| 1 | version: v1 |
| 2 | context_namespace: powervault |
| 3 | groups: |
| 4 | - family: System Health |
| 5 | metrics: |
| 6 | - system_health |
| 7 | charts: |
| 8 | - id: system_health |
| 9 | title: System Health Status |
| 10 | context: system_health |
| 11 | units: status |
| 12 | algorithm: absolute |
| 13 | dimensions: |
| 14 | - selector: system_health |
| 15 | name: health |
| 16 | |
| 17 | - family: Hardware Health |
| 18 | metrics: |
| 19 | - hw_controller_ok |
| 20 | - hw_controller_degraded |
| 21 | - hw_controller_fault |
| 22 | - hw_controller_unknown |
| 23 | - hw_drive_ok |
| 24 | - hw_drive_degraded |
| 25 | - hw_drive_fault |
| 26 | - hw_drive_unknown |
| 27 | - hw_fan_ok |
| 28 | - hw_fan_degraded |
| 29 | - hw_fan_fault |
| 30 | - hw_fan_unknown |
| 31 | - hw_psu_ok |
| 32 | - hw_psu_degraded |
| 33 | - hw_psu_fault |
| 34 | - hw_psu_unknown |
| 35 | - hw_fru_ok |
| 36 | - hw_fru_degraded |
| 37 | - hw_fru_fault |
| 38 | - hw_fru_unknown |
| 39 | - hw_port_ok |
| 40 | - hw_port_degraded |
| 41 | - hw_port_fault |
| 42 | - hw_port_unknown |
| 43 | charts: |
| 44 | - id: hw_health_controller |
| 45 | title: Controller Health Status |
| 46 | context: hw_health_controller |
| 47 | units: controllers |
| 48 | type: stacked |
| 49 | algorithm: absolute |
| 50 | dimensions: |
| 51 | - selector: hw_controller_ok |
| 52 | name: ok |
| 53 | - selector: hw_controller_degraded |
| 54 | name: degraded |
| 55 | - selector: hw_controller_fault |
| 56 | name: fault |
| 57 | - selector: hw_controller_unknown |
| 58 | name: unknown |
| 59 | - id: hw_health_drive |
| 60 | title: Drive Health Status |
| 61 | context: hw_health_drive |
| 62 | units: drives |
| 63 | type: stacked |
| 64 | algorithm: absolute |
| 65 | dimensions: |
| 66 | - selector: hw_drive_ok |
| 67 | name: ok |
| 68 | - selector: hw_drive_degraded |
| 69 | name: degraded |
| 70 | - selector: hw_drive_fault |
| 71 | name: fault |
| 72 | - selector: hw_drive_unknown |
| 73 | name: unknown |
| 74 | - id: hw_health_fan |
| 75 | title: Fan Health Status |
| 76 | context: hw_health_fan |
| 77 | units: fans |
| 78 | type: stacked |
| 79 | algorithm: absolute |
| 80 | dimensions: |
| 81 | - selector: hw_fan_ok |
| 82 | name: ok |
| 83 | - selector: hw_fan_degraded |
| 84 | name: degraded |
| 85 | - selector: hw_fan_fault |
| 86 | name: fault |
| 87 | - selector: hw_fan_unknown |
| 88 | name: unknown |
| 89 | - id: hw_health_psu |
| 90 | title: Power Supply Health Status |
| 91 | context: hw_health_psu |
| 92 | units: PSUs |
| 93 | type: stacked |
| 94 | algorithm: absolute |
| 95 | dimensions: |
| 96 | - selector: hw_psu_ok |
| 97 | name: ok |
| 98 | - selector: hw_psu_degraded |
| 99 | name: degraded |
| 100 | - selector: hw_psu_fault |
| 101 | name: fault |
| 102 | - selector: hw_psu_unknown |
| 103 | name: unknown |
| 104 | - id: hw_health_fru |
| 105 | title: FRU Health Status |
| 106 | context: hw_health_fru |
| 107 | units: FRUs |
| 108 | type: stacked |
| 109 | algorithm: absolute |
| 110 | dimensions: |
| 111 | - selector: hw_fru_ok |
| 112 | name: ok |
| 113 | - selector: hw_fru_degraded |
| 114 | name: degraded |
| 115 | - selector: hw_fru_fault |
| 116 | name: fault |
| 117 | - selector: hw_fru_unknown |
| 118 | name: unknown |
| 119 | - id: hw_health_port |
| 120 | title: Port Health Status |
| 121 | context: hw_health_port |
| 122 | units: ports |
| 123 | type: stacked |
| 124 | algorithm: absolute |
| 125 | dimensions: |
| 126 | - selector: hw_port_ok |
| 127 | name: ok |
| 128 | - selector: hw_port_degraded |
| 129 | name: degraded |
| 130 | - selector: hw_port_fault |
| 131 | name: fault |
| 132 | - selector: hw_port_unknown |
| 133 | name: unknown |
| 134 | |
| 135 | - family: Controller |
| 136 | metrics: |
| 137 | - controller_iops |
| 138 | - controller_throughput |
| 139 | - controller_cpu_load |
| 140 | - controller_write_cache_used |
| 141 | - controller_forwarded_cmds |
| 142 | - controller_data_read |
| 143 | - controller_data_written |
| 144 | - controller_read_ops |
| 145 | - controller_write_ops |
| 146 | - controller_write_cache_hits |
| 147 | - controller_write_cache_misses |
| 148 | - controller_read_cache_hits |
| 149 | - controller_read_cache_misses |
| 150 | charts: |
| 151 | - id: controller_iops |
| 152 | title: Controller IOPS |
| 153 | context: controller_iops |
| 154 | units: ops/s |
| 155 | algorithm: absolute |
| 156 | instances: |
| 157 | by_labels: [controller] |
| 158 | dimensions: |
| 159 | - selector: controller_iops |
| 160 | name: iops |
| 161 | - id: controller_throughput |
| 162 | title: Controller Throughput |
| 163 | context: controller_throughput |
| 164 | units: bytes/s |
| 165 | algorithm: absolute |
| 166 | instances: |
| 167 | by_labels: [controller] |
| 168 | dimensions: |
| 169 | - selector: controller_throughput |
| 170 | name: throughput |
| 171 | - id: controller_cpu_load |
| 172 | title: Controller CPU Load |
| 173 | context: controller_cpu_load |
| 174 | units: percentage |
| 175 | algorithm: absolute |
| 176 | instances: |
| 177 | by_labels: [controller] |
| 178 | dimensions: |
| 179 | - selector: controller_cpu_load |
| 180 | name: cpu_load |
| 181 | options: |
| 182 | float: true |
| 183 | - id: controller_write_cache_used |
| 184 | title: Controller Write Cache Usage |
| 185 | context: controller_write_cache_used |
| 186 | units: percentage |
| 187 | algorithm: absolute |
| 188 | instances: |
| 189 | by_labels: [controller] |
| 190 | dimensions: |
| 191 | - selector: controller_write_cache_used |
| 192 | name: write_cache_used |
| 193 | - id: controller_forwarded_cmds |
| 194 | title: Controller Forwarded Commands |
| 195 | context: controller_forwarded_cmds |
| 196 | units: commands |
| 197 | algorithm: incremental |
| 198 | instances: |
| 199 | by_labels: [controller] |
| 200 | dimensions: |
| 201 | - selector: controller_forwarded_cmds |
| 202 | name: forwarded |
| 203 | - id: controller_io |
| 204 | title: Controller I/O |
| 205 | context: controller_io |
| 206 | units: bytes |
| 207 | type: area |
| 208 | algorithm: incremental |
| 209 | instances: |
| 210 | by_labels: [controller] |
| 211 | dimensions: |
| 212 | - selector: controller_data_read |
| 213 | name: read |
| 214 | - selector: controller_data_written |
| 215 | name: written |
| 216 | options: |
| 217 | multiplier: -1 |
| 218 | - id: controller_ops |
| 219 | title: Controller Operations |
| 220 | context: controller_ops |
| 221 | units: ops |
| 222 | type: area |
| 223 | algorithm: incremental |
| 224 | instances: |
| 225 | by_labels: [controller] |
| 226 | dimensions: |
| 227 | - selector: controller_read_ops |
| 228 | name: read |
| 229 | - selector: controller_write_ops |
| 230 | name: write |
| 231 | options: |
| 232 | multiplier: -1 |
| 233 | - id: controller_cache_hits |
| 234 | title: Controller Cache Hits and Misses |
| 235 | context: controller_cache_hits |
| 236 | units: hits |
| 237 | type: stacked |
| 238 | algorithm: incremental |
| 239 | instances: |
| 240 | by_labels: [controller] |
| 241 | dimensions: |
| 242 | - selector: controller_read_cache_hits |
| 243 | name: read_hits |
| 244 | - selector: controller_read_cache_misses |
| 245 | name: read_misses |
| 246 | - selector: controller_write_cache_hits |
| 247 | name: write_hits |
| 248 | - selector: controller_write_cache_misses |
| 249 | name: write_misses |
| 250 | |
| 251 | - family: Volume |
| 252 | metrics: |
| 253 | - volume_iops |
| 254 | - volume_throughput |
| 255 | - volume_write_cache_percent |
| 256 | - volume_data_read |
| 257 | - volume_data_written |
| 258 | - volume_read_ops |
| 259 | - volume_write_ops |
| 260 | - volume_write_cache_hits |
| 261 | - volume_write_cache_misses |
| 262 | - volume_read_cache_hits |
| 263 | - volume_read_cache_misses |
| 264 | - volume_tier_ssd |
| 265 | - volume_tier_sas |
| 266 | - volume_tier_sata |
| 267 | charts: |
| 268 | - id: volume_iops |
| 269 | title: Volume IOPS |
| 270 | context: volume_iops |
| 271 | units: ops/s |
| 272 | algorithm: absolute |
| 273 | instances: |
| 274 | by_labels: [volume] |
| 275 | dimensions: |
| 276 | - selector: volume_iops |
| 277 | name: iops |
| 278 | - id: volume_throughput |
| 279 | title: Volume Throughput |
| 280 | context: volume_throughput |
| 281 | units: bytes/s |
| 282 | algorithm: absolute |
| 283 | instances: |
| 284 | by_labels: [volume] |
| 285 | dimensions: |
| 286 | - selector: volume_throughput |
| 287 | name: throughput |
| 288 | - id: volume_write_cache_percent |
| 289 | title: Volume Write Cache Usage |
| 290 | context: volume_write_cache_percent |
| 291 | units: percentage |
| 292 | algorithm: absolute |
| 293 | instances: |
| 294 | by_labels: [volume] |
| 295 | dimensions: |
| 296 | - selector: volume_write_cache_percent |
| 297 | name: write_cache |
| 298 | - id: volume_io |
| 299 | title: Volume I/O |
| 300 | context: volume_io |
| 301 | units: bytes |
| 302 | type: area |
| 303 | algorithm: incremental |
| 304 | instances: |
| 305 | by_labels: [volume] |
| 306 | dimensions: |
| 307 | - selector: volume_data_read |
| 308 | name: read |
| 309 | - selector: volume_data_written |
| 310 | name: written |
| 311 | options: |
| 312 | multiplier: -1 |
| 313 | - id: volume_ops |
| 314 | title: Volume Operations |
| 315 | context: volume_ops |
| 316 | units: ops |
| 317 | type: area |
| 318 | algorithm: incremental |
| 319 | instances: |
| 320 | by_labels: [volume] |
| 321 | dimensions: |
| 322 | - selector: volume_read_ops |
| 323 | name: read |
| 324 | - selector: volume_write_ops |
| 325 | name: write |
| 326 | options: |
| 327 | multiplier: -1 |
| 328 | - id: volume_cache_hits |
| 329 | title: Volume Cache Hits and Misses |
| 330 | context: volume_cache_hits |
| 331 | units: hits |
| 332 | type: stacked |
| 333 | algorithm: incremental |
| 334 | instances: |
| 335 | by_labels: [volume] |
| 336 | dimensions: |
| 337 | - selector: volume_read_cache_hits |
| 338 | name: read_hits |
| 339 | - selector: volume_read_cache_misses |
| 340 | name: read_misses |
| 341 | - selector: volume_write_cache_hits |
| 342 | name: write_hits |
| 343 | - selector: volume_write_cache_misses |
| 344 | name: write_misses |
| 345 | - id: volume_tier_distribution |
| 346 | title: Volume Storage Tier Distribution |
| 347 | context: volume_tier_distribution |
| 348 | units: percentage |
| 349 | type: stacked |
| 350 | algorithm: absolute |
| 351 | instances: |
| 352 | by_labels: [volume] |
| 353 | dimensions: |
| 354 | - selector: volume_tier_ssd |
| 355 | name: ssd |
| 356 | - selector: volume_tier_sas |
| 357 | name: sas |
| 358 | - selector: volume_tier_sata |
| 359 | name: sata |
| 360 | |
| 361 | - family: Port |
| 362 | metrics: |
| 363 | - port_read_ops |
| 364 | - port_write_ops |
| 365 | - port_data_read |
| 366 | - port_data_written |
| 367 | charts: |
| 368 | - id: port_io |
| 369 | title: Port I/O |
| 370 | context: port_io |
| 371 | units: bytes |
| 372 | type: area |
| 373 | algorithm: incremental |
| 374 | instances: |
| 375 | by_labels: [port] |
| 376 | dimensions: |
| 377 | - selector: port_data_read |
| 378 | name: read |
| 379 | - selector: port_data_written |
| 380 | name: written |
| 381 | options: |
| 382 | multiplier: -1 |
| 383 | - id: port_ops |
| 384 | title: Port Operations |
| 385 | context: port_ops |
| 386 | units: ops |
| 387 | type: area |
| 388 | algorithm: incremental |
| 389 | instances: |
| 390 | by_labels: [port] |
| 391 | dimensions: |
| 392 | - selector: port_read_ops |
| 393 | name: read |
| 394 | - selector: port_write_ops |
| 395 | name: write |
| 396 | options: |
| 397 | multiplier: -1 |
| 398 | |
| 399 | - family: PHY Errors |
| 400 | metrics: |
| 401 | - phy_disparity_errors |
| 402 | - phy_lost_dwords |
| 403 | - phy_invalid_dwords |
| 404 | charts: |
| 405 | - id: phy_errors |
| 406 | title: SAS PHY Errors |
| 407 | context: phy_errors |
| 408 | units: errors |
| 409 | type: stacked |
| 410 | algorithm: incremental |
| 411 | instances: |
| 412 | by_labels: [port] |
| 413 | dimensions: |
| 414 | - selector: phy_disparity_errors |
| 415 | name: disparity |
| 416 | - selector: phy_lost_dwords |
| 417 | name: lost_dwords |
| 418 | - selector: phy_invalid_dwords |
| 419 | name: invalid_dwords |
| 420 | |
| 421 | - family: Pool |
| 422 | metrics: |
| 423 | - pool_total_bytes |
| 424 | - pool_available_bytes |
| 425 | charts: |
| 426 | - id: pool_capacity |
| 427 | title: Pool Capacity |
| 428 | context: pool_capacity |
| 429 | units: bytes |
| 430 | algorithm: absolute |
| 431 | instances: |
| 432 | by_labels: [pool] |
| 433 | dimensions: |
| 434 | - selector: pool_total_bytes |
| 435 | name: total |
| 436 | - selector: pool_available_bytes |
| 437 | name: available |
| 438 | |
| 439 | - family: Drive |
| 440 | metrics: |
| 441 | - drive_temperature |
| 442 | - drive_power_on_hours |
| 443 | - drive_ssd_life_left |
| 444 | charts: |
| 445 | - id: drive_temperature |
| 446 | title: Drive Temperature |
| 447 | context: drive_temperature |
| 448 | units: Celsius |
| 449 | algorithm: absolute |
| 450 | instances: |
| 451 | by_labels: [drive] |
| 452 | dimensions: |
| 453 | - selector: drive_temperature |
| 454 | name: temperature |
| 455 | - id: drive_power_on_hours |
| 456 | title: Drive Power-On Hours |
| 457 | context: drive_power_on_hours |
| 458 | units: hours |
| 459 | algorithm: absolute |
| 460 | instances: |
| 461 | by_labels: [drive] |
| 462 | dimensions: |
| 463 | - selector: drive_power_on_hours |
| 464 | name: power_on_hours |
| 465 | - id: drive_ssd_life_left |
| 466 | title: Drive SSD Life Remaining |
| 467 | context: drive_ssd_life_left |
| 468 | units: percentage |
| 469 | algorithm: absolute |
| 470 | instances: |
| 471 | by_labels: [drive] |
| 472 | dimensions: |
| 473 | - selector: drive_ssd_life_left |
| 474 | name: life_left |
| 475 | |
| 476 | - family: Sensor |
| 477 | metrics: |
| 478 | - sensor_temperature |
| 479 | - sensor_voltage |
| 480 | - sensor_current |
| 481 | - sensor_charge_capacity |
| 482 | charts: |
| 483 | - id: sensor_temperature |
| 484 | title: Sensor Temperature |
| 485 | context: sensor_temperature |
| 486 | units: Celsius |
| 487 | algorithm: absolute |
| 488 | instances: |
| 489 | by_labels: [sensor] |
| 490 | dimensions: |
| 491 | - selector: sensor_temperature |
| 492 | name: temperature |
| 493 | - id: sensor_voltage |
| 494 | title: Sensor Voltage |
| 495 | context: sensor_voltage |
| 496 | units: millivolts |
| 497 | algorithm: absolute |
| 498 | instances: |
| 499 | by_labels: [sensor] |
| 500 | dimensions: |
| 501 | - selector: sensor_voltage |
| 502 | name: voltage |
| 503 | options: |
| 504 | float: true |
| 505 | - id: sensor_current |
| 506 | title: Sensor Current |
| 507 | context: sensor_current |
| 508 | units: milliamps |
| 509 | algorithm: absolute |
| 510 | instances: |
| 511 | by_labels: [sensor] |
| 512 | dimensions: |
| 513 | - selector: sensor_current |
| 514 | name: current |
| 515 | options: |
| 516 | float: true |
| 517 | - id: sensor_charge_capacity |
| 518 | title: Sensor Charge Capacity |
| 519 | context: sensor_charge_capacity |
| 520 | units: percentage |
| 521 | algorithm: absolute |
| 522 | instances: |
| 523 | by_labels: [sensor] |
| 524 | dimensions: |
| 525 | - selector: sensor_charge_capacity |
| 526 | name: charge_capacity |