName] { |
| 584 | c.seenDatabases[dbName] = true |
| 585 | c.addDatabaseCharts(dbName) |
| 586 | } |
| 587 | |
| 588 | dbID := cleanDatabaseName(dbName) |
| 589 | mx[fmt.Sprintf("database_%s_data_file_size", dbID)] = size |
| 590 | } |
| 591 | |
| 592 | return rows.Err() |
| 593 | } |
| 594 | |
| 595 | func (c *Collector) collectLockMetrics(mx map[string]int64) error { |
| 596 | ctx, cancel := context.WithTimeout(context.Background(), c.Timeout.Duration()) |
| 597 | defer cancel() |
| 598 | |
| 599 | rows, err := c.db.QueryContext(ctx, queryLocksByResource) |
| 600 | if err != nil { |
| 601 | return fmt.Errorf("locks by resource query failed: %v", err) |
| 602 | } |
| 603 | defer rows.Close() |
| 604 | |
| 605 | for rows.Next() { |
| 606 | var resourceType string |
| 607 | var count int64 |
| 608 | if err := rows.Scan(&resourceType, &count); err != nil { |
| 609 | continue |
| 610 | } |
| 611 | |
| 612 | resourceType = strings.TrimSpace(resourceType) |
| 613 | |
| 614 | if !c.seenLockTypes[resourceType] { |
| 615 | c.seenLockTypes[resourceType] = true |
| 616 | c.addLockResourceCharts(resourceType) |
| 617 | } |
| 618 | |
| 619 | resID := cleanResourceTypeName(resourceType) |
| 620 | mx[fmt.Sprintf("locks_%s_count", resID)] = count |
| 621 | } |
| 622 | |
| 623 | return rows.Err() |
| 624 | } |
| 625 | |
| 626 | func (c *Collector) collectWaitStats(mx map[string]int64) error { |
| 627 | ctx, cancel := context.WithTimeout(context.Background(), c.Timeout.Duration()) |
| 628 | defer cancel() |
| 629 | |
| 630 | rows, err := c.db.QueryContext(ctx, queryWaitStats) |
| 631 | if err != nil { |
| 632 | return fmt.Errorf("wait stats query failed: %v", err) |
| 633 | } |
| 634 | defer rows.Close() |
| 635 | |
| 636 | for rows.Next() { |
| 637 | var waitType string |
| 638 | var totalWait, resourceWait, signalWait, maxWait, waitingTasks int64 |
| 639 | if err := rows.Scan(&waitType, &totalWait, &resourceWait, &signalWait, &maxWait, &waitingTasks); err != nil { |
| 640 | continue |
| 641 | } |
| 642 | |
| 643 | waitType = strings.TrimSpace(waitType) |
| 644 | waitCategory := getWaitCategory(waitType) |
| 645 | |
| 646 | if !c.seenWaitTypes[waitType] { |
| 647 | c.seenWaitTypes[waitType] = true |
| 648 | c.addWaitTypeCharts(waitType, waitCategory) |
| 649 | } |
| 650 | |
| 651 | waitID := cleanWaitTypeName(waitType) |
| 652 | mx[fmt.Sprintf("wait_%s_total_ms", waitID)] = totalWait |
| 653 | mx[fmt.Sprintf("wait_%s_resource_ms", waitID)] = resourceWait |
| 654 | mx[fmt.Sprintf("wait_%s_signal_ms", waitID)] = signalWait |
| 655 | mx[fmt.Sprintf("wait_%s_max_ms", waitID)] = maxWait |
| 656 | mx[fmt.Sprintf("wait_%s_tasks", waitID)] = waitingTasks |
| 657 | } |
| 658 | |
| 659 | return rows.Err() |
| 660 | } |
| 661 | |
| 662 | func (c *Collector) collectJobStatus(mx map[string]int64) error { |
| 663 | ctx, cancel := context.WithTimeout(context.Background(), c.Timeout.Duration()) |
| 664 | defer cancel() |
| 665 | |
| 666 | rows, err := c.db.QueryContext(ctx, queryJobs) |
| 667 | if err != nil { |
| 668 | return fmt.Errorf("jobs query failed: %v", err) |
| 669 | } |
| 670 | defer rows.Close() |
| 671 | |
| 672 | for rows.Next() { |
| 673 | var jobName string |
| 674 | var enabled int64 |
| 675 | if err := rows.Scan(&jobName, &enabled); err != nil { |
| 676 | continue |
| 677 | } |
| 678 | |
| 679 | jobName = strings.TrimSpace(jobName) |
| 680 | |
| 681 | if !c.seenJobs[jobName] { |
| 682 | c.seenJobs[jobName] = true |
| 683 | c.addJobCharts(jobName) |
| 684 | } |
| 685 | |
| 686 | jobID := cleanJobName(jobName) |
| 687 | if enabled == 1 { |
| 688 | mx[fmt.Sprintf("job_%s_enabled", jobID)] = 1 |
| 689 | mx[fmt.Sprintf("job_%s_disabled", jobID)] = 0 |
| 690 | } else { |
| 691 | mx[fmt.Sprintf("job_%s_enabled", jobID)] = 0 |
| 692 | mx[fmt.Sprintf("job_%s_disabled", jobID)] = 1 |
| 693 | } |
| 694 | } |
| 695 | |
| 696 | return rows.Err() |
| 697 | } |
| 698 | |
| 699 | func (c *Collector) collectSQLErrors(mx map[string]int64) error { |
| 700 | ctx, cancel := context.WithTimeout(context.Background(), c.Timeout.Duration()) |
| 701 | defer cancel() |
| 702 | |
| 703 | var value int64 |
| 704 | err := c.db.QueryRowContext(ctx, querySQLErrors).Scan(&value) |
| 705 | if err != nil { |
| 706 | return fmt.Errorf("sql errors query failed: %v", err) |
| 707 | } |
| 708 | |
| 709 | mx["sql_errors_total"] = value |
| 710 | |
| 711 | return nil |
| 712 | } |
| 713 | |
| 714 | func (c *Collector) collectDatabaseStatus(mx map[string]int64) error { |
| 715 | ctx, cancel := context.WithTimeout(context.Background(), c.Timeout.Duration()) |
| 716 | defer cancel() |
| 717 | |
| 718 | rows, err := c.db.QueryContext(ctx, queryDatabaseStatus) |
| 719 | if err != nil { |
| 720 | return fmt.Errorf("database status query failed: %v", err) |
| 721 | } |
| 722 | defer rows.Close() |
| 723 | |
| 724 | for rows.Next() { |
| 725 | var dbName string |
| 726 | var state int64 |
| 727 | var isReadOnly bool |
| 728 | if err := rows.Scan(&dbName, &state, &isReadOnly); err != nil { |
| 729 | continue |
| 730 | } |
| 731 | |
| 732 | dbName = strings.TrimSpace(dbName) |
| 733 | |
| 734 | if !c.seenDatabases[dbName] { |
| 735 | c.seenDatabases[dbName] = true |
| 736 | c.addDatabaseCharts(dbName) |
| 737 | } |
| 738 | |
| 739 | dbID := cleanDatabaseName(dbName) |
| 740 | |
| 741 | // Database state values: |
| 742 | // 0 = ONLINE, 1 = RESTORING, 2 = RECOVERING, 3 = RECOVERY_PENDING |
| 743 | // 4 = SUSPECT, 5 = EMERGENCY, 6 = OFFLINE |
| 744 | mx[fmt.Sprintf("database_%s_state_online", dbID)] = boolToInt(state == 0) |
| 745 | mx[fmt.Sprintf("database_%s_state_restoring", dbID)] = boolToInt(state == 1) |
| 746 | mx[fmt.Sprintf("database_%s_state_recovering", dbID)] = boolToInt(state == 2) |
| 747 | mx[fmt.Sprintf("database_%s_state_pending", dbID)] = boolToInt(state == 3) |
| 748 | mx[fmt.Sprintf("database_%s_state_suspect", dbID)] = boolToInt(state == 4) |
| 749 | mx[fmt.Sprintf("database_%s_state_emergency", dbID)] = boolToInt(state == 5) |
| 750 | mx[fmt.Sprintf("database_%s_state_offline", dbID)] = boolToInt(state == 6) |
| 751 | |
| 752 | // Read-only status |
| 753 | mx[fmt.Sprintf("database_%s_read_only", dbID)] = boolToInt(isReadOnly) |
| 754 | mx[fmt.Sprintf("database_%s_read_write", dbID)] = boolToInt(!isReadOnly) |
| 755 | } |
| 756 | |
| 757 | return rows.Err() |
| 758 | } |
| 759 | |
| 760 | func (c *Collector) collectReplicationStatus(mx map[string]int64) error { |
| 761 | ctx, cancel := context.WithTimeout(context.Background(), c.Timeout.Duration()) |
| 762 | defer cancel() |
| 763 | |
| 764 | // First collect monitor data (status, latency, etc.) |
| 765 | rows, err := c.db.QueryContext(ctx, queryReplicationStatus) |
| 766 | if err != nil { |
| 767 | // Replication may not be configured, don't treat as error |
| 768 | c.Debugf("replication status query failed (may not be configured): %v", err) |
| 769 | return nil |
| 770 | } |
| 771 | defer rows.Close() |
| 772 | |
| 773 | for rows.Next() { |
| 774 | var pubDB, publication string |
| 775 | var status, warning, worstLatency, bestLatency, avgLatency, runningAgents int64 |
| 776 | if err := rows.Scan(&pubDB, &publication, &status, &warning, &worstLatency, &bestLatency, &avgLatency, &runningAgents); err != nil { |
| 777 | c.Debugf("replication scan error: %v", err) |
| 778 | continue |
| 779 | } |
| 780 | |
| 781 | pubDB = strings.TrimSpace(pubDB) |
| 782 | publication = strings.TrimSpace(publication) |
| 783 | |
| 784 | pubKey := pubDB + "_" + publication |
| 785 | if !c.seenReplications[pubKey] { |
| 786 | c.seenReplications[pubKey] = true |
| 787 | c.addReplicationCharts(pubDB, publication) |
| 788 | } |
| 789 | |
| 790 | pubID := cleanPublicationName(pubDB, publication) |
| 791 | |
| 792 | // Decode status into 6 discrete states (matching C implementation) |
| 793 | // 1=started, 2=succeeded, 3=in_progress, 4=idle, 5=retrying, 6=failed |
| 794 | mx[fmt.Sprintf("replication_%s_status_started", pubID)] = boolToInt(status == 1) |
| 795 | mx[fmt.Sprintf("replication_%s_status_succeeded", pubID)] = boolToInt(status == 2) |
| 796 | mx[fmt.Sprintf("replication_%s_status_in_progress", pubID)] = boolToInt(status == 3) |
| 797 | mx[fmt.Sprintf("replication_%s_status_idle", pubID)] = boolToInt(status == 4) |
| 798 | mx[fmt.Sprintf("replication_%s_status_retrying", pubID)] = boolToInt(status == 5) |
| 799 | mx[fmt.Sprintf("replication_%s_status_failed", pubID)] = boolToInt(status == 6) |
| 800 | |
| 801 | // Decode warning into 7 individual flags (bitfield) |
| 802 | // Bit 0x01: expiration, 0x02: latency, 0x04: mergeexpiration |
| 803 | // 0x08: mergeslowrunduration, 0x10: mergefastrunduration |
| 804 | // 0x20: mergefastrunspeed, 0x40: mergeslowrunspeed |
| 805 | mx[fmt.Sprintf("replication_%s_warning_expiration", pubID)] = boolToInt(warning&0x01 != 0) |
| 806 | mx[fmt.Sprintf("replication_%s_warning_latency", pubID)] = boolToInt(warning&0x02 != 0) |
| 807 | mx[fmt.Sprintf("replication_%s_warning_mergeexpiration", pubID)] = boolToInt(warning&0x04 != 0) |
| 808 | mx[fmt.Sprintf("replication_%s_warning_mergeslowrunduration", pubID)] = boolToInt(warning&0x08 != 0) |
| 809 | mx[fmt.Sprintf("replication_%s_warning_mergefastrunduration", pubID)] = boolToInt(warning&0x10 != 0) |
| 810 | mx[fmt.Sprintf("replication_%s_warning_mergefastrunspeed", pubID)] = boolToInt(warning&0x20 != 0) |
| 811 | mx[fmt.Sprintf("replication_%s_warning_mergeslowrunspeed", pubID)] = boolToInt(warning&0x40 != 0) |
| 812 | |
| 813 | mx[fmt.Sprintf("replication_%s_latency_avg", pubID)] = avgLatency |
| 814 | // Handle the noLatencySentinel for "no value" (only bestLatency uses sentinel in query) |
| 815 | if bestLatency == noLatencySentinel { |
| 816 | bestLatency = 0 |
| 817 | } |
| 818 | mx[fmt.Sprintf("replication_%s_latency_best", pubID)] = bestLatency |
| 819 | mx[fmt.Sprintf("replication_%s_latency_worst", pubID)] = worstLatency |
| 820 | mx[fmt.Sprintf("replication_%s_agents_running", pubID)] = runningAgents |
| 821 | } |
| 822 | |
| 823 | if err := rows.Err(); err != nil { |
| 824 | return err |
| 825 | } |
| 826 | |
| 827 | // Now collect subscription counts from MSpublications/MSsubscriptions |
| 828 | rows2, err := c.db.QueryContext(ctx, querySubscriptionCount) |
| 829 | if err != nil { |
| 830 | c.Debugf("subscription count query failed: %v", err) |
| 831 | return nil |
| 832 | } |
| 833 | defer rows2.Close() |
| 834 | |
| 835 | for rows2.Next() { |
| 836 | var pubDB, publication string |
| 837 | var subCount int64 |
| 838 | if err := rows2.Scan(&pubDB, &publication, &subCount); err != nil { |
| 839 | continue |
| 840 | } |
| 841 | |
| 842 | pubDB = strings.TrimSpace(pubDB) |
| 843 | publication = strings.TrimSpace(publication) |
| 844 | pubID := cleanPublicationName(pubDB, publication) |
| 845 | mx[fmt.Sprintf("replication_%s_subscriptions", pubID)] = subCount |
| 846 | } |
| 847 | |
| 848 | return rows2.Err() |
| 849 | } |
| 850 | |
| 851 | func boolToInt(b bool) int64 { |
| 852 | if b { |
| 853 | return 1 |
| 854 | } |
| 855 | return 0 |
| 856 | } |
| 857 | |
| 858 | // parseMajorVersion extracts the major version from a string like "16.0.4175.1" |
| 859 | func parseMajorVersion(version string) int { |
| 860 | parts := strings.SplitN(version, ".", 2) |
| 861 | if len(parts) == 0 { |
| 862 | return 0 |
| 863 | } |
| 864 | v, err := strconv.Atoi(parts[0]) |
| 865 | if err != nil { |
| 866 | return 0 |
| 867 | } |
| 868 | return v |
| 869 | } |
| 870 | |
| 871 | func (c *Collector) checkHadrEnabled() error { |
| 872 | ctx, cancel := context.WithTimeout(context.Background(), c.Timeout.Duration()) |
| 873 | defer cancel() |
| 874 | |
| 875 | var enabled sql.NullInt64 |
| 876 | err := c.db.QueryRowContext(ctx, queryHadrEnabled).Scan(&enabled) |
| 877 | if err != nil { |
| 878 | return fmt.Errorf("HADR enabled check failed: %v", err) |
| 879 | } |
| 880 | |
| 881 | c.hadrEnabled = enabled.Valid && enabled.Int64 == 1 |
| 882 | |
| 883 | if c.hadrEnabled { |
| 884 | c.Debugf("Always On Availability Groups is enabled") |
| 885 | } else { |
| 886 | c.Debugf("Always On Availability Groups is not enabled, skipping AG metrics") |
| 887 | } |
| 888 | |
| 889 | return nil |
| 890 | } |
| 891 | |
| 892 | // agDatabaseReplicaQuery returns the appropriate query for the SQL Server version |
| 893 | func agDatabaseReplicaQuery(majorVersion int) string { |
| 894 | if majorVersion >= 13 { // SQL Server 2016+ |
| 895 | return queryAGDatabaseReplicas16 |
| 896 | } |
| 897 | return queryAGDatabaseReplicasPre16 // SQL Server 2012-2014 |
| 898 | } |
| 899 | |
| 900 | func (c *Collector) collectAvailabilityGroups(mx map[string]int64) error { |
| 901 | if err := c.collectAGHealth(mx); err != nil { |
| 902 | return err |
| 903 | } |
| 904 | if err := c.collectAGReplicaStates(mx); err != nil { |
| 905 | return err |
| 906 | } |
| 907 | if err := c.collectAGDatabaseReplicas(mx); err != nil { |
| 908 | return err |
| 909 | } |
| 910 | if err := c.collectAGCluster(mx); err != nil { |
| 911 | c.Debugf("AG cluster query failed (WSFC may not be configured): %v", err) |
| 912 | } |
| 913 | if err := c.collectAGClusterMembers(mx); err != nil { |
| 914 | c.Debugf("AG cluster members query failed: %v", err) |
| 915 | } |
| 916 | if err := c.collectAGFailoverReadiness(mx); err != nil { |
| 917 | c.Debugf("AG failover readiness query failed: %v", err) |
| 918 | } |
| 919 | if err := c.collectAGAutoPageRepair(mx); err != nil { |
| 920 | c.Debugf("AG auto page repair query failed: %v", err) |
| 921 | } |
| 922 | if c.majorVersion >= 15 { // SQL Server 2019+ |
| 923 | if err := c.collectAGThreads(mx); err != nil { |
| 924 | c.Debugf("AG threads query failed: %v", err) |
| 925 | } |
| 926 | } |
| 927 | return nil |
| 928 | } |
| 929 | |
| 930 | func (c *Collector) collectAGHealth(mx map[string]int64) error { |
| 931 | ctx, cancel := context.WithTimeout(context.Background(), c.Timeout.Duration()) |
| 932 | defer cancel() |
| 933 | |
| 934 | rows, err := c.db.QueryContext(ctx, queryAGHealth) |
| 935 | if err != nil { |
| 936 | return fmt.Errorf("AG health query failed: %v", err) |
| 937 | } |
| 938 | defer rows.Close() |
| 939 | |
| 940 | for rows.Next() { |
| 941 | var agName string |
| 942 | var syncHealth, primaryRecoveryHealth, secondaryRecoveryHealth int64 |
| 943 | if err := rows.Scan(&agName, &syncHealth, &primaryRecoveryHealth, &secondaryRecoveryHealth); err != nil { |
| 944 | c.Debugf("AG health scan failed: %v", err) |
| 945 | continue |
| 946 | } |
| 947 | |
| 948 | agName = strings.TrimSpace(agName) |
| 949 | |
| 950 | if !c.seenAGs[agName] { |
| 951 | c.seenAGs[agName] = true |
| 952 | c.addAGCharts(agName) |
| 953 | } |
| 954 | |
| 955 | agID := cleanAGName(agName) |
| 956 | |
| 957 | // sync health: 0=not_healthy, 1=partially_healthy, 2=healthy |
| 958 | mx[fmt.Sprintf("ag_%s_sync_health_not_healthy", agID)] = boolToInt(syncHealth == 0) |
| 959 | mx[fmt.Sprintf("ag_%s_sync_health_partially_healthy", agID)] = boolToInt(syncHealth == 1) |
| 960 | mx[fmt.Sprintf("ag_%s_sync_health_healthy", agID)] = boolToInt(syncHealth == 2) |
| 961 | |
| 962 | // primary recovery health: -1=N/A (on secondary), 0=in_progress, 1=online |
| 963 | mx[fmt.Sprintf("ag_%s_primary_recovery_online", agID)] = boolToInt(primaryRecoveryHealth == 1) |
| 964 | mx[fmt.Sprintf("ag_%s_primary_recovery_in_progress", agID)] = boolToInt(primaryRecoveryHealth == 0) |
| 965 | |
| 966 | // secondary recovery health: -1=N/A (on primary), 0=in_progress, 1=online |
| 967 | mx[fmt.Sprintf("ag_%s_secondary_recovery_online", agID)] = boolToInt(secondaryRecoveryHealth == 1) |
| 968 | mx[fmt.Sprintf("ag_%s_secondary_recovery_in_progress", agID)] = boolToInt(secondaryRecoveryHealth == 0) |
| 969 | } |
| 970 | |
| 971 | return rows.Err() |
| 972 | } |
| 973 | |
| 974 | func (c *Collector) collectAGReplicaStates(mx map[string]int64) error { |
| 975 | ctx, cancel := context.WithTimeout(context.Background(), c.Timeout.Duration()) |
| 976 | defer cancel() |
| 977 | |
| 978 | rows, err := c.db.QueryContext(ctx, queryAGReplicaStates) |
| 979 | if err != nil { |
| 980 | return fmt.Errorf("AG replica states query failed: %v", err) |
| 981 | } |
| 982 | defer rows.Close() |
| 983 | |
| 984 | for rows.Next() { |
| 985 | var agName, replicaServer, availMode, failoverMode string |
| 986 | var role, connState, syncHealth int64 |
| 987 | if err := rows.Scan(&agName, &replicaServer, &availMode, &failoverMode, |
| 988 | &role, &connState, &syncHealth); err != nil { |
| 989 | c.Debugf("AG replica state scan failed: %v", err) |
| 990 | continue |
| 991 | } |
| 992 | |
| 993 | agName = strings.TrimSpace(agName) |
| 994 | replicaServer = strings.TrimSpace(replicaServer) |
| 995 | |
| 996 | replicaKey := agName + "_" + replicaServer |
| 997 | if !c.seenAGReplicas[replicaKey] { |
| 998 | c.seenAGReplicas[replicaKey] = true |
| 999 | c.addAGReplicaCharts(agName, replicaServer, availMode, failoverMode) |
| 1000 | } |
| 1001 | |
| 1002 | rID := cleanAGReplicaName(agName, replicaServer) |
| 1003 | |
| 1004 | // role: -1=unknown, 0=resolving, 1=primary, 2=secondary |
| 1005 | mx[fmt.Sprintf("ag_replica_%s_role_resolving", rID)] = boolToInt(role == 0) |
| 1006 | mx[fmt.Sprintf("ag_replica_%s_role_primary", rID)] = boolToInt(role == 1) |
| 1007 | mx[fmt.Sprintf("ag_replica_%s_role_secondary", rID)] = boolToInt(role == 2) |
| 1008 | mx[fmt.Sprintf("ag_replica_%s_role_unknown", rID)] = boolToInt(role == -1) |
| 1009 | |
| 1010 | // connected state: -1=unknown, 0=disconnected, 1=connected |
| 1011 | mx[fmt.Sprintf("ag_replica_%s_connected", rID)] = boolToInt(connState == 1) |
| 1012 | mx[fmt.Sprintf("ag_replica_%s_disconnected", rID)] = boolToInt(connState == 0) |
| 1013 | mx[fmt.Sprintf("ag_replica_%s_conn_unknown", rID)] = boolToInt(connState == -1) |
| 1014 | |
| 1015 | // sync health: 0=not_healthy, 1=partially_healthy, 2=healthy |
| 1016 | mx[fmt.Sprintf("ag_replica_%s_sync_health_not_healthy", rID)] = boolToInt(syncHealth == 0) |
| 1017 | mx[fmt.Sprintf("ag_replica_%s_sync_health_partially_healthy", rID)] = boolToInt(syncHealth == 1) |
| 1018 | mx[fmt.Sprintf("ag_replica_%s_sync_health_healthy", rID)] = boolToInt(syncHealth == 2) |
| 1019 | } |
| 1020 | |
| 1021 | return rows.Err() |
| 1022 | } |
| 1023 | |
| 1024 | func (c *Collector) collectAGDatabaseReplicas(mx map[string]int64) error { |
| 1025 | ctx, cancel := context.WithTimeout(context.Background(), c.Timeout.Duration()) |
| 1026 | defer cancel() |
| 1027 | |
| 1028 | query := agDatabaseReplicaQuery(c.majorVersion) |
| 1029 | rows, err := c.db.QueryContext(ctx, query) |
| 1030 | if err != nil { |
| 1031 | return fmt.Errorf("AG database replicas query failed: %v", err) |
| 1032 | } |
| 1033 | defer rows.Close() |
| 1034 | |
| 1035 | for rows.Next() { |
| 1036 | var agName, replicaServer, dbName string |
| 1037 | var syncState, isSuspended int64 |
| 1038 | var logSendQueue, logSendRate, redoQueue, redoRate, filestreamRate int64 |
| 1039 | var secondaryLag int64 |
| 1040 | |
| 1041 | if err := rows.Scan( |
| 1042 | &agName, &replicaServer, &dbName, |
| 1043 | &syncState, &isSuspended, |
| 1044 | &logSendQueue, &logSendRate, |
| 1045 | &redoQueue, &redoRate, &filestreamRate, |
| 1046 | &secondaryLag, |
| 1047 | ); err != nil { |
| 1048 | c.Debugf("AG database replica scan failed: %v", err) |
| 1049 | continue |
| 1050 | } |
| 1051 | |
| 1052 | agName = strings.TrimSpace(agName) |
| 1053 | replicaServer = strings.TrimSpace(replicaServer) |
| 1054 | dbName = strings.TrimSpace(dbName) |
| 1055 | |
| 1056 | if dbName == "" { |
| 1057 | continue |
| 1058 | } |
| 1059 | |
| 1060 | drKey := agName + "_" + replicaServer + "_" + dbName |
| 1061 | if !c.seenAGDatabaseReplicas[drKey] { |
| 1062 | c.seenAGDatabaseReplicas[drKey] = true |
| 1063 | c.addAGDatabaseReplicaCharts(agName, replicaServer, dbName) |
| 1064 | } |
| 1065 | |
| 1066 | drID := cleanAGDatabaseReplicaName(agName, replicaServer, dbName) |
| 1067 | |
| 1068 | // sync state: 0=not_synchronizing, 1=synchronizing, 2=synchronized, 3=reverting, 4=initializing |
| 1069 | mx[fmt.Sprintf("ag_db_%s_sync_state_not_synchronizing", drID)] = boolToInt(syncState == 0) |
| 1070 | mx[fmt.Sprintf("ag_db_%s_sync_state_synchronizing", drID)] = boolToInt(syncState == 1) |
| 1071 | mx[fmt.Sprintf("ag_db_%s_sync_state_synchronized", drID)] = boolToInt(syncState == 2) |
| 1072 | mx[fmt.Sprintf("ag_db_%s_sync_state_reverting", drID)] = boolToInt(syncState == 3) |
| 1073 | mx[fmt.Sprintf("ag_db_%s_sync_state_initializing", drID)] = boolToInt(syncState == 4) |
| 1074 | |
| 1075 | // queue sizes and rates (already converted to bytes in SQL) |
| 1076 | mx[fmt.Sprintf("ag_db_%s_log_send_queue_size", drID)] = logSendQueue |
| 1077 | mx[fmt.Sprintf("ag_db_%s_log_send_rate", drID)] = logSendRate |
| 1078 | mx[fmt.Sprintf("ag_db_%s_redo_queue_size", drID)] = redoQueue |
| 1079 | mx[fmt.Sprintf("ag_db_%s_redo_rate", drID)] = redoRate |
| 1080 | mx[fmt.Sprintf("ag_db_%s_filestream_send_rate", drID)] = filestreamRate |
| 1081 | |
| 1082 | // suspended |
| 1083 | mx[fmt.Sprintf("ag_db_%s_suspended", drID)] = isSuspended |
| 1084 | mx[fmt.Sprintf("ag_db_%s_not_suspended", drID)] = boolToInt(isSuspended == 0) |
| 1085 | |
| 1086 | // secondary lag (only meaningful on SQL 2016+, -1 = not available) |
| 1087 | if secondaryLag >= 0 { |
| 1088 | mx[fmt.Sprintf("ag_db_%s_secondary_lag_seconds", drID)] = secondaryLag |
| 1089 | } |
| 1090 | } |
| 1091 | |
| 1092 | return rows.Err() |
| 1093 | } |
| 1094 | |
| 1095 | func (c *Collector) collectAGCluster(mx map[string]int64) error { |
| 1096 | ctx, cancel := context.WithTimeout(context.Background(), c.Timeout.Duration()) |
| 1097 | defer cancel() |
| 1098 | |
| 1099 | var quorumState int64 |
| 1100 | err := c.db.QueryRowContext(ctx, queryAGCluster).Scan(&quorumState) |
| 1101 | if err != nil { |
| 1102 | return fmt.Errorf("AG cluster query failed: %v", err) |
| 1103 | } |
| 1104 | |
| 1105 | if !c.agClusterChartAdded { |
| 1106 | c.agClusterChartAdded = true |
| 1107 | if err := c.Charts().Add(agClusterQuorumStateChart.Copy()); err != nil { |
| 1108 | c.Warning(err) |
| 1109 | } |
| 1110 | } |
| 1111 | |
| 1112 | // quorum state: 0=unknown, 1=normal, 2=forced |
| 1113 | mx["ag_cluster_quorum_state_unknown"] = boolToInt(quorumState == 0) |
| 1114 | mx["ag_cluster_quorum_state_normal"] = boolToInt(quorumState == 1) |
| 1115 | mx["ag_cluster_quorum_state_forced"] = boolToInt(quorumState == 2) |
| 1116 | |
| 1117 | return nil |
| 1118 | } |
| 1119 | |
| 1120 | func (c *Collector) collectAGClusterMembers(mx map[string]int64) error { |
| 1121 | ctx, cancel := context.WithTimeout(context.Background(), c.Timeout.Duration()) |
| 1122 | defer cancel() |
| 1123 | |
| 1124 | rows, err := c.db.QueryContext(ctx, queryAGClusterMembers) |
| 1125 | if err != nil { |
| 1126 | return fmt.Errorf("AG cluster members query failed: %v", err) |
| 1127 | } |
| 1128 | defer rows.Close() |
| 1129 | |
| 1130 | for rows.Next() { |
| 1131 | var memberName string |
| 1132 | var memberState, quorumVotes int64 |
| 1133 | if err := rows.Scan(&memberName, &memberState, &quorumVotes); err != nil { |
| 1134 | c.Debugf("AG cluster member scan failed: %v", err) |
| 1135 | continue |
| 1136 | } |
| 1137 | |
| 1138 | memberName = strings.TrimSpace(memberName) |
| 1139 | |
| 1140 | if !c.seenAGClusterMembers[memberName] { |
| 1141 | c.seenAGClusterMembers[memberName] = true |
| 1142 | c.addAGClusterMemberCharts(memberName) |
| 1143 | } |
| 1144 | |
| 1145 | mID := cleanAGName(memberName) |
| 1146 | |
| 1147 | // member state: 0=offline, 1=online |
| 1148 | mx[fmt.Sprintf("ag_cluster_member_%s_up", mID)] = boolToInt(memberState == 1) |
| 1149 | mx[fmt.Sprintf("ag_cluster_member_%s_down", mID)] = boolToInt(memberState == 0) |
| 1150 | mx[fmt.Sprintf("ag_cluster_member_%s_quorum_votes", mID)] = quorumVotes |
| 1151 | } |
| 1152 | |
| 1153 | return rows.Err() |
| 1154 | } |
| 1155 | |
| 1156 | func (c *Collector) collectAGFailoverReadiness(mx map[string]int64) error { |
| 1157 | ctx, cancel := context.WithTimeout(context.Background(), c.Timeout.Duration()) |
| 1158 | defer cancel() |
| 1159 | |
| 1160 | rows, err := c.db.QueryContext(ctx, queryAGFailoverReadiness) |
| 1161 | if err != nil { |
| 1162 | return fmt.Errorf("AG failover readiness query failed: %v", err) |
| 1163 | } |
| 1164 | defer rows.Close() |
| 1165 | |
| 1166 | for rows.Next() { |
| 1167 | var agName, replicaServer, dbName string |
| 1168 | var isFailoverReady, isDatabaseJoined int64 |
| 1169 | if err := rows.Scan(&agName, &replicaServer, &dbName, &isFailoverReady, &isDatabaseJoined); err != nil { |
| 1170 | c.Debugf("AG failover readiness scan failed: %v", err) |
| 1171 | continue |
| 1172 | } |
| 1173 | |
| 1174 | agName = strings.TrimSpace(agName) |
| 1175 | replicaServer = strings.TrimSpace(replicaServer) |
| 1176 | dbName = strings.TrimSpace(dbName) |
| 1177 | |
| 1178 | drID := cleanAGDatabaseReplicaName(agName, replicaServer, dbName) |
| 1179 | |
| 1180 | mx[fmt.Sprintf("ag_db_%s_failover_ready", drID)] = isFailoverReady |
| 1181 | mx[fmt.Sprintf("ag_db_%s_failover_not_ready", drID)] = boolToInt(isFailoverReady == 0) |
| 1182 | mx[fmt.Sprintf("ag_db_%s_joined", drID)] = isDatabaseJoined |
| 1183 | mx[fmt.Sprintf("ag_db_%s_not_joined", drID)] = boolToInt(isDatabaseJoined == 0) |
| 1184 | } |
| 1185 | |
| 1186 | return rows.Err() |
| 1187 | } |
| 1188 | |
| 1189 | func (c *Collector) collectAGAutoPageRepair(mx map[string]int64) error { |
| 1190 | ctx, cancel := context.WithTimeout(context.Background(), c.Timeout.Duration()) |
| 1191 | defer cancel() |
| 1192 | |
| 1193 | rows, err := c.db.QueryContext(ctx, queryAGAutoPageRepair) |
| 1194 | if err != nil { |
| 1195 | return fmt.Errorf("AG auto page repair query failed: %v", err) |
| 1196 | } |
| 1197 | defer rows.Close() |
| 1198 | |
| 1199 | for rows.Next() { |
| 1200 | var dbName string |
| 1201 | var successfulRepairs, failedRepairs int64 |
| 1202 | if err := rows.Scan(&dbName, &successfulRepairs, &failedRepairs); err != nil { |
| 1203 | c.Debugf("AG page repair scan failed: %v", err) |
| 1204 | continue |
| 1205 | } |
| 1206 | |
| 1207 | dbName = strings.TrimSpace(dbName) |
| 1208 | |
| 1209 | if dbName == "" { |
| 1210 | continue |
| 1211 | } |
| 1212 | |
| 1213 | if !c.seenAGPageRepairDBs[dbName] { |
| 1214 | c.seenAGPageRepairDBs[dbName] = true |
| 1215 | c.addAGPageRepairCharts(dbName) |
| 1216 | } |
| 1217 | |
| 1218 | dbID := cleanDatabaseName(dbName) |
| 1219 | |
| 1220 | mx[fmt.Sprintf("ag_page_repair_%s_successful", dbID)] = successfulRepairs |
| 1221 | mx[fmt.Sprintf("ag_page_repair_%s_failed", dbID)] = failedRepairs |
| 1222 | } |
| 1223 | |
| 1224 | return rows.Err() |
| 1225 | } |
| 1226 | |
| 1227 | func (c *Collector) collectAGThreads(mx map[string]int64) error { |
| 1228 | ctx, cancel := context.WithTimeout(context.Background(), c.Timeout.Duration()) |
| 1229 | defer cancel() |
| 1230 | |
| 1231 | rows, err := c.db.QueryContext(ctx, queryAGThreads) |
| 1232 | if err != nil { |
| 1233 | return fmt.Errorf("AG threads query failed: %v", err) |
| 1234 | } |
| 1235 | defer rows.Close() |
| 1236 | |
| 1237 | for rows.Next() { |
| 1238 | var agName string |
| 1239 | var captureThreads, redoThreads, parallelRedoThreads int64 |
| 1240 | if err := rows.Scan(&agName, &captureThreads, &redoThreads, ¶llelRedoThreads); err != nil { |
| 1241 | c.Debugf("AG threads scan failed: %v", err) |
| 1242 | continue |
| 1243 | } |
| 1244 | |
| 1245 | agID := cleanAGName(strings.TrimSpace(agName)) |
| 1246 | |
| 1247 | mx[fmt.Sprintf("ag_%s_capture_threads", agID)] = captureThreads |
| 1248 | mx[fmt.Sprintf("ag_%s_redo_threads", agID)] = redoThreads |
| 1249 | mx[fmt.Sprintf("ag_%s_parallel_redo_threads", agID)] = parallelRedoThreads |
| 1250 | } |
| 1251 | |
| 1252 | return rows.Err() |
| 1253 | } |
| 1254 | |
| 1255 | func (c *Collector) collectProcessMemory(mx map[string]int64) error { |
| 1256 | ctx, cancel := context.WithTimeout(context.Background(), c.Timeout.Duration()) |
| 1257 | defer cancel() |
| 1258 | |
| 1259 | var resident, virtual, utilization, pageFaults int64 |
| 1260 | err := c.db.QueryRowContext(ctx, queryProcessMemory).Scan(&resident, &virtual, &utilization, &pageFaults) |
| 1261 | if err != nil { |
| 1262 | return fmt.Errorf("process memory query failed: %v", err) |
| 1263 | } |
| 1264 | |
| 1265 | mx["process_memory_resident"] = resident |
| 1266 | mx["process_memory_virtual"] = virtual |
| 1267 | mx["process_memory_utilization"] = utilization |
| 1268 | mx["process_page_faults"] = pageFaults |
| 1269 | |
| 1270 | return nil |
| 1271 | } |
| 1272 | |
| 1273 | func (c *Collector) collectOSMemory(mx map[string]int64) error { |
| 1274 | ctx, cancel := context.WithTimeout(context.Background(), c.Timeout.Duration()) |
| 1275 | defer cancel() |
| 1276 | |
| 1277 | var memUsed, memAvailable, pagefileUsed, pagefileAvailable int64 |
| 1278 | err := c.db.QueryRowContext(ctx, queryOSMemory).Scan(&memUsed, &memAvailable, &pagefileUsed, &pagefileAvailable) |
| 1279 | if err != nil { |
| 1280 | return fmt.Errorf("OS memory query failed: %v", err) |
| 1281 | } |
| 1282 | |
| 1283 | mx["os_memory_used"] = memUsed |
| 1284 | mx["os_memory_available"] = memAvailable |
| 1285 | mx["os_pagefile_used"] = pagefileUsed |
| 1286 | mx["os_pagefile_available"] = pagefileAvailable |
| 1287 | |
| 1288 | return nil |
| 1289 | } |
| 1290 | |
| 1291 | func (c *Collector) collectIOStall(mx map[string]int64) error { |
| 1292 | ctx, cancel := context.WithTimeout(context.Background(), c.Timeout.Duration()) |
| 1293 | defer cancel() |
| 1294 | |
| 1295 | rows, err := c.db.QueryContext(ctx, queryIOStall) |
| 1296 | if err != nil { |
| 1297 | return fmt.Errorf("IO stall query failed: %v", err) |
| 1298 | } |
| 1299 | defer rows.Close() |
| 1300 | |
| 1301 | for rows.Next() { |
| 1302 | var dbName string |
| 1303 | var readMs, writeMs, totalMs int64 |
| 1304 | if err := rows.Scan(&dbName, &readMs, &writeMs, &totalMs); err != nil { |
| 1305 | continue |
| 1306 | } |
| 1307 | |
| 1308 | dbName = strings.TrimSpace(dbName) |
| 1309 | if dbName == "" { |
| 1310 | continue |
| 1311 | } |
| 1312 | |
| 1313 | if !c.seenDatabases[dbName] { |
| 1314 | c.seenDatabases[dbName] = true |
| 1315 | c.addDatabaseCharts(dbName) |
| 1316 | } |
| 1317 | |
| 1318 | dbID := cleanDatabaseName(dbName) |
| 1319 | mx[fmt.Sprintf("database_%s_io_stall_read", dbID)] = readMs |
| 1320 | mx[fmt.Sprintf("database_%s_io_stall_write", dbID)] = writeMs |
| 1321 | } |
| 1322 | |
| 1323 | return rows.Err() |
| 1324 | } |
| 1325 | |
| 1326 | func (c *Collector) collectLogGrowths(mx map[string]int64) error { |
| 1327 | ctx, cancel := context.WithTimeout(context.Background(), c.Timeout.Duration()) |
| 1328 | defer cancel() |
| 1329 | |
| 1330 | rows, err := c.db.QueryContext(ctx, queryLogGrowths) |
| 1331 | if err != nil { |
| 1332 | return fmt.Errorf("log growths query failed: %v", err) |
| 1333 | } |
| 1334 | defer rows.Close() |
| 1335 | |
| 1336 | for rows.Next() { |
| 1337 | var dbName string |
| 1338 | var growths int64 |
| 1339 | if err := rows.Scan(&dbName, &growths); err != nil { |
| 1340 | continue |
| 1341 | } |
| 1342 | |
| 1343 | dbName = strings.TrimSpace(dbName) |
| 1344 | if dbName == "" { |
| 1345 | continue |
| 1346 | } |
| 1347 | |
| 1348 | if !c.seenDatabases[dbName] { |
| 1349 | c.seenDatabases[dbName] = true |
| 1350 | c.addDatabaseCharts(dbName) |
| 1351 | } |
| 1352 | |
| 1353 | dbID := cleanDatabaseName(dbName) |
| 1354 | mx[fmt.Sprintf("database_%s_log_growths", dbID)] = growths |
| 1355 | } |
| 1356 | |
| 1357 | return rows.Err() |
| 1358 | } |